bpf_verifier.h

   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 tmp_str_buf in bpf_verifier.
  22 * we need at least 306 bytes to fit full stack mask representation
  23 * (in the "-8,-16,...,-512" form)
  24 */
  25#define TMP_STR_BUF_LEN 320
  26/* Patch buffer size */
  27#define INSN_BUF_SIZE 32
  28
  29#define ITER_PREFIX "bpf_iter_"
  30
  31enum bpf_iter_state {
  32	BPF_ITER_STATE_INVALID, /* for non-first slot */
  33	BPF_ITER_STATE_ACTIVE,
  34	BPF_ITER_STATE_DRAINED,
  35};
  36
  37struct bpf_reg_state {
  38	/* Ordering of fields matters.  See states_equal() */
  39	enum bpf_reg_type type;
  40	/*
  41	 * Fixed part of pointer offset, pointer types only.
  42	 * Or constant delta between "linked" scalars with the same ID.
  43	 */
  44	s32 off;
  45	union {
  46		/* valid when type == PTR_TO_PACKET */
  47		int range;
  48
  49		/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
  50		 *   PTR_TO_MAP_VALUE_OR_NULL
  51		 */
  52		struct {
  53			struct bpf_map *map_ptr;
  54			/* To distinguish map lookups from outer map
  55			 * the map_uid is non-zero for registers
  56			 * pointing to inner maps.
  57			 */
  58			u32 map_uid;
  59		};
  60
  61		/* for PTR_TO_BTF_ID */
  62		struct {
  63			struct btf *btf;
  64			u32 btf_id;
  65		};
  66
  67		struct { /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
  68			u32 mem_size;
  69			u32 dynptr_id; /* for dynptr slices */
  70		};
  71
  72		/* For dynptr stack slots */
  73		struct {
  74			enum bpf_dynptr_type type;
  75			/* A dynptr is 16 bytes so it takes up 2 stack slots.
  76			 * We need to track which slot is the first slot
  77			 * to protect against cases where the user may try to
  78			 * pass in an address starting at the second slot of the
  79			 * dynptr.
  80			 */
  81			bool first_slot;
  82		} dynptr;
  83
  84		/* For bpf_iter stack slots */
  85		struct {
  86			/* BTF container and BTF type ID describing
  87			 * struct bpf_iter_<type> of an iterator state
  88			 */
  89			struct btf *btf;
  90			u32 btf_id;
  91			/* packing following two fields to fit iter state into 16 bytes */
  92			enum bpf_iter_state state:2;
  93			int depth:30;
  94		} iter;
  95
  96		/* For irq stack slots */
  97		struct {
  98			enum {
  99				IRQ_NATIVE_KFUNC,
 100				IRQ_LOCK_KFUNC,
 101			} kfunc_class;
 102		} irq;
 103
 104		/* Max size from any of the above. */
 105		struct {
 106			unsigned long raw1;
 107			unsigned long raw2;
 108		} raw;
 109
 110		u32 subprogno; /* for PTR_TO_FUNC */
 111	};
 112	/* For scalar types (SCALAR_VALUE), this represents our knowledge of
 113	 * the actual value.
 114	 * For pointer types, this represents the variable part of the offset
 115	 * from the pointed-to object, and is shared with all bpf_reg_states
 116	 * with the same id as us.
 117	 */
 118	struct tnum var_off;
 119	/* Used to determine if any memory access using this register will
 120	 * result in a bad access.
 121	 * These refer to the same value as var_off, not necessarily the actual
 122	 * contents of the register.
 123	 */
 124	s64 smin_value; /* minimum possible (s64)value */
 125	s64 smax_value; /* maximum possible (s64)value */
 126	u64 umin_value; /* minimum possible (u64)value */
 127	u64 umax_value; /* maximum possible (u64)value */
 128	s32 s32_min_value; /* minimum possible (s32)value */
 129	s32 s32_max_value; /* maximum possible (s32)value */
 130	u32 u32_min_value; /* minimum possible (u32)value */
 131	u32 u32_max_value; /* maximum possible (u32)value */
 132	/* For PTR_TO_PACKET, used to find other pointers with the same variable
 133	 * offset, so they can share range knowledge.
 134	 * For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
 135	 * came from, when one is tested for != NULL.
 136	 * For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
 137	 * for the purpose of tracking that it's freed.
 138	 * For PTR_TO_SOCKET this is used to share which pointers retain the
 139	 * same reference to the socket, to determine proper reference freeing.
 140	 * For stack slots that are dynptrs, this is used to track references to
 141	 * the dynptr to determine proper reference freeing.
 142	 * Similarly to dynptrs, we use ID to track "belonging" of a reference
 143	 * to a specific instance of bpf_iter.
 144	 */
 145	/*
 146	 * Upper bit of ID is used to remember relationship between "linked"
 147	 * registers. Example:
 148	 * r1 = r2;    both will have r1->id == r2->id == N
 149	 * r1 += 10;   r1->id == N | BPF_ADD_CONST and r1->off == 10
 150	 */
 151#define BPF_ADD_CONST (1U << 31)
 152	u32 id;
 153	/* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
 154	 * from a pointer-cast helper, bpf_sk_fullsock() and
 155	 * bpf_tcp_sock().
 156	 *
 157	 * Consider the following where "sk" is a reference counted
 158	 * pointer returned from "sk = bpf_sk_lookup_tcp();":
 159	 *
 160	 * 1: sk = bpf_sk_lookup_tcp();
 161	 * 2: if (!sk) { return 0; }
 162	 * 3: fullsock = bpf_sk_fullsock(sk);
 163	 * 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
 164	 * 5: tp = bpf_tcp_sock(fullsock);
 165	 * 6: if (!tp) { bpf_sk_release(sk); return 0; }
 166	 * 7: bpf_sk_release(sk);
 167	 * 8: snd_cwnd = tp->snd_cwnd;  // verifier will complain
 168	 *
 169	 * After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
 170	 * "tp" ptr should be invalidated also.  In order to do that,
 171	 * the reg holding "fullsock" and "sk" need to remember
 172	 * the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
 173	 * such that the verifier can reset all regs which have
 174	 * ref_obj_id matching the sk_reg->id.
 175	 *
 176	 * sk_reg->ref_obj_id is set to sk_reg->id at line 1.
 177	 * sk_reg->id will stay as NULL-marking purpose only.
 178	 * After NULL-marking is done, sk_reg->id can be reset to 0.
 179	 *
 180	 * After "fullsock = bpf_sk_fullsock(sk);" at line 3,
 181	 * fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
 182	 *
 183	 * After "tp = bpf_tcp_sock(fullsock);" at line 5,
 184	 * tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
 185	 * which is the same as sk_reg->ref_obj_id.
 186	 *
 187	 * From the verifier perspective, if sk, fullsock and tp
 188	 * are not NULL, they are the same ptr with different
 189	 * reg->type.  In particular, bpf_sk_release(tp) is also
 190	 * allowed and has the same effect as bpf_sk_release(sk).
 191	 */
 192	u32 ref_obj_id;
 193	/* Inside the callee two registers can be both PTR_TO_STACK like
 194	 * R1=fp-8 and R2=fp-8, but one of them points to this function stack
 195	 * while another to the caller's stack. To differentiate them 'frameno'
 196	 * is used which is an index in bpf_verifier_state->frame[] array
 197	 * pointing to bpf_func_state.
 198	 */
 199	u32 frameno;
 200	/* Tracks subreg definition. The stored value is the insn_idx of the
 201	 * writing insn. This is safe because subreg_def is used before any insn
 202	 * patching which only happens after main verification finished.
 203	 */
 204	s32 subreg_def;
 205	/* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
 206	bool precise;
 207};
 208
 209enum bpf_stack_slot_type {
 210	STACK_INVALID,    /* nothing was stored in this stack slot */
 211	STACK_SPILL,      /* register spilled into stack */
 212	STACK_MISC,	  /* BPF program wrote some data into this slot */
 213	STACK_ZERO,	  /* BPF program wrote constant zero */
 214	/* A dynptr is stored in this stack slot. The type of dynptr
 215	 * is stored in bpf_stack_state->spilled_ptr.dynptr.type
 216	 */
 217	STACK_DYNPTR,
 218	STACK_ITER,
 219	STACK_IRQ_FLAG,
 220};
 221
 222#define BPF_REG_SIZE 8	/* size of eBPF register in bytes */
 223
 224#define BPF_REGMASK_ARGS ((1 << BPF_REG_1) | (1 << BPF_REG_2) | \
 225			  (1 << BPF_REG_3) | (1 << BPF_REG_4) | \
 226			  (1 << BPF_REG_5))
 227
 228#define BPF_DYNPTR_SIZE		sizeof(struct bpf_dynptr_kern)
 229#define BPF_DYNPTR_NR_SLOTS		(BPF_DYNPTR_SIZE / BPF_REG_SIZE)
 230
 231struct bpf_stack_state {
 232	struct bpf_reg_state spilled_ptr;
 233	u8 slot_type[BPF_REG_SIZE];
 234};
 235
 236struct bpf_reference_state {
 237	/* Each reference object has a type. Ensure REF_TYPE_PTR is zero to
 238	 * default to pointer reference on zero initialization of a state.
 239	 */
 240	enum ref_state_type {
 241		REF_TYPE_PTR		= (1 << 1),
 242		REF_TYPE_IRQ		= (1 << 2),
 243		REF_TYPE_LOCK		= (1 << 3),
 244		REF_TYPE_RES_LOCK 	= (1 << 4),
 245		REF_TYPE_RES_LOCK_IRQ	= (1 << 5),
 246		REF_TYPE_LOCK_MASK	= REF_TYPE_LOCK | REF_TYPE_RES_LOCK | REF_TYPE_RES_LOCK_IRQ,
 247	} type;
 248	/* Track each reference created with a unique id, even if the same
 249	 * instruction creates the reference multiple times (eg, via CALL).
 250	 */
 251	int id;
 252	/* Instruction where the allocation of this reference occurred. This
 253	 * is used purely to inform the user of a reference leak.
 254	 */
 255	int insn_idx;
 256	/* Use to keep track of the source object of a lock, to ensure
 257	 * it matches on unlock.
 258	 */
 259	void *ptr;
 260};
 261
 262struct bpf_retval_range {
 263	s32 minval;
 264	s32 maxval;
 265};
 266
 267/* state of the program:
 268 * type of all registers and stack info
 269 */
 270struct bpf_func_state {
 271	struct bpf_reg_state regs[MAX_BPF_REG];
 272	/* index of call instruction that called into this func */
 273	int callsite;
 274	/* stack frame number of this function state from pov of
 275	 * enclosing bpf_verifier_state.
 276	 * 0 = main function, 1 = first callee.
 277	 */
 278	u32 frameno;
 279	/* subprog number == index within subprog_info
 280	 * zero == main subprog
 281	 */
 282	u32 subprogno;
 283	/* Every bpf_timer_start will increment async_entry_cnt.
 284	 * It's used to distinguish:
 285	 * void foo(void) { for(;;); }
 286	 * void foo(void) { bpf_timer_set_callback(,foo); }
 287	 */
 288	u32 async_entry_cnt;
 289	struct bpf_retval_range callback_ret_range;
 290	bool in_callback_fn;
 291	bool in_async_callback_fn;
 292	bool in_exception_callback_fn;
 293	/* For callback calling functions that limit number of possible
 294	 * callback executions (e.g. bpf_loop) keeps track of current
 295	 * simulated iteration number.
 296	 * Value in frame N refers to number of times callback with frame
 297	 * N+1 was simulated, e.g. for the following call:
 298	 *
 299	 *   bpf_loop(..., fn, ...); | suppose current frame is N
 300	 *                           | fn would be simulated in frame N+1
 301	 *                           | number of simulations is tracked in frame N
 302	 */
 303	u32 callback_depth;
 304
 305	/* The following fields should be last. See copy_func_state() */
 306	/* The state of the stack. Each element of the array describes BPF_REG_SIZE
 307	 * (i.e. 8) bytes worth of stack memory.
 308	 * stack[0] represents bytes [*(r10-8)..*(r10-1)]
 309	 * stack[1] represents bytes [*(r10-16)..*(r10-9)]
 310	 * ...
 311	 * stack[allocated_stack/8 - 1] represents [*(r10-allocated_stack)..*(r10-allocated_stack+7)]
 312	 */
 313	struct bpf_stack_state *stack;
 314	/* Size of the current stack, in bytes. The stack state is tracked below, in
 315	 * `stack`. allocated_stack is always a multiple of BPF_REG_SIZE.
 316	 */
 317	int allocated_stack;
 318};
 319
 320#define MAX_CALL_FRAMES 8
 321
 322/* instruction history flags, used in bpf_jmp_history_entry.flags field */
 323enum {
 324	/* instruction references stack slot through PTR_TO_STACK register;
 325	 * we also store stack's frame number in lower 3 bits (MAX_CALL_FRAMES is 8)
 326	 * and accessed stack slot's index in next 6 bits (MAX_BPF_STACK is 512,
 327	 * 8 bytes per slot, so slot index (spi) is [0, 63])
 328	 */
 329	INSN_F_FRAMENO_MASK = 0x7, /* 3 bits */
 330
 331	INSN_F_SPI_MASK = 0x3f, /* 6 bits */
 332	INSN_F_SPI_SHIFT = 3, /* shifted 3 bits to the left */
 333
 334	INSN_F_STACK_ACCESS = BIT(9),
 335
 336	INSN_F_DST_REG_STACK = BIT(10), /* dst_reg is PTR_TO_STACK */
 337	INSN_F_SRC_REG_STACK = BIT(11), /* src_reg is PTR_TO_STACK */
 338	/* total 12 bits are used now. */
 339};
 340
 341static_assert(INSN_F_FRAMENO_MASK + 1 >= MAX_CALL_FRAMES);
 342static_assert(INSN_F_SPI_MASK + 1 >= MAX_BPF_STACK / 8);
 343
 344struct bpf_jmp_history_entry {
 345	u32 idx;
 346	/* insn idx can't be bigger than 1 million */
 347	u32 prev_idx : 20;
 348	/* special INSN_F_xxx flags */
 349	u32 flags : 12;
 350	/* additional registers that need precision tracking when this
 351	 * jump is backtracked, vector of six 10-bit records
 352	 */
 353	u64 linked_regs;
 354};
 355
 356/* Maximum number of register states that can exist at once */
 357#define BPF_ID_MAP_SIZE ((MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) * MAX_CALL_FRAMES)
 358struct bpf_verifier_state {
 359	/* call stack tracking */
 360	struct bpf_func_state *frame[MAX_CALL_FRAMES];
 361	struct bpf_verifier_state *parent;
 362	/* Acquired reference states */
 363	struct bpf_reference_state *refs;
 364	/*
 365	 * 'branches' field is the number of branches left to explore:
 366	 * 0 - all possible paths from this state reached bpf_exit or
 367	 * were safely pruned
 368	 * 1 - at least one path is being explored.
 369	 * This state hasn't reached bpf_exit
 370	 * 2 - at least two paths are being explored.
 371	 * This state is an immediate parent of two children.
 372	 * One is fallthrough branch with branches==1 and another
 373	 * state is pushed into stack (to be explored later) also with
 374	 * branches==1. The parent of this state has branches==1.
 375	 * The verifier state tree connected via 'parent' pointer looks like:
 376	 * 1
 377	 * 1
 378	 * 2 -> 1 (first 'if' pushed into stack)
 379	 * 1
 380	 * 2 -> 1 (second 'if' pushed into stack)
 381	 * 1
 382	 * 1
 383	 * 1 bpf_exit.
 384	 *
 385	 * Once do_check() reaches bpf_exit, it calls update_branch_counts()
 386	 * and the verifier state tree will look:
 387	 * 1
 388	 * 1
 389	 * 2 -> 1 (first 'if' pushed into stack)
 390	 * 1
 391	 * 1 -> 1 (second 'if' pushed into stack)
 392	 * 0
 393	 * 0
 394	 * 0 bpf_exit.
 395	 * After pop_stack() the do_check() will resume at second 'if'.
 396	 *
 397	 * If is_state_visited() sees a state with branches > 0 it means
 398	 * there is a loop. If such state is exactly equal to the current state
 399	 * it's an infinite loop. Note states_equal() checks for states
 400	 * equivalency, so two states being 'states_equal' does not mean
 401	 * infinite loop. The exact comparison is provided by
 402	 * states_maybe_looping() function. It's a stronger pre-check and
 403	 * much faster than states_equal().
 404	 *
 405	 * This algorithm may not find all possible infinite loops or
 406	 * loop iteration count may be too high.
 407	 * In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
 408	 */
 409	u32 branches;
 410	u32 insn_idx;
 411	u32 curframe;
 412
 413	u32 acquired_refs;
 414	u32 active_locks;
 415	u32 active_preempt_locks;
 416	u32 active_irq_id;
 417	u32 active_lock_id;
 418	void *active_lock_ptr;
 419	u32 active_rcu_locks;
 420
 421	bool speculative;
 422	bool in_sleepable;
 423	bool cleaned;
 424
 425	/* first and last insn idx of this verifier state */
 426	u32 first_insn_idx;
 427	u32 last_insn_idx;
 428	/* if this state is a backedge state then equal_state
 429	 * records cached state to which this state is equal.
 430	 */
 431	struct bpf_verifier_state *equal_state;
 432	/* jmp history recorded from first to last.
 433	 * backtracking is using it to go from last to first.
 434	 * For most states jmp_history_cnt is [0-3].
 435	 * For loops can go up to ~40.
 436	 */
 437	struct bpf_jmp_history_entry *jmp_history;
 438	u32 jmp_history_cnt;
 439	u32 dfs_depth;
 440	u32 callback_unroll_depth;
 441	u32 may_goto_depth;
 442};
 443
 444#define bpf_get_spilled_reg(slot, frame, mask)				\
 445	(((slot < frame->allocated_stack / BPF_REG_SIZE) &&		\
 446	  ((1 << frame->stack[slot].slot_type[BPF_REG_SIZE - 1]) & (mask))) \
 447	 ? &frame->stack[slot].spilled_ptr : NULL)
 448
 449/* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
 450#define bpf_for_each_spilled_reg(iter, frame, reg, mask)			\
 451	for (iter = 0, reg = bpf_get_spilled_reg(iter, frame, mask);		\
 452	     iter < frame->allocated_stack / BPF_REG_SIZE;		\
 453	     iter++, reg = bpf_get_spilled_reg(iter, frame, mask))
 454
 455#define bpf_for_each_reg_in_vstate_mask(__vst, __state, __reg, __mask, __expr)   \
 456	({                                                               \
 457		struct bpf_verifier_state *___vstate = __vst;            \
 458		int ___i, ___j;                                          \
 459		for (___i = 0; ___i <= ___vstate->curframe; ___i++) {    \
 460			struct bpf_reg_state *___regs;                   \
 461			__state = ___vstate->frame[___i];                \
 462			___regs = __state->regs;                         \
 463			for (___j = 0; ___j < MAX_BPF_REG; ___j++) {     \
 464				__reg = &___regs[___j];                  \
 465				(void)(__expr);                          \
 466			}                                                \
 467			bpf_for_each_spilled_reg(___j, __state, __reg, __mask) { \
 468				if (!__reg)                              \
 469					continue;                        \
 470				(void)(__expr);                          \
 471			}                                                \
 472		}                                                        \
 473	})
 474
 475/* Invoke __expr over regsiters in __vst, setting __state and __reg */
 476#define bpf_for_each_reg_in_vstate(__vst, __state, __reg, __expr) \
 477	bpf_for_each_reg_in_vstate_mask(__vst, __state, __reg, 1 << STACK_SPILL, __expr)
 478
 479/* linked list of verifier states used to prune search */
 480struct bpf_verifier_state_list {
 481	struct bpf_verifier_state state;
 482	struct list_head node;
 483	u32 miss_cnt;
 484	u32 hit_cnt:31;
 485	u32 in_free_list:1;
 486};
 487
 488struct bpf_loop_inline_state {
 489	unsigned int initialized:1; /* set to true upon first entry */
 490	unsigned int fit_for_inline:1; /* true if callback function is the same
 491					* at each call and flags are always zero
 492					*/
 493	u32 callback_subprogno; /* valid when fit_for_inline is true */
 494};
 495
 496/* pointer and state for maps */
 497struct bpf_map_ptr_state {
 498	struct bpf_map *map_ptr;
 499	bool poison;
 500	bool unpriv;
 501};
 502
 503/* Possible states for alu_state member. */
 504#define BPF_ALU_SANITIZE_SRC		(1U << 0)
 505#define BPF_ALU_SANITIZE_DST		(1U << 1)
 506#define BPF_ALU_NEG_VALUE		(1U << 2)
 507#define BPF_ALU_NON_POINTER		(1U << 3)
 508#define BPF_ALU_IMMEDIATE		(1U << 4)
 509#define BPF_ALU_SANITIZE		(BPF_ALU_SANITIZE_SRC | \
 510					 BPF_ALU_SANITIZE_DST)
 511
 512/*
 513 * An array of BPF instructions.
 514 * Primary usage: return value of bpf_insn_successors.
 515 */
 516struct bpf_iarray {
 517	int cnt;
 518	u32 items[];
 519};
 520
 521struct bpf_insn_aux_data {
 522	union {
 523		enum bpf_reg_type ptr_type;	/* pointer type for load/store insns */
 524		struct bpf_map_ptr_state map_ptr_state;
 525		s32 call_imm;			/* saved imm field of call insn */
 526		u32 alu_limit;			/* limit for add/sub register with pointer */
 527		struct {
 528			u32 map_index;		/* index into used_maps[] */
 529			u32 map_off;		/* offset from value base address */
 530		};
 531		struct {
 532			enum bpf_reg_type reg_type;	/* type of pseudo_btf_id */
 533			union {
 534				struct {
 535					struct btf *btf;
 536					u32 btf_id;	/* btf_id for struct typed var */
 537				};
 538				u32 mem_size;	/* mem_size for non-struct typed var */
 539			};
 540		} btf_var;
 541		/* if instruction is a call to bpf_loop this field tracks
 542		 * the state of the relevant registers to make decision about inlining
 543		 */
 544		struct bpf_loop_inline_state loop_inline_state;
 545	};
 546	union {
 547		/* remember the size of type passed to bpf_obj_new to rewrite R1 */
 548		u64 obj_new_size;
 549		/* remember the offset of node field within type to rewrite */
 550		u64 insert_off;
 551	};
 552	struct bpf_iarray *jt;	/* jump table for gotox or bpf_tailcall call instruction */
 553	struct btf_struct_meta *kptr_struct_meta;
 554	u64 map_key_state; /* constant (32 bit) key tracking for maps */
 555	int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
 556	u32 seen; /* this insn was processed by the verifier at env->pass_cnt */
 557	bool nospec; /* do not execute this instruction speculatively */
 558	bool nospec_result; /* result is unsafe under speculation, nospec must follow */
 559	bool zext_dst; /* this insn zero extends dst reg */
 560	bool needs_zext; /* alu op needs to clear upper bits */
 561	bool non_sleepable; /* helper/kfunc may be called from non-sleepable context */
 562	bool is_iter_next; /* bpf_iter_<type>_next() kfunc call */
 563	bool call_with_percpu_alloc_ptr; /* {this,per}_cpu_ptr() with prog percpu alloc */
 564	u8 alu_state; /* used in combination with alu_limit */
 565	/* true if STX or LDX instruction is a part of a spill/fill
 566	 * pattern for a bpf_fastcall call.
 567	 */
 568	u8 fastcall_pattern:1;
 569	/* for CALL instructions, a number of spill/fill pairs in the
 570	 * bpf_fastcall pattern.
 571	 */
 572	u8 fastcall_spills_num:3;
 573	u8 arg_prog:4;
 574
 575	/* below fields are initialized once */
 576	unsigned int orig_idx; /* original instruction index */
 577	bool jmp_point;
 578	bool prune_point;
 579	/* ensure we check state equivalence and save state checkpoint and
 580	 * this instruction, regardless of any heuristics
 581	 */
 582	bool force_checkpoint;
 583	/* true if instruction is a call to a helper function that
 584	 * accepts callback function as a parameter.
 585	 */
 586	bool calls_callback;
 587	/*
 588	 * CFG strongly connected component this instruction belongs to,
 589	 * zero if it is a singleton SCC.
 590	 */
 591	u32 scc;
 592	/* registers alive before this instruction. */
 593	u16 live_regs_before;
 594};
 595
 596#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
 597#define MAX_USED_BTFS 64 /* max number of BTFs accessed by one BPF program */
 598
 599#define BPF_VERIFIER_TMP_LOG_SIZE	1024
 600
 601struct bpf_verifier_log {
 602	/* Logical start and end positions of a "log window" of the verifier log.
 603	 * start_pos == 0 means we haven't truncated anything.
 604	 * Once truncation starts to happen, start_pos + len_total == end_pos,
 605	 * except during log reset situations, in which (end_pos - start_pos)
 606	 * might get smaller than len_total (see bpf_vlog_reset()).
 607	 * Generally, (end_pos - start_pos) gives number of useful data in
 608	 * user log buffer.
 609	 */
 610	u64 start_pos;
 611	u64 end_pos;
 612	char __user *ubuf;
 613	u32 level;
 614	u32 len_total;
 615	u32 len_max;
 616	char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
 617};
 618
 619#define BPF_LOG_LEVEL1	1
 620#define BPF_LOG_LEVEL2	2
 621#define BPF_LOG_STATS	4
 622#define BPF_LOG_FIXED	8
 623#define BPF_LOG_LEVEL	(BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
 624#define BPF_LOG_MASK	(BPF_LOG_LEVEL | BPF_LOG_STATS | BPF_LOG_FIXED)
 625#define BPF_LOG_KERNEL	(BPF_LOG_MASK + 1) /* kernel internal flag */
 626#define BPF_LOG_MIN_ALIGNMENT 8U
 627#define BPF_LOG_ALIGNMENT 40U
 628
 629static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
 630{
 631	return log && log->level;
 632}
 633
 634#define BPF_MAX_SUBPROGS 256
 635
 636struct bpf_subprog_arg_info {
 637	enum bpf_arg_type arg_type;
 638	union {
 639		u32 mem_size;
 640		u32 btf_id;
 641	};
 642};
 643
 644enum priv_stack_mode {
 645	PRIV_STACK_UNKNOWN,
 646	NO_PRIV_STACK,
 647	PRIV_STACK_ADAPTIVE,
 648};
 649
 650struct bpf_subprog_info {
 651	/* 'start' has to be the first field otherwise find_subprog() won't work */
 652	u32 start; /* insn idx of function entry point */
 653	u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
 654	u32 postorder_start; /* The idx to the env->cfg.insn_postorder */
 655	u32 exit_idx; /* Index of one of the BPF_EXIT instructions in this subprogram */
 656	u16 stack_depth; /* max. stack depth used by this function */
 657	u16 stack_extra;
 658	/* offsets in range [stack_depth .. fastcall_stack_off)
 659	 * are used for bpf_fastcall spills and fills.
 660	 */
 661	s16 fastcall_stack_off;
 662	bool has_tail_call: 1;
 663	bool tail_call_reachable: 1;
 664	bool has_ld_abs: 1;
 665	bool is_cb: 1;
 666	bool is_async_cb: 1;
 667	bool is_exception_cb: 1;
 668	bool args_cached: 1;
 669	/* true if bpf_fastcall stack region is used by functions that can't be inlined */
 670	bool keep_fastcall_stack: 1;
 671	bool changes_pkt_data: 1;
 672	bool might_sleep: 1;
 673	u8 arg_cnt:3;
 674
 675	enum priv_stack_mode priv_stack_mode;
 676	struct bpf_subprog_arg_info args[MAX_BPF_FUNC_REG_ARGS];
 677};
 678
 679struct bpf_verifier_env;
 680
 681struct backtrack_state {
 682	struct bpf_verifier_env *env;
 683	u32 frame;
 684	u32 reg_masks[MAX_CALL_FRAMES];
 685	u64 stack_masks[MAX_CALL_FRAMES];
 686};
 687
 688struct bpf_id_pair {
 689	u32 old;
 690	u32 cur;
 691};
 692
 693struct bpf_idmap {
 694	u32 tmp_id_gen;
 695	struct bpf_id_pair map[BPF_ID_MAP_SIZE];
 696};
 697
 698struct bpf_idset {
 699	u32 count;
 700	u32 ids[BPF_ID_MAP_SIZE];
 701};
 702
 703/* see verifier.c:compute_scc_callchain() */
 704struct bpf_scc_callchain {
 705	/* call sites from bpf_verifier_state->frame[*]->callsite leading to this SCC */
 706	u32 callsites[MAX_CALL_FRAMES - 1];
 707	/* last frame in a chain is identified by SCC id */
 708	u32 scc;
 709};
 710
 711/* verifier state waiting for propagate_backedges() */
 712struct bpf_scc_backedge {
 713	struct bpf_scc_backedge *next;
 714	struct bpf_verifier_state state;
 715};
 716
 717struct bpf_scc_visit {
 718	struct bpf_scc_callchain callchain;
 719	/* first state in current verification path that entered SCC
 720	 * identified by the callchain
 721	 */
 722	struct bpf_verifier_state *entry_state;
 723	struct bpf_scc_backedge *backedges; /* list of backedges */
 724	u32 num_backedges;
 725};
 726
 727/* An array of bpf_scc_visit structs sharing tht same bpf_scc_callchain->scc
 728 * but having different bpf_scc_callchain->callsites.
 729 */
 730struct bpf_scc_info {
 731	u32 num_visits;
 732	struct bpf_scc_visit visits[];
 733};
 734
 735struct bpf_liveness;
 736
 737/* single container for all structs
 738 * one verifier_env per bpf_check() call
 739 */
 740struct bpf_verifier_env {
 741	u32 insn_idx;
 742	u32 prev_insn_idx;
 743	struct bpf_prog *prog;		/* eBPF program being verified */
 744	const struct bpf_verifier_ops *ops;
 745	struct module *attach_btf_mod;	/* The owner module of prog->aux->attach_btf */
 746	struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
 747	int stack_size;			/* number of states to be processed */
 748	bool strict_alignment;		/* perform strict pointer alignment checks */
 749	bool test_state_freq;		/* test verifier with different pruning frequency */
 750	bool test_reg_invariants;	/* fail verification on register invariants violations */
 751	struct bpf_verifier_state *cur_state; /* current verifier state */
 752	/* Search pruning optimization, array of list_heads for
 753	 * lists of struct bpf_verifier_state_list.
 754	 */
 755	struct list_head *explored_states;
 756	struct list_head free_list;	/* list of struct bpf_verifier_state_list */
 757	struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
 758	struct btf_mod_pair used_btfs[MAX_USED_BTFS]; /* array of BTF's used by BPF program */
 759	struct bpf_map *insn_array_maps[MAX_USED_MAPS]; /* array of INSN_ARRAY map's to be relocated */
 760	u32 used_map_cnt;		/* number of used maps */
 761	u32 used_btf_cnt;		/* number of used BTF objects */
 762	u32 insn_array_map_cnt;		/* number of used maps of type BPF_MAP_TYPE_INSN_ARRAY */
 763	u32 id_gen;			/* used to generate unique reg IDs */
 764	u32 hidden_subprog_cnt;		/* number of hidden subprogs */
 765	int exception_callback_subprog;
 766	bool explore_alu_limits;
 767	bool allow_ptr_leaks;
 768	/* Allow access to uninitialized stack memory. Writes with fixed offset are
 769	 * always allowed, so this refers to reads (with fixed or variable offset),
 770	 * to writes with variable offset and to indirect (helper) accesses.
 771	 */
 772	bool allow_uninit_stack;
 773	bool bpf_capable;
 774	bool bypass_spec_v1;
 775	bool bypass_spec_v4;
 776	bool seen_direct_write;
 777	bool seen_exception;
 778	struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
 779	const struct bpf_line_info *prev_linfo;
 780	struct bpf_verifier_log log;
 781	struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 2]; /* max + 2 for the fake and exception subprogs */
 782	union {
 783		struct bpf_idmap idmap_scratch;
 784		struct bpf_idset idset_scratch;
 785	};
 786	struct {
 787		int *insn_state;
 788		int *insn_stack;
 789		/*
 790		 * vector of instruction indexes sorted in post-order, grouped by subprogram,
 791		 * see bpf_subprog_info->postorder_start.
 792		 */
 793		int *insn_postorder;
 794		int cur_stack;
 795		/* current position in the insn_postorder vector */
 796		int cur_postorder;
 797	} cfg;
 798	struct backtrack_state bt;
 799	struct bpf_jmp_history_entry *cur_hist_ent;
 800	u32 pass_cnt; /* number of times do_check() was called */
 801	u32 subprog_cnt;
 802	/* number of instructions analyzed by the verifier */
 803	u32 prev_insn_processed, insn_processed;
 804	/* number of jmps, calls, exits analyzed so far */
 805	u32 prev_jmps_processed, jmps_processed;
 806	/* total verification time */
 807	u64 verification_time;
 808	/* maximum number of verifier states kept in 'branching' instructions */
 809	u32 max_states_per_insn;
 810	/* total number of allocated verifier states */
 811	u32 total_states;
 812	/* some states are freed during program analysis.
 813	 * this is peak number of states. this number dominates kernel
 814	 * memory consumption during verification
 815	 */
 816	u32 peak_states;
 817	/* longest register parentage chain walked for liveness marking */
 818	u32 longest_mark_read_walk;
 819	u32 free_list_size;
 820	u32 explored_states_size;
 821	u32 num_backedges;
 822	bpfptr_t fd_array;
 823
 824	/* bit mask to keep track of whether a register has been accessed
 825	 * since the last time the function state was printed
 826	 */
 827	u32 scratched_regs;
 828	/* Same as scratched_regs but for stack slots */
 829	u64 scratched_stack_slots;
 830	u64 prev_log_pos, prev_insn_print_pos;
 831	/* buffer used to temporary hold constants as scalar registers */
 832	struct bpf_reg_state fake_reg[2];
 833	/* buffer used to generate temporary string representations,
 834	 * e.g., in reg_type_str() to generate reg_type string
 835	 */
 836	char tmp_str_buf[TMP_STR_BUF_LEN];
 837	struct bpf_insn insn_buf[INSN_BUF_SIZE];
 838	struct bpf_insn epilogue_buf[INSN_BUF_SIZE];
 839	struct bpf_scc_callchain callchain_buf;
 840	struct bpf_liveness *liveness;
 841	/* array of pointers to bpf_scc_info indexed by SCC id */
 842	struct bpf_scc_info **scc_info;
 843	u32 scc_cnt;
 844	struct bpf_iarray *succ;
 845	struct bpf_iarray *gotox_tmp_buf;
 846};
 847
 848static inline struct bpf_func_info_aux *subprog_aux(struct bpf_verifier_env *env, int subprog)
 849{
 850	return &env->prog->aux->func_info_aux[subprog];
 851}
 852
 853static inline struct bpf_subprog_info *subprog_info(struct bpf_verifier_env *env, int subprog)
 854{
 855	return &env->subprog_info[subprog];
 856}
 857
 858__printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
 859				      const char *fmt, va_list args);
 860__printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
 861					   const char *fmt, ...);
 862__printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
 863			    const char *fmt, ...);
 864int bpf_vlog_init(struct bpf_verifier_log *log, u32 log_level,
 865		  char __user *log_buf, u32 log_size);
 866void bpf_vlog_reset(struct bpf_verifier_log *log, u64 new_pos);
 867int bpf_vlog_finalize(struct bpf_verifier_log *log, u32 *log_size_actual);
 868
 869__printf(3, 4) void verbose_linfo(struct bpf_verifier_env *env,
 870				  u32 insn_off,
 871				  const char *prefix_fmt, ...);
 872
 873#define verifier_bug_if(cond, env, fmt, args...)						\
 874	({											\
 875		bool __cond = (cond);								\
 876		if (unlikely(__cond))								\
 877			verifier_bug(env, fmt " (" #cond ")", ##args);				\
 878		(__cond);									\
 879	})
 880#define verifier_bug(env, fmt, args...)								\
 881	({											\
 882		BPF_WARN_ONCE(1, "verifier bug: " fmt "\n", ##args);				\
 883		bpf_log(&env->log, "verifier bug: " fmt "\n", ##args);				\
 884	})
 885
 886static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
 887{
 888	struct bpf_verifier_state *cur = env->cur_state;
 889
 890	return cur->frame[cur->curframe];
 891}
 892
 893static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
 894{
 895	return cur_func(env)->regs;
 896}
 897
 898int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
 899int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
 900				 int insn_idx, int prev_insn_idx);
 901int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
 902void
 903bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
 904			      struct bpf_insn *insn);
 905void
 906bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
 907
 908/* this lives here instead of in bpf.h because it needs to dereference tgt_prog */
 909static inline u64 bpf_trampoline_compute_key(const struct bpf_prog *tgt_prog,
 910					     struct btf *btf, u32 btf_id)
 911{
 912	if (tgt_prog)
 913		return ((u64)tgt_prog->aux->id << 32) | btf_id;
 914	else
 915		return ((u64)btf_obj_id(btf) << 32) | 0x80000000 | btf_id;
 916}
 917
 918/* unpack the IDs from the key as constructed above */
 919static inline void bpf_trampoline_unpack_key(u64 key, u32 *obj_id, u32 *btf_id)
 920{
 921	if (obj_id)
 922		*obj_id = key >> 32;
 923	if (btf_id)
 924		*btf_id = key & 0x7FFFFFFF;
 925}
 926
 927int bpf_check_attach_target(struct bpf_verifier_log *log,
 928			    const struct bpf_prog *prog,
 929			    const struct bpf_prog *tgt_prog,
 930			    u32 btf_id,
 931			    struct bpf_attach_target_info *tgt_info);
 932void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab);
 933
 934int mark_chain_precision(struct bpf_verifier_env *env, int regno);
 935
 936#define BPF_BASE_TYPE_MASK	GENMASK(BPF_BASE_TYPE_BITS - 1, 0)
 937
 938/* extract base type from bpf_{arg, return, reg}_type. */
 939static inline u32 base_type(u32 type)
 940{
 941	return type & BPF_BASE_TYPE_MASK;
 942}
 943
 944/* extract flags from an extended type. See bpf_type_flag in bpf.h. */
 945static inline u32 type_flag(u32 type)
 946{
 947	return type & ~BPF_BASE_TYPE_MASK;
 948}
 949
 950/* only use after check_attach_btf_id() */
 951static inline enum bpf_prog_type resolve_prog_type(const struct bpf_prog *prog)
 952{
 953	return (prog->type == BPF_PROG_TYPE_EXT && prog->aux->saved_dst_prog_type) ?
 954		prog->aux->saved_dst_prog_type : prog->type;
 955}
 956
 957static inline bool bpf_prog_check_recur(const struct bpf_prog *prog)
 958{
 959	switch (resolve_prog_type(prog)) {
 960	case BPF_PROG_TYPE_TRACING:
 961		return prog->expected_attach_type != BPF_TRACE_ITER;
 962	case BPF_PROG_TYPE_STRUCT_OPS:
 963		return prog->aux->jits_use_priv_stack;
 964	case BPF_PROG_TYPE_LSM:
 965	case BPF_PROG_TYPE_SYSCALL:
 966		return false;
 967	default:
 968		return true;
 969	}
 970}
 971
 972#define BPF_REG_TRUSTED_MODIFIERS (MEM_ALLOC | PTR_TRUSTED | NON_OWN_REF)
 973
 974static inline bool bpf_type_has_unsafe_modifiers(u32 type)
 975{
 976	return type_flag(type) & ~BPF_REG_TRUSTED_MODIFIERS;
 977}
 978
 979static inline bool type_is_ptr_alloc_obj(u32 type)
 980{
 981	return base_type(type) == PTR_TO_BTF_ID && type_flag(type) & MEM_ALLOC;
 982}
 983
 984static inline bool type_is_non_owning_ref(u32 type)
 985{
 986	return type_is_ptr_alloc_obj(type) && type_flag(type) & NON_OWN_REF;
 987}
 988
 989static inline bool type_is_pkt_pointer(enum bpf_reg_type type)
 990{
 991	type = base_type(type);
 992	return type == PTR_TO_PACKET ||
 993	       type == PTR_TO_PACKET_META;
 994}
 995
 996static inline bool type_is_sk_pointer(enum bpf_reg_type type)
 997{
 998	return type == PTR_TO_SOCKET ||
 999		type == PTR_TO_SOCK_COMMON ||
1000		type == PTR_TO_TCP_SOCK ||
1001		type == PTR_TO_XDP_SOCK;
1002}
1003
1004static inline bool type_may_be_null(u32 type)
1005{
1006	return type & PTR_MAYBE_NULL;
1007}
1008
1009static inline void mark_reg_scratched(struct bpf_verifier_env *env, u32 regno)
1010{
1011	env->scratched_regs |= 1U << regno;
1012}
1013
1014static inline void mark_stack_slot_scratched(struct bpf_verifier_env *env, u32 spi)
1015{
1016	env->scratched_stack_slots |= 1ULL << spi;
1017}
1018
1019static inline bool reg_scratched(const struct bpf_verifier_env *env, u32 regno)
1020{
1021	return (env->scratched_regs >> regno) & 1;
1022}
1023
1024static inline bool stack_slot_scratched(const struct bpf_verifier_env *env, u64 regno)
1025{
1026	return (env->scratched_stack_slots >> regno) & 1;
1027}
1028
1029static inline bool verifier_state_scratched(const struct bpf_verifier_env *env)
1030{
1031	return env->scratched_regs || env->scratched_stack_slots;
1032}
1033
1034static inline void mark_verifier_state_clean(struct bpf_verifier_env *env)
1035{
1036	env->scratched_regs = 0U;
1037	env->scratched_stack_slots = 0ULL;
1038}
1039
1040/* Used for printing the entire verifier state. */
1041static inline void mark_verifier_state_scratched(struct bpf_verifier_env *env)
1042{
1043	env->scratched_regs = ~0U;
1044	env->scratched_stack_slots = ~0ULL;
1045}
1046
1047static inline bool bpf_stack_narrow_access_ok(int off, int fill_size, int spill_size)
1048{
1049#ifdef __BIG_ENDIAN
1050	off -= spill_size - fill_size;
1051#endif
1052
1053	return !(off % BPF_REG_SIZE);
1054}
1055
1056static inline bool insn_is_gotox(struct bpf_insn *insn)
1057{
1058	return BPF_CLASS(insn->code) == BPF_JMP &&
1059	       BPF_OP(insn->code) == BPF_JA &&
1060	       BPF_SRC(insn->code) == BPF_X;
1061}
1062
1063const char *reg_type_str(struct bpf_verifier_env *env, enum bpf_reg_type type);
1064const char *dynptr_type_str(enum bpf_dynptr_type type);
1065const char *iter_type_str(const struct btf *btf, u32 btf_id);
1066const char *iter_state_str(enum bpf_iter_state state);
1067
1068void print_verifier_state(struct bpf_verifier_env *env, const struct bpf_verifier_state *vstate,
1069			  u32 frameno, bool print_all);
1070void print_insn_state(struct bpf_verifier_env *env, const struct bpf_verifier_state *vstate,
1071		      u32 frameno);
1072
1073struct bpf_subprog_info *bpf_find_containing_subprog(struct bpf_verifier_env *env, int off);
1074int bpf_jmp_offset(struct bpf_insn *insn);
1075struct bpf_iarray *bpf_insn_successors(struct bpf_verifier_env *env, u32 idx);
1076void bpf_fmt_stack_mask(char *buf, ssize_t buf_sz, u64 stack_mask);
1077bool bpf_calls_callback(struct bpf_verifier_env *env, int insn_idx);
1078
1079int bpf_stack_liveness_init(struct bpf_verifier_env *env);
1080void bpf_stack_liveness_free(struct bpf_verifier_env *env);
1081int bpf_update_live_stack(struct bpf_verifier_env *env);
1082int bpf_mark_stack_read(struct bpf_verifier_env *env, u32 frameno, u32 insn_idx, u64 mask);
1083void bpf_mark_stack_write(struct bpf_verifier_env *env, u32 frameno, u64 mask);
1084int bpf_reset_stack_write_marks(struct bpf_verifier_env *env, u32 insn_idx);
1085int bpf_commit_stack_write_marks(struct bpf_verifier_env *env);
1086int bpf_live_stack_query_init(struct bpf_verifier_env *env, struct bpf_verifier_state *st);
1087bool bpf_stack_slot_alive(struct bpf_verifier_env *env, u32 frameno, u32 spi);
1088void bpf_reset_live_stack_callchain(struct bpf_verifier_env *env);
1089
1090#endif /* _LINUX_BPF_VERIFIER_H */