[rest_api][aptos_vm] Prevent running move code on too stale of a state

api/src/basic.rs

@@ -5,23 +5,17 @@ use crate::{
     accept_type::AcceptType,
     context::{api_spawn_blocking, Context},
     generate_error_response, generate_success_response,
-    response::{InternalError, ServiceUnavailableError},
+    response::ServiceUnavailableError,
     ApiTags,
 };
-use anyhow::Context as AnyhowContext;
 use aptos_api_types::AptosErrorCode;
 use poem_openapi::{
     param::Query,
     payload::{Html, Json},
     Object, OpenApi,
 };
 use serde::{Deserialize, Serialize};
-use std::{
-    collections::HashMap,
-    ops::Sub,
-    sync::Arc,
-    time::{Duration, SystemTime, UNIX_EPOCH},
-};
+use std::{collections::HashMap, sync::Arc, time::Duration};
 
 const OPEN_API_HTML: &str = include_str!("../doc/spec.html");
 
@@ -143,22 +137,10 @@ impl BasicApi {
 
         // If we have a duration, check that it's close to the current time, otherwise it's ok
         if let Some(max_skew) = duration_secs.0 {
-            let ledger_timestamp = Duration::from_micros(ledger_info.timestamp());
-            let skew_threshold = SystemTime::now()
-                .sub(Duration::from_secs(max_skew as u64))
-                .duration_since(UNIX_EPOCH)
-                .context("Failed to determine absolute unix time based on given duration")
-                .map_err(|err| {
-                    HealthCheckError::internal_with_code(
-                        err,
-                        AptosErrorCode::InternalError,
-                        &ledger_info,
-                    )
-                })?;
-
-            if ledger_timestamp < skew_threshold {
+            let duration_since_onchain_timestamp = ledger_info.duration_since_timestamp();
+            if duration_since_onchain_timestamp > Duration::from_secs(max_skew as u64) {
                 return Err(HealthCheckError::service_unavailable_with_code(
-                    format!("The latest ledger info timestamp is {:?}, which is beyond the allowed skew ({}s).", ledger_timestamp, max_skew),
+                    format!("The latest ledger info timestamp is {:?} old, which is beyond the allowed skew ({}s).", duration_since_onchain_timestamp, max_skew),
                     AptosErrorCode::HealthCheckFailed,
                     &ledger_info,
                 ));

api/src/context.rs

@@ -64,7 +64,7 @@ use std::{
         atomic::{AtomicU64, AtomicUsize, Ordering},
         Arc, RwLock, RwLockWriteGuard,
     },
-    time::Instant,
+    time::{Duration, Instant},
 };
 
 // Context holds application scope context
@@ -1554,6 +1554,27 @@ impl Context {
     pub fn simulate_txn_stats(&self) -> &FunctionStats {
         &self.simulate_txn_stats
     }
+
+    pub fn validate_ledger_for_move_code_execution(
+        &self,
+        ledger_info: &LedgerInfo,
+    ) -> Result<(), String> {
+        let max_stale = Duration::from_secs(
+            self.node_config
+                .api
+                .max_stale_state_to_execute_move_code_on_s,
+        );
+        let duration_since_onchain_timestamp = ledger_info.duration_since_timestamp();
+        if max_stale < duration_since_onchain_timestamp {
+            Err(format!(
+                "Node is too stale ({}s > {}s limit) to be able to execute move code",
+                duration_since_onchain_timestamp.as_secs(),
+                max_stale.as_secs(),
+            ))
+        } else {
+            Ok(())
+        }
+    }
 }
 
 pub struct GasScheduleCache {

api/src/transactions.rs

@@ -509,6 +509,17 @@ impl TransactionsApi {
         let context = self.context.clone();
         api_spawn_blocking(move || {
             let ledger_info = context.get_latest_ledger_info()?;
+
+            context
+                .validate_ledger_for_move_code_execution(&ledger_info)
+                .map_err(|message| {
+                    SubmitTransactionError::forbidden_with_code(
+                        message,
+                        AptosErrorCode::InternalError,
+                        &ledger_info,
+                    )
+                })?;
+
             let mut signed_transaction = api.get_signed_transaction(&ledger_info, data)?;
 
             // Confirm the simulation filter allows the transaction. We use HashValue::zero()

api/src/view_function.rs

@@ -83,6 +83,16 @@ fn view_request(
     let (ledger_info, requested_version) = context
         .get_latest_ledger_info_and_verify_lookup_version(ledger_version.map(|inner| inner.0))?;
 
+    context
+        .validate_ledger_for_move_code_execution(&ledger_info)
+        .map_err(|message| {
+            BasicErrorWith404::forbidden_with_code(
+                message,
+                AptosErrorCode::InternalError,
+                &ledger_info,
+            )
+        })?;
+
     let state_view = context
         .state_view_at_version(requested_version)
         .map_err(|err| {

api/types/Cargo.toml

@@ -17,6 +17,7 @@ anyhow = { workspace = true }
 aptos-config = { workspace = true }
 aptos-crypto = { workspace = true }
 aptos-framework = { workspace = true }
+aptos-infallible = { workspace = true }
 aptos-logger = { workspace = true }
 aptos-openapi = { workspace = true }
 aptos-resource-viewer = { workspace = true }

api/types/src/ledger_info.rs

@@ -6,6 +6,7 @@ use crate::U64;
 use aptos_types::{chain_id::ChainId, ledger_info::LedgerInfoWithSignatures};
 use poem_openapi::Object as PoemObject;
 use serde::{Deserialize, Serialize};
+use std::time::Duration;
 
 /// The Ledger information representing the current state of the chain
 #[derive(Clone, Debug, Deserialize, Serialize, PartialEq, Eq, PoemObject)]
@@ -75,4 +76,13 @@ impl LedgerInfo {
     pub fn timestamp(&self) -> u64 {
         self.ledger_timestamp.into()
     }
+
+    pub fn timestamp_duration(&self) -> Duration {
+        Duration::from_micros(self.ledger_timestamp.into())
+    }
+
+    pub fn duration_since_timestamp(&self) -> Duration {
+        let now = aptos_infallible::duration_since_epoch();
+        now.saturating_sub(self.timestamp_duration())
+    }
 }

config/src/config/api_config.rs

@@ -84,6 +84,12 @@ pub struct ApiConfig {
     pub simulation_filter: Filter,
     /// Configuration to filter view function requests.
     pub view_filter: ViewFilter,
+    /// Maximum time state (onchain timestamp) is allowed to be stale, for the node
+    /// to respond to requests that require executing move code.
+    /// Preventing running code on extremely stale state avoids triggering code paths
+    /// that should otherwise never be triggered.
+    /// This is on top of above `simulation_filter` and `view_filter`.
+    pub max_stale_state_to_execute_move_code_on_s: u64,
     /// Periodically log stats for view function and simulate transaction usage
     pub periodic_function_stats_sec: Option<u64>,
     /// The time wait_by_hash will wait before returning 404.
@@ -145,6 +151,8 @@ impl Default for ApiConfig {
             wait_by_hash_timeout_ms: 1_000,
             wait_by_hash_poll_interval_ms: 20,
             wait_by_hash_max_active_connections: 100,
+            // by default, don't execute move code, if state is more than 1 day stale
+            max_stale_state_to_execute_move_code_on_s: 24 * 3600,
         }
     }
 }