gyuheon0h/report-unhandled-exceptions

Author: gyuheon0h

bin_tests/src/bin/crashtracker_bin_test.rs

@@ -25,6 +25,7 @@ mod unix {
     use libdd_common::{tag, Endpoint};
     use libdd_crashtracker::{
         self as crashtracker, CrashtrackerConfiguration, CrashtrackerReceiverConfig, Metadata,
+        StackFrame, StackTrace,
     };
 
     const TEST_COLLECTOR_TIMEOUT: Duration = Duration::from_secs(15);
@@ -154,6 +155,31 @@ mod unix {
             "raise_sigill" => raise(Signal::SIGILL)?,
             "raise_sigbus" => raise(Signal::SIGBUS)?,
             "raise_sigsegv" => raise(Signal::SIGSEGV)?,
+            "unhandled_exception" => {
+                let mut stacktrace = StackTrace::new_incomplete();
+                let mut stackframe1 = StackFrame::new();
+                stackframe1.with_ip(1234);
+                stackframe1.with_function("test_function1".to_string());
+                stackframe1.with_file("test_file1".to_string());
+
+                let mut stackframe2 = StackFrame::new();
+                stackframe2.with_ip(5678);
+                stackframe2.with_function("test_function2".to_string());
+                stackframe2.with_file("test_file2".to_string());
+
+                stacktrace.push_frame(stackframe1, true).unwrap();
+                stacktrace.push_frame(stackframe2, true).unwrap();
+
+                stacktrace.set_complete().unwrap();
+
+                crashtracker::report_unhandled_exception(
+                    Some("RuntimeException"),
+                    Some("an exception occured"),
+                    stacktrace,
+                )?;
+
+                process::exit(0);
+            }
             _ => anyhow::bail!("Unexpected crash_typ: {crash_typ}"),
         }
         crashtracker::end_op(crashtracker::OpTypes::ProfilerCollectingSample)?;

bin_tests/src/test_types.rs

@@ -114,6 +114,8 @@ pub enum CrashType {
     RaiseSigBus,
     /// Raise SIGSEGV
     RaiseSigSegv,
+    /// Unhandled Exception
+    UnhandledException,
 }
 
 impl CrashType {
@@ -129,6 +131,7 @@ impl CrashType {
             Self::RaiseSigIll => "raise_sigill",
             Self::RaiseSigBus => "raise_sigbus",
             Self::RaiseSigSegv => "raise_sigsegv",
+            Self::UnhandledException => "unhandled_exception",
         }
     }
 
@@ -138,7 +141,11 @@ impl CrashType {
     pub const fn expects_success(self) -> bool {
         matches!(
             self,
-            Self::KillSigBus | Self::KillSigSegv | Self::RaiseSigBus | Self::RaiseSigSegv
+            Self::KillSigBus
+                | Self::KillSigSegv
+                | Self::RaiseSigBus
+                | Self::RaiseSigSegv
+                | Self::UnhandledException
         )
     }
 
@@ -150,6 +157,7 @@ impl CrashType {
             Self::KillSigAbrt | Self::RaiseSigAbrt => 6,                    // SIGABRT
             Self::KillSigIll | Self::RaiseSigIll => 4,                      // SIGILL
             Self::KillSigBus | Self::RaiseSigBus => 7,                      // SIGBUS
+            Self::UnhandledException => 0,                                  // no signal
         }
     }
 
@@ -160,6 +168,7 @@ impl CrashType {
             Self::KillSigAbrt | Self::RaiseSigAbrt => "SIGABRT",
             Self::KillSigIll | Self::RaiseSigIll => "SIGILL",
             Self::KillSigBus | Self::RaiseSigBus => "SIGBUS",
+            Self::UnhandledException => "Unhandled Exception",
         }
     }
 }
@@ -184,6 +193,7 @@ impl std::str::FromStr for CrashType {
             "raise_sigill" => Ok(Self::RaiseSigIll),
             "raise_sigbus" => Ok(Self::RaiseSigBus),
             "raise_sigsegv" => Ok(Self::RaiseSigSegv),
+            "unhandled_exception" => Ok(Self::UnhandledException),
             _ => Err(format!("Unknown crash type: {}", s)),
         }
     }
@@ -220,5 +230,6 @@ mod tests {
         assert!(!CrashType::KillSigAbrt.expects_success());
         assert!(CrashType::KillSigBus.expects_success());
         assert!(CrashType::KillSigSegv.expects_success());
+        assert!(CrashType::UnhandledException.expects_success());
     }
 }

bin_tests/tests/crashtracker_bin_test.rs

@@ -96,6 +96,39 @@ fn run_standard_crash_test_refactored(
 // These tests below use the new infrastructure but require custom validation logic
 // that doesn't fit the simple macro-generated pattern.
 
+#[test]
+#[cfg_attr(miri, ignore)]
+fn test_crash_tracking_bin_unhandled_exception() {
+    let config = CrashTestConfig::new(
+        BuildProfile::Release,
+        TestMode::DoNothing,
+        CrashType::UnhandledException,
+    );
+    let artifacts = StandardArtifacts::new(config.profile);
+    let artifacts_map = build_artifacts(&artifacts.as_slice()).unwrap();
+
+    let validator: ValidatorFn = Box::new(|payload, _fixtures| {
+        PayloadValidator::new(payload)
+            .validate_counters()?
+            .validate_error_kind("UnixSignal")?
+            .validate_error_message_contains("Process was terminated due to an unhandled exception of type 'RuntimeException'. Message: \"an exception occured\"")?
+            // The two frames emitted in the bin: test_function1 and test_function2
+            .validate_callstack_functions(&["test_function1", "test_function2"])?;
+
+        // Unhandled exceptions have no signal info
+        let sig_info = &payload["sig_info"];
+        assert!(
+            sig_info.is_null()
+                || sig_info.is_object() && sig_info.as_object().is_none_or(|m| m.is_empty()),
+            "Expected no sig_info for unhandled exception, got: {sig_info:?}"
+        );
+
+        Ok(())
+    });
+
+    run_crash_test_with_artifacts(&config, &artifacts_map, &artifacts, validator).unwrap();
+}
+
 #[test]
 #[cfg_attr(miri, ignore)]
 fn test_crash_tracking_bin_runtime_callback_frame() {
@@ -1027,6 +1060,12 @@ fn assert_siginfo_message(sig_info: &Value, crash_typ: &str) {
             assert_eq!(sig_info["si_signo"], libc::SIGILL);
             assert_eq!(sig_info["si_signo_human_readable"], "SIGILL");
         }
+        "unhandled_exception" => {
+            assert!(
+                sig_info.is_null()
+                    || sig_info.is_object() && sig_info.as_object().is_none_or(|m| m.is_empty())
+            );
+        }
         _ => panic!("unexpected crash_typ {crash_typ}"),
     }
 }

libdd-crashtracker/src/collector/collector_manager.rs

@@ -118,8 +118,9 @@ pub(crate) fn run_collector_child(
         config_str,
         metadata_str,
         message_ptr,
-        sig_info,
-        ucontext,
+        Some(sig_info),
+        None, // stacktrace is none; this is collected in the signal handler
+        Some(ucontext),
         ppid,
         crashing_tid,
     );

libdd-crashtracker/src/collector/crash_handler.rs

@@ -8,8 +8,10 @@ use super::receiver_manager::Receiver;
 use super::signal_handler_manager::chain_signal_handler;
 use crate::crash_info::Metadata;
 use crate::shared::configuration::CrashtrackerConfiguration;
+use crate::StackTrace;
 use libc::{c_void, siginfo_t, ucontext_t};
 use libdd_common::timeout::TimeoutManager;
+use std::os::unix::{io::FromRawFd, net::UnixStream};
 use std::panic;
 use std::panic::PanicHookInfo;
 use std::ptr;
@@ -301,6 +303,121 @@ fn handle_posix_signal_impl(
     Ok(())
 }
 
+/// Gets a clone of the current metadata, if set.
+/// Unlike the signal handler path, this reads without consuming the stored value.
+///
+/// SAFETY:
+///     This function must not be called concurrently with `update_metadata`.
+fn get_metadata() -> Option<(crate::crash_info::Metadata, String)> {
+    let ptr = METADATA.load(SeqCst);
+    if ptr.is_null() {
+        None
+    } else {
+        // Safety: ptr was created by Box::into_raw in update_metadata
+        let (metadata, metadata_string) = unsafe { &*ptr };
+        Some((metadata.clone(), metadata_string.clone()))
+    }
+}
+
+/// Gets a clone of the current config, if set.
+/// Unlike the signal handler path, this reads without consuming the stored value.
+///
+/// SAFETY:
+///     This function must not be called concurrently with `update_config`.
+fn get_config() -> Option<(
+    crate::shared::configuration::CrashtrackerConfiguration,
+    String,
+)> {
+    let ptr = CONFIG.load(SeqCst);
+    if ptr.is_null() {
+        None
+    } else {
+        // Safety: ptr was created by Box::into_raw in update_config
+        let (config, config_string) = unsafe { &*ptr };
+        Some((config.clone(), config_string.clone()))
+    }
+}
+
+/// This function is designed to be when a program is at a terminal state
+/// and the application wants to report an unhandled exception to the crashtracker
+///
+/// Preconditions:
+/// - The crashtracker must be started
+/// - The stacktrace must be valid
+///
+/// This function will spawn the receiver process and call an emit function to pipe over
+/// the crash data. We don't use the collector process because we are not in a signal handler
+/// Rather, we call emit_crashreport directly and pipe over data to the receiver
+pub fn report_unhandled_exception(
+    exception_type: Option<&str>,
+    exception_message: Option<&str>,
+    stacktrace: StackTrace,
+) -> Result<(), CrashHandlerError> {
+    let Some((config, config_str)) = get_config() else {
+        return Err(CrashHandlerError::NoConfig);
+    };
+    let Some((_metadata, metadata_str)) = get_metadata() else {
+        return Err(CrashHandlerError::NoMetadata);
+    };
+
+    // Turn crashtracker off to prevent a recursive crash report emission
+    // We do not turn it back on because this function is not intended to be used as
+    // a recurring mechanism to report exceptions. We expect the application to exit
+    // after
+    disable();
+
+    let unix_socket_path = config.unix_socket_path().as_deref().unwrap_or_default();
+
+    let receiver = if unix_socket_path.is_empty() {
+        Receiver::spawn_from_stored_config()?
+    } else {
+        Receiver::from_socket(unix_socket_path)?
+    };
+
+    let timeout_manager = TimeoutManager::new(config.timeout());
+
+    let pid = unsafe { libc::getpid() };
+    let tid = libdd_common::threading::get_current_thread_id() as libc::pid_t;
+
+    let error_type_str = exception_type.unwrap_or("<unknown>");
+    let error_message_str = exception_message.unwrap_or("<no message>");
+    let message = format!(
+        "Process was terminated due to an unhandled exception of type '{error_type_str}'. \
+         Message: \"{error_message_str}\""
+    );
+
+    let message_ptr = Box::into_raw(Box::new(message));
+
+    // Duplicate the socket fd so we can poll for receiver completion after we close the write end.
+    // UnixStream::from_raw_fd takes ownership of uds_fd, so we need a separate fd to poll.
+    let poll_fd = unsafe { libc::dup(receiver.handle.uds_fd) };
+    let receiver_pid = receiver.handle.pid;
+
+    {
+        let mut unix_stream = unsafe { UnixStream::from_raw_fd(receiver.handle.uds_fd) };
+        let _ = super::emitters::emit_crashreport(
+            &mut unix_stream,
+            &config,
+            &config_str,
+            &metadata_str,
+            message_ptr,
+            None,
+            Some(stacktrace),
+            None,
+            pid,
+            tid,
+        );
+        // unix_stream is dropped here, closing the write end of the socket.
+        // This signals EOF to the receiver so it can finish writing the crash report.
+    }
+
+    // Wait for the receiver to signal it is done (POLLHUP on the dup'd fd), then reap it.
+    let finish_handle = super::process_handle::ProcessHandle::new(poll_fd, receiver_pid);
+    finish_handle.finish(&timeout_manager);
+    unsafe { libc::close(poll_fd) };
+
+    Ok(())
+}
 #[cfg(test)]
 mod tests {
     use super::*;