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                 if (unlikely(!skb_mac_header_was_set(skb)))
  67                         return NULL;
  68                 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
  69         }
  70         if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
  71                 return ptr;
  72
  73         return NULL;
  74 }
  75
  76 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
  77 {
  78         gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO |
  79                           gfp_extra_flags;
  80         struct bpf_prog_aux *aux;
  81         struct bpf_prog *fp;
  82
  83         size = round_up(size, PAGE_SIZE);
  84         fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
  85         if (fp == NULL)
  86                 return NULL;
  87
  88         kmemcheck_annotate_bitfield(fp, meta);
  89
  90         aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
  91         if (aux == NULL) {
  92                 vfree(fp);
  93                 return NULL;
  94         }
  95
  96         fp->pages = size / PAGE_SIZE;
  97         fp->aux = aux;
  98         fp->aux->prog = fp;
  99
 100         return fp;
 101 }
 102 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
 103
 104 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
 105                                   gfp_t gfp_extra_flags)
 106 {
 107         gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO |
 108                           gfp_extra_flags;
 109         struct bpf_prog *fp;
 110         u32 pages, delta;
 111         int ret;
 112
 113         BUG_ON(fp_old == NULL);
 114
 115         size = round_up(size, PAGE_SIZE);
 116         pages = size / PAGE_SIZE;
 117         if (pages <= fp_old->pages)
 118                 return fp_old;
 119
 120         delta = pages - fp_old->pages;
 121         ret = __bpf_prog_charge(fp_old->aux->user, delta);
 122         if (ret)
 123                 return NULL;
 124
 125         fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
 126         if (fp == NULL) {
 127                 __bpf_prog_uncharge(fp_old->aux->user, delta);
 128         } else {
 129                 kmemcheck_annotate_bitfield(fp, meta);
 130
 131                 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
 132                 fp->pages = pages;
 133                 fp->aux->prog = fp;
 134
 135                 /* We keep fp->aux from fp_old around in the new
 136                  * reallocated structure.
 137                  */
 138                 fp_old->aux = NULL;
 139                 __bpf_prog_free(fp_old);
 140         }
 141
 142         return fp;
 143 }
 144
 145 void __bpf_prog_free(struct bpf_prog *fp)
 146 {
 147         kfree(fp->aux);
 148         vfree(fp);
 149 }
 150
 151 static bool bpf_is_jmp_and_has_target(const struct bpf_insn *insn)
 152 {
 153         return BPF_CLASS(insn->code) == BPF_JMP  &&
 154                /* Call and Exit are both special jumps with no
 155                 * target inside the BPF instruction image.
 156                 */
 157                BPF_OP(insn->code) != BPF_CALL &&
 158                BPF_OP(insn->code) != BPF_EXIT;
 159 }
 160
 161 static void bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta)
 162 {
 163         struct bpf_insn *insn = prog->insnsi;
 164         u32 i, insn_cnt = prog->len;
 165
 166         for (i = 0; i < insn_cnt; i++, insn++) {
 167                 if (!bpf_is_jmp_and_has_target(insn))
 168                         continue;
 169
 170                 /* Adjust offset of jmps if we cross boundaries. */
 171                 if (i < pos && i + insn->off + 1 > pos)
 172                         insn->off += delta;
 173                 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
 174                         insn->off -= delta;
 175         }
 176 }
 177
 178 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 179                                        const struct bpf_insn *patch, u32 len)
 180 {
 181         u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
 182         struct bpf_prog *prog_adj;
 183
 184         /* Since our patchlet doesn't expand the image, we're done. */
 185         if (insn_delta == 0) {
 186                 memcpy(prog->insnsi + off, patch, sizeof(*patch));
 187                 return prog;
 188         }
 189
 190         insn_adj_cnt = prog->len + insn_delta;
 191
 192         /* Several new instructions need to be inserted. Make room
 193          * for them. Likely, there's no need for a new allocation as
 194          * last page could have large enough tailroom.
 195          */
 196         prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
 197                                     GFP_USER);
 198         if (!prog_adj)
 199                 return NULL;
 200
 201         prog_adj->len = insn_adj_cnt;
 202
 203         /* Patching happens in 3 steps:
 204          *
 205          * 1) Move over tail of insnsi from next instruction onwards,
 206          *    so we can patch the single target insn with one or more
 207          *    new ones (patching is always from 1 to n insns, n > 0).
 208          * 2) Inject new instructions at the target location.
 209          * 3) Adjust branch offsets if necessary.
 210          */
 211         insn_rest = insn_adj_cnt - off - len;
 212
 213         memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
 214                 sizeof(*patch) * insn_rest);
 215         memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
 216
 217         bpf_adj_branches(prog_adj, off, insn_delta);
 218
 219         return prog_adj;
 220 }
 221
 222 #ifdef CONFIG_BPF_JIT
 223 /* All BPF JIT sysctl knobs here. */
 224 int bpf_jit_enable   __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
 225 int bpf_jit_harden   __read_mostly;
 226 long bpf_jit_limit   __read_mostly;
 227 long bpf_jit_limit_max __read_mostly;
 228
 229 static atomic_long_t bpf_jit_current;
 230
 231 /* Can be overridden by an arch's JIT compiler if it has a custom,
 232  * dedicated BPF backend memory area, or if neither of the two
 233  * below apply.
 234  */
 235 u64 __weak bpf_jit_alloc_exec_limit(void)
 236 {
 237 #if defined(MODULES_VADDR)
 238         return MODULES_END - MODULES_VADDR;
 239 #else
 240         return VMALLOC_END - VMALLOC_START;
 241 #endif
 242 }
 243
 244 static int __init bpf_jit_charge_init(void)
 245 {
 246         /* Only used as heuristic here to derive limit. */
 247         bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
 248         bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
 249                                             PAGE_SIZE), LONG_MAX);
 250         return 0;
 251 }
 252 pure_initcall(bpf_jit_charge_init);
 253
 254 static int bpf_jit_charge_modmem(u32 pages)
 255 {
 256         if (atomic_long_add_return(pages, &bpf_jit_current) >
 257             (bpf_jit_limit >> PAGE_SHIFT)) {
 258                 if (!capable(CAP_SYS_ADMIN)) {
 259                         atomic_long_sub(pages, &bpf_jit_current);
 260                         return -EPERM;
 261                 }
 262         }
 263
 264         return 0;
 265 }
 266
 267 static void bpf_jit_uncharge_modmem(u32 pages)
 268 {
 269         atomic_long_sub(pages, &bpf_jit_current);
 270 }
 271
 272 struct bpf_binary_header *
 273 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 274                      unsigned int alignment,
 275                      bpf_jit_fill_hole_t bpf_fill_ill_insns)
 276 {
 277         struct bpf_binary_header *hdr;
 278         u32 size, hole, start, pages;
 279
 280         /* Most of BPF filters are really small, but if some of them
 281          * fill a page, allow at least 128 extra bytes to insert a
 282          * random section of illegal instructions.
 283          */
 284         size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
 285         pages = size / PAGE_SIZE;
 286
 287         if (bpf_jit_charge_modmem(pages))
 288                 return NULL;
 289         hdr = module_alloc(size);
 290         if (!hdr) {
 291                 bpf_jit_uncharge_modmem(pages);
 292                 return NULL;
 293         }
 294
 295         /* Fill space with illegal/arch-dep instructions. */
 296         bpf_fill_ill_insns(hdr, size);
 297
 298         hdr->pages = pages;
 299         hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
 300                      PAGE_SIZE - sizeof(*hdr));
 301         start = (get_random_int() % hole) & ~(alignment - 1);
 302
 303         /* Leave a random number of instructions before BPF code. */
 304         *image_ptr = &hdr->image[start];
 305
 306         return hdr;
 307 }
 308
 309 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 310 {
 311         u32 pages = hdr->pages;
 312
 313         module_memfree(hdr);
 314         bpf_jit_uncharge_modmem(pages);
 315 }
 316
 317 static int bpf_jit_blind_insn(const struct bpf_insn *from,
 318                               const struct bpf_insn *aux,
 319                               struct bpf_insn *to_buff)
 320 {
 321         struct bpf_insn *to = to_buff;
 322         u32 imm_rnd = get_random_int();
 323         s16 off;
 324
 325         BUILD_BUG_ON(BPF_REG_AX  + 1 != MAX_BPF_JIT_REG);
 326         BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
 327
 328         if (from->imm == 0 &&
 329             (from->code == (BPF_ALU   | BPF_MOV | BPF_K) ||
 330              from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
 331                 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
 332                 goto out;
 333         }
 334
 335         switch (from->code) {
 336         case BPF_ALU | BPF_ADD | BPF_K:
 337         case BPF_ALU | BPF_SUB | BPF_K:
 338         case BPF_ALU | BPF_AND | BPF_K:
 339         case BPF_ALU | BPF_OR  | BPF_K:
 340         case BPF_ALU | BPF_XOR | BPF_K:
 341         case BPF_ALU | BPF_MUL | BPF_K:
 342         case BPF_ALU | BPF_MOV | BPF_K:
 343         case BPF_ALU | BPF_DIV | BPF_K:
 344         case BPF_ALU | BPF_MOD | BPF_K:
 345                 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 346                 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 347                 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
 348                 break;
 349
 350         case BPF_ALU64 | BPF_ADD | BPF_K:
 351         case BPF_ALU64 | BPF_SUB | BPF_K:
 352         case BPF_ALU64 | BPF_AND | BPF_K:
 353         case BPF_ALU64 | BPF_OR  | BPF_K:
 354         case BPF_ALU64 | BPF_XOR | BPF_K:
 355         case BPF_ALU64 | BPF_MUL | BPF_K:
 356         case BPF_ALU64 | BPF_MOV | BPF_K:
 357         case BPF_ALU64 | BPF_DIV | BPF_K:
 358         case BPF_ALU64 | BPF_MOD | BPF_K:
 359                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 360                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 361                 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
 362                 break;
 363
 364         case BPF_JMP | BPF_JEQ  | BPF_K:
 365         case BPF_JMP | BPF_JNE  | BPF_K:
 366         case BPF_JMP | BPF_JGT  | BPF_K:
 367         case BPF_JMP | BPF_JGE  | BPF_K:
 368         case BPF_JMP | BPF_JSGT | BPF_K:
 369         case BPF_JMP | BPF_JSGE | BPF_K:
 370         case BPF_JMP | BPF_JSET | BPF_K:
 371                 /* Accommodate for extra offset in case of a backjump. */
 372                 off = from->off;
 373                 if (off < 0)
 374                         off -= 2;
 375                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 376                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 377                 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
 378                 break;
 379
 380         case BPF_LD | BPF_ABS | BPF_W:
 381         case BPF_LD | BPF_ABS | BPF_H:
 382         case BPF_LD | BPF_ABS | BPF_B:
 383                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 384                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 385                 *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0);
 386                 break;
 387
 388         case BPF_LD | BPF_IND | BPF_W:
 389         case BPF_LD | BPF_IND | BPF_H:
 390         case BPF_LD | BPF_IND | BPF_B:
 391                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 392                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 393                 *to++ = BPF_ALU32_REG(BPF_ADD, BPF_REG_AX, from->src_reg);
 394                 *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0);
 395                 break;
 396
 397         case BPF_LD | BPF_IMM | BPF_DW:
 398                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
 399                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 400                 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
 401                 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
 402                 break;
 403         case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
 404                 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
 405                 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 406                 *to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
 407                 break;
 408
 409         case BPF_ST | BPF_MEM | BPF_DW:
 410         case BPF_ST | BPF_MEM | BPF_W:
 411         case BPF_ST | BPF_MEM | BPF_H:
 412         case BPF_ST | BPF_MEM | BPF_B:
 413                 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
 414                 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
 415                 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
 416                 break;
 417         }
 418 out:
 419         return to - to_buff;
 420 }
 421
 422 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
 423                                               gfp_t gfp_extra_flags)
 424 {
 425         gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO |
 426                           gfp_extra_flags;
 427         struct bpf_prog *fp;
 428
 429         fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
 430         if (fp != NULL) {
 431                 kmemcheck_annotate_bitfield(fp, meta);
 432
 433                 /* aux->prog still points to the fp_other one, so
 434                  * when promoting the clone to the real program,
 435                  * this still needs to be adapted.
 436                  */
 437                 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
 438         }
 439
 440         return fp;
 441 }
 442
 443 static void bpf_prog_clone_free(struct bpf_prog *fp)
 444 {
 445         /* aux was stolen by the other clone, so we cannot free
 446          * it from this path! It will be freed eventually by the
 447          * other program on release.
 448          *
 449          * At this point, we don't need a deferred release since
 450          * clone is guaranteed to not be locked.
 451          */
 452         fp->aux = NULL;
 453         __bpf_prog_free(fp);
 454 }
 455
 456 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
 457 {
 458         /* We have to repoint aux->prog to self, as we don't
 459          * know whether fp here is the clone or the original.
 460          */
 461         fp->aux->prog = fp;
 462         bpf_prog_clone_free(fp_other);
 463 }
 464
 465 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
 466 {
 467         struct bpf_insn insn_buff[16], aux[2];
 468         struct bpf_prog *clone, *tmp;
 469         int insn_delta, insn_cnt;
 470         struct bpf_insn *insn;
 471         int i, rewritten;
 472
 473         if (!bpf_jit_blinding_enabled())
 474                 return prog;
 475
 476         clone = bpf_prog_clone_create(prog, GFP_USER);
 477         if (!clone)
 478                 return ERR_PTR(-ENOMEM);
 479
 480         insn_cnt = clone->len;
 481         insn = clone->insnsi;
 482
 483         for (i = 0; i < insn_cnt; i++, insn++) {
 484                 /* We temporarily need to hold the original ld64 insn
 485                  * so that we can still access the first part in the
 486                  * second blinding run.
 487                  */
 488                 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
 489                     insn[1].code == 0)
 490                         memcpy(aux, insn, sizeof(aux));
 491
 492                 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
 493                 if (!rewritten)
 494                         continue;
 495
 496                 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
 497                 if (!tmp) {
 498                         /* Patching may have repointed aux->prog during
 499                          * realloc from the original one, so we need to
 500                          * fix it up here on error.
 501                          */
 502                         bpf_jit_prog_release_other(prog, clone);
 503                         return ERR_PTR(-ENOMEM);
 504                 }
 505
 506                 clone = tmp;
 507                 insn_delta = rewritten - 1;
 508
 509                 /* Walk new program and skip insns we just inserted. */
 510                 insn = clone->insnsi + i + insn_delta;
 511                 insn_cnt += insn_delta;
 512                 i        += insn_delta;
 513         }
 514
 515         return clone;
 516 }
 517 #endif /* CONFIG_BPF_JIT */
 518
 519 /* Base function for offset calculation. Needs to go into .text section,
 520  * therefore keeping it non-static as well; will also be used by JITs
 521  * anyway later on, so do not let the compiler omit it.
 522  */
 523 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 524 {
 525         return 0;
 526 }
 527 EXPORT_SYMBOL_GPL(__bpf_call_base);
 528
 529 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
 530 /**
 531  *      __bpf_prog_run - run eBPF program on a given context
 532  *      @ctx: is the data we are operating on
 533  *      @insn: is the array of eBPF instructions
 534  *
 535  * Decode and execute eBPF instructions.
 536  */
 537 static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 538 {
 539         u64 stack[MAX_BPF_STACK / sizeof(u64)];
 540         u64 regs[MAX_BPF_REG], tmp;
 541         static const void *jumptable[256] = {
 542                 [0 ... 255] = &&default_label,
 543                 /* Now overwrite non-defaults ... */
 544                 /* 32 bit ALU operations */
 545                 [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X,
 546                 [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K,
 547                 [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X,
 548                 [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K,
 549                 [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X,
 550                 [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K,
 551                 [BPF_ALU | BPF_OR | BPF_X]  = &&ALU_OR_X,
 552                 [BPF_ALU | BPF_OR | BPF_K]  = &&ALU_OR_K,
 553                 [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X,
 554                 [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K,
 555                 [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X,
 556                 [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K,
 557                 [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X,
 558                 [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K,
 559                 [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X,
 560                 [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K,
 561                 [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X,
 562                 [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K,
 563                 [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X,
 564                 [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K,
 565                 [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X,
 566                 [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K,
 567                 [BPF_ALU | BPF_NEG] = &&ALU_NEG,
 568                 [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE,
 569                 [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE,
 570                 /* 64 bit ALU operations */
 571                 [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X,
 572                 [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K,
 573                 [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X,
 574                 [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K,
 575                 [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X,
 576                 [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K,
 577                 [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X,
 578                 [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K,
 579                 [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X,
 580                 [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K,
 581                 [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X,
 582                 [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K,
 583                 [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X,
 584                 [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K,
 585                 [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X,
 586                 [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K,
 587                 [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X,
 588                 [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K,
 589                 [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X,
 590                 [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K,
 591                 [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X,
 592                 [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K,
 593                 [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X,
 594                 [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K,
 595                 [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG,
 596                 /* Call instruction */
 597                 [BPF_JMP | BPF_CALL] = &&JMP_CALL,
 598                 [BPF_JMP | BPF_CALL | BPF_X] = &&JMP_TAIL_CALL,
 599                 /* Jumps */
 600                 [BPF_JMP | BPF_JA] = &&JMP_JA,
 601                 [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X,
 602                 [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K,
 603                 [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X,
 604                 [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K,
 605                 [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X,
 606                 [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K,
 607                 [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X,
 608                 [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K,
 609                 [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X,
 610                 [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K,
 611                 [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X,
 612                 [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K,
 613                 [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X,
 614                 [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K,
 615                 /* Program return */
 616                 [BPF_JMP | BPF_EXIT] = &&JMP_EXIT,
 617                 /* Store instructions */
 618                 [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B,
 619                 [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H,
 620                 [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W,
 621                 [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW,
 622                 [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W,
 623                 [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW,
 624                 [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B,
 625                 [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H,
 626                 [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W,
 627                 [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW,
 628                 /* Load instructions */
 629                 [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B,
 630                 [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H,
 631                 [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W,
 632                 [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW,
 633                 [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W,
 634                 [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H,
 635                 [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B,
 636                 [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W,
 637                 [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H,
 638                 [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B,
 639                 [BPF_LD | BPF_IMM | BPF_DW] = &&LD_IMM_DW,
 640         };
 641         u32 tail_call_cnt = 0;
 642         void *ptr;
 643         int off;
 644
 645 #define CONT     ({ insn++; goto select_insn; })
 646 #define CONT_JMP ({ insn++; goto select_insn; })
 647
 648         FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
 649         ARG1 = (u64) (unsigned long) ctx;
 650
 651 select_insn:
 652         goto *jumptable[insn->code];
 653
 654         /* ALU */
 655 #define ALU(OPCODE, OP)                 \
 656         ALU64_##OPCODE##_X:             \
 657                 DST = DST OP SRC;       \
 658                 CONT;                   \
 659         ALU_##OPCODE##_X:               \
 660                 DST = (u32) DST OP (u32) SRC;   \
 661                 CONT;                   \
 662         ALU64_##OPCODE##_K:             \
 663                 DST = DST OP IMM;               \
 664                 CONT;                   \
 665         ALU_##OPCODE##_K:               \
 666                 DST = (u32) DST OP (u32) IMM;   \
 667                 CONT;
 668
 669         ALU(ADD,  +)
 670         ALU(SUB,  -)
 671         ALU(AND,  &)
 672         ALU(OR,   |)
 673         ALU(LSH, <<)
 674         ALU(RSH, >>)
 675         ALU(XOR,  ^)
 676         ALU(MUL,  *)
 677 #undef ALU
 678         ALU_NEG:
 679                 DST = (u32) -DST;
 680                 CONT;
 681         ALU64_NEG:
 682                 DST = -DST;
 683                 CONT;
 684         ALU_MOV_X:
 685                 DST = (u32) SRC;
 686                 CONT;
 687         ALU_MOV_K:
 688                 DST = (u32) IMM;
 689                 CONT;
 690         ALU64_MOV_X:
 691                 DST = SRC;
 692                 CONT;
 693         ALU64_MOV_K:
 694                 DST = IMM;
 695                 CONT;
 696         LD_IMM_DW:
 697                 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
 698                 insn++;
 699                 CONT;
 700         ALU64_ARSH_X:
 701                 (*(s64 *) &DST) >>= SRC;
 702                 CONT;
 703         ALU64_ARSH_K:
 704                 (*(s64 *) &DST) >>= IMM;
 705                 CONT;
 706         ALU64_MOD_X:
 707                 if (unlikely(SRC == 0))
 708                         return 0;
 709                 div64_u64_rem(DST, SRC, &tmp);
 710                 DST = tmp;
 711                 CONT;
 712         ALU_MOD_X:
 713                 if (unlikely((u32)SRC == 0))
 714                         return 0;
 715                 tmp = (u32) DST;
 716                 DST = do_div(tmp, (u32) SRC);
 717                 CONT;
 718         ALU64_MOD_K:
 719                 div64_u64_rem(DST, IMM, &tmp);
 720                 DST = tmp;
 721                 CONT;
 722         ALU_MOD_K:
 723                 tmp = (u32) DST;
 724                 DST = do_div(tmp, (u32) IMM);
 725                 CONT;
 726         ALU64_DIV_X:
 727                 if (unlikely(SRC == 0))
 728                         return 0;
 729                 DST = div64_u64(DST, SRC);
 730                 CONT;
 731         ALU_DIV_X:
 732                 if (unlikely((u32)SRC == 0))
 733                         return 0;
 734                 tmp = (u32) DST;
 735                 do_div(tmp, (u32) SRC);
 736                 DST = (u32) tmp;
 737                 CONT;
 738         ALU64_DIV_K:
 739                 DST = div64_u64(DST, IMM);
 740                 CONT;
 741         ALU_DIV_K:
 742                 tmp = (u32) DST;
 743                 do_div(tmp, (u32) IMM);
 744                 DST = (u32) tmp;
 745                 CONT;
 746         ALU_END_TO_BE:
 747                 switch (IMM) {
 748                 case 16:
 749                         DST = (__force u16) cpu_to_be16(DST);
 750                         break;
 751                 case 32:
 752                         DST = (__force u32) cpu_to_be32(DST);
 753                         break;
 754                 case 64:
 755                         DST = (__force u64) cpu_to_be64(DST);
 756                         break;
 757                 }
 758                 CONT;
 759         ALU_END_TO_LE:
 760                 switch (IMM) {
 761                 case 16:
 762                         DST = (__force u16) cpu_to_le16(DST);
 763                         break;
 764                 case 32:
 765                         DST = (__force u32) cpu_to_le32(DST);
 766                         break;
 767                 case 64:
 768                         DST = (__force u64) cpu_to_le64(DST);
 769                         break;
 770                 }
 771                 CONT;
 772
 773         /* CALL */
 774         JMP_CALL:
 775                 /* Function call scratches BPF_R1-BPF_R5 registers,
 776                  * preserves BPF_R6-BPF_R9, and stores return value
 777                  * into BPF_R0.
 778                  */
 779                 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
 780                                                        BPF_R4, BPF_R5);
 781                 CONT;
 782
 783         JMP_TAIL_CALL: {
 784                 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
 785                 struct bpf_array *array = container_of(map, struct bpf_array, map);
 786                 struct bpf_prog *prog;
 787                 u32 index = BPF_R3;
 788
 789                 if (unlikely(index >= array->map.max_entries))
 790                         goto out;
 791                 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
 792                         goto out;
 793
 794                 tail_call_cnt++;
 795
 796                 prog = READ_ONCE(array->ptrs[index]);
 797                 if (!prog)
 798                         goto out;
 799
 800                 /* ARG1 at this point is guaranteed to point to CTX from
 801                  * the verifier side due to the fact that the tail call is
 802                  * handeled like a helper, that is, bpf_tail_call_proto,
 803                  * where arg1_type is ARG_PTR_TO_CTX.
 804                  */
 805                 insn = prog->insnsi;
 806                 goto select_insn;
 807 out:
 808                 CONT;
 809         }
 810         /* JMP */
 811         JMP_JA:
 812                 insn += insn->off;
 813                 CONT;
 814         JMP_JEQ_X:
 815                 if (DST == SRC) {
 816                         insn += insn->off;
 817                         CONT_JMP;
 818                 }
 819                 CONT;
 820         JMP_JEQ_K:
 821                 if (DST == IMM) {
 822                         insn += insn->off;
 823                         CONT_JMP;
 824                 }
 825                 CONT;
 826         JMP_JNE_X:
 827                 if (DST != SRC) {
 828                         insn += insn->off;
 829                         CONT_JMP;
 830                 }
 831                 CONT;
 832         JMP_JNE_K:
 833                 if (DST != IMM) {
 834                         insn += insn->off;
 835                         CONT_JMP;
 836                 }
 837                 CONT;
 838         JMP_JGT_X:
 839                 if (DST > SRC) {
 840                         insn += insn->off;
 841                         CONT_JMP;
 842                 }
 843                 CONT;
 844         JMP_JGT_K:
 845                 if (DST > IMM) {
 846                         insn += insn->off;
 847                         CONT_JMP;
 848                 }
 849                 CONT;
 850         JMP_JGE_X:
 851                 if (DST >= SRC) {
 852                         insn += insn->off;
 853                         CONT_JMP;
 854                 }
 855                 CONT;
 856         JMP_JGE_K:
 857                 if (DST >= IMM) {
 858                         insn += insn->off;
 859                         CONT_JMP;
 860                 }
 861                 CONT;
 862         JMP_JSGT_X:
 863                 if (((s64) DST) > ((s64) SRC)) {
 864                         insn += insn->off;
 865                         CONT_JMP;
 866                 }
 867                 CONT;
 868         JMP_JSGT_K:
 869                 if (((s64) DST) > ((s64) IMM)) {
 870                         insn += insn->off;
 871                         CONT_JMP;
 872                 }
 873                 CONT;
 874         JMP_JSGE_X:
 875                 if (((s64) DST) >= ((s64) SRC)) {
 876                         insn += insn->off;
 877                         CONT_JMP;
 878                 }
 879                 CONT;
 880         JMP_JSGE_K:
 881                 if (((s64) DST) >= ((s64) IMM)) {
 882                         insn += insn->off;
 883                         CONT_JMP;
 884                 }
 885                 CONT;
 886         JMP_JSET_X:
 887                 if (DST & SRC) {
 888                         insn += insn->off;
 889                         CONT_JMP;
 890                 }
 891                 CONT;
 892         JMP_JSET_K:
 893                 if (DST & IMM) {
 894                         insn += insn->off;
 895                         CONT_JMP;
 896                 }
 897                 CONT;
 898         JMP_EXIT:
 899                 return BPF_R0;
 900
 901         /* STX and ST and LDX*/
 902 #define LDST(SIZEOP, SIZE)                                              \
 903         STX_MEM_##SIZEOP:                                               \
 904                 *(SIZE *)(unsigned long) (DST + insn->off) = SRC;       \
 905                 CONT;                                                   \
 906         ST_MEM_##SIZEOP:                                                \
 907                 *(SIZE *)(unsigned long) (DST + insn->off) = IMM;       \
 908                 CONT;                                                   \
 909         LDX_MEM_##SIZEOP:                                               \
 910                 DST = *(SIZE *)(unsigned long) (SRC + insn->off);       \
 911                 CONT;
 912
 913         LDST(B,   u8)
 914         LDST(H,  u16)
 915         LDST(W,  u32)
 916         LDST(DW, u64)
 917 #undef LDST
 918         STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
 919                 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
 920                            (DST + insn->off));
 921                 CONT;
 922         STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
 923                 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
 924                              (DST + insn->off));
 925                 CONT;
 926         LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */
 927                 off = IMM;
 928 load_word:
 929                 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are
 930                  * only appearing in the programs where ctx ==
 931                  * skb. All programs keep 'ctx' in regs[BPF_REG_CTX]
 932                  * == BPF_R6, bpf_convert_filter() saves it in BPF_R6,
 933                  * internal BPF verifier will check that BPF_R6 ==
 934                  * ctx.
 935                  *
 936                  * BPF_ABS and BPF_IND are wrappers of function calls,
 937                  * so they scratch BPF_R1-BPF_R5 registers, preserve
 938                  * BPF_R6-BPF_R9, and store return value into BPF_R0.
 939                  *
 940                  * Implicit input:
 941                  *   ctx == skb == BPF_R6 == CTX
 942                  *
 943                  * Explicit input:
 944                  *   SRC == any register
 945                  *   IMM == 32-bit immediate
 946                  *
 947                  * Output:
 948                  *   BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
 949                  */
 950
 951                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 4, &tmp);
 952                 if (likely(ptr != NULL)) {
 953                         BPF_R0 = get_unaligned_be32(ptr);
 954                         CONT;
 955                 }
 956
 957                 return 0;
 958         LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */
 959                 off = IMM;
 960 load_half:
 961                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 2, &tmp);
 962                 if (likely(ptr != NULL)) {
 963                         BPF_R0 = get_unaligned_be16(ptr);
 964                         CONT;
 965                 }
 966
 967                 return 0;
 968         LD_ABS_B: /* BPF_R0 = *(u8 *) (skb->data + imm32) */
 969                 off = IMM;
 970 load_byte:
 971                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 1, &tmp);
 972                 if (likely(ptr != NULL)) {
 973                         BPF_R0 = *(u8 *)ptr;
 974                         CONT;
 975                 }
 976
 977                 return 0;
 978         LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */
 979                 off = IMM + SRC;
 980                 goto load_word;
 981         LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */
 982                 off = IMM + SRC;
 983                 goto load_half;
 984         LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */
 985                 off = IMM + SRC;
 986                 goto load_byte;
 987
 988         default_label:
 989                 /* If we ever reach this, we have a bug somewhere. */
 990                 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
 991                 return 0;
 992 }
 993 STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */
 994
 995 #else
 996 static unsigned int __bpf_prog_ret0_warn(void *ctx,
 997                                          const struct bpf_insn *insn)
 998 {
 999         /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1000          * is not working properly, so warn about it!
1001          */
1002         WARN_ON_ONCE(1);
1003         return 0;
1004 }
1005 #endif
1006
1007 bool bpf_prog_array_compatible(struct bpf_array *array,
1008                                const struct bpf_prog *fp)
1009 {
1010         if (!array->owner_prog_type) {
1011                 /* There's no owner yet where we could check for
1012                  * compatibility.
1013                  */
1014                 array->owner_prog_type = fp->type;
1015                 array->owner_jited = fp->jited;
1016
1017                 return true;
1018         }
1019
1020         return array->owner_prog_type == fp->type &&
1021                array->owner_jited == fp->jited;
1022 }
1023
1024 static int bpf_check_tail_call(const struct bpf_prog *fp)
1025 {
1026         struct bpf_prog_aux *aux = fp->aux;
1027         int i;
1028
1029         for (i = 0; i < aux->used_map_cnt; i++) {
1030                 struct bpf_map *map = aux->used_maps[i];
1031                 struct bpf_array *array;
1032
1033                 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1034                         continue;
1035
1036                 array = container_of(map, struct bpf_array, map);
1037                 if (!bpf_prog_array_compatible(array, fp))
1038                         return -EINVAL;
1039         }
1040
1041         return 0;
1042 }
1043
1044 /**
1045  *      bpf_prog_select_runtime - select exec runtime for BPF program
1046  *      @fp: bpf_prog populated with internal BPF program
1047  *      @err: pointer to error variable
1048  *
1049  * Try to JIT eBPF program, if JIT is not available, use interpreter.
1050  * The BPF program will be executed via BPF_PROG_RUN() macro.
1051  */
1052 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1053 {
1054 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1055         fp->bpf_func = (void *) __bpf_prog_run;
1056 #else
1057         fp->bpf_func = (void *) __bpf_prog_ret0_warn;
1058 #endif
1059
1060         /* eBPF JITs can rewrite the program in case constant
1061          * blinding is active. However, in case of error during
1062          * blinding, bpf_int_jit_compile() must always return a
1063          * valid program, which in this case would simply not
1064          * be JITed, but falls back to the interpreter.
1065          */
1066         fp = bpf_int_jit_compile(fp);
1067 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1068         if (!fp->jited) {
1069                 *err = -ENOTSUPP;
1070                 return fp;
1071         }
1072 #endif
1073         bpf_prog_lock_ro(fp);
1074
1075         /* The tail call compatibility check can only be done at
1076          * this late stage as we need to determine, if we deal
1077          * with JITed or non JITed program concatenations and not
1078          * all eBPF JITs might immediately support all features.
1079          */
1080         *err = bpf_check_tail_call(fp);
1081
1082         return fp;
1083 }
1084 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1085
1086 static void bpf_prog_free_deferred(struct work_struct *work)
1087 {
1088         struct bpf_prog_aux *aux;
1089
1090         aux = container_of(work, struct bpf_prog_aux, work);
1091         bpf_jit_free(aux->prog);
1092 }
1093
1094 /* Free internal BPF program */
1095 void bpf_prog_free(struct bpf_prog *fp)
1096 {
1097         struct bpf_prog_aux *aux = fp->aux;
1098
1099         INIT_WORK(&aux->work, bpf_prog_free_deferred);
1100         schedule_work(&aux->work);
1101 }
1102 EXPORT_SYMBOL_GPL(bpf_prog_free);
1103
1104 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
1105 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
1106
1107 void bpf_user_rnd_init_once(void)
1108 {
1109         prandom_init_once(&bpf_user_rnd_state);
1110 }
1111
1112 BPF_CALL_0(bpf_user_rnd_u32)
1113 {
1114         /* Should someone ever have the rather unwise idea to use some
1115          * of the registers passed into this function, then note that
1116          * this function is called from native eBPF and classic-to-eBPF
1117          * transformations. Register assignments from both sides are
1118          * different, f.e. classic always sets fn(ctx, A, X) here.
1119          */
1120         struct rnd_state *state;
1121         u32 res;
1122
1123         state = &get_cpu_var(bpf_user_rnd_state);
1124         res = prandom_u32_state(state);
1125         put_cpu_var(bpf_user_rnd_state);
1126
1127         return res;
1128 }
1129
1130 /* Weak definitions of helper functions in case we don't have bpf syscall. */
1131 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
1132 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
1133 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
1134
1135 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
1136 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
1137 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
1138
1139 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
1140 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
1141 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
1142
1143 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
1144 {
1145         return NULL;
1146 }
1147
1148 u64 __weak
1149 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
1150                  void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
1151 {
1152         return -ENOTSUPP;
1153 }
1154
1155 /* Always built-in helper functions. */
1156 const struct bpf_func_proto bpf_tail_call_proto = {
1157         .func           = NULL,
1158         .gpl_only       = false,
1159         .ret_type       = RET_VOID,
1160         .arg1_type      = ARG_PTR_TO_CTX,
1161         .arg2_type      = ARG_CONST_MAP_PTR,
1162         .arg3_type      = ARG_ANYTHING,
1163 };
1164
1165 /* For classic BPF JITs that don't implement bpf_int_jit_compile(). */
1166 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
1167 {
1168         return prog;
1169 }
1170
1171 bool __weak bpf_helper_changes_skb_data(void *func)
1172 {
1173         return false;
1174 }
1175
1176 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
1177  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
1178  */
1179 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
1180                          int len)
1181 {
1182         return -EFAULT;
1183 }