eBPF-verifier源码分析

bpf_check

在bpf_check函数中，会按照顺序依次做如下check：

check_subprogs

先看check_subprogs：

static int check_subprogs(struct bpf_verifier_env *env)
{
	int i, ret, subprog_start, subprog_end, off, cur_subprog = 0;
	struct bpf_subprog_info *subprog = env->subprog_info;
	struct bpf_insn *insn = env->prog->insnsi;
	int insn_cnt = env->prog->len;

	/* Add entry function. */
	ret = add_subprog(env, 0);
	if (ret < 0)
		return ret;

	/* determine subprog starts. The end is one before the next starts */
	for (i = 0; i < insn_cnt; i++) {
		if (insn[i].code != (BPF_JMP | BPF_CALL))
			continue;
		if (insn[i].src_reg != BPF_PSEUDO_CALL)
			continue;
		if (!env->bpf_capable) {
			verbose(env,
				"function calls to other bpf functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n");
			return -EPERM;
		}
		ret = add_subprog(env, i + insn[i].imm + 1); //所以说如果多个点对同一个函数进行调用，那么就会多次检查这个函数，为了提升效率就有必要进行prune的判断；
		if (ret < 0)
			return ret;
	}

	/* Add a fake 'exit' subprog which could simplify subprog iteration
	 * logic. 'subprog_cnt' should not be increased.
	 */
	subprog[env->subprog_cnt].start = insn_cnt;

	if (env->log.level & BPF_LOG_LEVEL2)
		for (i = 0; i < env->subprog_cnt; i++)
			verbose(env, "func#%d @%d\n", i, subprog[i].start);

	/* now check that all jumps are within the same subprog */
	subprog_start = subprog[cur_subprog].start;
	subprog_end = subprog[cur_subprog + 1].start;
	for (i = 0; i < insn_cnt; i++) {
		u8 code = insn[i].code;

		if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32)
			goto next;
		if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL)
			goto next;
		off = i + insn[i].off + 1;
		if (off < subprog_start || off >= subprog_end) {
			verbose(env, "jump out of range from insn %d to %d\n", i, off);
			return -EINVAL;
		}
next:
		if (i == subprog_end - 1) {
			/* to avoid fall-through from one subprog into another
			 * the last insn of the subprog should be either exit
			 * or unconditional jump back
			 */
			if (code != (BPF_JMP | BPF_EXIT) &&
			    code != (BPF_JMP | BPF_JA)) {
				verbose(env, "last insn is not an exit or jmp\n");
				return -EINVAL;
			}
			subprog_start = subprog_end;
			cur_subprog++;
			if (cur_subprog < env->subprog_cnt)
				subprog_end = subprog[cur_subprog + 1].start;
		}
	}
	return 0;
}

先是对整个程序进行逐条指令检查，遇到 jmp|call 之后就要add_subprog；（注意最开始把主程序入口0add进去了）具体来看，感觉就是根据jmp|call划分基本块；

然后针对每一个subprog（基本块）为一个基本单位，逐条指令进行分析，如果不是跳转类指令（jmp、call）就跳过，分析下一条指令；

如果是，则判断目标地址有没有超出当前基本块的范围；

最后在每个subprog的结尾处判断其结尾指令是否是exit或者跳转；

check_cfg

从头开始分析指令，维护一个insn_stack用于DFS搜索，维护一个insn_state用于表示每一条指令的状态；然后开始DFS搜索，每次取栈顶元素进行分析，每一类跳转指令都要考虑将后续的点进行push_insn，然后如果之前discover过，那么就要标记这个地方是可考虑prune的。

/* non-recursive depth-first-search to detect loops in BPF program
 * loop == back-edge in directed graph
 */
static int check_cfg(struct bpf_verifier_env *env)
{
	struct bpf_insn *insns = env->prog->insnsi;
	int insn_cnt = env->prog->len;
	int *insn_stack, *insn_state;
	int ret = 0;
	int i, t;

	insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
	if (!insn_state)
		return -ENOMEM;

	insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
	if (!insn_stack) {
		kvfree(insn_state);
		return -ENOMEM;
	}

	insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
	insn_stack[0] = 0; /* 0 is the first instruction */
	env->cfg.cur_stack = 1;

peek_stack:
	if (env->cfg.cur_stack == 0) //如果栈为空，就去检查所有的state是否都被探索过了
		goto check_state;
	t = insn_stack[env->cfg.cur_stack - 1];//取栈顶元素

	if (BPF_CLASS(insns[t].code) == BPF_JMP ||
	    BPF_CLASS(insns[t].code) == BPF_JMP32) {
		u8 opcode = BPF_OP(insns[t].code);

		if (opcode == BPF_EXIT) {
			goto mark_explored;  //将当前指令状态标记为 EXPLORED
		} else if (opcode == BPF_CALL) {
			ret = push_insn(t, t + 1, FALLTHROUGH, env, false);
			if (ret == 1)
				goto peek_stack;
			else if (ret < 0)
				goto err_free;
			if (t + 1 < insn_cnt) //ret == 0 说明call了一个已经discover过的函数，所以要标记一下这个点可以考虑优化掉
				init_explored_state(env, t + 1);
			if (insns[t].src_reg == BPF_PSEUDO_CALL) {
				init_explored_state(env, t);
				ret = push_insn(t, t + insns[t].imm + 1, BRANCH,
						env, false);
				if (ret == 1)
					goto peek_stack;
				else if (ret < 0)
					goto err_free;
			}
		} else if (opcode == BPF_JA) {
			if (BPF_SRC(insns[t].code) != BPF_K) {
				ret = -EINVAL;
				goto err_free;
			}
			/* unconditional jump with single edge */
			ret = push_insn(t, t + insns[t].off + 1,
					FALLTHROUGH, env, true);
			if (ret == 1)
				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
			 */
			if (t + 1 < insn_cnt)
				init_explored_state(env, t + 1);
		} else {
			/* conditional jump with two edges */
			init_explored_state(env, t);  //为什么直接调用？
			ret = push_insn(t, t + 1, FALLTHROUGH, env, true); //失败的情况
			if (ret == 1)
				goto peek_stack;
			else if (ret < 0)
				goto err_free;

			ret = push_insn(t, t + insns[t].off + 1, BRANCH, env, true); //成功的情况
			if (ret == 1)
				goto peek_stack;
			else if (ret < 0)
				goto err_free;
		}
	} else {
		/* all other non-branch instructions with single
		 * fall-through edge
		 */
		ret = push_insn(t, t + 1, FALLTHROUGH, env, false); //非跳转指令直接push进去；
		if (ret == 1)
			goto peek_stack;
		else if (ret < 0)
			goto err_free;
	}

mark_explored:
	insn_state[t] = EXPLORED;
	if (env->cfg.cur_stack-- <= 0) {
		verbose(env, "pop stack internal bug\n");
		ret = -EFAULT;
		goto err_free;
	}
	goto peek_stack;

check_state:
	for (i = 0; i < insn_cnt; i++) {
		if (insn_state[i] != EXPLORED) {
			verbose(env, "unreachable insn %d\n", i);
			ret = -EINVAL;
			goto err_free;
		}
	}
	ret = 0; /* cfg looks good */

err_free:
	kvfree(insn_state);
	kvfree(insn_stack);
	env->cfg.insn_state = env->cfg.insn_stack = NULL;
	return ret;
}

调用push_insn如果返回0，表示遇到了一个DISCOVERED的地址，那么就要调用init_explored_state，标记为可考虑优化；

do_check_subprogs

static int do_check_subprogs(struct bpf_verifier_env *env)
{
	struct bpf_prog_aux *aux = env->prog->aux;
	int i, ret;

	if (!aux->func_info)
		return 0;

	for (i = 1; i < env->subprog_cnt; i++) {
		if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL)
			continue;
		env->insn_idx = env->subprog_info[i].start;
		WARN_ON_ONCE(env->insn_idx == 0);
		ret = do_check_common(env, i);
		if (ret) {
			return ret;
		} else if (env->log.level & BPF_LOG_LEVEL) {
			verbose(env,
				"Func#%d is safe for any args that match its prototype\n",
				i);
		}
	}
	return 0;
}

do_check_common

遍历所有的subprog，然后调用do_check_common：

static int do_check_common(struct bpf_verifier_env *env, int subprog)
{
	bool pop_log = !(env->log.level & BPF_LOG_LEVEL2);
	struct bpf_verifier_state *state;
	struct bpf_reg_state *regs;
	int ret, i;

	env->prev_linfo = NULL;
	env->pass_cnt++;

	state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL);
	if (!state)
		return -ENOMEM;
	state->curframe = 0;
	state->speculative = false;
	state->branches = 1;
	state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL);
	if (!state->frame[0]) {
		kfree(state);
		return -ENOMEM;
	}
	env->cur_state = state;
	init_func_state(env, state->frame[0],
			BPF_MAIN_FUNC /* callsite */,
			0 /* frameno */,
			subprog);

	regs = state->frame[state->curframe]->regs;
	if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) {
		ret = btf_prepare_func_args(env, subprog, regs);
		if (ret)
			goto out;
		for (i = BPF_REG_1; i <= BPF_REG_5; i++) {
			if (regs[i].type == PTR_TO_CTX)
				mark_reg_known_zero(env, regs, i);
			else if (regs[i].type == SCALAR_VALUE)
				mark_reg_unknown(env, regs, i);  //参数传递，什么都有可能
		}
	} else {
		/* 1st arg to a function */
		regs[BPF_REG_1].type = PTR_TO_CTX;
		mark_reg_known_zero(env, regs, BPF_REG_1);
		ret = btf_check_func_arg_match(env, subprog, regs);
		if (ret == -EFAULT)
			/* unlikely verifier bug. abort.
			 * ret == 0 and ret < 0 are sadly acceptable for
			 * main() function due to backward compatibility.
			 * Like socket filter program may be written as:
			 * int bpf_prog(struct pt_regs *ctx)
			 * and never dereference that ctx in the program.
			 * 'struct pt_regs' is a type mismatch for socket
			 * filter that should be using 'struct __sk_buff'.
			 */
			goto out;
	}

	ret = do_check(env);
out:
	/* check for NULL is necessary, since cur_state can be freed inside
	 * do_check() under memory pressure.
	 */
	if (env->cur_state) {
		free_verifier_state(env->cur_state, true);
		env->cur_state = NULL;
	}
	while (!pop_stack(env, NULL, NULL, false));
	if (!ret && pop_log)
		bpf_vlog_reset(&env->log, 0);
	free_states(env);
	if (ret)
		/* clean aux data in case subprog was rejected */
		sanitize_insn_aux_data(env);
	return ret;
}

首先准备一个state；然后根据函数类型和参数类型设置reg；

do_check

之后调用do_check：

static int do_check(struct bpf_verifier_env *env)
{
	bool pop_log = !(env->log.level & BPF_LOG_LEVEL2);
	struct bpf_verifier_state *state = env->cur_state;
	struct bpf_insn *insns = env->prog->insnsi;
	struct bpf_reg_state *regs;
	int insn_cnt = env->prog->len;
	bool do_print_state = false;
	int prev_insn_idx = -1;

	for (;;) {
		struct bpf_insn *insn;
		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);
			return -EFAULT;
		}

		insn = &insns[env->insn_idx];
		class = BPF_CLASS(insn->code);

		if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
			verbose(env,
				"BPF program is too large. Processed %d insn\n",
				env->insn_processed);
			return -E2BIG;
		}

		err = is_state_visited(env, env->insn_idx); /////  如果这个state是visited，就直接prune掉
		if (err < 0)
			return err;
		if (err == 1) {
			/* found equivalent state, can prune the search */
			if (env->log.level & BPF_LOG_LEVEL) {
				if (do_print_state)
					verbose(env, "\nfrom %d to %d%s: safe\n",
						env->prev_insn_idx, env->insn_idx,
						env->cur_state->speculative ?
						" (speculative execution)" : "");
				else
					verbose(env, "%d: safe\n", env->insn_idx); /////
			}
			goto process_bpf_exit;
		}

		if (signal_pending(current))
			return -EAGAIN;

		if (need_resched())
			cond_resched();

		if (env->log.level & BPF_LOG_LEVEL2 ||
		    (env->log.level & BPF_LOG_LEVEL && do_print_state)) {
			if (env->log.level & BPF_LOG_LEVEL2)
				verbose(env, "%d:", env->insn_idx);
			else
				verbose(env, "\nfrom %d to %d%s:",
					env->prev_insn_idx, env->insn_idx,
					env->cur_state->speculative ?
					" (speculative execution)" : "");
			print_verifier_state(env, state->frame[state->curframe]);
			do_print_state = false;
		}

		if (env->log.level & BPF_LOG_LEVEL) {
			const struct bpf_insn_cbs cbs = {
				.cb_print	= verbose,
				.private_data	= env,
			};

			verbose_linfo(env, env->insn_idx, "; ");
			verbose(env, "%d: ", env->insn_idx);
			print_bpf_insn(&cbs, insn, env->allow_ptr_leaks);
		}

		if (bpf_prog_is_dev_bound(env->prog->aux)) {
			err = bpf_prog_offload_verify_insn(env, env->insn_idx,
							   env->prev_insn_idx);
			if (err)
				return err;
		}

		regs = cur_regs(env);
		env->insn_aux_data[env->insn_idx].seen = env->pass_cnt;
		prev_insn_idx = env->insn_idx;

		if (class == BPF_ALU || class == BPF_ALU64) {
			err = check_alu_op(env, insn);
			if (err)
				return err;

		} else if (class == BPF_LDX) {
			enum bpf_reg_type *prev_src_type, src_reg_type;

			/* check for reserved fields is already done */

			/* check src operand */
			err = check_reg_arg(env, insn->src_reg, SRC_OP);
			if (err)
				return err;

			err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
			if (err)
				return err;

			src_reg_type = regs[insn->src_reg].type;

			/* check that memory (src_reg + off) is readable,
			 * the state of dst_reg will be updated by this func
			 */
			err = check_mem_access(env, env->insn_idx, insn->src_reg,
					       insn->off, BPF_SIZE(insn->code),
					       BPF_READ, insn->dst_reg, false);
			if (err)
				return err;

			prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type;

			if (*prev_src_type == NOT_INIT) {
				/* saw a valid insn
				 * dst_reg = *(u32 *)(src_reg + off)
				 * save type to validate intersecting paths
				 */
				*prev_src_type = src_reg_type;

			} else if (reg_type_mismatch(src_reg_type, *prev_src_type)) {
				/* ABuser program is trying to use the same insn
				 * dst_reg = *(u32*) (src_reg + off)
				 * with different pointer types:
				 * src_reg == ctx in one branch and
				 * src_reg == stack|map in some other branch.
				 * Reject it.
				 */
				verbose(env, "same insn cannot be used with different pointers\n");
				return -EINVAL;
			}

		} else if (class == BPF_STX) {
			enum bpf_reg_type *prev_dst_type, dst_reg_type;

			if (BPF_MODE(insn->code) == BPF_XADD) {
				err = check_xadd(env, env->insn_idx, insn);
				if (err)
					return err;
				env->insn_idx++;
				continue;
			}

			/* check src1 operand */
			err = check_reg_arg(env, insn->src_reg, SRC_OP);
			if (err)
				return err;
			/* check src2 operand */
			err = check_reg_arg(env, insn->dst_reg, SRC_OP);
			if (err)
				return err;

			dst_reg_type = regs[insn->dst_reg].type;

			/* check that memory (dst_reg + off) is writeable */
			err = check_mem_access(env, env->insn_idx, insn->dst_reg,
					       insn->off, BPF_SIZE(insn->code),
					       BPF_WRITE, insn->src_reg, false);
			if (err)
				return err;

			prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type;

			if (*prev_dst_type == NOT_INIT) {
				*prev_dst_type = dst_reg_type;
			} else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) {
				verbose(env, "same insn cannot be used with different pointers\n");
				return -EINVAL;
			}

		} else if (class == BPF_ST) {
			if (BPF_MODE(insn->code) != BPF_MEM ||
			    insn->src_reg != BPF_REG_0) {
				verbose(env, "BPF_ST uses reserved fields\n");
				return -EINVAL;
			}
			/* check src operand */
			err = check_reg_arg(env, insn->dst_reg, SRC_OP);
			if (err)
				return err;

			if (is_ctx_reg(env, insn->dst_reg)) {
				verbose(env, "BPF_ST stores into R%d %s is not allowed\n",
					insn->dst_reg,
					reg_type_str[reg_state(env, insn->dst_reg)->type]);
				return -EACCES;
			}

			/* check that memory (dst_reg + off) is writeable */
			err = check_mem_access(env, env->insn_idx, insn->dst_reg,
					       insn->off, BPF_SIZE(insn->code),
					       BPF_WRITE, -1, false);
			if (err)
				return err;

		} else if (class == BPF_JMP || class == BPF_JMP32) {
			u8 opcode = BPF_OP(insn->code);

			env->jmps_processed++;
			if (opcode == BPF_CALL) {
				if (BPF_SRC(insn->code) != BPF_K ||
				    insn->off != 0 ||
				    (insn->src_reg != BPF_REG_0 &&
				     insn->src_reg != BPF_PSEUDO_CALL) ||
				    insn->dst_reg != BPF_REG_0 ||
				    class == BPF_JMP32) {
					verbose(env, "BPF_CALL uses reserved fields\n");
					return -EINVAL;
				}

				if (env->cur_state->active_spin_lock &&
				    (insn->src_reg == BPF_PSEUDO_CALL ||
				     insn->imm != BPF_FUNC_spin_unlock)) {
					verbose(env, "function calls are not allowed while holding a lock\n");
					return -EINVAL;
				}
				if (insn->src_reg == BPF_PSEUDO_CALL)
					err = check_func_call(env, insn, &env->insn_idx);
				else
					err = check_helper_call(env, insn->imm, env->insn_idx);
				if (err)
					return err;

			} else if (opcode == BPF_JA) {
				if (BPF_SRC(insn->code) != BPF_K ||
				    insn->imm != 0 ||
				    insn->src_reg != BPF_REG_0 ||
				    insn->dst_reg != BPF_REG_0 ||
				    class == BPF_JMP32) {
					verbose(env, "BPF_JA uses reserved fields\n");
					return -EINVAL;
				}

				env->insn_idx += insn->off + 1;
				continue;

			} else if (opcode == BPF_EXIT) {
				if (BPF_SRC(insn->code) != BPF_K ||
				    insn->imm != 0 ||
				    insn->src_reg != BPF_REG_0 ||
				    insn->dst_reg != BPF_REG_0 ||
				    class == BPF_JMP32) {
					verbose(env, "BPF_EXIT uses reserved fields\n");
					return -EINVAL;
				}

				if (env->cur_state->active_spin_lock) {
					verbose(env, "bpf_spin_unlock is missing\n");
					return -EINVAL;
				}

				if (state->curframe) {
					/* exit from nested function */
					err = prepare_func_exit(env, &env->insn_idx);
					if (err)
						return err;
					do_print_state = true;
					continue;
				}

				err = check_reference_leak(env);
				if (err)
					return err;

				err = check_return_code(env);
				if (err)
					return err;
process_bpf_exit:
				update_branch_counts(env, env->cur_state);
				err = pop_stack(env, &prev_insn_idx,
						&env->insn_idx, pop_log);
				if (err < 0) {
					if (err != -ENOENT)
						return err;
					break;
				} else {
					do_print_state = true;
					continue;
				}
			} else {
				err = check_cond_jmp_op(env, insn, &env->insn_idx);
				if (err)
					return err;
			}
		} else if (class == BPF_LD) {
			u8 mode = BPF_MODE(insn->code);

			if (mode == BPF_ABS || mode == BPF_IND) {
				err = check_ld_abs(env, insn);
				if (err)
					return err;

			} else if (mode == BPF_IMM) {
				err = check_ld_imm(env, insn);
				if (err)
					return err;

				env->insn_idx++;
				env->insn_aux_data[env->insn_idx].seen = env->pass_cnt;
			} else {
				verbose(env, "invalid BPF_LD mode\n");
				return -EINVAL;
			}
		} else {
			verbose(env, "unknown insn class %d\n", class);
			return -EINVAL;
		}

		env->insn_idx++;
	}

	return 0;
}

do_check_common是针对每个subprog只调用一次的，但后在do_check_common中调用了do_check，即do_check也是对于每个subprog都只调用一次的；

在do_check内部，会从头开始遍历依次所有指令，首先进入is_state_visited函数，这个函数仅对prune_point有效，不是prune_point的会立即退出；然后这里的遍历指令不是简单的顺序遍历就完事了，这要充分考虑多种情况：条件跳转、直接跳转、exit；

在env中会维护一个stack，然后通过env->stack_size表示栈中元素的个数，env->head表示栈头，然后通过单链表的形式组织elem们进入栈；

当do_check遇到了条件跳转会进入到check_cond_jmp_op函数中，如果确定两个分支都会走到，则会先通过mark_chain_precision函数来修改precise标记，然后将true分支压入栈中，

上面这段代码，走到exit时会走下来，出栈一个指令继续分析；比如is_state_visited返回为true，那么就调到这里，直接分析下一个；同样，如果是条件跳转的false分支分析完了，进入到exit了，那么下面pop出来的就是true分支了；

is_state_visited

static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
{
	struct bpf_verifier_state_list *new_sl;
	struct bpf_verifier_state_list *sl, **pprev;
	struct bpf_verifier_state *cur = env->cur_state, *new;
	int i, j, err, states_cnt = 0;
	bool add_new_state = env->test_state_freq ? true : false;

	cur->last_insn_idx = env->prev_insn_idx;
	if (!env->insn_aux_data[insn_idx].prune_point) //必须是可prune的点才可以继续分析
		/* this 'insn_idx' instruction wasn't marked, so we will not
		 * be doing state search here
		 */
		return 0;

	/* bpf progs typically have pruning point every 4 instructions
	 * http://vger.kernel.org/bpfconf2019.html#session-1
	 * Do not add new state for future pruning if the verifier hasn't seen
	 * at least 2 jumps and at least 8 instructions.
	 * This heuristics helps decrease 'total_states' and 'peak_states' metric.
	 * In tests that amounts to up to 50% reduction into total verifier
	 * memory consumption and 20% verifier time speedup.
	 */
	if (env->jmps_processed - env->prev_jmps_processed >= 2 &&
	    env->insn_processed - env->prev_insn_processed >= 8)
		add_new_state = true;

	pprev = explored_state(env, insn_idx); //直接拿到一个链表，最开始可能直接就是空的；
	sl = *pprev;

	clean_live_states(env, insn_idx, cur);
	
	while (sl) {
		states_cnt++;
		if (sl->state.insn_idx != insn_idx)
			goto next;
		if (sl->state.branches) {
			if (states_maybe_looping(&sl->state, cur) &&
			    states_equal(env, &sl->state, cur)) {
				verbose_linfo(env, insn_idx, "; ");
				verbose(env, "infinite loop detected at insn %d\n", insn_idx);
				return -EINVAL;
			}
			/* if the verifier is processing a loop, avoid adding new state
			 * too often, since different loop iterations have distinct
			 * states and may not help future pruning.
			 * This threshold shouldn't be too low to make sure that
			 * a loop with large bound will be rejected quickly.
			 * The most abusive loop will be:
			 * r1 += 1
			 * if r1 < 1000000 goto pc-2
			 * 1M insn_procssed limit / 100 == 10k peak states.
			 * This threshold shouldn't be too high either, since states
			 * at the end of the loop are likely to be useful in pruning.
			 */
			if (env->jmps_processed - env->prev_jmps_processed < 20 &&
			    env->insn_processed - env->prev_insn_processed < 100)
				add_new_state = false;
			goto miss;
		}
		if (states_equal(env, &sl->state, cur)) {
			sl->hit_cnt++;
			/* reached equivalent register/stack state,
			 * prune the search.
			 * Registers read by the continuation are read by us.
			 * If we have any write marks in env->cur_state, they
			 * will prevent corresponding reads in the continuation
			 * from reaching our parent (an explored_state).  Our
			 * own state will get the read marks recorded, but
			 * they'll be immediately forgotten as we're pruning
			 * this state and will pop a new one.
			 */
			err = propagate_liveness(env, &sl->state, cur);

			/* if previous state reached the exit with precision and
			 * current state is equivalent to it (except precsion marks)
			 * the precision needs to be propagated back in
			 * the current state.
			 */
			err = err ? : push_jmp_history(env, cur);
			err = err ? : propagate_precision(env, &sl->state);
			if (err)
				return err;
			return 1;
		}
miss:
		/* when new state is not going to be added do not increase miss count.
		 * Otherwise several loop iterations will remove the state
		 * recorded earlier. The goal of these heuristics is to have
		 * states from some iterations of the loop (some in the beginning
		 * and some at the end) to help pruning.
		 */
		if (add_new_state)
			sl->miss_cnt++;
		/* heuristic to determine whether this state is beneficial
		 * to keep checking from state equivalence point of view.
		 * Higher numbers increase max_states_per_insn and verification time,
		 * but do not meaningfully decrease insn_processed.
		 */
		if (sl->miss_cnt > sl->hit_cnt * 3 + 3) {
			/* the state is unlikely to be useful. Remove it to
			 * speed up verification
			 */
			*pprev = sl->next;
			if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) {
				u32 br = sl->state.branches;

				WARN_ONCE(br,
					  "BUG live_done but branches_to_explore %d\n",
					  br);
				free_verifier_state(&sl->state, false);
				kfree(sl);
				env->peak_states--;
			} else {
				/* cannot free this state, since parentage chain may
				 * walk it later. Add it for free_list instead to
				 * be freed at the end of verification
				 */
				sl->next = env->free_list;
				env->free_list = sl;
			}
			sl = *pprev;
			continue;
		}
next:
		pprev = &sl->next;
		sl = *pprev;
	}

	if (env->max_states_per_insn < states_cnt)
		env->max_states_per_insn = states_cnt;

	if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES)
		return push_jmp_history(env, cur);

	if (!add_new_state)
		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,
	 * but it will either reach outer most bpf_exit (which means it's safe)
	 * or it will be rejected. When there are no loops the verifier won't be
	 * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx)
	 * again on the way to bpf_exit.
	 * When looping the sl->state.branches will be > 0 and this state
	 * will not be considered for equivalence until branches == 0.
	 */
	new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL);
	if (!new_sl)
		return -ENOMEM;
	env->total_states++;
	env->peak_states++;
	env->prev_jmps_processed = env->jmps_processed;
	env->prev_insn_processed = env->insn_processed;

	/* add new state to the head of linked list */
	new = &new_sl->state;
	err = copy_verifier_state(new, cur);  //创建一个只有当前state的state_list
	if (err) {
		free_verifier_state(new, false);
		kfree(new_sl);
		return err;
	}
	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); //再次获取这个链表，并采用头插法链接进去
	*explored_state(env, insn_idx) = new_sl;
	/* connect new state to parentage chain. Current frame needs all
	 * registers connected. Only r6 - r9 of the callers are alive (pushed
	 * to the stack implicitly by JITs) so in callers' frames connect just
	 * r6 - r9 as an optimization. Callers will have r1 - r5 connected to
	 * the state of the call instruction (with WRITTEN set), and r0 comes
	 * from callee with its full parentage chain, anyway.
	 */
	/* clear write marks in current state: the writes we did are not writes
	 * our child did, so they don't screen off its reads from us.
	 * (There are no read marks in current state, because reads always mark
	 * their parent and current state never has children yet.  Only
	 * explored_states can get read marks.)
	 */
	for (j = 0; j <= cur->curframe; j++) {
		for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++)
			cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i];
		for (i = 0; i < BPF_REG_FP; i++)
			cur->frame[j]->regs[i].live = REG_LIVE_NONE;
	}

	/* all stack frames are accessible from callee, clear them all */
	for (j = 0; j <= cur->curframe; j++) {
		struct bpf_func_state *frame = cur->frame[j];
		struct bpf_func_state *newframe = new->frame[j];

		for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) {
			frame->stack[i].spilled_ptr.live = REG_LIVE_NONE;
			frame->stack[i].spilled_ptr.parent =
						&newframe->stack[i].spilled_ptr;
		}
	}
	return 0;
}

对于is_state_visited这个函数的理解，重点在于bpf_verifier_state_list这个结构，首先在env中存在一个名为explored_states的bpf_verifier_state_list类型数组，该数组的每一项要对应一个调用者和一个被调用者，因此搜索空间非常巨大，所以要采用哈希表的形式来存储；

在最开始的情况下，获取到一个空链表项，直接跳过前面的代码，看add_new，先分配一个新的链表项new_sl，然后将当前的state情况copy进去，链进链表中：

进入到is_state_visited函数中，首先就要获取这样一个链表，（当然它可能为空），然后开始遍历其中的每一个state，因为这些state其实都是相同的调用者和相同的被调用者了；

copy

看一下这个new_sl是如何创建与初始化的：

传递new进去copy：

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]);
		dst_state->frame[i] = NULL;
	}
	dst_state->speculative = src->speculative;
	dst_state->curframe = src->curframe;
	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) {
			dst = kzalloc(sizeof(*dst), GFP_KERNEL);
			if (!dst)
				return -ENOMEM;
			dst_state->frame[i] = dst;
		}
		err = copy_func_state(dst, src->frame[i]);
		if (err)
			return err;
	}
	return 0;
}

之后回来继续：

关于frame的定义如下：

states_equal

好下面看如何判断两个state相同：

static int is_state_visited(env, env->insn_idx){
    pprev = explored_state(env, insn_idx);
	sl = *pprev;
    //sl 遍历每一个相同的调用者和被调用者的state_list
	states_equal(env, &sl->state, cur){
        //old = sl->state, cur = cur
        for (i = 0; i <= old->curframe; i++) {
			func_states_equal(old->frame[i], cur->frame[i]){
                //old = old-frame[i], cur = cur->frame[i]
                regsafe(&old->regs[i], &cur->regs[i], idmap){
                    //rold = old->regs[i], rcur = &cur->regs[i]
                    switch (rold->type) {
						case SCALAR_VALUE:
                            if (!rold->precise && !rcur->precise)
								return true;
                }
            }
			
		}
    }
	
}

总结

bpf_check(){
	check_subprogs();//根据jmp|call划分基本块，加入到subprog数组中
	check_cfg();//对所有指令DFS搜索，探索不可达指令，标记prune_point
	do_check_common(){//检查某一个subprog是否可以优化掉
		do_check(){
            pprev = explored_state(env, insn_idx);
			sl = *pprev;
			is_state_visited(){//遍历pprev链表，检查是否是visit过，
                //先获取state_list，callsite^callee->hashtab
                //逐个state进行比较
                while(){
                    states_equal(env, &sl->state, cur){
        			//old = sl->state, cur = cur
        			for (i = 0; i <= old->curframe; i++) {
						func_states_equal(old->frame[i], cur->frame[i]){
                			//old = old-frame[i], cur = cur->frame[i]
                			regsafe(&old->regs[i], &cur->regs[i], idmap){
                    			//rold = old->regs[i], rcur = &cur->regs[i]
                    			switch (rold->type) {
									case SCALAR_VALUE:
                                        //sl->state->frame[i]->regs->precise
                            			if (!rold->precise && !rcur->precise)
											return true;
                				}
            				}
			
						}
    				}
                }
				//添加新的sl到链表中 copy
                new = &new_sl->state;
				err = copy_verifier_state(new, cur){
                    //dst_state=new, src=cur
                    dst = dst_state->frame[i];
					if (!dst) {
						dst = kzalloc(sizeof(*dst), GFP_KERNEL);
						if (!dst)
							return -ENOMEM;
						dst_state->frame[i] = dst;
					}
                    err = copy_func_state(dst, src->frame[i]){
                        //...
                        memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs));
                        //...
                    }
                }
                
			}
            //对所有指令进行switch，逐个检查：
            switch(op){
                case JMP:
                    //各种情况处理
                    check_cond_jmp_op(){//条件跳转，会对寄存器的precise进行标记
                        if(){//如果存在分支
                            mark_chain_precision(env, insn->dst_reg);//当前env的cur_state，然后找到curframe，将其regs[regno]设置为相应的precision等；后需要回溯的
                            //env->cur_state->frame[st->curframe]->reg[regno]
                            //...
                            mark_chain_precision(env, insn->src_reg);
                        }
                    }
            }
		}
	}
}

precise的理解

首先在regsafe的检查中是先看reg的所有估计值和var_off的，如果完全一样直接认为是safe的（除了stack还要再检查一个frameno）；如果存在不一样的地方，则检查type，这个precise是只针对标量有检查意义的；（因此，可以将precise理解为一个分支优化时对scalar的检查标准，如果是precise=false说明不必较真）

执行do_check_common->do_check->is_state_visited发现没有visit过，就要将当前state加入到链表中，这个时候就有了这次跳转（刚jmp，还没有执行任何新的指令）时的状态（可以理解为拍了一个快照一样保存到了sl中）；

至于在对具体的指令进行分析，例如在条件跳转时要给src_reg标注精确，这个时候到了跳转进新的基本块的时候，这个标注就会带来，然后就会对src_reg的值进行严格检查，不精确就不让优化掉；

对于frame的理解

首先看eBPF的官方文档：

https://www.kernel.org/doc/html/latest/bpf/verifier.html

主要就是说了一个state，如果call一个函数了，那么就会创建一个新的frame，通常frame0都是给自己的，然后新的frame1给被调用者，此时复制frame0的r1r5给frame1作为参数传递，r0/r6r9对frame1来说是null，

看check_func_call函数：

可以看到创建一个新的frame之后，将state的curframe切换到新的栈中，PC也到了被调用函数中了；

state只有subprog才会创建新的（在do_check_common函数中切换env->cur_state）;

而subprog的标准是call | jmp32 以及src_reg == BPF_PSEUDO_CALL; 即eBPF到eBPF的函数调用，区别于辅助函数；所以只有内部的call才会被切分成subprog；

再看prune：而所有的跳转分支则也会标记为prune_point，所以prune_point是subprog的超集；

所以就明朗了，其实 is_state_visited 中的比较、失败后的new_sl的copy、以及__mark_chain_precision都是围绕着frame来的！

只有func_state中才会有regs，bpf_verifier_state中包s含func_state，但是没有regs；

mark_chain_precision调用情况总结

check_stack_write：写栈上写8字节；

static int __mark_chain_precision(struct bpf_verifier_env *env, int regno,
				  int spi)
{
	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 = regno >= 0 ? 1u << regno : 0;
	u64 stack_mask = spi >= 0 ? 1ull << spi : 0;
	bool skip_first = true;
	bool new_marks = false;
	int i, err;

	if (!env->bpf_capable)
		return 0;
///===============找出当前func frame，找出需要标记的reg==================================
	func = st->frame[st->curframe];
	if (regno >= 0) {
		reg = &func->regs[regno];
		if (reg->type != SCALAR_VALUE) {
			WARN_ONCE(1, "backtracing misuse");
			return -EFAULT;
		}
		if (!reg->precise)
			new_marks = true;
		else
			reg_mask = 0;
		reg->precise = true;
	}
///=============设置堆栈precise的情况==================================================
	while (spi >= 0) {
		if (func->stack[spi].slot_type[0] != STACK_SPILL) {
			stack_mask = 0;
			break;
		}
		reg = &func->stack[spi].spilled_ptr;
		if (reg->type != SCALAR_VALUE) {
			stack_mask = 0;
			break;
		}
		if (!reg->precise)
			new_marks = true;
		else
			stack_mask = 0;
		reg->precise = true;
		break;
	}

	if (!new_marks)
		return 0;
	if (!reg_mask && !stack_mask)
		return 0;
//===============================开始回溯================================================
	for (;;) {
		DECLARE_BITMAP(mask, 64);
		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; //跳过第一条指令
			} else {
				err = backtrack_insn(env, i, &reg_mask, &stack_mask);//回溯一条指令，就是我现在加了一个precise=true，可能会对前面其他的指令有影响，所以要回溯
			}
			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);//向前回溯一条指令
			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;
			}
		}//一个state的
		st = st->parent; //向前一个状态
		if (!st)
			break;

		new_marks = false;
		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) {
				reg_mask &= ~(1u << i);
				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) {
				/* the sequence of instructions:
				 * 2: (bf) r3 = r10
				 * 3: (7b) *(u64 *)(r3 -8) = r0
				 * 4: (79) r4 = *(u64 *)(r10 -8)
				 * doesn't contain jmps. It's backtracked
				 * as a single block.
				 * During backtracking insn 3 is not recognized as
				 * stack access, so at the end of backtracking
				 * stack slot fp-8 is still marked in stack_mask.
				 * However the parent state may not have accessed
				 * fp-8 and it's "unallocated" stack space.
				 * In such case fallback to conservative.
				 */
				mark_all_scalars_precise(env, st);
				return 0;
			}

			if (func->stack[i].slot_type[0] != STACK_SPILL) {
				stack_mask &= ~(1ull << i);
				continue;
			}
			reg = &func->stack[i].spilled_ptr;
			if (reg->type != SCALAR_VALUE) {
				stack_mask &= ~(1ull << i);
				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 (!reg_mask && !stack_mask)
			break;
		if (!new_marks)
			break;

		last_idx = st->last_insn_idx;
		first_idx = st->first_insn_idx;
	}
	return 0;
}//就是不断的回溯，只要reg_mask和stack_mask不为0就要继续向上回溯，

在当前state中先回溯，到第一条指令后对reg们进行标记，reg_mask和stack_mask还不为0，就得继续向上回溯parent；