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
 133 void __bpf_prog_free(struct bpf_prog *fp)
 134 {
 135         kfree(fp->aux);
 136         vfree(fp);
 137 }
 138
 139 static bool bpf_is_jmp_and_has_target(const struct bpf_insn *insn)
 140 {
 141         return BPF_CLASS(insn->code) == BPF_JMP  &&
 142                /* Call and Exit are both special jumps with no
 143                 * target inside the BPF instruction image.
 144                 */
 145                BPF_OP(insn->code) != BPF_CALL &&
 146                BPF_OP(insn->code) != BPF_EXIT;
 147 }
 148
 149 static void bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta)
 150 {
 151         struct bpf_insn *insn = prog->insnsi;
 152         u32 i, insn_cnt = prog->len;
 153
 154         for (i = 0; i < insn_cnt; i++, insn++) {
 155                 if (!bpf_is_jmp_and_has_target(insn))
 156                         continue;
 157
 158                 /* Adjust offset of jmps if we cross boundaries. */
 159                 if (i < pos && i + insn->off + 1 > pos)
 160                         insn->off += delta;
 161                 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
 162                         insn->off -= delta;
 163         }
 164 }
 165
 166 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 167                                        const struct bpf_insn *patch, u32 len)
 168 {
 169         u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
 170         struct bpf_prog *prog_adj;
 171
 172         /* Since our patchlet doesn't expand the image, we're done. */
 173         if (insn_delta == 0) {
 174                 memcpy(prog->insnsi + off, patch, sizeof(*patch));
 175                 return prog;
 176         }
 177
 178         insn_adj_cnt = prog->len + insn_delta;
 179
 180         /* Several new instructions need to be inserted. Make room
 181          * for them. Likely, there's no need for a new allocation as
 182          * last page could have large enough tailroom.
 183          */
 184         prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
 185                                     GFP_USER);
 186         if (!prog_adj)
 187                 return NULL;
 188
 189         prog_adj->len = insn_adj_cnt;
 190
 191         /* Patching happens in 3 steps:
 192          *
 193          * 1) Move over tail of insnsi from next instruction onwards,
 194          *    so we can patch the single target insn with one or more
 195          *    new ones (patching is always from 1 to n insns, n > 0).
 196          * 2) Inject new instructions at the target location.
 197          * 3) Adjust branch offsets if necessary.
 198          */
 199         insn_rest = insn_adj_cnt - off - len;
 200
 201         memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
 202                 sizeof(*patch) * insn_rest);
 203         memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
 204
 205         bpf_adj_branches(prog_adj, off, insn_delta);
 206
 207         return prog_adj;
 208 }
 209
 210 #ifdef CONFIG_BPF_JIT
 211 /* All BPF JIT sysctl knobs here. */
 212 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
 213 int bpf_jit_harden   __read_mostly;
 214 long bpf_jit_limit   __read_mostly;
 215 long bpf_jit_limit_max __read_mostly;
 216
 217 static atomic_long_t bpf_jit_current;
 218
 219 /* Can be overridden by an arch's JIT compiler if it has a custom,
 220  * dedicated BPF backend memory area, or if neither of the two
 221  * below apply.
 222  */
 223 u64 __weak bpf_jit_alloc_exec_limit(void)
 224 {
 225 #if defined(MODULES_VADDR)
 226         return MODULES_END - MODULES_VADDR;
 227 #else
 228         return VMALLOC_END - VMALLOC_START;
 229 #endif
 230 }
 231
 232 static int __init bpf_jit_charge_init(void)
 233 {
 234         /* Only used as heuristic here to derive limit. */
 235         bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
 236         bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
 237                                             PAGE_SIZE), LONG_MAX);
 238         return 0;
 239 }
 240 pure_initcall(bpf_jit_charge_init);
 241
 242 static int bpf_jit_charge_modmem(u32 pages)
 243 {
 244         if (atomic_long_add_return(pages, &bpf_jit_current) >
 245             (bpf_jit_limit >> PAGE_SHIFT)) {
 246                 if (!capable(CAP_SYS_ADMIN)) {
 247                         atomic_long_sub(pages, &bpf_jit_current);
 248                         return -EPERM;
 249                 }
 250         }
 251
 252         return 0;
 253 }
 254
 255 static void bpf_jit_uncharge_modmem(u32 pages)
 256 {
 257         atomic_long_sub(pages, &bpf_jit_current);
 258 }
 259
 260 struct bpf_binary_header *
 261 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 262                      unsigned int alignment,
 263                      bpf_jit_fill_hole_t bpf_fill_ill_insns)
 264 {
 265         struct bpf_binary_header *hdr;
 266         u32 size, hole, start, pages;
 267
 268         /* Most of BPF filters are really small, but if some of them
 269          * fill a page, allow at least 128 extra bytes to insert a
 270          * random section of illegal instructions.
 271          */
 272         size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
 273         pages = size / PAGE_SIZE;
 274
 275         if (bpf_jit_charge_modmem(pages))
 276                 return NULL;
 277         hdr = module_alloc(size);
 278         if (!hdr) {
 279                 bpf_jit_uncharge_modmem(pages);
 280                 return NULL;
 281         }
 282
 283         /* Fill space with illegal/arch-dep instructions. */
 284         bpf_fill_ill_insns(hdr, size);
 285
 286         hdr->pages = pages;
 287         hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
 288                      PAGE_SIZE - sizeof(*hdr));
 289         start = (get_random_int() % hole) & ~(alignment - 1);
 290
 291         /* Leave a random number of instructions before BPF code. */
 292         *image_ptr = &hdr->image[start];
 293
 294         return hdr;
 295 }
 296
 297 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 298 {
 299         u32 pages = hdr->pages;
 300
 301         module_memfree(hdr);
 302         bpf_jit_uncharge_modmem(pages);
 303 }
 304
 305 static int bpf_jit_blind_insn(const struct bpf_insn *from,
 306                               const struct bpf_insn *aux,
 307                               struct bpf_insn *to_buff)
 308 {
 309         struct bpf_insn *to = to_buff;
 310         u32 imm_rnd = get_random_int();
 311         s16 off;
 312
 313         BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
 314         BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
 315
 316         if (from->imm == 0 &&
 317             (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
 318              from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
 319                 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
 320                 goto out;
 321         }
 322
 323         switch (from->code) {
 324         case BPF_ALU | BPF_ADD | BPF_K:
 325         case BPF_ALU | BPF_SUB | BPF_K:
 326         case BPF_ALU | BPF_AND | BPF_K:
 327         case BPF_ALU | BPF_OR  | BPF_K:
 328         case BPF_ALU | BPF_XOR | BPF_K:
 329         case BPF_ALU | BPF_MUL | BPF_K:
 330         case BPF_ALU | BPF_MOV | BPF_K:
 331         case BPF_ALU | BPF_DIV | BPF_K:
 332         case BPF_ALU | BPF_MOD | BPF_K:
 333                 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 334                 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 335                 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
 336                 break;
 337
 338         case BPF_ALU64 | BPF_ADD | BPF_K:
 339         case BPF_ALU64 | BPF_SUB | BPF_K:
 340         case BPF_ALU64 | BPF_AND | BPF_K:
 341         case BPF_ALU64 | BPF_OR  | BPF_K:
 342         case BPF_ALU64 | BPF_XOR | BPF_K:
 343         case BPF_ALU64 | BPF_MUL | BPF_K:
 344         case BPF_ALU64 | BPF_MOV | BPF_K:
 345         case BPF_ALU64 | BPF_DIV | BPF_K:
 346         case BPF_ALU64 | BPF_MOD | BPF_K:
 347                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 348                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 349                 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
 350                 break;
 351
 352         case BPF_JMP | BPF_JEQ  | BPF_K:
 353         case BPF_JMP | BPF_JNE  | BPF_K:
 354         case BPF_JMP | BPF_JGT  | BPF_K:
 355         case BPF_JMP | BPF_JGE  | BPF_K:
 356         case BPF_JMP | BPF_JSGT | BPF_K:
 357         case BPF_JMP | BPF_JSGE | BPF_K:
 358         case BPF_JMP | BPF_JSET | BPF_K:
 359                 /* Accommodate for extra offset in case of a backjump. */
 360                 off = from->off;
 361                 if (off < 0)
 362                         off -= 2;
 363                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 364                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 365                 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
 366                 break;
 367
 368         case BPF_LD | BPF_ABS | BPF_W:
 369         case BPF_LD | BPF_ABS | BPF_H:
 370         case BPF_LD | BPF_ABS | BPF_B:
 371                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 372                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 373                 *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0);
 374                 break;
 375
 376         case BPF_LD | BPF_IND | BPF_W:
 377         case BPF_LD | BPF_IND | BPF_H:
 378         case BPF_LD | BPF_IND | BPF_B:
 379                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 380                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 381                 *to++ = BPF_ALU32_REG(BPF_ADD, BPF_REG_AX, from->src_reg);
 382                 *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0);
 383                 break;
 384
 385         case BPF_LD | BPF_IMM | BPF_DW:
 386                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
 387                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 388                 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
 389                 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
 390                 break;
 391         case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
 392                 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
 393                 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 394                 *to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
 395                 break;
 396
 397         case BPF_ST | BPF_MEM | BPF_DW:
 398         case BPF_ST | BPF_MEM | BPF_W:
 399         case BPF_ST | BPF_MEM | BPF_H:
 400         case BPF_ST | BPF_MEM | BPF_B:
 401                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 402                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 403                 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
 404                 break;
 405         }
 406 out:
 407         return to - to_buff;
 408 }
 409
 410 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
 411                                               gfp_t gfp_extra_flags)
 412 {
 413         gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO |
 414                           gfp_extra_flags;
 415         struct bpf_prog *fp;
 416
 417         fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
 418         if (fp != NULL) {
 419                 kmemcheck_annotate_bitfield(fp, meta);
 420
 421                 /* aux->prog still points to the fp_other one, so
 422                  * when promoting the clone to the real program,
 423                  * this still needs to be adapted.
 424                  */
 425                 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
 426         }
 427
 428         return fp;
 429 }
 430
 431 static void bpf_prog_clone_free(struct bpf_prog *fp)
 432 {
 433         /* aux was stolen by the other clone, so we cannot free
 434          * it from this path! It will be freed eventually by the
 435          * other program on release.
 436          *
 437          * At this point, we don't need a deferred release since
 438          * clone is guaranteed to not be locked.
 439          */
 440         fp->aux = NULL;
 441         __bpf_prog_free(fp);
 442 }
 443
 444 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
 445 {
 446         /* We have to repoint aux->prog to self, as we don't
 447          * know whether fp here is the clone or the original.
 448          */
 449         fp->aux->prog = fp;
 450         bpf_prog_clone_free(fp_other);
 451 }
 452
 453 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
 454 {
 455         struct bpf_insn insn_buff[16], aux[2];
 456         struct bpf_prog *clone, *tmp;
 457         int insn_delta, insn_cnt;
 458         struct bpf_insn *insn;
 459         int i, rewritten;
 460
 461         if (!bpf_jit_blinding_enabled())
 462                 return prog;
 463
 464         clone = bpf_prog_clone_create(prog, GFP_USER);
 465         if (!clone)
 466                 return ERR_PTR(-ENOMEM);
 467
 468         insn_cnt = clone->len;
 469         insn = clone->insnsi;
 470
 471         for (i = 0; i < insn_cnt; i++, insn++) {
 472                 /* We temporarily need to hold the original ld64 insn
 473                  * so that we can still access the first part in the
 474                  * second blinding run.
 475                  */
 476                 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
 477                     insn[1].code == 0)
 478                         memcpy(aux, insn, sizeof(aux));
 479
 480                 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
 481                 if (!rewritten)
 482                         continue;
 483
 484                 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
 485                 if (!tmp) {
 486                         /* Patching may have repointed aux->prog during
 487                          * realloc from the original one, so we need to
 488                          * fix it up here on error.
 489                          */
 490                         bpf_jit_prog_release_other(prog, clone);
 491                         return ERR_PTR(-ENOMEM);
 492                 }
 493
 494                 clone = tmp;
 495                 insn_delta = rewritten - 1;
 496
 497                 /* Walk new program and skip insns we just inserted. */
 498                 insn = clone->insnsi + i + insn_delta;
 499                 insn_cnt += insn_delta;
 500                 i        += insn_delta;
 501         }
 502
 503         return clone;
 504 }
 505 #endif /* CONFIG_BPF_JIT */
 506
 507 /* Base function for offset calculation. Needs to go into .text section,
 508  * therefore keeping it non-static as well; will also be used by JITs
 509  * anyway later on, so do not let the compiler omit it.
 510  */
 511 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 512 {
 513         return 0;
 514 }
 515 EXPORT_SYMBOL_GPL(__bpf_call_base);
 516
 517 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
 518 /**
 519  *      __bpf_prog_run - run eBPF program on a given context
 520  *      @ctx: is the data we are operating on
 521  *      @insn: is the array of eBPF instructions
 522  *
 523  * Decode and execute eBPF instructions.
 524  */
 525 static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 526 {
 527         u64 stack[MAX_BPF_STACK / sizeof(u64)];
 528         u64 regs[MAX_BPF_REG], tmp;
 529         static const void *jumptable[256] = {
 530                 [0 ... 255] = &&default_label,
 531                 /* Now overwrite non-defaults ... */
 532                 /* 32 bit ALU operations */
 533                 [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X,
 534                 [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K,
 535                 [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X,
 536                 [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K,
 537                 [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X,
 538                 [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K,
 539                 [BPF_ALU | BPF_OR | BPF_X]  = &&ALU_OR_X,
 540                 [BPF_ALU | BPF_OR | BPF_K]  = &&ALU_OR_K,
 541                 [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X,
 542                 [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K,
 543                 [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X,
 544                 [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K,
 545                 [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X,
 546                 [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K,
 547                 [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X,
 548                 [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K,
 549                 [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X,
 550                 [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K,
 551                 [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X,
 552                 [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K,
 553                 [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X,
 554                 [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K,
 555                 [BPF_ALU | BPF_NEG] = &&ALU_NEG,
 556                 [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE,
 557                 [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE,
 558                 /* 64 bit ALU operations */
 559                 [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X,
 560                 [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K,
 561                 [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X,
 562                 [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K,
 563                 [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X,
 564                 [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K,
 565                 [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X,
 566                 [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K,
 567                 [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X,
 568                 [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K,
 569                 [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X,
 570                 [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K,
 571                 [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X,
 572                 [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K,
 573                 [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X,
 574                 [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K,
 575                 [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X,
 576                 [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K,
 577                 [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X,
 578                 [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K,
 579                 [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X,
 580                 [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K,
 581                 [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X,
 582                 [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K,
 583                 [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG,
 584                 /* Call instruction */
 585                 [BPF_JMP | BPF_CALL] = &&JMP_CALL,
 586                 [BPF_JMP | BPF_CALL | BPF_X] = &&JMP_TAIL_CALL,
 587                 /* Jumps */
 588                 [BPF_JMP | BPF_JA] = &&JMP_JA,
 589                 [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X,
 590                 [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K,
 591                 [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X,
 592                 [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K,
 593                 [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X,
 594                 [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K,
 595                 [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X,
 596                 [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K,
 597                 [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X,
 598                 [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K,
 599                 [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X,
 600                 [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K,
 601                 [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X,
 602                 [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K,
 603                 /* Program return */
 604                 [BPF_JMP | BPF_EXIT] = &&JMP_EXIT,
 605                 /* Store instructions */
 606                 [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B,
 607                 [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H,
 608                 [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W,
 609                 [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW,
 610                 [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W,
 611                 [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW,
 612                 [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B,
 613                 [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H,
 614                 [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W,
 615                 [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW,
 616                 /* Load instructions */
 617                 [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B,
 618                 [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H,
 619                 [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W,
 620                 [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW,
 621                 [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W,
 622                 [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H,
 623                 [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B,
 624                 [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W,
 625                 [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H,
 626                 [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B,
 627                 [BPF_LD | BPF_IMM | BPF_DW] = &&LD_IMM_DW,
 628         };
 629         u32 tail_call_cnt = 0;
 630         void *ptr;
 631         int off;
 632
 633 #define CONT     ({ insn++; goto select_insn; })
 634 #define CONT_JMP ({ insn++; goto select_insn; })
 635
 636         FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
 637         ARG1 = (u64) (unsigned long) ctx;
 638
 639 select_insn:
 640         goto *jumptable[insn->code];
 641
 642         /* ALU */
 643 #define ALU(OPCODE, OP)                 \
 644         ALU64_##OPCODE##_X:             \
 645                 DST = DST OP SRC;       \
 646                 CONT;                   \
 647         ALU_##OPCODE##_X:               \
 648                 DST = (u32) DST OP (u32) SRC;   \
 649                 CONT;                   \
 650         ALU64_##OPCODE##_K:             \
 651                 DST = DST OP IMM;               \
 652                 CONT;                   \
 653         ALU_##OPCODE##_K:               \
 654                 DST = (u32) DST OP (u32) IMM;   \
 655                 CONT;
 656
 657         ALU(ADD,  +)
 658         ALU(SUB,  -)
 659         ALU(AND,  &)
 660         ALU(OR,   |)
 661         ALU(LSH, <<)
 662         ALU(RSH, >>)
 663         ALU(XOR,  ^)
 664         ALU(MUL,  *)
 665 #undef ALU
 666         ALU_NEG:
 667                 DST = (u32) -DST;
 668                 CONT;
 669         ALU64_NEG:
 670                 DST = -DST;
 671                 CONT;
 672         ALU_MOV_X:
 673                 DST = (u32) SRC;
 674                 CONT;
 675         ALU_MOV_K:
 676                 DST = (u32) IMM;
 677                 CONT;
 678         ALU64_MOV_X:
 679                 DST = SRC;
 680                 CONT;
 681         ALU64_MOV_K:
 682                 DST = IMM;
 683                 CONT;
 684         LD_IMM_DW:
 685                 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
 686                 insn++;
 687                 CONT;
 688         ALU64_ARSH_X:
 689                 (*(s64 *) &DST) >>= SRC;
 690                 CONT;
 691         ALU64_ARSH_K:
 692                 (*(s64 *) &DST) >>= IMM;
 693                 CONT;
 694         ALU64_MOD_X:
 695                 if (unlikely(SRC == 0))
 696                         return 0;
 697                 div64_u64_rem(DST, SRC, &tmp);
 698                 DST = tmp;
 699                 CONT;
 700         ALU_MOD_X:
 701                 if (unlikely((u32)SRC == 0))
 702                         return 0;
 703                 tmp = (u32) DST;
 704                 DST = do_div(tmp, (u32) SRC);
 705                 CONT;
 706         ALU64_MOD_K:
 707                 div64_u64_rem(DST, IMM, &tmp);
 708                 DST = tmp;
 709                 CONT;
 710         ALU_MOD_K:
 711                 tmp = (u32) DST;
 712                 DST = do_div(tmp, (u32) IMM);
 713                 CONT;
 714         ALU64_DIV_X:
 715                 if (unlikely(SRC == 0))
 716                         return 0;
 717                 DST = div64_u64(DST, SRC);
 718                 CONT;
 719         ALU_DIV_X:
 720                 if (unlikely((u32)SRC == 0))
 721                         return 0;
 722                 tmp = (u32) DST;
 723                 do_div(tmp, (u32) SRC);
 724                 DST = (u32) tmp;
 725                 CONT;
 726         ALU64_DIV_K:
 727                 DST = div64_u64(DST, IMM);
 728                 CONT;
 729         ALU_DIV_K:
 730                 tmp = (u32) DST;
 731                 do_div(tmp, (u32) IMM);
 732                 DST = (u32) tmp;
 733                 CONT;
 734         ALU_END_TO_BE:
 735                 switch (IMM) {
 736                 case 16:
 737                         DST = (__force u16) cpu_to_be16(DST);
 738                         break;
 739                 case 32:
 740                         DST = (__force u32) cpu_to_be32(DST);
 741                         break;
 742                 case 64:
 743                         DST = (__force u64) cpu_to_be64(DST);
 744                         break;
 745                 }
 746                 CONT;
 747         ALU_END_TO_LE:
 748                 switch (IMM) {
 749                 case 16:
 750                         DST = (__force u16) cpu_to_le16(DST);
 751                         break;
 752                 case 32:
 753                         DST = (__force u32) cpu_to_le32(DST);
 754                         break;
 755                 case 64:
 756                         DST = (__force u64) cpu_to_le64(DST);
 757                         break;
 758                 }
 759                 CONT;
 760
 761         /* CALL */
 762         JMP_CALL:
 763                 /* Function call scratches BPF_R1-BPF_R5 registers,
 764                  * preserves BPF_R6-BPF_R9, and stores return value
 765                  * into BPF_R0.
 766                  */
 767                 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
 768                                                        BPF_R4, BPF_R5);
 769                 CONT;
 770
 771         JMP_TAIL_CALL: {
 772                 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
 773                 struct bpf_array *array = container_of(map, struct bpf_array, map);
 774                 struct bpf_prog *prog;
 775                 u32 index = BPF_R3;
 776
 777                 if (unlikely(index >= array->map.max_entries))
 778                         goto out;
 779                 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
 780                         goto out;
 781
 782                 tail_call_cnt++;
 783
 784                 prog = READ_ONCE(array->ptrs[index]);
 785                 if (!prog)
 786                         goto out;
 787
 788                 /* ARG1 at this point is guaranteed to point to CTX from
 789                  * the verifier side due to the fact that the tail call is
 790                  * handeled like a helper, that is, bpf_tail_call_proto,
 791                  * where arg1_type is ARG_PTR_TO_CTX.
 792                  */
 793                 insn = prog->insnsi;
 794                 goto select_insn;
 795 out:
 796                 CONT;
 797         }
 798         /* JMP */
 799         JMP_JA:
 800                 insn += insn->off;
 801                 CONT;
 802         JMP_JEQ_X:
 803                 if (DST == SRC) {
 804                         insn += insn->off;
 805                         CONT_JMP;
 806                 }
 807                 CONT;
 808         JMP_JEQ_K:
 809                 if (DST == IMM) {
 810                         insn += insn->off;
 811                         CONT_JMP;
 812                 }
 813                 CONT;
 814         JMP_JNE_X:
 815                 if (DST != SRC) {
 816                         insn += insn->off;
 817                         CONT_JMP;
 818                 }
 819                 CONT;
 820         JMP_JNE_K:
 821                 if (DST != IMM) {
 822                         insn += insn->off;
 823                         CONT_JMP;
 824                 }
 825                 CONT;
 826         JMP_JGT_X:
 827                 if (DST > SRC) {
 828                         insn += insn->off;
 829                         CONT_JMP;
 830                 }
 831                 CONT;
 832         JMP_JGT_K:
 833                 if (DST > IMM) {
 834                         insn += insn->off;
 835                         CONT_JMP;
 836                 }
 837                 CONT;
 838         JMP_JGE_X:
 839                 if (DST >= SRC) {
 840                         insn += insn->off;
 841                         CONT_JMP;
 842                 }
 843                 CONT;
 844         JMP_JGE_K:
 845                 if (DST >= IMM) {
 846                         insn += insn->off;
 847                         CONT_JMP;
 848                 }
 849                 CONT;
 850         JMP_JSGT_X:
 851                 if (((s64) DST) > ((s64) SRC)) {
 852                         insn += insn->off;
 853                         CONT_JMP;
 854                 }
 855                 CONT;
 856         JMP_JSGT_K:
 857                 if (((s64) DST) > ((s64) IMM)) {
 858                         insn += insn->off;
 859                         CONT_JMP;
 860                 }
 861                 CONT;
 862         JMP_JSGE_X:
 863                 if (((s64) DST) >= ((s64) SRC)) {
 864                         insn += insn->off;
 865                         CONT_JMP;
 866                 }
 867                 CONT;
 868         JMP_JSGE_K:
 869                 if (((s64) DST) >= ((s64) IMM)) {
 870                         insn += insn->off;
 871                         CONT_JMP;
 872                 }
 873                 CONT;
 874         JMP_JSET_X:
 875                 if (DST & SRC) {
 876                         insn += insn->off;
 877                         CONT_JMP;
 878                 }
 879                 CONT;
 880         JMP_JSET_K:
 881                 if (DST & IMM) {
 882                         insn += insn->off;
 883                         CONT_JMP;
 884                 }
 885                 CONT;
 886         JMP_EXIT:
 887                 return BPF_R0;
 888
 889         /* STX and ST and LDX*/
 890 #define LDST(SIZEOP, SIZE)                                              \
 891         STX_MEM_##SIZEOP:                                               \
 892                 *(SIZE *)(unsigned long) (DST + insn->off) = SRC;       \
 893                 CONT;                                                   \
 894         ST_MEM_##SIZEOP:                                                \
 895                 *(SIZE *)(unsigned long) (DST + insn->off) = IMM;       \
 896                 CONT;                                                   \
 897         LDX_MEM_##SIZEOP:                                               \
 898                 DST = *(SIZE *)(unsigned long) (SRC + insn->off);       \
 899                 CONT;
 900
 901         LDST(B,   u8)
 902         LDST(H,  u16)
 903         LDST(W,  u32)
 904         LDST(DW, u64)
 905 #undef LDST
 906         STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
 907                 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
 908                            (DST + insn->off));
 909                 CONT;
 910         STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
 911                 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
 912                              (DST + insn->off));
 913                 CONT;
 914         LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */
 915                 off = IMM;
 916 load_word:
 917                 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are
 918                  * only appearing in the programs where ctx ==
 919                  * skb. All programs keep 'ctx' in regs[BPF_REG_CTX]
 920                  * == BPF_R6, bpf_convert_filter() saves it in BPF_R6,
 921                  * internal BPF verifier will check that BPF_R6 ==
 922                  * ctx.
 923                  *
 924                  * BPF_ABS and BPF_IND are wrappers of function calls,
 925                  * so they scratch BPF_R1-BPF_R5 registers, preserve
 926                  * BPF_R6-BPF_R9, and store return value into BPF_R0.
 927                  *
 928                  * Implicit input:
 929                  *   ctx == skb == BPF_R6 == CTX
 930                  *
 931                  * Explicit input:
 932                  *   SRC == any register
 933                  *   IMM == 32-bit immediate
 934                  *
 935                  * Output:
 936                  *   BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
 937                  */
 938
 939                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 4, &tmp);
 940                 if (likely(ptr != NULL)) {
 941                         BPF_R0 = get_unaligned_be32(ptr);
 942                         CONT;
 943                 }
 944
 945                 return 0;
 946         LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */
 947                 off = IMM;
 948 load_half:
 949                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 2, &tmp);
 950                 if (likely(ptr != NULL)) {
 951                         BPF_R0 = get_unaligned_be16(ptr);
 952                         CONT;
 953                 }
 954
 955                 return 0;
 956         LD_ABS_B: /* BPF_R0 = *(u8 *) (skb->data + imm32) */
 957                 off = IMM;
 958 load_byte:
 959                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 1, &tmp);
 960                 if (likely(ptr != NULL)) {
 961                         BPF_R0 = *(u8 *)ptr;
 962                         CONT;
 963                 }
 964
 965                 return 0;
 966         LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */
 967                 off = IMM + SRC;
 968                 goto load_word;
 969         LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */
 970                 off = IMM + SRC;
 971                 goto load_half;
 972         LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */
 973                 off = IMM + SRC;
 974                 goto load_byte;
 975
 976         default_label:
 977                 /* If we ever reach this, we have a bug somewhere. */
 978                 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
 979                 return 0;
 980 }
 981 STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */
 982
 983 #else
 984 static unsigned int __bpf_prog_ret0_warn(void *ctx,
 985                                          const struct bpf_insn *insn)
 986 {
 987         /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
 988          * is not working properly, so warn about it!
 989          */
 990         WARN_ON_ONCE(1);
 991         return 0;
 992 }
 993 #endif
 994
 995 bool bpf_prog_array_compatible(struct bpf_array *array,
 996                                const struct bpf_prog *fp)
 997 {
 998         if (!array->owner_prog_type) {
 999                 /* There's no owner yet where we could check for
1000                  * compatibility.
1001                  */
1002                 array->owner_prog_type = fp->type;
1003                 array->owner_jited = fp->jited;
1004
1005                 return true;
1006         }
1007
1008         return array->owner_prog_type == fp->type &&
1009                array->owner_jited == fp->jited;
1010 }
1011
1012 static int bpf_check_tail_call(const struct bpf_prog *fp)
1013 {
1014         struct bpf_prog_aux *aux = fp->aux;
1015         int i;
1016
1017         for (i = 0; i < aux->used_map_cnt; i++) {
1018                 struct bpf_map *map = aux->used_maps[i];
1019                 struct bpf_array *array;
1020
1021                 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1022                         continue;
1023
1024                 array = container_of(map, struct bpf_array, map);
1025                 if (!bpf_prog_array_compatible(array, fp))
1026                         return -EINVAL;
1027         }
1028
1029         return 0;
1030 }
1031
1032 /**
1033  *      bpf_prog_select_runtime - select exec runtime for BPF program
1034  *      @fp: bpf_prog populated with internal BPF program
1035  *      @err: pointer to error variable
1036  *
1037  * Try to JIT eBPF program, if JIT is not available, use interpreter.
1038  * The BPF program will be executed via BPF_PROG_RUN() macro.
1039  */
1040 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1041 {
1042 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1043         fp->bpf_func = (void *) __bpf_prog_run;
1044 #else
1045         fp->bpf_func = (void *) __bpf_prog_ret0_warn;
1046 #endif
1047
1048         /* eBPF JITs can rewrite the program in case constant
1049          * blinding is active. However, in case of error during
1050          * blinding, bpf_int_jit_compile() must always return a
1051          * valid program, which in this case would simply not
1052          * be JITed, but falls back to the interpreter.
1053          */
1054         fp = bpf_int_jit_compile(fp);
1055 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1056         if (!fp->jited) {
1057                 *err = -ENOTSUPP;
1058                 return fp;
1059         }
1060 #endif
1061         bpf_prog_lock_ro(fp);
1062
1063         /* The tail call compatibility check can only be done at
1064          * this late stage as we need to determine, if we deal
1065          * with JITed or non JITed program concatenations and not
1066          * all eBPF JITs might immediately support all features.
1067          */
1068         *err = bpf_check_tail_call(fp);
1069
1070         return fp;
1071 }
1072 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1073
1074 static void bpf_prog_free_deferred(struct work_struct *work)
1075 {
1076         struct bpf_prog_aux *aux;
1077
1078         aux = container_of(work, struct bpf_prog_aux, work);
1079         bpf_jit_free(aux->prog);
1080 }
1081
1082 /* Free internal BPF program */
1083 void bpf_prog_free(struct bpf_prog *fp)
1084 {
1085         struct bpf_prog_aux *aux = fp->aux;
1086
1087         INIT_WORK(&aux->work, bpf_prog_free_deferred);
1088         schedule_work(&aux->work);
1089 }
1090 EXPORT_SYMBOL_GPL(bpf_prog_free);
1091
1092 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
1093 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
1094
1095 void bpf_user_rnd_init_once(void)
1096 {
1097         prandom_init_once(&bpf_user_rnd_state);
1098 }
1099
1100 BPF_CALL_0(bpf_user_rnd_u32)
1101 {
1102         /* Should someone ever have the rather unwise idea to use some
1103          * of the registers passed into this function, then note that
1104          * this function is called from native eBPF and classic-to-eBPF
1105          * transformations. Register assignments from both sides are
1106          * different, f.e. classic always sets fn(ctx, A, X) here.
1107          */
1108         struct rnd_state *state;
1109         u32 res;
1110
1111         state = &get_cpu_var(bpf_user_rnd_state);
1112         res = prandom_u32_state(state);
1113         put_cpu_var(bpf_user_rnd_state);
1114
1115         return res;
1116 }
1117
1118 /* Weak definitions of helper functions in case we don't have bpf syscall. */
1119 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
1120 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
1121 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
1122
1123 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
1124 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
1125 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
1126
1127 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
1128 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
1129 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
1130
1131 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
1132 {
1133         return NULL;
1134 }
1135
1136 u64 __weak
1137 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
1138                  void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
1139 {
1140         return -ENOTSUPP;
1141 }
1142
1143 /* Always built-in helper functions. */
1144 const struct bpf_func_proto bpf_tail_call_proto = {
1145         .func           = NULL,
1146         .gpl_only       = false,
1147         .ret_type       = RET_VOID,
1148         .arg1_type      = ARG_PTR_TO_CTX,
1149         .arg2_type      = ARG_CONST_MAP_PTR,
1150         .arg3_type      = ARG_ANYTHING,
1151 };
1152
1153 /* For classic BPF JITs that don't implement bpf_int_jit_compile(). */
1154 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
1155 {
1156         return prog;
1157 }
1158
1159 bool __weak bpf_helper_changes_skb_data(void *func)
1160 {
1161         return false;
1162 }
1163
1164 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
1165  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
1166  */
1167 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
1168                          int len)
1169 {
1170         return -EFAULT;
1171 }