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 <uapi/linux/btf.h> 25 #include <linux/filter.h> 26 #include <linux/skbuff.h> 27 #include <linux/vmalloc.h> 28 #include <linux/random.h> 29 #include <linux/moduleloader.h> 30 #include <linux/bpf.h> 31 #include <linux/btf.h> 32 #include <linux/frame.h> 33 #include <linux/rbtree_latch.h> 34 #include <linux/kallsyms.h> 35 #include <linux/rcupdate.h> 36 #include <linux/perf_event.h> 37 38 #include <asm/unaligned.h> 39 40 /* Registers */ 41 #define BPF_R0 regs[BPF_REG_0] 42 #define BPF_R1 regs[BPF_REG_1] 43 #define BPF_R2 regs[BPF_REG_2] 44 #define BPF_R3 regs[BPF_REG_3] 45 #define BPF_R4 regs[BPF_REG_4] 46 #define BPF_R5 regs[BPF_REG_5] 47 #define BPF_R6 regs[BPF_REG_6] 48 #define BPF_R7 regs[BPF_REG_7] 49 #define BPF_R8 regs[BPF_REG_8] 50 #define BPF_R9 regs[BPF_REG_9] 51 #define BPF_R10 regs[BPF_REG_10] 52 53 /* Named registers */ 54 #define DST regs[insn->dst_reg] 55 #define SRC regs[insn->src_reg] 56 #define FP regs[BPF_REG_FP] 57 #define AX regs[BPF_REG_AX] 58 #define ARG1 regs[BPF_REG_ARG1] 59 #define CTX regs[BPF_REG_CTX] 60 #define IMM insn->imm 61 62 /* No hurry in this branch 63 * 64 * Exported for the bpf jit load helper. 65 */ 66 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size) 67 { 68 u8 *ptr = NULL; 69 70 if (k >= SKF_NET_OFF) 71 ptr = skb_network_header(skb) + k - SKF_NET_OFF; 72 else if (k >= SKF_LL_OFF) 73 ptr = skb_mac_header(skb) + k - SKF_LL_OFF; 74 75 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb)) 76 return ptr; 77 78 return NULL; 79 } 80 81 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags) 82 { 83 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; 84 struct bpf_prog_aux *aux; 85 struct bpf_prog *fp; 86 87 size = round_up(size, PAGE_SIZE); 88 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); 89 if (fp == NULL) 90 return NULL; 91 92 aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags); 93 if (aux == NULL) { 94 vfree(fp); 95 return NULL; 96 } 97 98 fp->pages = size / PAGE_SIZE; 99 fp->aux = aux; 100 fp->aux->prog = fp; 101 fp->jit_requested = ebpf_jit_enabled(); 102 103 INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode); 104 105 return fp; 106 } 107 108 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags) 109 { 110 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; 111 struct bpf_prog *prog; 112 int cpu; 113 114 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags); 115 if (!prog) 116 return NULL; 117 118 prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags); 119 if (!prog->aux->stats) { 120 kfree(prog->aux); 121 vfree(prog); 122 return NULL; 123 } 124 125 for_each_possible_cpu(cpu) { 126 struct bpf_prog_stats *pstats; 127 128 pstats = per_cpu_ptr(prog->aux->stats, cpu); 129 u64_stats_init(&pstats->syncp); 130 } 131 return prog; 132 } 133 EXPORT_SYMBOL_GPL(bpf_prog_alloc); 134 135 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog) 136 { 137 if (!prog->aux->nr_linfo || !prog->jit_requested) 138 return 0; 139 140 prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo, 141 sizeof(*prog->aux->jited_linfo), 142 GFP_KERNEL | __GFP_NOWARN); 143 if (!prog->aux->jited_linfo) 144 return -ENOMEM; 145 146 return 0; 147 } 148 149 void bpf_prog_free_jited_linfo(struct bpf_prog *prog) 150 { 151 kfree(prog->aux->jited_linfo); 152 prog->aux->jited_linfo = NULL; 153 } 154 155 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog) 156 { 157 if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0]) 158 bpf_prog_free_jited_linfo(prog); 159 } 160 161 /* The jit engine is responsible to provide an array 162 * for insn_off to the jited_off mapping (insn_to_jit_off). 163 * 164 * The idx to this array is the insn_off. Hence, the insn_off 165 * here is relative to the prog itself instead of the main prog. 166 * This array has one entry for each xlated bpf insn. 167 * 168 * jited_off is the byte off to the last byte of the jited insn. 169 * 170 * Hence, with 171 * insn_start: 172 * The first bpf insn off of the prog. The insn off 173 * here is relative to the main prog. 174 * e.g. if prog is a subprog, insn_start > 0 175 * linfo_idx: 176 * The prog's idx to prog->aux->linfo and jited_linfo 177 * 178 * jited_linfo[linfo_idx] = prog->bpf_func 179 * 180 * For i > linfo_idx, 181 * 182 * jited_linfo[i] = prog->bpf_func + 183 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1] 184 */ 185 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 186 const u32 *insn_to_jit_off) 187 { 188 u32 linfo_idx, insn_start, insn_end, nr_linfo, i; 189 const struct bpf_line_info *linfo; 190 void **jited_linfo; 191 192 if (!prog->aux->jited_linfo) 193 /* Userspace did not provide linfo */ 194 return; 195 196 linfo_idx = prog->aux->linfo_idx; 197 linfo = &prog->aux->linfo[linfo_idx]; 198 insn_start = linfo[0].insn_off; 199 insn_end = insn_start + prog->len; 200 201 jited_linfo = &prog->aux->jited_linfo[linfo_idx]; 202 jited_linfo[0] = prog->bpf_func; 203 204 nr_linfo = prog->aux->nr_linfo - linfo_idx; 205 206 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++) 207 /* The verifier ensures that linfo[i].insn_off is 208 * strictly increasing 209 */ 210 jited_linfo[i] = prog->bpf_func + 211 insn_to_jit_off[linfo[i].insn_off - insn_start - 1]; 212 } 213 214 void bpf_prog_free_linfo(struct bpf_prog *prog) 215 { 216 bpf_prog_free_jited_linfo(prog); 217 kvfree(prog->aux->linfo); 218 } 219 220 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 221 gfp_t gfp_extra_flags) 222 { 223 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; 224 struct bpf_prog *fp; 225 u32 pages, delta; 226 int ret; 227 228 BUG_ON(fp_old == NULL); 229 230 size = round_up(size, PAGE_SIZE); 231 pages = size / PAGE_SIZE; 232 if (pages <= fp_old->pages) 233 return fp_old; 234 235 delta = pages - fp_old->pages; 236 ret = __bpf_prog_charge(fp_old->aux->user, delta); 237 if (ret) 238 return NULL; 239 240 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); 241 if (fp == NULL) { 242 __bpf_prog_uncharge(fp_old->aux->user, delta); 243 } else { 244 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE); 245 fp->pages = pages; 246 fp->aux->prog = fp; 247 248 /* We keep fp->aux from fp_old around in the new 249 * reallocated structure. 250 */ 251 fp_old->aux = NULL; 252 __bpf_prog_free(fp_old); 253 } 254 255 return fp; 256 } 257 258 void __bpf_prog_free(struct bpf_prog *fp) 259 { 260 if (fp->aux) { 261 free_percpu(fp->aux->stats); 262 kfree(fp->aux); 263 } 264 vfree(fp); 265 } 266 267 int bpf_prog_calc_tag(struct bpf_prog *fp) 268 { 269 const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64); 270 u32 raw_size = bpf_prog_tag_scratch_size(fp); 271 u32 digest[SHA_DIGEST_WORDS]; 272 u32 ws[SHA_WORKSPACE_WORDS]; 273 u32 i, bsize, psize, blocks; 274 struct bpf_insn *dst; 275 bool was_ld_map; 276 u8 *raw, *todo; 277 __be32 *result; 278 __be64 *bits; 279 280 raw = vmalloc(raw_size); 281 if (!raw) 282 return -ENOMEM; 283 284 sha_init(digest); 285 memset(ws, 0, sizeof(ws)); 286 287 /* We need to take out the map fd for the digest calculation 288 * since they are unstable from user space side. 289 */ 290 dst = (void *)raw; 291 for (i = 0, was_ld_map = false; i < fp->len; i++) { 292 dst[i] = fp->insnsi[i]; 293 if (!was_ld_map && 294 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) && 295 (dst[i].src_reg == BPF_PSEUDO_MAP_FD || 296 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) { 297 was_ld_map = true; 298 dst[i].imm = 0; 299 } else if (was_ld_map && 300 dst[i].code == 0 && 301 dst[i].dst_reg == 0 && 302 dst[i].src_reg == 0 && 303 dst[i].off == 0) { 304 was_ld_map = false; 305 dst[i].imm = 0; 306 } else { 307 was_ld_map = false; 308 } 309 } 310 311 psize = bpf_prog_insn_size(fp); 312 memset(&raw[psize], 0, raw_size - psize); 313 raw[psize++] = 0x80; 314 315 bsize = round_up(psize, SHA_MESSAGE_BYTES); 316 blocks = bsize / SHA_MESSAGE_BYTES; 317 todo = raw; 318 if (bsize - psize >= sizeof(__be64)) { 319 bits = (__be64 *)(todo + bsize - sizeof(__be64)); 320 } else { 321 bits = (__be64 *)(todo + bsize + bits_offset); 322 blocks++; 323 } 324 *bits = cpu_to_be64((psize - 1) << 3); 325 326 while (blocks--) { 327 sha_transform(digest, todo, ws); 328 todo += SHA_MESSAGE_BYTES; 329 } 330 331 result = (__force __be32 *)digest; 332 for (i = 0; i < SHA_DIGEST_WORDS; i++) 333 result[i] = cpu_to_be32(digest[i]); 334 memcpy(fp->tag, result, sizeof(fp->tag)); 335 336 vfree(raw); 337 return 0; 338 } 339 340 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old, 341 s32 end_new, u32 curr, const bool probe_pass) 342 { 343 const s64 imm_min = S32_MIN, imm_max = S32_MAX; 344 s32 delta = end_new - end_old; 345 s64 imm = insn->imm; 346 347 if (curr < pos && curr + imm + 1 >= end_old) 348 imm += delta; 349 else if (curr >= end_new && curr + imm + 1 < end_new) 350 imm -= delta; 351 if (imm < imm_min || imm > imm_max) 352 return -ERANGE; 353 if (!probe_pass) 354 insn->imm = imm; 355 return 0; 356 } 357 358 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old, 359 s32 end_new, u32 curr, const bool probe_pass) 360 { 361 const s32 off_min = S16_MIN, off_max = S16_MAX; 362 s32 delta = end_new - end_old; 363 s32 off = insn->off; 364 365 if (curr < pos && curr + off + 1 >= end_old) 366 off += delta; 367 else if (curr >= end_new && curr + off + 1 < end_new) 368 off -= delta; 369 if (off < off_min || off > off_max) 370 return -ERANGE; 371 if (!probe_pass) 372 insn->off = off; 373 return 0; 374 } 375 376 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old, 377 s32 end_new, const bool probe_pass) 378 { 379 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0); 380 struct bpf_insn *insn = prog->insnsi; 381 int ret = 0; 382 383 for (i = 0; i < insn_cnt; i++, insn++) { 384 u8 code; 385 386 /* In the probing pass we still operate on the original, 387 * unpatched image in order to check overflows before we 388 * do any other adjustments. Therefore skip the patchlet. 389 */ 390 if (probe_pass && i == pos) { 391 i = end_new; 392 insn = prog->insnsi + end_old; 393 } 394 code = insn->code; 395 if ((BPF_CLASS(code) != BPF_JMP && 396 BPF_CLASS(code) != BPF_JMP32) || 397 BPF_OP(code) == BPF_EXIT) 398 continue; 399 /* Adjust offset of jmps if we cross patch boundaries. */ 400 if (BPF_OP(code) == BPF_CALL) { 401 if (insn->src_reg != BPF_PSEUDO_CALL) 402 continue; 403 ret = bpf_adj_delta_to_imm(insn, pos, end_old, 404 end_new, i, probe_pass); 405 } else { 406 ret = bpf_adj_delta_to_off(insn, pos, end_old, 407 end_new, i, probe_pass); 408 } 409 if (ret) 410 break; 411 } 412 413 return ret; 414 } 415 416 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta) 417 { 418 struct bpf_line_info *linfo; 419 u32 i, nr_linfo; 420 421 nr_linfo = prog->aux->nr_linfo; 422 if (!nr_linfo || !delta) 423 return; 424 425 linfo = prog->aux->linfo; 426 427 for (i = 0; i < nr_linfo; i++) 428 if (off < linfo[i].insn_off) 429 break; 430 431 /* Push all off < linfo[i].insn_off by delta */ 432 for (; i < nr_linfo; i++) 433 linfo[i].insn_off += delta; 434 } 435 436 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 437 const struct bpf_insn *patch, u32 len) 438 { 439 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1; 440 const u32 cnt_max = S16_MAX; 441 struct bpf_prog *prog_adj; 442 int err; 443 444 /* Since our patchlet doesn't expand the image, we're done. */ 445 if (insn_delta == 0) { 446 memcpy(prog->insnsi + off, patch, sizeof(*patch)); 447 return prog; 448 } 449 450 insn_adj_cnt = prog->len + insn_delta; 451 452 /* Reject anything that would potentially let the insn->off 453 * target overflow when we have excessive program expansions. 454 * We need to probe here before we do any reallocation where 455 * we afterwards may not fail anymore. 456 */ 457 if (insn_adj_cnt > cnt_max && 458 (err = bpf_adj_branches(prog, off, off + 1, off + len, true))) 459 return ERR_PTR(err); 460 461 /* Several new instructions need to be inserted. Make room 462 * for them. Likely, there's no need for a new allocation as 463 * last page could have large enough tailroom. 464 */ 465 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt), 466 GFP_USER); 467 if (!prog_adj) 468 return ERR_PTR(-ENOMEM); 469 470 prog_adj->len = insn_adj_cnt; 471 472 /* Patching happens in 3 steps: 473 * 474 * 1) Move over tail of insnsi from next instruction onwards, 475 * so we can patch the single target insn with one or more 476 * new ones (patching is always from 1 to n insns, n > 0). 477 * 2) Inject new instructions at the target location. 478 * 3) Adjust branch offsets if necessary. 479 */ 480 insn_rest = insn_adj_cnt - off - len; 481 482 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1, 483 sizeof(*patch) * insn_rest); 484 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len); 485 486 /* We are guaranteed to not fail at this point, otherwise 487 * the ship has sailed to reverse to the original state. An 488 * overflow cannot happen at this point. 489 */ 490 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false)); 491 492 bpf_adj_linfo(prog_adj, off, insn_delta); 493 494 return prog_adj; 495 } 496 497 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt) 498 { 499 /* Branch offsets can't overflow when program is shrinking, no need 500 * to call bpf_adj_branches(..., true) here 501 */ 502 memmove(prog->insnsi + off, prog->insnsi + off + cnt, 503 sizeof(struct bpf_insn) * (prog->len - off - cnt)); 504 prog->len -= cnt; 505 506 return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false)); 507 } 508 509 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp) 510 { 511 int i; 512 513 for (i = 0; i < fp->aux->func_cnt; i++) 514 bpf_prog_kallsyms_del(fp->aux->func[i]); 515 } 516 517 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp) 518 { 519 bpf_prog_kallsyms_del_subprogs(fp); 520 bpf_prog_kallsyms_del(fp); 521 } 522 523 #ifdef CONFIG_BPF_JIT 524 /* All BPF JIT sysctl knobs here. */ 525 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON); 526 int bpf_jit_harden __read_mostly; 527 int bpf_jit_kallsyms __read_mostly; 528 long bpf_jit_limit __read_mostly; 529 530 static __always_inline void 531 bpf_get_prog_addr_region(const struct bpf_prog *prog, 532 unsigned long *symbol_start, 533 unsigned long *symbol_end) 534 { 535 const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog); 536 unsigned long addr = (unsigned long)hdr; 537 538 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog)); 539 540 *symbol_start = addr; 541 *symbol_end = addr + hdr->pages * PAGE_SIZE; 542 } 543 544 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym) 545 { 546 const char *end = sym + KSYM_NAME_LEN; 547 const struct btf_type *type; 548 const char *func_name; 549 550 BUILD_BUG_ON(sizeof("bpf_prog_") + 551 sizeof(prog->tag) * 2 + 552 /* name has been null terminated. 553 * We should need +1 for the '_' preceding 554 * the name. However, the null character 555 * is double counted between the name and the 556 * sizeof("bpf_prog_") above, so we omit 557 * the +1 here. 558 */ 559 sizeof(prog->aux->name) > KSYM_NAME_LEN); 560 561 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_"); 562 sym = bin2hex(sym, prog->tag, sizeof(prog->tag)); 563 564 /* prog->aux->name will be ignored if full btf name is available */ 565 if (prog->aux->func_info_cnt) { 566 type = btf_type_by_id(prog->aux->btf, 567 prog->aux->func_info[prog->aux->func_idx].type_id); 568 func_name = btf_name_by_offset(prog->aux->btf, type->name_off); 569 snprintf(sym, (size_t)(end - sym), "_%s", func_name); 570 return; 571 } 572 573 if (prog->aux->name[0]) 574 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name); 575 else 576 *sym = 0; 577 } 578 579 static __always_inline unsigned long 580 bpf_get_prog_addr_start(struct latch_tree_node *n) 581 { 582 unsigned long symbol_start, symbol_end; 583 const struct bpf_prog_aux *aux; 584 585 aux = container_of(n, struct bpf_prog_aux, ksym_tnode); 586 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); 587 588 return symbol_start; 589 } 590 591 static __always_inline bool bpf_tree_less(struct latch_tree_node *a, 592 struct latch_tree_node *b) 593 { 594 return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b); 595 } 596 597 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n) 598 { 599 unsigned long val = (unsigned long)key; 600 unsigned long symbol_start, symbol_end; 601 const struct bpf_prog_aux *aux; 602 603 aux = container_of(n, struct bpf_prog_aux, ksym_tnode); 604 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); 605 606 if (val < symbol_start) 607 return -1; 608 if (val >= symbol_end) 609 return 1; 610 611 return 0; 612 } 613 614 static const struct latch_tree_ops bpf_tree_ops = { 615 .less = bpf_tree_less, 616 .comp = bpf_tree_comp, 617 }; 618 619 static DEFINE_SPINLOCK(bpf_lock); 620 static LIST_HEAD(bpf_kallsyms); 621 static struct latch_tree_root bpf_tree __cacheline_aligned; 622 623 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux) 624 { 625 WARN_ON_ONCE(!list_empty(&aux->ksym_lnode)); 626 list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms); 627 latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops); 628 } 629 630 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux) 631 { 632 if (list_empty(&aux->ksym_lnode)) 633 return; 634 635 latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops); 636 list_del_rcu(&aux->ksym_lnode); 637 } 638 639 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp) 640 { 641 return fp->jited && !bpf_prog_was_classic(fp); 642 } 643 644 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 645 { 646 return list_empty(&fp->aux->ksym_lnode) || 647 fp->aux->ksym_lnode.prev == LIST_POISON2; 648 } 649 650 void bpf_prog_kallsyms_add(struct bpf_prog *fp) 651 { 652 if (!bpf_prog_kallsyms_candidate(fp) || 653 !capable(CAP_SYS_ADMIN)) 654 return; 655 656 spin_lock_bh(&bpf_lock); 657 bpf_prog_ksym_node_add(fp->aux); 658 spin_unlock_bh(&bpf_lock); 659 } 660 661 void bpf_prog_kallsyms_del(struct bpf_prog *fp) 662 { 663 if (!bpf_prog_kallsyms_candidate(fp)) 664 return; 665 666 spin_lock_bh(&bpf_lock); 667 bpf_prog_ksym_node_del(fp->aux); 668 spin_unlock_bh(&bpf_lock); 669 } 670 671 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr) 672 { 673 struct latch_tree_node *n; 674 675 if (!bpf_jit_kallsyms_enabled()) 676 return NULL; 677 678 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops); 679 return n ? 680 container_of(n, struct bpf_prog_aux, ksym_tnode)->prog : 681 NULL; 682 } 683 684 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 685 unsigned long *off, char *sym) 686 { 687 unsigned long symbol_start, symbol_end; 688 struct bpf_prog *prog; 689 char *ret = NULL; 690 691 rcu_read_lock(); 692 prog = bpf_prog_kallsyms_find(addr); 693 if (prog) { 694 bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end); 695 bpf_get_prog_name(prog, sym); 696 697 ret = sym; 698 if (size) 699 *size = symbol_end - symbol_start; 700 if (off) 701 *off = addr - symbol_start; 702 } 703 rcu_read_unlock(); 704 705 return ret; 706 } 707 708 bool is_bpf_text_address(unsigned long addr) 709 { 710 bool ret; 711 712 rcu_read_lock(); 713 ret = bpf_prog_kallsyms_find(addr) != NULL; 714 rcu_read_unlock(); 715 716 return ret; 717 } 718 719 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 720 char *sym) 721 { 722 struct bpf_prog_aux *aux; 723 unsigned int it = 0; 724 int ret = -ERANGE; 725 726 if (!bpf_jit_kallsyms_enabled()) 727 return ret; 728 729 rcu_read_lock(); 730 list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) { 731 if (it++ != symnum) 732 continue; 733 734 bpf_get_prog_name(aux->prog, sym); 735 736 *value = (unsigned long)aux->prog->bpf_func; 737 *type = BPF_SYM_ELF_TYPE; 738 739 ret = 0; 740 break; 741 } 742 rcu_read_unlock(); 743 744 return ret; 745 } 746 747 static atomic_long_t bpf_jit_current; 748 749 /* Can be overridden by an arch's JIT compiler if it has a custom, 750 * dedicated BPF backend memory area, or if neither of the two 751 * below apply. 752 */ 753 u64 __weak bpf_jit_alloc_exec_limit(void) 754 { 755 #if defined(MODULES_VADDR) 756 return MODULES_END - MODULES_VADDR; 757 #else 758 return VMALLOC_END - VMALLOC_START; 759 #endif 760 } 761 762 static int __init bpf_jit_charge_init(void) 763 { 764 /* Only used as heuristic here to derive limit. */ 765 bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2, 766 PAGE_SIZE), LONG_MAX); 767 return 0; 768 } 769 pure_initcall(bpf_jit_charge_init); 770 771 static int bpf_jit_charge_modmem(u32 pages) 772 { 773 if (atomic_long_add_return(pages, &bpf_jit_current) > 774 (bpf_jit_limit >> PAGE_SHIFT)) { 775 if (!capable(CAP_SYS_ADMIN)) { 776 atomic_long_sub(pages, &bpf_jit_current); 777 return -EPERM; 778 } 779 } 780 781 return 0; 782 } 783 784 static void bpf_jit_uncharge_modmem(u32 pages) 785 { 786 atomic_long_sub(pages, &bpf_jit_current); 787 } 788 789 void *__weak bpf_jit_alloc_exec(unsigned long size) 790 { 791 return module_alloc(size); 792 } 793 794 void __weak bpf_jit_free_exec(void *addr) 795 { 796 module_memfree(addr); 797 } 798 799 struct bpf_binary_header * 800 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 801 unsigned int alignment, 802 bpf_jit_fill_hole_t bpf_fill_ill_insns) 803 { 804 struct bpf_binary_header *hdr; 805 u32 size, hole, start, pages; 806 807 /* Most of BPF filters are really small, but if some of them 808 * fill a page, allow at least 128 extra bytes to insert a 809 * random section of illegal instructions. 810 */ 811 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE); 812 pages = size / PAGE_SIZE; 813 814 if (bpf_jit_charge_modmem(pages)) 815 return NULL; 816 hdr = bpf_jit_alloc_exec(size); 817 if (!hdr) { 818 bpf_jit_uncharge_modmem(pages); 819 return NULL; 820 } 821 822 /* Fill space with illegal/arch-dep instructions. */ 823 bpf_fill_ill_insns(hdr, size); 824 825 hdr->pages = pages; 826 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)), 827 PAGE_SIZE - sizeof(*hdr)); 828 start = (get_random_int() % hole) & ~(alignment - 1); 829 830 /* Leave a random number of instructions before BPF code. */ 831 *image_ptr = &hdr->image[start]; 832 833 return hdr; 834 } 835 836 void bpf_jit_binary_free(struct bpf_binary_header *hdr) 837 { 838 u32 pages = hdr->pages; 839 840 bpf_jit_free_exec(hdr); 841 bpf_jit_uncharge_modmem(pages); 842 } 843 844 /* This symbol is only overridden by archs that have different 845 * requirements than the usual eBPF JITs, f.e. when they only 846 * implement cBPF JIT, do not set images read-only, etc. 847 */ 848 void __weak bpf_jit_free(struct bpf_prog *fp) 849 { 850 if (fp->jited) { 851 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp); 852 853 bpf_jit_binary_unlock_ro(hdr); 854 bpf_jit_binary_free(hdr); 855 856 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp)); 857 } 858 859 bpf_prog_unlock_free(fp); 860 } 861 862 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 863 const struct bpf_insn *insn, bool extra_pass, 864 u64 *func_addr, bool *func_addr_fixed) 865 { 866 s16 off = insn->off; 867 s32 imm = insn->imm; 868 u8 *addr; 869 870 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL; 871 if (!*func_addr_fixed) { 872 /* Place-holder address till the last pass has collected 873 * all addresses for JITed subprograms in which case we 874 * can pick them up from prog->aux. 875 */ 876 if (!extra_pass) 877 addr = NULL; 878 else if (prog->aux->func && 879 off >= 0 && off < prog->aux->func_cnt) 880 addr = (u8 *)prog->aux->func[off]->bpf_func; 881 else 882 return -EINVAL; 883 } else { 884 /* Address of a BPF helper call. Since part of the core 885 * kernel, it's always at a fixed location. __bpf_call_base 886 * and the helper with imm relative to it are both in core 887 * kernel. 888 */ 889 addr = (u8 *)__bpf_call_base + imm; 890 } 891 892 *func_addr = (unsigned long)addr; 893 return 0; 894 } 895 896 static int bpf_jit_blind_insn(const struct bpf_insn *from, 897 const struct bpf_insn *aux, 898 struct bpf_insn *to_buff) 899 { 900 struct bpf_insn *to = to_buff; 901 u32 imm_rnd = get_random_int(); 902 s16 off; 903 904 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG); 905 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG); 906 907 /* Constraints on AX register: 908 * 909 * AX register is inaccessible from user space. It is mapped in 910 * all JITs, and used here for constant blinding rewrites. It is 911 * typically "stateless" meaning its contents are only valid within 912 * the executed instruction, but not across several instructions. 913 * There are a few exceptions however which are further detailed 914 * below. 915 * 916 * Constant blinding is only used by JITs, not in the interpreter. 917 * The interpreter uses AX in some occasions as a local temporary 918 * register e.g. in DIV or MOD instructions. 919 * 920 * In restricted circumstances, the verifier can also use the AX 921 * register for rewrites as long as they do not interfere with 922 * the above cases! 923 */ 924 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX) 925 goto out; 926 927 if (from->imm == 0 && 928 (from->code == (BPF_ALU | BPF_MOV | BPF_K) || 929 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) { 930 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg); 931 goto out; 932 } 933 934 switch (from->code) { 935 case BPF_ALU | BPF_ADD | BPF_K: 936 case BPF_ALU | BPF_SUB | BPF_K: 937 case BPF_ALU | BPF_AND | BPF_K: 938 case BPF_ALU | BPF_OR | BPF_K: 939 case BPF_ALU | BPF_XOR | BPF_K: 940 case BPF_ALU | BPF_MUL | BPF_K: 941 case BPF_ALU | BPF_MOV | BPF_K: 942 case BPF_ALU | BPF_DIV | BPF_K: 943 case BPF_ALU | BPF_MOD | BPF_K: 944 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); 945 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 946 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX); 947 break; 948 949 case BPF_ALU64 | BPF_ADD | BPF_K: 950 case BPF_ALU64 | BPF_SUB | BPF_K: 951 case BPF_ALU64 | BPF_AND | BPF_K: 952 case BPF_ALU64 | BPF_OR | BPF_K: 953 case BPF_ALU64 | BPF_XOR | BPF_K: 954 case BPF_ALU64 | BPF_MUL | BPF_K: 955 case BPF_ALU64 | BPF_MOV | BPF_K: 956 case BPF_ALU64 | BPF_DIV | BPF_K: 957 case BPF_ALU64 | BPF_MOD | BPF_K: 958 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); 959 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 960 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX); 961 break; 962 963 case BPF_JMP | BPF_JEQ | BPF_K: 964 case BPF_JMP | BPF_JNE | BPF_K: 965 case BPF_JMP | BPF_JGT | BPF_K: 966 case BPF_JMP | BPF_JLT | BPF_K: 967 case BPF_JMP | BPF_JGE | BPF_K: 968 case BPF_JMP | BPF_JLE | BPF_K: 969 case BPF_JMP | BPF_JSGT | BPF_K: 970 case BPF_JMP | BPF_JSLT | BPF_K: 971 case BPF_JMP | BPF_JSGE | BPF_K: 972 case BPF_JMP | BPF_JSLE | BPF_K: 973 case BPF_JMP | BPF_JSET | BPF_K: 974 /* Accommodate for extra offset in case of a backjump. */ 975 off = from->off; 976 if (off < 0) 977 off -= 2; 978 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); 979 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 980 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off); 981 break; 982 983 case BPF_JMP32 | BPF_JEQ | BPF_K: 984 case BPF_JMP32 | BPF_JNE | BPF_K: 985 case BPF_JMP32 | BPF_JGT | BPF_K: 986 case BPF_JMP32 | BPF_JLT | BPF_K: 987 case BPF_JMP32 | BPF_JGE | BPF_K: 988 case BPF_JMP32 | BPF_JLE | BPF_K: 989 case BPF_JMP32 | BPF_JSGT | BPF_K: 990 case BPF_JMP32 | BPF_JSLT | BPF_K: 991 case BPF_JMP32 | BPF_JSGE | BPF_K: 992 case BPF_JMP32 | BPF_JSLE | BPF_K: 993 case BPF_JMP32 | BPF_JSET | BPF_K: 994 /* Accommodate for extra offset in case of a backjump. */ 995 off = from->off; 996 if (off < 0) 997 off -= 2; 998 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); 999 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 1000 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX, 1001 off); 1002 break; 1003 1004 case BPF_LD | BPF_IMM | BPF_DW: 1005 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm); 1006 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 1007 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); 1008 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX); 1009 break; 1010 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */ 1011 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm); 1012 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 1013 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX); 1014 break; 1015 1016 case BPF_ST | BPF_MEM | BPF_DW: 1017 case BPF_ST | BPF_MEM | BPF_W: 1018 case BPF_ST | BPF_MEM | BPF_H: 1019 case BPF_ST | BPF_MEM | BPF_B: 1020 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); 1021 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); 1022 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off); 1023 break; 1024 } 1025 out: 1026 return to - to_buff; 1027 } 1028 1029 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other, 1030 gfp_t gfp_extra_flags) 1031 { 1032 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; 1033 struct bpf_prog *fp; 1034 1035 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL); 1036 if (fp != NULL) { 1037 /* aux->prog still points to the fp_other one, so 1038 * when promoting the clone to the real program, 1039 * this still needs to be adapted. 1040 */ 1041 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE); 1042 } 1043 1044 return fp; 1045 } 1046 1047 static void bpf_prog_clone_free(struct bpf_prog *fp) 1048 { 1049 /* aux was stolen by the other clone, so we cannot free 1050 * it from this path! It will be freed eventually by the 1051 * other program on release. 1052 * 1053 * At this point, we don't need a deferred release since 1054 * clone is guaranteed to not be locked. 1055 */ 1056 fp->aux = NULL; 1057 __bpf_prog_free(fp); 1058 } 1059 1060 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other) 1061 { 1062 /* We have to repoint aux->prog to self, as we don't 1063 * know whether fp here is the clone or the original. 1064 */ 1065 fp->aux->prog = fp; 1066 bpf_prog_clone_free(fp_other); 1067 } 1068 1069 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog) 1070 { 1071 struct bpf_insn insn_buff[16], aux[2]; 1072 struct bpf_prog *clone, *tmp; 1073 int insn_delta, insn_cnt; 1074 struct bpf_insn *insn; 1075 int i, rewritten; 1076 1077 if (!bpf_jit_blinding_enabled(prog) || prog->blinded) 1078 return prog; 1079 1080 clone = bpf_prog_clone_create(prog, GFP_USER); 1081 if (!clone) 1082 return ERR_PTR(-ENOMEM); 1083 1084 insn_cnt = clone->len; 1085 insn = clone->insnsi; 1086 1087 for (i = 0; i < insn_cnt; i++, insn++) { 1088 /* We temporarily need to hold the original ld64 insn 1089 * so that we can still access the first part in the 1090 * second blinding run. 1091 */ 1092 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) && 1093 insn[1].code == 0) 1094 memcpy(aux, insn, sizeof(aux)); 1095 1096 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff); 1097 if (!rewritten) 1098 continue; 1099 1100 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten); 1101 if (IS_ERR(tmp)) { 1102 /* Patching may have repointed aux->prog during 1103 * realloc from the original one, so we need to 1104 * fix it up here on error. 1105 */ 1106 bpf_jit_prog_release_other(prog, clone); 1107 return tmp; 1108 } 1109 1110 clone = tmp; 1111 insn_delta = rewritten - 1; 1112 1113 /* Walk new program and skip insns we just inserted. */ 1114 insn = clone->insnsi + i + insn_delta; 1115 insn_cnt += insn_delta; 1116 i += insn_delta; 1117 } 1118 1119 clone->blinded = 1; 1120 return clone; 1121 } 1122 #endif /* CONFIG_BPF_JIT */ 1123 1124 /* Base function for offset calculation. Needs to go into .text section, 1125 * therefore keeping it non-static as well; will also be used by JITs 1126 * anyway later on, so do not let the compiler omit it. This also needs 1127 * to go into kallsyms for correlation from e.g. bpftool, so naming 1128 * must not change. 1129 */ 1130 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) 1131 { 1132 return 0; 1133 } 1134 EXPORT_SYMBOL_GPL(__bpf_call_base); 1135 1136 /* All UAPI available opcodes. */ 1137 #define BPF_INSN_MAP(INSN_2, INSN_3) \ 1138 /* 32 bit ALU operations. */ \ 1139 /* Register based. */ \ 1140 INSN_3(ALU, ADD, X), \ 1141 INSN_3(ALU, SUB, X), \ 1142 INSN_3(ALU, AND, X), \ 1143 INSN_3(ALU, OR, X), \ 1144 INSN_3(ALU, LSH, X), \ 1145 INSN_3(ALU, RSH, X), \ 1146 INSN_3(ALU, XOR, X), \ 1147 INSN_3(ALU, MUL, X), \ 1148 INSN_3(ALU, MOV, X), \ 1149 INSN_3(ALU, ARSH, X), \ 1150 INSN_3(ALU, DIV, X), \ 1151 INSN_3(ALU, MOD, X), \ 1152 INSN_2(ALU, NEG), \ 1153 INSN_3(ALU, END, TO_BE), \ 1154 INSN_3(ALU, END, TO_LE), \ 1155 /* Immediate based. */ \ 1156 INSN_3(ALU, ADD, K), \ 1157 INSN_3(ALU, SUB, K), \ 1158 INSN_3(ALU, AND, K), \ 1159 INSN_3(ALU, OR, K), \ 1160 INSN_3(ALU, LSH, K), \ 1161 INSN_3(ALU, RSH, K), \ 1162 INSN_3(ALU, XOR, K), \ 1163 INSN_3(ALU, MUL, K), \ 1164 INSN_3(ALU, MOV, K), \ 1165 INSN_3(ALU, ARSH, K), \ 1166 INSN_3(ALU, DIV, K), \ 1167 INSN_3(ALU, MOD, K), \ 1168 /* 64 bit ALU operations. */ \ 1169 /* Register based. */ \ 1170 INSN_3(ALU64, ADD, X), \ 1171 INSN_3(ALU64, SUB, X), \ 1172 INSN_3(ALU64, AND, X), \ 1173 INSN_3(ALU64, OR, X), \ 1174 INSN_3(ALU64, LSH, X), \ 1175 INSN_3(ALU64, RSH, X), \ 1176 INSN_3(ALU64, XOR, X), \ 1177 INSN_3(ALU64, MUL, X), \ 1178 INSN_3(ALU64, MOV, X), \ 1179 INSN_3(ALU64, ARSH, X), \ 1180 INSN_3(ALU64, DIV, X), \ 1181 INSN_3(ALU64, MOD, X), \ 1182 INSN_2(ALU64, NEG), \ 1183 /* Immediate based. */ \ 1184 INSN_3(ALU64, ADD, K), \ 1185 INSN_3(ALU64, SUB, K), \ 1186 INSN_3(ALU64, AND, K), \ 1187 INSN_3(ALU64, OR, K), \ 1188 INSN_3(ALU64, LSH, K), \ 1189 INSN_3(ALU64, RSH, K), \ 1190 INSN_3(ALU64, XOR, K), \ 1191 INSN_3(ALU64, MUL, K), \ 1192 INSN_3(ALU64, MOV, K), \ 1193 INSN_3(ALU64, ARSH, K), \ 1194 INSN_3(ALU64, DIV, K), \ 1195 INSN_3(ALU64, MOD, K), \ 1196 /* Call instruction. */ \ 1197 INSN_2(JMP, CALL), \ 1198 /* Exit instruction. */ \ 1199 INSN_2(JMP, EXIT), \ 1200 /* 32-bit Jump instructions. */ \ 1201 /* Register based. */ \ 1202 INSN_3(JMP32, JEQ, X), \ 1203 INSN_3(JMP32, JNE, X), \ 1204 INSN_3(JMP32, JGT, X), \ 1205 INSN_3(JMP32, JLT, X), \ 1206 INSN_3(JMP32, JGE, X), \ 1207 INSN_3(JMP32, JLE, X), \ 1208 INSN_3(JMP32, JSGT, X), \ 1209 INSN_3(JMP32, JSLT, X), \ 1210 INSN_3(JMP32, JSGE, X), \ 1211 INSN_3(JMP32, JSLE, X), \ 1212 INSN_3(JMP32, JSET, X), \ 1213 /* Immediate based. */ \ 1214 INSN_3(JMP32, JEQ, K), \ 1215 INSN_3(JMP32, JNE, K), \ 1216 INSN_3(JMP32, JGT, K), \ 1217 INSN_3(JMP32, JLT, K), \ 1218 INSN_3(JMP32, JGE, K), \ 1219 INSN_3(JMP32, JLE, K), \ 1220 INSN_3(JMP32, JSGT, K), \ 1221 INSN_3(JMP32, JSLT, K), \ 1222 INSN_3(JMP32, JSGE, K), \ 1223 INSN_3(JMP32, JSLE, K), \ 1224 INSN_3(JMP32, JSET, K), \ 1225 /* Jump instructions. */ \ 1226 /* Register based. */ \ 1227 INSN_3(JMP, JEQ, X), \ 1228 INSN_3(JMP, JNE, X), \ 1229 INSN_3(JMP, JGT, X), \ 1230 INSN_3(JMP, JLT, X), \ 1231 INSN_3(JMP, JGE, X), \ 1232 INSN_3(JMP, JLE, X), \ 1233 INSN_3(JMP, JSGT, X), \ 1234 INSN_3(JMP, JSLT, X), \ 1235 INSN_3(JMP, JSGE, X), \ 1236 INSN_3(JMP, JSLE, X), \ 1237 INSN_3(JMP, JSET, X), \ 1238 /* Immediate based. */ \ 1239 INSN_3(JMP, JEQ, K), \ 1240 INSN_3(JMP, JNE, K), \ 1241 INSN_3(JMP, JGT, K), \ 1242 INSN_3(JMP, JLT, K), \ 1243 INSN_3(JMP, JGE, K), \ 1244 INSN_3(JMP, JLE, K), \ 1245 INSN_3(JMP, JSGT, K), \ 1246 INSN_3(JMP, JSLT, K), \ 1247 INSN_3(JMP, JSGE, K), \ 1248 INSN_3(JMP, JSLE, K), \ 1249 INSN_3(JMP, JSET, K), \ 1250 INSN_2(JMP, JA), \ 1251 /* Store instructions. */ \ 1252 /* Register based. */ \ 1253 INSN_3(STX, MEM, B), \ 1254 INSN_3(STX, MEM, H), \ 1255 INSN_3(STX, MEM, W), \ 1256 INSN_3(STX, MEM, DW), \ 1257 INSN_3(STX, XADD, W), \ 1258 INSN_3(STX, XADD, DW), \ 1259 /* Immediate based. */ \ 1260 INSN_3(ST, MEM, B), \ 1261 INSN_3(ST, MEM, H), \ 1262 INSN_3(ST, MEM, W), \ 1263 INSN_3(ST, MEM, DW), \ 1264 /* Load instructions. */ \ 1265 /* Register based. */ \ 1266 INSN_3(LDX, MEM, B), \ 1267 INSN_3(LDX, MEM, H), \ 1268 INSN_3(LDX, MEM, W), \ 1269 INSN_3(LDX, MEM, DW), \ 1270 /* Immediate based. */ \ 1271 INSN_3(LD, IMM, DW) 1272 1273 bool bpf_opcode_in_insntable(u8 code) 1274 { 1275 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true 1276 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true 1277 static const bool public_insntable[256] = { 1278 [0 ... 255] = false, 1279 /* Now overwrite non-defaults ... */ 1280 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL), 1281 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */ 1282 [BPF_LD | BPF_ABS | BPF_B] = true, 1283 [BPF_LD | BPF_ABS | BPF_H] = true, 1284 [BPF_LD | BPF_ABS | BPF_W] = true, 1285 [BPF_LD | BPF_IND | BPF_B] = true, 1286 [BPF_LD | BPF_IND | BPF_H] = true, 1287 [BPF_LD | BPF_IND | BPF_W] = true, 1288 }; 1289 #undef BPF_INSN_3_TBL 1290 #undef BPF_INSN_2_TBL 1291 return public_insntable[code]; 1292 } 1293 1294 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 1295 /** 1296 * __bpf_prog_run - run eBPF program on a given context 1297 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers 1298 * @insn: is the array of eBPF instructions 1299 * @stack: is the eBPF storage stack 1300 * 1301 * Decode and execute eBPF instructions. 1302 */ 1303 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack) 1304 { 1305 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y 1306 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z 1307 static const void *jumptable[256] = { 1308 [0 ... 255] = &&default_label, 1309 /* Now overwrite non-defaults ... */ 1310 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL), 1311 /* Non-UAPI available opcodes. */ 1312 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS, 1313 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL, 1314 }; 1315 #undef BPF_INSN_3_LBL 1316 #undef BPF_INSN_2_LBL 1317 u32 tail_call_cnt = 0; 1318 1319 #define CONT ({ insn++; goto select_insn; }) 1320 #define CONT_JMP ({ insn++; goto select_insn; }) 1321 1322 select_insn: 1323 goto *jumptable[insn->code]; 1324 1325 /* ALU */ 1326 #define ALU(OPCODE, OP) \ 1327 ALU64_##OPCODE##_X: \ 1328 DST = DST OP SRC; \ 1329 CONT; \ 1330 ALU_##OPCODE##_X: \ 1331 DST = (u32) DST OP (u32) SRC; \ 1332 CONT; \ 1333 ALU64_##OPCODE##_K: \ 1334 DST = DST OP IMM; \ 1335 CONT; \ 1336 ALU_##OPCODE##_K: \ 1337 DST = (u32) DST OP (u32) IMM; \ 1338 CONT; 1339 1340 ALU(ADD, +) 1341 ALU(SUB, -) 1342 ALU(AND, &) 1343 ALU(OR, |) 1344 ALU(LSH, <<) 1345 ALU(RSH, >>) 1346 ALU(XOR, ^) 1347 ALU(MUL, *) 1348 #undef ALU 1349 ALU_NEG: 1350 DST = (u32) -DST; 1351 CONT; 1352 ALU64_NEG: 1353 DST = -DST; 1354 CONT; 1355 ALU_MOV_X: 1356 DST = (u32) SRC; 1357 CONT; 1358 ALU_MOV_K: 1359 DST = (u32) IMM; 1360 CONT; 1361 ALU64_MOV_X: 1362 DST = SRC; 1363 CONT; 1364 ALU64_MOV_K: 1365 DST = IMM; 1366 CONT; 1367 LD_IMM_DW: 1368 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32; 1369 insn++; 1370 CONT; 1371 ALU_ARSH_X: 1372 DST = (u64) (u32) ((*(s32 *) &DST) >> SRC); 1373 CONT; 1374 ALU_ARSH_K: 1375 DST = (u64) (u32) ((*(s32 *) &DST) >> IMM); 1376 CONT; 1377 ALU64_ARSH_X: 1378 (*(s64 *) &DST) >>= SRC; 1379 CONT; 1380 ALU64_ARSH_K: 1381 (*(s64 *) &DST) >>= IMM; 1382 CONT; 1383 ALU64_MOD_X: 1384 div64_u64_rem(DST, SRC, &AX); 1385 DST = AX; 1386 CONT; 1387 ALU_MOD_X: 1388 AX = (u32) DST; 1389 DST = do_div(AX, (u32) SRC); 1390 CONT; 1391 ALU64_MOD_K: 1392 div64_u64_rem(DST, IMM, &AX); 1393 DST = AX; 1394 CONT; 1395 ALU_MOD_K: 1396 AX = (u32) DST; 1397 DST = do_div(AX, (u32) IMM); 1398 CONT; 1399 ALU64_DIV_X: 1400 DST = div64_u64(DST, SRC); 1401 CONT; 1402 ALU_DIV_X: 1403 AX = (u32) DST; 1404 do_div(AX, (u32) SRC); 1405 DST = (u32) AX; 1406 CONT; 1407 ALU64_DIV_K: 1408 DST = div64_u64(DST, IMM); 1409 CONT; 1410 ALU_DIV_K: 1411 AX = (u32) DST; 1412 do_div(AX, (u32) IMM); 1413 DST = (u32) AX; 1414 CONT; 1415 ALU_END_TO_BE: 1416 switch (IMM) { 1417 case 16: 1418 DST = (__force u16) cpu_to_be16(DST); 1419 break; 1420 case 32: 1421 DST = (__force u32) cpu_to_be32(DST); 1422 break; 1423 case 64: 1424 DST = (__force u64) cpu_to_be64(DST); 1425 break; 1426 } 1427 CONT; 1428 ALU_END_TO_LE: 1429 switch (IMM) { 1430 case 16: 1431 DST = (__force u16) cpu_to_le16(DST); 1432 break; 1433 case 32: 1434 DST = (__force u32) cpu_to_le32(DST); 1435 break; 1436 case 64: 1437 DST = (__force u64) cpu_to_le64(DST); 1438 break; 1439 } 1440 CONT; 1441 1442 /* CALL */ 1443 JMP_CALL: 1444 /* Function call scratches BPF_R1-BPF_R5 registers, 1445 * preserves BPF_R6-BPF_R9, and stores return value 1446 * into BPF_R0. 1447 */ 1448 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3, 1449 BPF_R4, BPF_R5); 1450 CONT; 1451 1452 JMP_CALL_ARGS: 1453 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2, 1454 BPF_R3, BPF_R4, 1455 BPF_R5, 1456 insn + insn->off + 1); 1457 CONT; 1458 1459 JMP_TAIL_CALL: { 1460 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2; 1461 struct bpf_array *array = container_of(map, struct bpf_array, map); 1462 struct bpf_prog *prog; 1463 u32 index = BPF_R3; 1464 1465 if (unlikely(index >= array->map.max_entries)) 1466 goto out; 1467 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) 1468 goto out; 1469 1470 tail_call_cnt++; 1471 1472 prog = READ_ONCE(array->ptrs[index]); 1473 if (!prog) 1474 goto out; 1475 1476 /* ARG1 at this point is guaranteed to point to CTX from 1477 * the verifier side due to the fact that the tail call is 1478 * handeled like a helper, that is, bpf_tail_call_proto, 1479 * where arg1_type is ARG_PTR_TO_CTX. 1480 */ 1481 insn = prog->insnsi; 1482 goto select_insn; 1483 out: 1484 CONT; 1485 } 1486 JMP_JA: 1487 insn += insn->off; 1488 CONT; 1489 JMP_EXIT: 1490 return BPF_R0; 1491 /* JMP */ 1492 #define COND_JMP(SIGN, OPCODE, CMP_OP) \ 1493 JMP_##OPCODE##_X: \ 1494 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \ 1495 insn += insn->off; \ 1496 CONT_JMP; \ 1497 } \ 1498 CONT; \ 1499 JMP32_##OPCODE##_X: \ 1500 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \ 1501 insn += insn->off; \ 1502 CONT_JMP; \ 1503 } \ 1504 CONT; \ 1505 JMP_##OPCODE##_K: \ 1506 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \ 1507 insn += insn->off; \ 1508 CONT_JMP; \ 1509 } \ 1510 CONT; \ 1511 JMP32_##OPCODE##_K: \ 1512 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \ 1513 insn += insn->off; \ 1514 CONT_JMP; \ 1515 } \ 1516 CONT; 1517 COND_JMP(u, JEQ, ==) 1518 COND_JMP(u, JNE, !=) 1519 COND_JMP(u, JGT, >) 1520 COND_JMP(u, JLT, <) 1521 COND_JMP(u, JGE, >=) 1522 COND_JMP(u, JLE, <=) 1523 COND_JMP(u, JSET, &) 1524 COND_JMP(s, JSGT, >) 1525 COND_JMP(s, JSLT, <) 1526 COND_JMP(s, JSGE, >=) 1527 COND_JMP(s, JSLE, <=) 1528 #undef COND_JMP 1529 /* STX and ST and LDX*/ 1530 #define LDST(SIZEOP, SIZE) \ 1531 STX_MEM_##SIZEOP: \ 1532 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \ 1533 CONT; \ 1534 ST_MEM_##SIZEOP: \ 1535 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \ 1536 CONT; \ 1537 LDX_MEM_##SIZEOP: \ 1538 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \ 1539 CONT; 1540 1541 LDST(B, u8) 1542 LDST(H, u16) 1543 LDST(W, u32) 1544 LDST(DW, u64) 1545 #undef LDST 1546 STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */ 1547 atomic_add((u32) SRC, (atomic_t *)(unsigned long) 1548 (DST + insn->off)); 1549 CONT; 1550 STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */ 1551 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long) 1552 (DST + insn->off)); 1553 CONT; 1554 1555 default_label: 1556 /* If we ever reach this, we have a bug somewhere. Die hard here 1557 * instead of just returning 0; we could be somewhere in a subprog, 1558 * so execution could continue otherwise which we do /not/ want. 1559 * 1560 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable(). 1561 */ 1562 pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code); 1563 BUG_ON(1); 1564 return 0; 1565 } 1566 STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */ 1567 1568 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size 1569 #define DEFINE_BPF_PROG_RUN(stack_size) \ 1570 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \ 1571 { \ 1572 u64 stack[stack_size / sizeof(u64)]; \ 1573 u64 regs[MAX_BPF_EXT_REG]; \ 1574 \ 1575 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \ 1576 ARG1 = (u64) (unsigned long) ctx; \ 1577 return ___bpf_prog_run(regs, insn, stack); \ 1578 } 1579 1580 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size 1581 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \ 1582 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \ 1583 const struct bpf_insn *insn) \ 1584 { \ 1585 u64 stack[stack_size / sizeof(u64)]; \ 1586 u64 regs[MAX_BPF_EXT_REG]; \ 1587 \ 1588 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \ 1589 BPF_R1 = r1; \ 1590 BPF_R2 = r2; \ 1591 BPF_R3 = r3; \ 1592 BPF_R4 = r4; \ 1593 BPF_R5 = r5; \ 1594 return ___bpf_prog_run(regs, insn, stack); \ 1595 } 1596 1597 #define EVAL1(FN, X) FN(X) 1598 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y) 1599 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y) 1600 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y) 1601 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y) 1602 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y) 1603 1604 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192); 1605 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384); 1606 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512); 1607 1608 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192); 1609 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384); 1610 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512); 1611 1612 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size), 1613 1614 static unsigned int (*interpreters[])(const void *ctx, 1615 const struct bpf_insn *insn) = { 1616 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192) 1617 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384) 1618 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512) 1619 }; 1620 #undef PROG_NAME_LIST 1621 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size), 1622 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, 1623 const struct bpf_insn *insn) = { 1624 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192) 1625 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384) 1626 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512) 1627 }; 1628 #undef PROG_NAME_LIST 1629 1630 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth) 1631 { 1632 stack_depth = max_t(u32, stack_depth, 1); 1633 insn->off = (s16) insn->imm; 1634 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] - 1635 __bpf_call_base_args; 1636 insn->code = BPF_JMP | BPF_CALL_ARGS; 1637 } 1638 1639 #else 1640 static unsigned int __bpf_prog_ret0_warn(const void *ctx, 1641 const struct bpf_insn *insn) 1642 { 1643 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON 1644 * is not working properly, so warn about it! 1645 */ 1646 WARN_ON_ONCE(1); 1647 return 0; 1648 } 1649 #endif 1650 1651 bool bpf_prog_array_compatible(struct bpf_array *array, 1652 const struct bpf_prog *fp) 1653 { 1654 if (fp->kprobe_override) 1655 return false; 1656 1657 if (!array->owner_prog_type) { 1658 /* There's no owner yet where we could check for 1659 * compatibility. 1660 */ 1661 array->owner_prog_type = fp->type; 1662 array->owner_jited = fp->jited; 1663 1664 return true; 1665 } 1666 1667 return array->owner_prog_type == fp->type && 1668 array->owner_jited == fp->jited; 1669 } 1670 1671 static int bpf_check_tail_call(const struct bpf_prog *fp) 1672 { 1673 struct bpf_prog_aux *aux = fp->aux; 1674 int i; 1675 1676 for (i = 0; i < aux->used_map_cnt; i++) { 1677 struct bpf_map *map = aux->used_maps[i]; 1678 struct bpf_array *array; 1679 1680 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) 1681 continue; 1682 1683 array = container_of(map, struct bpf_array, map); 1684 if (!bpf_prog_array_compatible(array, fp)) 1685 return -EINVAL; 1686 } 1687 1688 return 0; 1689 } 1690 1691 static void bpf_prog_select_func(struct bpf_prog *fp) 1692 { 1693 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 1694 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1); 1695 1696 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1]; 1697 #else 1698 fp->bpf_func = __bpf_prog_ret0_warn; 1699 #endif 1700 } 1701 1702 /** 1703 * bpf_prog_select_runtime - select exec runtime for BPF program 1704 * @fp: bpf_prog populated with internal BPF program 1705 * @err: pointer to error variable 1706 * 1707 * Try to JIT eBPF program, if JIT is not available, use interpreter. 1708 * The BPF program will be executed via BPF_PROG_RUN() macro. 1709 */ 1710 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err) 1711 { 1712 /* In case of BPF to BPF calls, verifier did all the prep 1713 * work with regards to JITing, etc. 1714 */ 1715 if (fp->bpf_func) 1716 goto finalize; 1717 1718 bpf_prog_select_func(fp); 1719 1720 /* eBPF JITs can rewrite the program in case constant 1721 * blinding is active. However, in case of error during 1722 * blinding, bpf_int_jit_compile() must always return a 1723 * valid program, which in this case would simply not 1724 * be JITed, but falls back to the interpreter. 1725 */ 1726 if (!bpf_prog_is_dev_bound(fp->aux)) { 1727 *err = bpf_prog_alloc_jited_linfo(fp); 1728 if (*err) 1729 return fp; 1730 1731 fp = bpf_int_jit_compile(fp); 1732 if (!fp->jited) { 1733 bpf_prog_free_jited_linfo(fp); 1734 #ifdef CONFIG_BPF_JIT_ALWAYS_ON 1735 *err = -ENOTSUPP; 1736 return fp; 1737 #endif 1738 } else { 1739 bpf_prog_free_unused_jited_linfo(fp); 1740 } 1741 } else { 1742 *err = bpf_prog_offload_compile(fp); 1743 if (*err) 1744 return fp; 1745 } 1746 1747 finalize: 1748 bpf_prog_lock_ro(fp); 1749 1750 /* The tail call compatibility check can only be done at 1751 * this late stage as we need to determine, if we deal 1752 * with JITed or non JITed program concatenations and not 1753 * all eBPF JITs might immediately support all features. 1754 */ 1755 *err = bpf_check_tail_call(fp); 1756 1757 return fp; 1758 } 1759 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime); 1760 1761 static unsigned int __bpf_prog_ret1(const void *ctx, 1762 const struct bpf_insn *insn) 1763 { 1764 return 1; 1765 } 1766 1767 static struct bpf_prog_dummy { 1768 struct bpf_prog prog; 1769 } dummy_bpf_prog = { 1770 .prog = { 1771 .bpf_func = __bpf_prog_ret1, 1772 }, 1773 }; 1774 1775 /* to avoid allocating empty bpf_prog_array for cgroups that 1776 * don't have bpf program attached use one global 'empty_prog_array' 1777 * It will not be modified the caller of bpf_prog_array_alloc() 1778 * (since caller requested prog_cnt == 0) 1779 * that pointer should be 'freed' by bpf_prog_array_free() 1780 */ 1781 static struct { 1782 struct bpf_prog_array hdr; 1783 struct bpf_prog *null_prog; 1784 } empty_prog_array = { 1785 .null_prog = NULL, 1786 }; 1787 1788 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags) 1789 { 1790 if (prog_cnt) 1791 return kzalloc(sizeof(struct bpf_prog_array) + 1792 sizeof(struct bpf_prog_array_item) * 1793 (prog_cnt + 1), 1794 flags); 1795 1796 return &empty_prog_array.hdr; 1797 } 1798 1799 void bpf_prog_array_free(struct bpf_prog_array __rcu *progs) 1800 { 1801 if (!progs || 1802 progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr) 1803 return; 1804 kfree_rcu(progs, rcu); 1805 } 1806 1807 int bpf_prog_array_length(struct bpf_prog_array __rcu *array) 1808 { 1809 struct bpf_prog_array_item *item; 1810 u32 cnt = 0; 1811 1812 rcu_read_lock(); 1813 item = rcu_dereference(array)->items; 1814 for (; item->prog; item++) 1815 if (item->prog != &dummy_bpf_prog.prog) 1816 cnt++; 1817 rcu_read_unlock(); 1818 return cnt; 1819 } 1820 1821 1822 static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array, 1823 u32 *prog_ids, 1824 u32 request_cnt) 1825 { 1826 struct bpf_prog_array_item *item; 1827 int i = 0; 1828 1829 item = rcu_dereference_check(array, 1)->items; 1830 for (; item->prog; item++) { 1831 if (item->prog == &dummy_bpf_prog.prog) 1832 continue; 1833 prog_ids[i] = item->prog->aux->id; 1834 if (++i == request_cnt) { 1835 item++; 1836 break; 1837 } 1838 } 1839 1840 return !!(item->prog); 1841 } 1842 1843 int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array, 1844 __u32 __user *prog_ids, u32 cnt) 1845 { 1846 unsigned long err = 0; 1847 bool nospc; 1848 u32 *ids; 1849 1850 /* users of this function are doing: 1851 * cnt = bpf_prog_array_length(); 1852 * if (cnt > 0) 1853 * bpf_prog_array_copy_to_user(..., cnt); 1854 * so below kcalloc doesn't need extra cnt > 0 check, but 1855 * bpf_prog_array_length() releases rcu lock and 1856 * prog array could have been swapped with empty or larger array, 1857 * so always copy 'cnt' prog_ids to the user. 1858 * In a rare race the user will see zero prog_ids 1859 */ 1860 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN); 1861 if (!ids) 1862 return -ENOMEM; 1863 rcu_read_lock(); 1864 nospc = bpf_prog_array_copy_core(array, ids, cnt); 1865 rcu_read_unlock(); 1866 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32)); 1867 kfree(ids); 1868 if (err) 1869 return -EFAULT; 1870 if (nospc) 1871 return -ENOSPC; 1872 return 0; 1873 } 1874 1875 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array, 1876 struct bpf_prog *old_prog) 1877 { 1878 struct bpf_prog_array_item *item = array->items; 1879 1880 for (; item->prog; item++) 1881 if (item->prog == old_prog) { 1882 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog); 1883 break; 1884 } 1885 } 1886 1887 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array, 1888 struct bpf_prog *exclude_prog, 1889 struct bpf_prog *include_prog, 1890 struct bpf_prog_array **new_array) 1891 { 1892 int new_prog_cnt, carry_prog_cnt = 0; 1893 struct bpf_prog_array_item *existing; 1894 struct bpf_prog_array *array; 1895 bool found_exclude = false; 1896 int new_prog_idx = 0; 1897 1898 /* Figure out how many existing progs we need to carry over to 1899 * the new array. 1900 */ 1901 if (old_array) { 1902 existing = old_array->items; 1903 for (; existing->prog; existing++) { 1904 if (existing->prog == exclude_prog) { 1905 found_exclude = true; 1906 continue; 1907 } 1908 if (existing->prog != &dummy_bpf_prog.prog) 1909 carry_prog_cnt++; 1910 if (existing->prog == include_prog) 1911 return -EEXIST; 1912 } 1913 } 1914 1915 if (exclude_prog && !found_exclude) 1916 return -ENOENT; 1917 1918 /* How many progs (not NULL) will be in the new array? */ 1919 new_prog_cnt = carry_prog_cnt; 1920 if (include_prog) 1921 new_prog_cnt += 1; 1922 1923 /* Do we have any prog (not NULL) in the new array? */ 1924 if (!new_prog_cnt) { 1925 *new_array = NULL; 1926 return 0; 1927 } 1928 1929 /* +1 as the end of prog_array is marked with NULL */ 1930 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL); 1931 if (!array) 1932 return -ENOMEM; 1933 1934 /* Fill in the new prog array */ 1935 if (carry_prog_cnt) { 1936 existing = old_array->items; 1937 for (; existing->prog; existing++) 1938 if (existing->prog != exclude_prog && 1939 existing->prog != &dummy_bpf_prog.prog) { 1940 array->items[new_prog_idx++].prog = 1941 existing->prog; 1942 } 1943 } 1944 if (include_prog) 1945 array->items[new_prog_idx++].prog = include_prog; 1946 array->items[new_prog_idx].prog = NULL; 1947 *new_array = array; 1948 return 0; 1949 } 1950 1951 int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array, 1952 u32 *prog_ids, u32 request_cnt, 1953 u32 *prog_cnt) 1954 { 1955 u32 cnt = 0; 1956 1957 if (array) 1958 cnt = bpf_prog_array_length(array); 1959 1960 *prog_cnt = cnt; 1961 1962 /* return early if user requested only program count or nothing to copy */ 1963 if (!request_cnt || !cnt) 1964 return 0; 1965 1966 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */ 1967 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC 1968 : 0; 1969 } 1970 1971 static void bpf_prog_free_deferred(struct work_struct *work) 1972 { 1973 struct bpf_prog_aux *aux; 1974 int i; 1975 1976 aux = container_of(work, struct bpf_prog_aux, work); 1977 if (bpf_prog_is_dev_bound(aux)) 1978 bpf_prog_offload_destroy(aux->prog); 1979 #ifdef CONFIG_PERF_EVENTS 1980 if (aux->prog->has_callchain_buf) 1981 put_callchain_buffers(); 1982 #endif 1983 for (i = 0; i < aux->func_cnt; i++) 1984 bpf_jit_free(aux->func[i]); 1985 if (aux->func_cnt) { 1986 kfree(aux->func); 1987 bpf_prog_unlock_free(aux->prog); 1988 } else { 1989 bpf_jit_free(aux->prog); 1990 } 1991 } 1992 1993 /* Free internal BPF program */ 1994 void bpf_prog_free(struct bpf_prog *fp) 1995 { 1996 struct bpf_prog_aux *aux = fp->aux; 1997 1998 INIT_WORK(&aux->work, bpf_prog_free_deferred); 1999 schedule_work(&aux->work); 2000 } 2001 EXPORT_SYMBOL_GPL(bpf_prog_free); 2002 2003 /* RNG for unpriviledged user space with separated state from prandom_u32(). */ 2004 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state); 2005 2006 void bpf_user_rnd_init_once(void) 2007 { 2008 prandom_init_once(&bpf_user_rnd_state); 2009 } 2010 2011 BPF_CALL_0(bpf_user_rnd_u32) 2012 { 2013 /* Should someone ever have the rather unwise idea to use some 2014 * of the registers passed into this function, then note that 2015 * this function is called from native eBPF and classic-to-eBPF 2016 * transformations. Register assignments from both sides are 2017 * different, f.e. classic always sets fn(ctx, A, X) here. 2018 */ 2019 struct rnd_state *state; 2020 u32 res; 2021 2022 state = &get_cpu_var(bpf_user_rnd_state); 2023 res = prandom_u32_state(state); 2024 put_cpu_var(bpf_user_rnd_state); 2025 2026 return res; 2027 } 2028 2029 /* Weak definitions of helper functions in case we don't have bpf syscall. */ 2030 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak; 2031 const struct bpf_func_proto bpf_map_update_elem_proto __weak; 2032 const struct bpf_func_proto bpf_map_delete_elem_proto __weak; 2033 const struct bpf_func_proto bpf_map_push_elem_proto __weak; 2034 const struct bpf_func_proto bpf_map_pop_elem_proto __weak; 2035 const struct bpf_func_proto bpf_map_peek_elem_proto __weak; 2036 const struct bpf_func_proto bpf_spin_lock_proto __weak; 2037 const struct bpf_func_proto bpf_spin_unlock_proto __weak; 2038 2039 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak; 2040 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak; 2041 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak; 2042 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak; 2043 2044 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak; 2045 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak; 2046 const struct bpf_func_proto bpf_get_current_comm_proto __weak; 2047 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak; 2048 const struct bpf_func_proto bpf_get_local_storage_proto __weak; 2049 2050 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void) 2051 { 2052 return NULL; 2053 } 2054 2055 u64 __weak 2056 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 2057 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) 2058 { 2059 return -ENOTSUPP; 2060 } 2061 EXPORT_SYMBOL_GPL(bpf_event_output); 2062 2063 /* Always built-in helper functions. */ 2064 const struct bpf_func_proto bpf_tail_call_proto = { 2065 .func = NULL, 2066 .gpl_only = false, 2067 .ret_type = RET_VOID, 2068 .arg1_type = ARG_PTR_TO_CTX, 2069 .arg2_type = ARG_CONST_MAP_PTR, 2070 .arg3_type = ARG_ANYTHING, 2071 }; 2072 2073 /* Stub for JITs that only support cBPF. eBPF programs are interpreted. 2074 * It is encouraged to implement bpf_int_jit_compile() instead, so that 2075 * eBPF and implicitly also cBPF can get JITed! 2076 */ 2077 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog) 2078 { 2079 return prog; 2080 } 2081 2082 /* Stub for JITs that support eBPF. All cBPF code gets transformed into 2083 * eBPF by the kernel and is later compiled by bpf_int_jit_compile(). 2084 */ 2085 void __weak bpf_jit_compile(struct bpf_prog *prog) 2086 { 2087 } 2088 2089 bool __weak bpf_helper_changes_pkt_data(void *func) 2090 { 2091 return false; 2092 } 2093 2094 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call 2095 * skb_copy_bits(), so provide a weak definition of it for NET-less config. 2096 */ 2097 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to, 2098 int len) 2099 { 2100 return -EFAULT; 2101 } 2102 2103 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 2104 EXPORT_SYMBOL(bpf_stats_enabled_key); 2105 int sysctl_bpf_stats_enabled __read_mostly; 2106 2107 /* All definitions of tracepoints related to BPF. */ 2108 #define CREATE_TRACE_POINTS 2109 #include <linux/bpf_trace.h> 2110 2111 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception); 2112