1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 3 */ 4 #ifndef _LINUX_BPF_VERIFIER_H 5 #define _LINUX_BPF_VERIFIER_H 1 6 7 #include <linux/bpf.h> /* for enum bpf_reg_type */ 8 #include <linux/btf.h> /* for struct btf and btf_id() */ 9 #include <linux/filter.h> /* for MAX_BPF_STACK */ 10 #include <linux/tnum.h> 11 12 /* Maximum variable offset umax_value permitted when resolving memory accesses. 13 * In practice this is far bigger than any realistic pointer offset; this limit 14 * ensures that umax_value + (int)off + (int)size cannot overflow a u64. 15 */ 16 #define BPF_MAX_VAR_OFF (1 << 29) 17 /* Maximum variable size permitted for ARG_CONST_SIZE[_OR_ZERO]. This ensures 18 * that converting umax_value to int cannot overflow. 19 */ 20 #define BPF_MAX_VAR_SIZ (1 << 29) 21 /* size of type_str_buf in bpf_verifier. */ 22 #define TYPE_STR_BUF_LEN 64 23 24 /* Liveness marks, used for registers and spilled-regs (in stack slots). 25 * Read marks propagate upwards until they find a write mark; they record that 26 * "one of this state's descendants read this reg" (and therefore the reg is 27 * relevant for states_equal() checks). 28 * Write marks collect downwards and do not propagate; they record that "the 29 * straight-line code that reached this state (from its parent) wrote this reg" 30 * (and therefore that reads propagated from this state or its descendants 31 * should not propagate to its parent). 32 * A state with a write mark can receive read marks; it just won't propagate 33 * them to its parent, since the write mark is a property, not of the state, 34 * but of the link between it and its parent. See mark_reg_read() and 35 * mark_stack_slot_read() in kernel/bpf/verifier.c. 36 */ 37 enum bpf_reg_liveness { 38 REG_LIVE_NONE = 0, /* reg hasn't been read or written this branch */ 39 REG_LIVE_READ32 = 0x1, /* reg was read, so we're sensitive to initial value */ 40 REG_LIVE_READ64 = 0x2, /* likewise, but full 64-bit content matters */ 41 REG_LIVE_READ = REG_LIVE_READ32 | REG_LIVE_READ64, 42 REG_LIVE_WRITTEN = 0x4, /* reg was written first, screening off later reads */ 43 REG_LIVE_DONE = 0x8, /* liveness won't be updating this register anymore */ 44 }; 45 46 struct bpf_reg_state { 47 /* Ordering of fields matters. See states_equal() */ 48 enum bpf_reg_type type; 49 /* Fixed part of pointer offset, pointer types only */ 50 s32 off; 51 union { 52 /* valid when type == PTR_TO_PACKET */ 53 int range; 54 55 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE | 56 * PTR_TO_MAP_VALUE_OR_NULL 57 */ 58 struct { 59 struct bpf_map *map_ptr; 60 /* To distinguish map lookups from outer map 61 * the map_uid is non-zero for registers 62 * pointing to inner maps. 63 */ 64 u32 map_uid; 65 }; 66 67 /* for PTR_TO_BTF_ID */ 68 struct { 69 struct btf *btf; 70 u32 btf_id; 71 }; 72 73 u32 mem_size; /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */ 74 75 /* Max size from any of the above. */ 76 struct { 77 unsigned long raw1; 78 unsigned long raw2; 79 } raw; 80 81 u32 subprogno; /* for PTR_TO_FUNC */ 82 }; 83 /* For PTR_TO_PACKET, used to find other pointers with the same variable 84 * offset, so they can share range knowledge. 85 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we 86 * came from, when one is tested for != NULL. 87 * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation 88 * for the purpose of tracking that it's freed. 89 * For PTR_TO_SOCKET this is used to share which pointers retain the 90 * same reference to the socket, to determine proper reference freeing. 91 */ 92 u32 id; 93 /* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned 94 * from a pointer-cast helper, bpf_sk_fullsock() and 95 * bpf_tcp_sock(). 96 * 97 * Consider the following where "sk" is a reference counted 98 * pointer returned from "sk = bpf_sk_lookup_tcp();": 99 * 100 * 1: sk = bpf_sk_lookup_tcp(); 101 * 2: if (!sk) { return 0; } 102 * 3: fullsock = bpf_sk_fullsock(sk); 103 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; } 104 * 5: tp = bpf_tcp_sock(fullsock); 105 * 6: if (!tp) { bpf_sk_release(sk); return 0; } 106 * 7: bpf_sk_release(sk); 107 * 8: snd_cwnd = tp->snd_cwnd; // verifier will complain 108 * 109 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and 110 * "tp" ptr should be invalidated also. In order to do that, 111 * the reg holding "fullsock" and "sk" need to remember 112 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id 113 * such that the verifier can reset all regs which have 114 * ref_obj_id matching the sk_reg->id. 115 * 116 * sk_reg->ref_obj_id is set to sk_reg->id at line 1. 117 * sk_reg->id will stay as NULL-marking purpose only. 118 * After NULL-marking is done, sk_reg->id can be reset to 0. 119 * 120 * After "fullsock = bpf_sk_fullsock(sk);" at line 3, 121 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id. 122 * 123 * After "tp = bpf_tcp_sock(fullsock);" at line 5, 124 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id 125 * which is the same as sk_reg->ref_obj_id. 126 * 127 * From the verifier perspective, if sk, fullsock and tp 128 * are not NULL, they are the same ptr with different 129 * reg->type. In particular, bpf_sk_release(tp) is also 130 * allowed and has the same effect as bpf_sk_release(sk). 131 */ 132 u32 ref_obj_id; 133 /* For scalar types (SCALAR_VALUE), this represents our knowledge of 134 * the actual value. 135 * For pointer types, this represents the variable part of the offset 136 * from the pointed-to object, and is shared with all bpf_reg_states 137 * with the same id as us. 138 */ 139 struct tnum var_off; 140 /* Used to determine if any memory access using this register will 141 * result in a bad access. 142 * These refer to the same value as var_off, not necessarily the actual 143 * contents of the register. 144 */ 145 s64 smin_value; /* minimum possible (s64)value */ 146 s64 smax_value; /* maximum possible (s64)value */ 147 u64 umin_value; /* minimum possible (u64)value */ 148 u64 umax_value; /* maximum possible (u64)value */ 149 s32 s32_min_value; /* minimum possible (s32)value */ 150 s32 s32_max_value; /* maximum possible (s32)value */ 151 u32 u32_min_value; /* minimum possible (u32)value */ 152 u32 u32_max_value; /* maximum possible (u32)value */ 153 /* parentage chain for liveness checking */ 154 struct bpf_reg_state *parent; 155 /* Inside the callee two registers can be both PTR_TO_STACK like 156 * R1=fp-8 and R2=fp-8, but one of them points to this function stack 157 * while another to the caller's stack. To differentiate them 'frameno' 158 * is used which is an index in bpf_verifier_state->frame[] array 159 * pointing to bpf_func_state. 160 */ 161 u32 frameno; 162 /* Tracks subreg definition. The stored value is the insn_idx of the 163 * writing insn. This is safe because subreg_def is used before any insn 164 * patching which only happens after main verification finished. 165 */ 166 s32 subreg_def; 167 enum bpf_reg_liveness live; 168 /* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */ 169 bool precise; 170 }; 171 172 enum bpf_stack_slot_type { 173 STACK_INVALID, /* nothing was stored in this stack slot */ 174 STACK_SPILL, /* register spilled into stack */ 175 STACK_MISC, /* BPF program wrote some data into this slot */ 176 STACK_ZERO, /* BPF program wrote constant zero */ 177 }; 178 179 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */ 180 181 struct bpf_stack_state { 182 struct bpf_reg_state spilled_ptr; 183 u8 slot_type[BPF_REG_SIZE]; 184 }; 185 186 struct bpf_reference_state { 187 /* Track each reference created with a unique id, even if the same 188 * instruction creates the reference multiple times (eg, via CALL). 189 */ 190 int id; 191 /* Instruction where the allocation of this reference occurred. This 192 * is used purely to inform the user of a reference leak. 193 */ 194 int insn_idx; 195 }; 196 197 /* state of the program: 198 * type of all registers and stack info 199 */ 200 struct bpf_func_state { 201 struct bpf_reg_state regs[MAX_BPF_REG]; 202 /* index of call instruction that called into this func */ 203 int callsite; 204 /* stack frame number of this function state from pov of 205 * enclosing bpf_verifier_state. 206 * 0 = main function, 1 = first callee. 207 */ 208 u32 frameno; 209 /* subprog number == index within subprog_info 210 * zero == main subprog 211 */ 212 u32 subprogno; 213 /* Every bpf_timer_start will increment async_entry_cnt. 214 * It's used to distinguish: 215 * void foo(void) { for(;;); } 216 * void foo(void) { bpf_timer_set_callback(,foo); } 217 */ 218 u32 async_entry_cnt; 219 bool in_callback_fn; 220 bool in_async_callback_fn; 221 222 /* The following fields should be last. See copy_func_state() */ 223 int acquired_refs; 224 struct bpf_reference_state *refs; 225 int allocated_stack; 226 struct bpf_stack_state *stack; 227 }; 228 229 struct bpf_idx_pair { 230 u32 prev_idx; 231 u32 idx; 232 }; 233 234 struct bpf_id_pair { 235 u32 old; 236 u32 cur; 237 }; 238 239 /* Maximum number of register states that can exist at once */ 240 #define BPF_ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) 241 #define MAX_CALL_FRAMES 8 242 struct bpf_verifier_state { 243 /* call stack tracking */ 244 struct bpf_func_state *frame[MAX_CALL_FRAMES]; 245 struct bpf_verifier_state *parent; 246 /* 247 * 'branches' field is the number of branches left to explore: 248 * 0 - all possible paths from this state reached bpf_exit or 249 * were safely pruned 250 * 1 - at least one path is being explored. 251 * This state hasn't reached bpf_exit 252 * 2 - at least two paths are being explored. 253 * This state is an immediate parent of two children. 254 * One is fallthrough branch with branches==1 and another 255 * state is pushed into stack (to be explored later) also with 256 * branches==1. The parent of this state has branches==1. 257 * The verifier state tree connected via 'parent' pointer looks like: 258 * 1 259 * 1 260 * 2 -> 1 (first 'if' pushed into stack) 261 * 1 262 * 2 -> 1 (second 'if' pushed into stack) 263 * 1 264 * 1 265 * 1 bpf_exit. 266 * 267 * Once do_check() reaches bpf_exit, it calls update_branch_counts() 268 * and the verifier state tree will look: 269 * 1 270 * 1 271 * 2 -> 1 (first 'if' pushed into stack) 272 * 1 273 * 1 -> 1 (second 'if' pushed into stack) 274 * 0 275 * 0 276 * 0 bpf_exit. 277 * After pop_stack() the do_check() will resume at second 'if'. 278 * 279 * If is_state_visited() sees a state with branches > 0 it means 280 * there is a loop. If such state is exactly equal to the current state 281 * it's an infinite loop. Note states_equal() checks for states 282 * equvalency, so two states being 'states_equal' does not mean 283 * infinite loop. The exact comparison is provided by 284 * states_maybe_looping() function. It's a stronger pre-check and 285 * much faster than states_equal(). 286 * 287 * This algorithm may not find all possible infinite loops or 288 * loop iteration count may be too high. 289 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in. 290 */ 291 u32 branches; 292 u32 insn_idx; 293 u32 curframe; 294 u32 active_spin_lock; 295 bool speculative; 296 297 /* first and last insn idx of this verifier state */ 298 u32 first_insn_idx; 299 u32 last_insn_idx; 300 /* jmp history recorded from first to last. 301 * backtracking is using it to go from last to first. 302 * For most states jmp_history_cnt is [0-3]. 303 * For loops can go up to ~40. 304 */ 305 struct bpf_idx_pair *jmp_history; 306 u32 jmp_history_cnt; 307 }; 308 309 #define bpf_get_spilled_reg(slot, frame) \ 310 (((slot < frame->allocated_stack / BPF_REG_SIZE) && \ 311 (frame->stack[slot].slot_type[0] == STACK_SPILL)) \ 312 ? &frame->stack[slot].spilled_ptr : NULL) 313 314 /* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */ 315 #define bpf_for_each_spilled_reg(iter, frame, reg) \ 316 for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \ 317 iter < frame->allocated_stack / BPF_REG_SIZE; \ 318 iter++, reg = bpf_get_spilled_reg(iter, frame)) 319 320 /* linked list of verifier states used to prune search */ 321 struct bpf_verifier_state_list { 322 struct bpf_verifier_state state; 323 struct bpf_verifier_state_list *next; 324 int miss_cnt, hit_cnt; 325 }; 326 327 /* Possible states for alu_state member. */ 328 #define BPF_ALU_SANITIZE_SRC (1U << 0) 329 #define BPF_ALU_SANITIZE_DST (1U << 1) 330 #define BPF_ALU_NEG_VALUE (1U << 2) 331 #define BPF_ALU_NON_POINTER (1U << 3) 332 #define BPF_ALU_IMMEDIATE (1U << 4) 333 #define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \ 334 BPF_ALU_SANITIZE_DST) 335 336 struct bpf_insn_aux_data { 337 union { 338 enum bpf_reg_type ptr_type; /* pointer type for load/store insns */ 339 unsigned long map_ptr_state; /* pointer/poison value for maps */ 340 s32 call_imm; /* saved imm field of call insn */ 341 u32 alu_limit; /* limit for add/sub register with pointer */ 342 struct { 343 u32 map_index; /* index into used_maps[] */ 344 u32 map_off; /* offset from value base address */ 345 }; 346 struct { 347 enum bpf_reg_type reg_type; /* type of pseudo_btf_id */ 348 union { 349 struct { 350 struct btf *btf; 351 u32 btf_id; /* btf_id for struct typed var */ 352 }; 353 u32 mem_size; /* mem_size for non-struct typed var */ 354 }; 355 } btf_var; 356 }; 357 u64 map_key_state; /* constant (32 bit) key tracking for maps */ 358 int ctx_field_size; /* the ctx field size for load insn, maybe 0 */ 359 u32 seen; /* this insn was processed by the verifier at env->pass_cnt */ 360 bool sanitize_stack_spill; /* subject to Spectre v4 sanitation */ 361 bool zext_dst; /* this insn zero extends dst reg */ 362 u8 alu_state; /* used in combination with alu_limit */ 363 364 /* below fields are initialized once */ 365 unsigned int orig_idx; /* original instruction index */ 366 bool prune_point; 367 }; 368 369 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */ 370 #define MAX_USED_BTFS 64 /* max number of BTFs accessed by one BPF program */ 371 372 #define BPF_VERIFIER_TMP_LOG_SIZE 1024 373 374 struct bpf_verifier_log { 375 u32 level; 376 char kbuf[BPF_VERIFIER_TMP_LOG_SIZE]; 377 char __user *ubuf; 378 u32 len_used; 379 u32 len_total; 380 }; 381 382 static inline bool bpf_verifier_log_full(const struct bpf_verifier_log *log) 383 { 384 return log->len_used >= log->len_total - 1; 385 } 386 387 #define BPF_LOG_LEVEL1 1 388 #define BPF_LOG_LEVEL2 2 389 #define BPF_LOG_STATS 4 390 #define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2) 391 #define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS) 392 #define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */ 393 #define BPF_LOG_MIN_ALIGNMENT 8U 394 #define BPF_LOG_ALIGNMENT 40U 395 396 static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log) 397 { 398 return log && 399 ((log->level && log->ubuf && !bpf_verifier_log_full(log)) || 400 log->level == BPF_LOG_KERNEL); 401 } 402 403 static inline bool 404 bpf_verifier_log_attr_valid(const struct bpf_verifier_log *log) 405 { 406 return log->len_total >= 128 && log->len_total <= UINT_MAX >> 2 && 407 log->level && log->ubuf && !(log->level & ~BPF_LOG_MASK); 408 } 409 410 #define BPF_MAX_SUBPROGS 256 411 412 struct bpf_subprog_info { 413 /* 'start' has to be the first field otherwise find_subprog() won't work */ 414 u32 start; /* insn idx of function entry point */ 415 u32 linfo_idx; /* The idx to the main_prog->aux->linfo */ 416 u16 stack_depth; /* max. stack depth used by this function */ 417 bool has_tail_call; 418 bool tail_call_reachable; 419 bool has_ld_abs; 420 bool is_async_cb; 421 }; 422 423 /* single container for all structs 424 * one verifier_env per bpf_check() call 425 */ 426 struct bpf_verifier_env { 427 u32 insn_idx; 428 u32 prev_insn_idx; 429 struct bpf_prog *prog; /* eBPF program being verified */ 430 const struct bpf_verifier_ops *ops; 431 struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */ 432 int stack_size; /* number of states to be processed */ 433 bool strict_alignment; /* perform strict pointer alignment checks */ 434 bool test_state_freq; /* test verifier with different pruning frequency */ 435 struct bpf_verifier_state *cur_state; /* current verifier state */ 436 struct bpf_verifier_state_list **explored_states; /* search pruning optimization */ 437 struct bpf_verifier_state_list *free_list; 438 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */ 439 struct btf_mod_pair used_btfs[MAX_USED_BTFS]; /* array of BTF's used by BPF program */ 440 u32 used_map_cnt; /* number of used maps */ 441 u32 used_btf_cnt; /* number of used BTF objects */ 442 u32 id_gen; /* used to generate unique reg IDs */ 443 bool explore_alu_limits; 444 bool allow_ptr_leaks; 445 bool allow_uninit_stack; 446 bool allow_ptr_to_map_access; 447 bool bpf_capable; 448 bool bypass_spec_v1; 449 bool bypass_spec_v4; 450 bool seen_direct_write; 451 struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */ 452 const struct bpf_line_info *prev_linfo; 453 struct bpf_verifier_log log; 454 struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 1]; 455 struct bpf_id_pair idmap_scratch[BPF_ID_MAP_SIZE]; 456 struct { 457 int *insn_state; 458 int *insn_stack; 459 int cur_stack; 460 } cfg; 461 u32 pass_cnt; /* number of times do_check() was called */ 462 u32 subprog_cnt; 463 /* number of instructions analyzed by the verifier */ 464 u32 prev_insn_processed, insn_processed; 465 /* number of jmps, calls, exits analyzed so far */ 466 u32 prev_jmps_processed, jmps_processed; 467 /* total verification time */ 468 u64 verification_time; 469 /* maximum number of verifier states kept in 'branching' instructions */ 470 u32 max_states_per_insn; 471 /* total number of allocated verifier states */ 472 u32 total_states; 473 /* some states are freed during program analysis. 474 * this is peak number of states. this number dominates kernel 475 * memory consumption during verification 476 */ 477 u32 peak_states; 478 /* longest register parentage chain walked for liveness marking */ 479 u32 longest_mark_read_walk; 480 bpfptr_t fd_array; 481 482 /* bit mask to keep track of whether a register has been accessed 483 * since the last time the function state was printed 484 */ 485 u32 scratched_regs; 486 /* Same as scratched_regs but for stack slots */ 487 u64 scratched_stack_slots; 488 u32 prev_log_len, prev_insn_print_len; 489 /* buffer used in reg_type_str() to generate reg_type string */ 490 char type_str_buf[TYPE_STR_BUF_LEN]; 491 }; 492 493 __printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log, 494 const char *fmt, va_list args); 495 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, 496 const char *fmt, ...); 497 __printf(2, 3) void bpf_log(struct bpf_verifier_log *log, 498 const char *fmt, ...); 499 500 static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env) 501 { 502 struct bpf_verifier_state *cur = env->cur_state; 503 504 return cur->frame[cur->curframe]; 505 } 506 507 static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env) 508 { 509 return cur_func(env)->regs; 510 } 511 512 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog); 513 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env, 514 int insn_idx, int prev_insn_idx); 515 int bpf_prog_offload_finalize(struct bpf_verifier_env *env); 516 void 517 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off, 518 struct bpf_insn *insn); 519 void 520 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt); 521 522 int check_ptr_off_reg(struct bpf_verifier_env *env, 523 const struct bpf_reg_state *reg, int regno); 524 int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, 525 u32 regno, u32 mem_size); 526 527 /* this lives here instead of in bpf.h because it needs to dereference tgt_prog */ 528 static inline u64 bpf_trampoline_compute_key(const struct bpf_prog *tgt_prog, 529 struct btf *btf, u32 btf_id) 530 { 531 if (tgt_prog) 532 return ((u64)tgt_prog->aux->id << 32) | btf_id; 533 else 534 return ((u64)btf_obj_id(btf) << 32) | 0x80000000 | btf_id; 535 } 536 537 /* unpack the IDs from the key as constructed above */ 538 static inline void bpf_trampoline_unpack_key(u64 key, u32 *obj_id, u32 *btf_id) 539 { 540 if (obj_id) 541 *obj_id = key >> 32; 542 if (btf_id) 543 *btf_id = key & 0x7FFFFFFF; 544 } 545 546 int bpf_check_attach_target(struct bpf_verifier_log *log, 547 const struct bpf_prog *prog, 548 const struct bpf_prog *tgt_prog, 549 u32 btf_id, 550 struct bpf_attach_target_info *tgt_info); 551 void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab); 552 553 #define BPF_BASE_TYPE_MASK GENMASK(BPF_BASE_TYPE_BITS - 1, 0) 554 555 /* extract base type from bpf_{arg, return, reg}_type. */ 556 static inline u32 base_type(u32 type) 557 { 558 return type & BPF_BASE_TYPE_MASK; 559 } 560 561 /* extract flags from an extended type. See bpf_type_flag in bpf.h. */ 562 static inline u32 type_flag(u32 type) 563 { 564 return type & ~BPF_BASE_TYPE_MASK; 565 } 566 567 #endif /* _LINUX_BPF_VERIFIER_H */ 568