bpf: precise scalar_value tracking

Introduce precision tracking logic that
helps cilium programs the most:
                  old clang  old clang    new clang  new clang
                          with all patches         with all patches
bpf_lb-DLB_L3.o      1838     2283         1923       1863
bpf_lb-DLB_L4.o      3218     2657         3077       2468
bpf_lb-DUNKNOWN.o    1064     545          1062       544
bpf_lxc-DDROP_ALL.o  26935    23045        166729     22629
bpf_lxc-DUNKNOWN.o   34439    35240        174607     28805
bpf_netdev.o         9721     8753         8407       6801
bpf_overlay.o        6184     7901         5420       4754
bpf_lxc_jit.o        39389    50925        39389      50925

Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
 R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1

At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.

The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1         // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2         // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8          // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1       // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
 // this is the end of verifier state. The following regs will be marked precised:
 R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks         // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0)   // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23    // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1  // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0                // here the assignment into R7
 // mark R0 to be precise:
 R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks                 // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
  // nothing further to backtrack

Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.

It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

include/linux/bpf_verifier.h

@@ -139,6 +139,8 @@ struct bpf_reg_state {
 	 */
 	s32 subreg_def;
 	enum bpf_reg_liveness live;
+	/* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
+	bool precise;
 };
 
 enum bpf_stack_slot_type {
@@ -190,6 +192,11 @@ struct bpf_func_state {
 	struct bpf_stack_state *stack;
 };
 
+struct bpf_idx_pair {
+	u32 prev_idx;
+	u32 idx;
+};
+
 #define MAX_CALL_FRAMES 8
 struct bpf_verifier_state {
 	/* call stack tracking */
@@ -245,6 +252,17 @@ struct bpf_verifier_state {
 	u32 curframe;
 	u32 active_spin_lock;
 	bool speculative;
+
+	/* first and last insn idx of this verifier state */
+	u32 first_insn_idx;
+	u32 last_insn_idx;
+	/* jmp history recorded from first to last.
+	 * backtracking is using it to go from last to first.
+	 * For most states jmp_history_cnt is [0-3].
+	 * For loops can go up to ~40.
+	 */
+	struct bpf_idx_pair *jmp_history;
+	u32 jmp_history_cnt;
 };
 
 #define bpf_get_spilled_reg(slot, frame)				\

kernel/bpf/verifier.c

@@ -455,12 +455,12 @@ static void print_verifier_state(struct bpf_verifier_env *env,
 		verbose(env, " R%d", i);
 		print_liveness(env, reg->live);
 		verbose(env, "=%s", reg_type_str[t]);
+		if (t == SCALAR_VALUE && reg->precise)
+			verbose(env, "P");
 		if ((t == SCALAR_VALUE || t == PTR_TO_STACK) &&
 		    tnum_is_const(reg->var_off)) {
 			/* reg->off should be 0 for SCALAR_VALUE */
 			verbose(env, "%lld", reg->var_off.value + reg->off);
-			if (t == PTR_TO_STACK)
-				verbose(env, ",call_%d", func(env, reg)->callsite);
 		} else {
 			verbose(env, "(id=%d", reg->id);
 			if (reg_type_may_be_refcounted_or_null(t))
@@ -522,12 +522,18 @@ static void print_verifier_state(struct bpf_verifier_env *env,
 			continue;
 		verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE);
 		print_liveness(env, state->stack[i].spilled_ptr.live);
-		if (state->stack[i].slot_type[0] == STACK_SPILL)
-			verbose(env, "=%s",
-				reg_type_str[state->stack[i].spilled_ptr.type]);
-		else
+		if (state->stack[i].slot_type[0] == STACK_SPILL) {
+			reg = &state->stack[i].spilled_ptr;
+			t = reg->type;
+			verbose(env, "=%s", reg_type_str[t]);
+			if (t == SCALAR_VALUE && reg->precise)
+				verbose(env, "P");
+			if (t == SCALAR_VALUE && tnum_is_const(reg->var_off))
+				verbose(env, "%lld", reg->var_off.value + reg->off);
+		} else {
 			verbose(env, "=%s", types_buf);
 		}
+	}
 	if (state->acquired_refs && state->refs[0].id) {
 		verbose(env, " refs=%d", state->refs[0].id);
 		for (i = 1; i < state->acquired_refs; i++)
@@ -675,6 +681,13 @@ static void free_func_state(struct bpf_func_state *state)
 	kfree(state);
 }
 
+static void clear_jmp_history(struct bpf_verifier_state *state)
+{
+	kfree(state->jmp_history);
+	state->jmp_history = NULL;
+	state->jmp_history_cnt = 0;
+}
+
 static void free_verifier_state(struct bpf_verifier_state *state,
 				bool free_self)
 {
@@ -684,6 +697,7 @@ static void free_verifier_state(struct bpf_verifier_state *state,
 		free_func_state(state->frame[i]);
 		state->frame[i] = NULL;
 	}
+	clear_jmp_history(state);
 	if (free_self)
 		kfree(state);
 }
@@ -711,8 +725,18 @@ static int copy_verifier_state(struct bpf_verifier_state *dst_state,
 			       const struct bpf_verifier_state *src)
 {
 	struct bpf_func_state *dst;
+	u32 jmp_sz = sizeof(struct bpf_idx_pair) * src->jmp_history_cnt;
 	int i, err;
 
+	if (dst_state->jmp_history_cnt < src->jmp_history_cnt) {
+		kfree(dst_state->jmp_history);
+		dst_state->jmp_history = kmalloc(jmp_sz, GFP_USER);
+		if (!dst_state->jmp_history)
+			return -ENOMEM;
+	}
+	memcpy(dst_state->jmp_history, src->jmp_history, jmp_sz);
+	dst_state->jmp_history_cnt = src->jmp_history_cnt;
+
 	/* if dst has more stack frames then src frame, free them */
 	for (i = src->curframe + 1; i <= dst_state->curframe; i++) {
 		free_func_state(dst_state->frame[i]);
@@ -723,6 +747,8 @@ static int copy_verifier_state(struct bpf_verifier_state *dst_state,
 	dst_state->active_spin_lock = src->active_spin_lock;
 	dst_state->branches = src->branches;
 	dst_state->parent = src->parent;
+	dst_state->first_insn_idx = src->first_insn_idx;
+	dst_state->last_insn_idx = src->last_insn_idx;
 	for (i = 0; i <= src->curframe; i++) {
 		dst = dst_state->frame[i];
 		if (!dst) {
@@ -967,6 +993,9 @@ static void __mark_reg_unbounded(struct bpf_reg_state *reg)
 	reg->smax_value = S64_MAX;
 	reg->umin_value = 0;
 	reg->umax_value = U64_MAX;
+
+	/* constant backtracking is enabled for root only for now */
+	reg->precise = capable(CAP_SYS_ADMIN) ? false : true;
 }
 
 /* Mark a register as having a completely unknown (scalar) value. */
@@ -1378,6 +1407,389 @@ static int check_reg_arg(struct bpf_verifier_env *env, u32 regno,
 	return 0;
 }
 
+/* for any branch, call, exit record the history of jmps in the given state */
+static int push_jmp_history(struct bpf_verifier_env *env,
+			    struct bpf_verifier_state *cur)
+{
+	u32 cnt = cur->jmp_history_cnt;
+	struct bpf_idx_pair *p;
+
+	cnt++;
+	p = krealloc(cur->jmp_history, cnt * sizeof(*p), GFP_USER);
+	if (!p)
+		return -ENOMEM;
+	p[cnt - 1].idx = env->insn_idx;
+	p[cnt - 1].prev_idx = env->prev_insn_idx;
+	cur->jmp_history = p;
+	cur->jmp_history_cnt = cnt;
+	return 0;
+}
+
+/* Backtrack one insn at a time. If idx is not at the top of recorded
+ * history then previous instruction came from straight line execution.
+ */
+static int get_prev_insn_idx(struct bpf_verifier_state *st, int i,
+			     u32 *history)
+{
+	u32 cnt = *history;
+
+	if (cnt && st->jmp_history[cnt - 1].idx == i) {
+		i = st->jmp_history[cnt - 1].prev_idx;
+		(*history)--;
+	} else {
+		i--;
+	}
+	return i;
+}
+
+/* For given verifier state backtrack_insn() is called from the last insn to
+ * the first insn. Its purpose is to compute a bitmask of registers and
+ * stack slots that needs precision in the parent verifier state.
+ */
+static int backtrack_insn(struct bpf_verifier_env *env, int idx,
+			  u32 *reg_mask, u64 *stack_mask)
+{
+	const struct bpf_insn_cbs cbs = {
+		.cb_print	= verbose,
+		.private_data	= env,
+	};
+	struct bpf_insn *insn = env->prog->insnsi + idx;
+	u8 class = BPF_CLASS(insn->code);
+	u8 opcode = BPF_OP(insn->code);
+	u8 mode = BPF_MODE(insn->code);
+	u32 dreg = 1u << insn->dst_reg;
+	u32 sreg = 1u << insn->src_reg;
+	u32 spi;
+
+	if (insn->code == 0)
+		return 0;
+	if (env->log.level & BPF_LOG_LEVEL) {
+		verbose(env, "regs=%x stack=%llx before ", *reg_mask, *stack_mask);
+		verbose(env, "%d: ", idx);
+		print_bpf_insn(&cbs, insn, env->allow_ptr_leaks);
+	}
+
+	if (class == BPF_ALU || class == BPF_ALU64) {
+		if (!(*reg_mask & dreg))
+			return 0;
+		if (opcode == BPF_MOV) {
+			if (BPF_SRC(insn->code) == BPF_X) {
+				/* dreg = sreg
+				 * dreg needs precision after this insn
+				 * sreg needs precision before this insn
+				 */
+				*reg_mask &= ~dreg;
+				*reg_mask |= sreg;
+			} else {
+				/* dreg = K
+				 * dreg needs precision after this insn.
+				 * Corresponding register is already marked
+				 * as precise=true in this verifier state.
+				 * No further markings in parent are necessary
+				 */
+				*reg_mask &= ~dreg;
+			}
+		} else {
+			if (BPF_SRC(insn->code) == BPF_X) {
+				/* dreg += sreg
+				 * both dreg and sreg need precision
+				 * before this insn
+				 */
+				*reg_mask |= sreg;
+			} /* else dreg += K
+			   * dreg still needs precision before this insn
+			   */
+		}
+	} else if (class == BPF_LDX) {
+		if (!(*reg_mask & dreg))
+			return 0;
+		*reg_mask &= ~dreg;
+
+		/* scalars can only be spilled into stack w/o losing precision.
+		 * Load from any other memory can be zero extended.
+		 * The desire to keep that precision is already indicated
+		 * by 'precise' mark in corresponding register of this state.
+		 * No further tracking necessary.
+		 */
+		if (insn->src_reg != BPF_REG_FP)
+			return 0;
+		if (BPF_SIZE(insn->code) != BPF_DW)
+			return 0;
+
+		/* dreg = *(u64 *)[fp - off] was a fill from the stack.
+		 * that [fp - off] slot contains scalar that needs to be
+		 * tracked with precision
+		 */
+		spi = (-insn->off - 1) / BPF_REG_SIZE;
+		if (spi >= 64) {
+			verbose(env, "BUG spi %d\n", spi);
+			WARN_ONCE(1, "verifier backtracking bug");
+			return -EFAULT;
+		}
+		*stack_mask |= 1ull << spi;
+	} else if (class == BPF_STX) {
+		if (*reg_mask & dreg)
+			/* stx shouldn't be using _scalar_ dst_reg
+			 * to access memory. It means backtracking
+			 * encountered a case of pointer subtraction.
+			 */
+			return -ENOTSUPP;
+		/* scalars can only be spilled into stack */
+		if (insn->dst_reg != BPF_REG_FP)
+			return 0;
+		if (BPF_SIZE(insn->code) != BPF_DW)
+			return 0;
+		spi = (-insn->off - 1) / BPF_REG_SIZE;
+		if (spi >= 64) {
+			verbose(env, "BUG spi %d\n", spi);
+			WARN_ONCE(1, "verifier backtracking bug");
+			return -EFAULT;
+		}
+		if (!(*stack_mask & (1ull << spi)))
+			return 0;
+		*stack_mask &= ~(1ull << spi);
+		*reg_mask |= sreg;
+	} else if (class == BPF_JMP || class == BPF_JMP32) {
+		if (opcode == BPF_CALL) {
+			if (insn->src_reg == BPF_PSEUDO_CALL)
+				return -ENOTSUPP;
+			/* regular helper call sets R0 */
+			*reg_mask &= ~1;
+			if (*reg_mask & 0x3f) {
+				/* if backtracing was looking for registers R1-R5
+				 * they should have been found already.
+				 */
+				verbose(env, "BUG regs %x\n", *reg_mask);
+				WARN_ONCE(1, "verifier backtracking bug");
+				return -EFAULT;
+			}
+		} else if (opcode == BPF_EXIT) {
+			return -ENOTSUPP;
+		}
+	} else if (class == BPF_LD) {
+		if (!(*reg_mask & dreg))
+			return 0;
+		*reg_mask &= ~dreg;
+		/* It's ld_imm64 or ld_abs or ld_ind.
+		 * For ld_imm64 no further tracking of precision
+		 * into parent is necessary
+		 */
+		if (mode == BPF_IND || mode == BPF_ABS)
+			/* to be analyzed */
+			return -ENOTSUPP;
+	} else if (class == BPF_ST) {
+		if (*reg_mask & dreg)
+			/* likely pointer subtraction */
+			return -ENOTSUPP;
+	}
+	return 0;
+}
+
+/* the scalar precision tracking algorithm:
+ * . at the start all registers have precise=false.
+ * . scalar ranges are tracked as normal through alu and jmp insns.
+ * . once precise value of the scalar register is used in:
+ *   .  ptr + scalar alu
+ *   . if (scalar cond K|scalar)
+ *   .  helper_call(.., scalar, ...) where ARG_CONST is expected
+ *   backtrack through the verifier states and mark all registers and
+ *   stack slots with spilled constants that these scalar regisers
+ *   should be precise.
+ * . during state pruning two registers (or spilled stack slots)
+ *   are equivalent if both are not precise.
+ *
+ * Note the verifier cannot simply walk register parentage chain,
+ * since many different registers and stack slots could have been
+ * used to compute single precise scalar.
+ *
+ * The approach of starting with precise=true for all registers and then
+ * backtrack to mark a register as not precise when the verifier detects
+ * that program doesn't care about specific value (e.g., when helper
+ * takes register as ARG_ANYTHING parameter) is not safe.
+ *
+ * It's ok to walk single parentage chain of the verifier states.
+ * It's possible that this backtracking will go all the way till 1st insn.
+ * All other branches will be explored for needing precision later.
+ *
+ * The backtracking needs to deal with cases like:
+ *   R8=map_value(id=0,off=0,ks=4,vs=1952,imm=0) R9_w=map_value(id=0,off=40,ks=4,vs=1952,imm=0)
+ * r9 -= r8
+ * r5 = r9
+ * if r5 > 0x79f goto pc+7
+ *    R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff))
+ * r5 += 1
+ * ...
+ * call bpf_perf_event_output#25
+ *   where .arg5_type = ARG_CONST_SIZE_OR_ZERO
+ *
+ * and this case:
+ * r6 = 1
+ * call foo // uses callee's r6 inside to compute r0
+ * r0 += r6
+ * if r0 == 0 goto
+ *
+ * to track above reg_mask/stack_mask needs to be independent for each frame.
+ *
+ * Also if parent's curframe > frame where backtracking started,
+ * the verifier need to mark registers in both frames, otherwise callees
+ * may incorrectly prune callers. This is similar to
+ * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences")
+ *
+ * For now backtracking falls back into conservative marking.
+ */
+static void mark_all_scalars_precise(struct bpf_verifier_env *env,
+				     struct bpf_verifier_state *st)
+{
+	struct bpf_func_state *func;
+	struct bpf_reg_state *reg;
+	int i, j;
+
+	/* big hammer: mark all scalars precise in this path.
+	 * pop_stack may still get !precise scalars.
+	 */
+	for (; st; st = st->parent)
+		for (i = 0; i <= st->curframe; i++) {
+			func = st->frame[i];
+			for (j = 0; j < BPF_REG_FP; j++) {
+				reg = &func->regs[j];
+				if (reg->type != SCALAR_VALUE)
+					continue;
+				reg->precise = true;
+			}
+			for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) {
+				if (func->stack[j].slot_type[0] != STACK_SPILL)
+					continue;
+				reg = &func->stack[j].spilled_ptr;
+				if (reg->type != SCALAR_VALUE)
+					continue;
+				reg->precise = true;
+			}
+		}
+}
+
+static int mark_chain_precision(struct bpf_verifier_env *env, int regno)
+{
+	struct bpf_verifier_state *st = env->cur_state;
+	int first_idx = st->first_insn_idx;
+	int last_idx = env->insn_idx;
+	struct bpf_func_state *func;
+	struct bpf_reg_state *reg;
+	u32 reg_mask = 1u << regno;
+	u64 stack_mask = 0;
+	bool skip_first = true;
+	int i, err;
+
+	if (!env->allow_ptr_leaks)
+		/* backtracking is root only for now */
+		return 0;
+
+	func = st->frame[st->curframe];
+	reg = &func->regs[regno];
+	if (reg->type != SCALAR_VALUE) {
+		WARN_ONCE(1, "backtracing misuse");
+		return -EFAULT;
+	}
+	if (reg->precise)
+		return 0;
+	func->regs[regno].precise = true;
+
+	for (;;) {
+		DECLARE_BITMAP(mask, 64);
+		bool new_marks = false;
+		u32 history = st->jmp_history_cnt;
+
+		if (env->log.level & BPF_LOG_LEVEL)
+			verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx);
+		for (i = last_idx;;) {
+			if (skip_first) {
+				err = 0;
+				skip_first = false;
+			} else {
+				err = backtrack_insn(env, i, &reg_mask, &stack_mask);
+			}
+			if (err == -ENOTSUPP) {
+				mark_all_scalars_precise(env, st);
+				return 0;
+			} else if (err) {
+				return err;
+			}
+			if (!reg_mask && !stack_mask)
+				/* Found assignment(s) into tracked register in this state.
+				 * Since this state is already marked, just return.
+				 * Nothing to be tracked further in the parent state.
+				 */
+				return 0;
+			if (i == first_idx)
+				break;
+			i = get_prev_insn_idx(st, i, &history);
+			if (i >= env->prog->len) {
+				/* This can happen if backtracking reached insn 0
+				 * and there are still reg_mask or stack_mask
+				 * to backtrack.
+				 * It means the backtracking missed the spot where
+				 * particular register was initialized with a constant.
+				 */
+				verbose(env, "BUG backtracking idx %d\n", i);
+				WARN_ONCE(1, "verifier backtracking bug");
+				return -EFAULT;
+			}
+		}
+		st = st->parent;
+		if (!st)
+			break;
+
+		func = st->frame[st->curframe];
+		bitmap_from_u64(mask, reg_mask);
+		for_each_set_bit(i, mask, 32) {
+			reg = &func->regs[i];
+			if (reg->type != SCALAR_VALUE)
+				continue;
+			if (!reg->precise)
+				new_marks = true;
+			reg->precise = true;
+		}
+
+		bitmap_from_u64(mask, stack_mask);
+		for_each_set_bit(i, mask, 64) {
+			if (i >= func->allocated_stack / BPF_REG_SIZE) {
+				/* This can happen if backtracking
+				 * is propagating stack precision where
+				 * caller has larger stack frame
+				 * than callee, but backtrack_insn() should
+				 * have returned -ENOTSUPP.
+				 */
+				verbose(env, "BUG spi %d stack_size %d\n",
+					i, func->allocated_stack);
+				WARN_ONCE(1, "verifier backtracking bug");
+				return -EFAULT;
+			}
+
+			if (func->stack[i].slot_type[0] != STACK_SPILL)
+				continue;
+			reg = &func->stack[i].spilled_ptr;
+			if (reg->type != SCALAR_VALUE)
+				continue;
+			if (!reg->precise)
+				new_marks = true;
+			reg->precise = true;
+		}
+		if (env->log.level & BPF_LOG_LEVEL) {
+			print_verifier_state(env, func);
+			verbose(env, "parent %s regs=%x stack=%llx marks\n",
+				new_marks ? "didn't have" : "already had",
+				reg_mask, stack_mask);
+		}
+
+		if (!new_marks)
+			break;
+
+		last_idx = st->last_insn_idx;
+		first_idx = st->first_insn_idx;
+	}
+	return 0;
+}
+
+
 static bool is_spillable_regtype(enum bpf_reg_type type)
 {
 	switch (type) {
@@ -1435,6 +1847,7 @@ static int check_stack_write(struct bpf_verifier_env *env,
 {
 	struct bpf_func_state *cur; /* state of the current function */
 	int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err;
+	u32 dst_reg = env->prog->insnsi[insn_idx].dst_reg;
 	struct bpf_reg_state *reg = NULL;
 
 	err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE),
@@ -1457,6 +1870,17 @@ static int check_stack_write(struct bpf_verifier_env *env,
 
 	if (reg && size == BPF_REG_SIZE && register_is_const(reg) &&
 	    !register_is_null(reg) && env->allow_ptr_leaks) {
+		if (dst_reg != BPF_REG_FP) {
+			/* The backtracking logic can only recognize explicit
+			 * stack slot address like [fp - 8]. Other spill of
+			 * scalar via different register has to be conervative.
+			 * Backtrack from here and mark all registers as precise
+			 * that contributed into 'reg' being a constant.
+			 */
+			err = mark_chain_precision(env, value_regno);
+			if (err)
+				return err;
+		}
 		save_register_state(state, spi, reg);
 	} else if (reg && is_spillable_regtype(reg->type)) {
 		/* register containing pointer is being spilled into stack */
@@ -1529,8 +1953,13 @@ static int check_stack_write(struct bpf_verifier_env *env,
 			state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN;
 
 		/* when we zero initialize stack slots mark them as such */
-		if (reg && register_is_null(reg))
+		if (reg && register_is_null(reg)) {
+			/* backtracking doesn't work for STACK_ZERO yet. */
+			err = mark_chain_precision(env, value_regno);
+			if (err)
+				return err;
 			type = STACK_ZERO;
+		}
 
 		/* Mark slots affected by this stack write. */
 		for (i = 0; i < size; i++)
@@ -1610,6 +2039,17 @@ static int check_stack_read(struct bpf_verifier_env *env,
 				 * so the whole register == const_zero
 				 */
 				__mark_reg_const_zero(&state->regs[value_regno]);
+				/* backtracking doesn't support STACK_ZERO yet,
+				 * so mark it precise here, so that later
+				 * backtracking can stop here.
+				 * Backtracking may not need this if this register
+				 * doesn't participate in pointer adjustment.
+				 * Forward propagation of precise flag is not
+				 * necessary either. This mark is only to stop
+				 * backtracking. Any register that contributed
+				 * to const 0 was marked precise before spill.
+				 */
+				state->regs[value_regno].precise = true;
 			} else {
 				/* have read misc data from the stack */
 				mark_reg_unknown(env, state->regs, value_regno);
@@ -2925,6 +3365,8 @@ static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
 		err = check_helper_mem_access(env, regno - 1,
 					      reg->umax_value,
 					      zero_size_allowed, meta);
+		if (!err)
+			err = mark_chain_precision(env, regno);
 	} else if (arg_type_is_int_ptr(arg_type)) {
 		int size = int_ptr_type_to_size(arg_type);
 
@@ -4361,6 +4803,7 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
 	struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg;
 	struct bpf_reg_state *ptr_reg = NULL, off_reg = {0};
 	u8 opcode = BPF_OP(insn->code);
+	int err;
 
 	dst_reg = &regs[insn->dst_reg];
 	src_reg = NULL;
@@ -4387,11 +4830,17 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
 				 * This is legal, but we have to reverse our
 				 * src/dest handling in computing the range
 				 */
+				err = mark_chain_precision(env, insn->dst_reg);
+				if (err)
+					return err;
 				return adjust_ptr_min_max_vals(env, insn,
 							       src_reg, dst_reg);
 			}
 		} else if (ptr_reg) {
 			/* pointer += scalar */
+			err = mark_chain_precision(env, insn->src_reg);
+			if (err)
+				return err;
 			return adjust_ptr_min_max_vals(env, insn,
 						       dst_reg, src_reg);
 		}
@@ -5348,6 +5797,13 @@ static int check_cond_jmp_op(struct bpf_verifier_env *env,
 		 tnum_is_const(src_reg->var_off))
 		pred = is_branch_taken(dst_reg, src_reg->var_off.value,
 				       opcode, is_jmp32);
+	if (pred >= 0) {
+		err = mark_chain_precision(env, insn->dst_reg);
+		if (BPF_SRC(insn->code) == BPF_X && !err)
+			err = mark_chain_precision(env, insn->src_reg);
+		if (err)
+			return err;
+	}
 	if (pred == 1) {
 		/* only follow the goto, ignore fall-through */
 		*insn_idx += insn->off;
@@ -5825,6 +6281,11 @@ static int check_cfg(struct bpf_verifier_env *env)
 				goto peek_stack;
 			else if (ret < 0)
 				goto err_free;
+			/* unconditional jmp is not a good pruning point,
+			 * but it's marked, since backtracking needs
+			 * to record jmp history in is_state_visited().
+			 */
+			init_explored_state(env, t + insns[t].off + 1);
 			/* tell verifier to check for equivalent states
 			 * after every call and jump
 			 */
@@ -6325,6 +6786,8 @@ static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur,
 	switch (rold->type) {
 	case SCALAR_VALUE:
 		if (rcur->type == SCALAR_VALUE) {
+			if (!rold->precise && !rcur->precise)
+				return true;
 			/* new val must satisfy old val knowledge */
 			return range_within(rold, rcur) &&
 			       tnum_in(rold->var_off, rcur->var_off);
@@ -6675,6 +7138,7 @@ static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
 	int i, j, err, states_cnt = 0;
 	bool add_new_state = false;
 
+	cur->last_insn_idx = env->prev_insn_idx;
 	if (!env->insn_aux_data[insn_idx].prune_point)
 		/* this 'insn_idx' instruction wasn't marked, so we will not
 		 * be doing state search here
@@ -6791,10 +7255,10 @@ static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
 		env->max_states_per_insn = states_cnt;
 
 	if (!env->allow_ptr_leaks && states_cnt > BPF_COMPLEXITY_LIMIT_STATES)
-		return 0;
+		return push_jmp_history(env, cur);
 
 	if (!add_new_state)
-		return 0;
+		return push_jmp_history(env, cur);
 
 	/* There were no equivalent states, remember the current one.
 	 * Technically the current state is not proven to be safe yet,
@@ -6824,7 +7288,10 @@ static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
 	new->insn_idx = insn_idx;
 	WARN_ONCE(new->branches != 1,
 		  "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx);
+
 	cur->parent = new;
+	cur->first_insn_idx = insn_idx;
+	clear_jmp_history(cur);
 	new_sl->next = *explored_state(env, insn_idx);
 	*explored_state(env, insn_idx) = new_sl;
 	/* connect new state to parentage chain. Current frame needs all
@@ -6904,6 +7371,7 @@ static int do_check(struct bpf_verifier_env *env)
 	struct bpf_reg_state *regs;
 	int insn_cnt = env->prog->len;
 	bool do_print_state = false;
+	int prev_insn_idx = -1;
 
 	env->prev_linfo = NULL;
 
@@ -6929,6 +7397,7 @@ static int do_check(struct bpf_verifier_env *env)
 		u8 class;
 		int err;
 
+		env->prev_insn_idx = prev_insn_idx;
 		if (env->insn_idx >= insn_cnt) {
 			verbose(env, "invalid insn idx %d insn_cnt %d\n",
 				env->insn_idx, insn_cnt);
@@ -7001,6 +7470,7 @@ static int do_check(struct bpf_verifier_env *env)
 
 		regs = cur_regs(env);
 		env->insn_aux_data[env->insn_idx].seen = true;
+		prev_insn_idx = env->insn_idx;
 
 		if (class == BPF_ALU || class == BPF_ALU64) {
 			err = check_alu_op(env, insn);
@@ -7174,7 +7644,6 @@ static int do_check(struct bpf_verifier_env *env)
 
 				if (state->curframe) {
 					/* exit from nested function */
-					env->prev_insn_idx = env->insn_idx;
 					err = prepare_func_exit(env, &env->insn_idx);
 					if (err)
 						return err;
@@ -7206,7 +7675,7 @@ static int do_check(struct bpf_verifier_env *env)
 					return err;
 process_bpf_exit:
 				update_branch_counts(env, env->cur_state);
-				err = pop_stack(env, &env->prev_insn_idx,
+				err = pop_stack(env, &prev_insn_idx,
 						&env->insn_idx);
 				if (err < 0) {
 					if (err != -ENOENT)