calltrace/

crash_handler.rs

//! Crash Handler Module
//!
//! This module provides crash detection and detailed crash reporting functionality.
//! It installs signal handlers for common crash signals only if no existing handlers are present,
//! and generates comprehensive crash reports including the current call stack.

use once_cell::sync::Lazy;
use std::arch::asm;
use std::collections::HashMap;
use std::ffi::{c_void, CStr};
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::{Arc, Mutex, Once};

use libc::{
    dladdr, sigaction, sigaddset, sigemptyset, siginfo_t, ucontext_t, Dl_info, SA_NODEFER,
    SA_RESTART, SA_SIGINFO, SIGABRT, SIGBUS, SIGFPE, SIGILL, SIGSEGV, SIGTRAP,
};

use crate::error::{CallTraceError, Result};
use crate::json_output::JsonOutputGenerator;
use crate::register_reader::RegisterContext;

/// Global crash handler state
static CRASH_HANDLER_INIT: Once = Once::new();
static CRASH_HANDLER_ENABLED: AtomicBool = AtomicBool::new(false);

/// Storage for original signal handlers
static ORIGINAL_HANDLERS: Lazy<Arc<Mutex<HashMap<i32, libc::sigaction>>>> =
    once_cell::sync::Lazy::new(|| Arc::new(Mutex::new(HashMap::new())));

/// Crash information structure
#[derive(Debug, Clone)]
pub struct CrashInfo {
    pub signal: i32,
    pub signal_name: String,
    pub thread_id: u64,
    pub fault_address: Option<u64>,
    pub instruction_pointer: Option<u64>,
    pub stack_pointer: Option<u64>,
    pub register_context: Option<RegisterContext>,
    pub backtrace: Vec<StackFrame>,
    pub call_tree_snapshot: Option<String>,
    pub crash_time: std::time::SystemTime,
}

/// Stack frame information
#[derive(Debug, Clone, serde::Serialize)]
pub struct StackFrame {
    pub address: u64,
    pub function_name: Option<String>,
    pub library_name: Option<String>,
    pub offset: Option<u64>,
}

/// Signals to handle for crash detection
const CRASH_SIGNALS: &[i32] = &[
    SIGSEGV, // Segmentation fault
    SIGABRT, // Abort signal
    SIGILL,  // Illegal instruction
    SIGFPE,  // Floating point exception
    SIGBUS,  // Bus error
    SIGTRAP, // Trace/breakpoint trap
];

/// Initialize crash handler
pub fn init_crash_handler() -> Result<()> {
    CRASH_HANDLER_INIT.call_once(|| {
        if let Err(e) = setup_crash_handlers() {
            eprintln!("CallTrace: Failed to initialize crash handler: {:?}", e);
        } else {
            CRASH_HANDLER_ENABLED.store(true, Ordering::Relaxed);
            if std::env::var("CALLTRACE_DEBUG").is_ok() {
                eprintln!("CallTrace: Crash handler initialized successfully");
            }
        }
    });
    Ok(())
}

/// Setup signal handlers for crash detection
fn setup_crash_handlers() -> Result<()> {
    let mut original_handlers = ORIGINAL_HANDLERS.lock().map_err(|_| {
        CallTraceError::InitializationError("Failed to lock original handlers".to_string())
    })?;

    for &signal in CRASH_SIGNALS {
        // Store the current handler for restoration during cleanup
        let mut existing_action: libc::sigaction = unsafe { std::mem::zeroed() };
        let result = unsafe { sigaction(signal, std::ptr::null(), &mut existing_action) };

        if result != 0 {
            continue; // Skip this signal if we can't query it
        }

        // Store the original handler for restoration later
        original_handlers.insert(signal, existing_action);

        // Install our crash handler (override any existing handler)
        let mut new_action: libc::sigaction = unsafe { std::mem::zeroed() };
        new_action.sa_sigaction = crash_signal_handler as *const () as usize;
        new_action.sa_flags = SA_SIGINFO | SA_RESTART | SA_NODEFER;

        // Initialize signal mask
        unsafe {
            sigemptyset(&mut new_action.sa_mask);
            for &other_signal in CRASH_SIGNALS {
                if other_signal != signal {
                    sigaddset(&mut new_action.sa_mask, other_signal);
                }
            }
        }

        let install_result = unsafe { sigaction(signal, &new_action, std::ptr::null_mut()) };

        if install_result != 0 {
            eprintln!("CallTrace: Failed to install handler for signal {}", signal);
        } else if std::env::var("CALLTRACE_DEBUG").is_ok() {
            eprintln!("CallTrace: Installed crash handler for signal {} (overriding any existing handler)", signal);
        }
    }

    Ok(())
}

/// Main crash signal handler
extern "C" fn crash_signal_handler(signal: i32, siginfo: *mut siginfo_t, context: *mut c_void) {
    // Avoid recursive crashes in the handler
    static HANDLER_RUNNING: AtomicBool = AtomicBool::new(false);

    if HANDLER_RUNNING.load(Ordering::Acquire) {
        unsafe {
            libc::_exit(128 + signal);
        }
    }
    HANDLER_RUNNING.store(true, Ordering::Release);

    // Generate crash report
    let crash_info = match generate_crash_info(signal, siginfo, context) {
        Ok(info) => info,
        Err(e) => {
            eprintln!("CallTrace: Failed to generate crash info: {:?}", e);
            unsafe {
                libc::_exit(128 + signal);
            }
        }
    };

    // Output crash report
    if let Err(e) = output_crash_report(&crash_info) {
        eprintln!("CallTrace: Failed to output crash report: {:?}", e);
    }

    // Restore original handler and re-raise the signal
    restore_and_reraise(signal);
}

/// Generate comprehensive crash information
fn generate_crash_info(
    signal: i32,
    siginfo: *mut siginfo_t,
    context: *mut c_void,
) -> Result<CrashInfo> {
    let signal_name = get_signal_name(signal);

    // Capture current thread ID
    let thread_id = unsafe {
        #[cfg(target_os = "linux")]
        {
            libc::gettid() as u64
        }
        #[cfg(not(target_os = "linux"))]
        {
            // Fallback to pthread thread ID on other systems
            libc::pthread_self() as u64
        }
    };

    // Extract fault address and context information
    let (fault_address, instruction_pointer, stack_pointer) =
        extract_context_info(siginfo, context);

    // Capture current register context (if possible)
    let register_context = unsafe { RegisterContext::capture().ok() };

    // Generate backtrace
    let backtrace = generate_backtrace()?;

    // Capture current call tree state
    let call_tree_snapshot = capture_call_tree_snapshot();

    Ok(CrashInfo {
        signal,
        signal_name,
        thread_id,
        fault_address,
        instruction_pointer,
        stack_pointer,
        register_context,
        backtrace,
        call_tree_snapshot,
        crash_time: std::time::SystemTime::now(),
    })
}

/// Extract context information from signal info and ucontext
fn extract_context_info(
    siginfo: *mut siginfo_t,
    context: *mut c_void,
) -> (Option<u64>, Option<u64>, Option<u64>) {
    if siginfo.is_null() || context.is_null() {
        return (None, None, None);
    }

    let fault_address = unsafe {
        let info = &*siginfo;
        if info.si_signo == SIGSEGV || info.si_signo == SIGBUS {
            // Access fault address from siginfo_t
            #[cfg(target_arch = "x86_64")]
            {
                // On x86_64, fault address is in si_addr field
                // For simplicity, we'll try to access it directly
                // This may not work on all platforms - a real implementation would use proper bindings
                None // Simplified for now - would need platform-specific offset calculation
            }
            #[cfg(not(target_arch = "x86_64"))]
            {
                None
            }
        } else {
            None
        }
    };

    let (instruction_pointer, stack_pointer) = unsafe {
        #[cfg(target_arch = "x86_64")]
        {
            let ucontext = &*(context as *const ucontext_t);
            // Access registers from ucontext - this is platform specific
            // On Linux x86_64, we can access them through mcontext
            let _mcontext_ptr = &ucontext.uc_mcontext as *const _ as *const u8;

            // These offsets are architecture and OS specific
            // For a production implementation, we'd use proper bindings
            // For now, we'll try a simpler approach
            (None, None) // Simplified for now
        }

        #[cfg(not(target_arch = "x86_64"))]
        {
            (None, None)
        }
    };

    (fault_address, instruction_pointer, stack_pointer)
}

/// Generate stack backtrace
/// This is a simplified implementation that captures basic stack information
fn generate_backtrace() -> Result<Vec<StackFrame>> {
    // For now, we'll implement a basic backtrace using frame pointers
    // In a production implementation, this would use libunwind or similar
    let mut frames = Vec::new();

    // Try to get some basic stack frames using unsafe frame pointer walking
    unsafe {
        let mut frame_ptr: *const *const c_void;

        // Get current frame pointer
        #[cfg(target_arch = "x86_64")]
        {
            asm!("mov {}, rbp", out(reg) frame_ptr, options(nomem, nostack));
        }

        #[cfg(not(target_arch = "x86_64"))]
        {
            // For non-x86_64, we can't easily walk the stack
            return Ok(frames);
        }

        // Walk up to 32 frames
        for _ in 0..32 {
            if frame_ptr.is_null() {
                break;
            }

            // Get return address (next pointer in frame)
            let return_addr_ptr = frame_ptr.offset(1);
            if return_addr_ptr.is_null() {
                break;
            }

            let return_addr = *return_addr_ptr as u64;
            if return_addr == 0 {
                break;
            }

            // Create frame info
            let frame = resolve_stack_frame(return_addr);
            frames.push(frame);

            // Move to next frame
            frame_ptr = *frame_ptr as *const *const c_void;

            // Safety check to prevent infinite loops
            if (frame_ptr as u64) < 0x1000 || (frame_ptr as u64) > 0x7fffffffffff {
                break;
            }
        }
    }

    Ok(frames)
}

/// Resolve stack frame information using dladdr
fn resolve_stack_frame(address: u64) -> StackFrame {
    let mut dl_info: Dl_info = unsafe { std::mem::zeroed() };
    let result = unsafe { dladdr(address as *const c_void, &mut dl_info) };

    let (function_name, library_name, offset) = if result != 0 {
        let func_name = if !dl_info.dli_sname.is_null() {
            let raw_name = unsafe {
                CStr::from_ptr(dl_info.dli_sname)
                    .to_string_lossy()
                    .into_owned()
            };

            // Try to demangle C++ function names
            Some(crate::dwarf_analyzer::demangle_function_name(&raw_name))
        } else {
            None
        };

        let lib_name = if !dl_info.dli_fname.is_null() {
            unsafe {
                Some(
                    CStr::from_ptr(dl_info.dli_fname)
                        .to_string_lossy()
                        .into_owned(),
                )
            }
        } else {
            None
        };

        let offset_val = if !dl_info.dli_saddr.is_null() {
            Some(address - (dl_info.dli_saddr as u64))
        } else {
            None
        };

        (func_name, lib_name, offset_val)
    } else {
        (None, None, None)
    };

    StackFrame {
        address,
        function_name,
        library_name,
        offset,
    }
}

/// Capture current call tree state as JSON string
fn capture_call_tree_snapshot() -> Option<String> {
    use crate::CALL_TREE_MANAGER;

    match JsonOutputGenerator::new().generate_output(&CALL_TREE_MANAGER) {
        Ok(trace_session) => serde_json::to_string_pretty(&trace_session).ok(),
        Err(_) => None,
    }
}

/// Output crash report to stderr and optional file
fn output_crash_report(crash_info: &CrashInfo) -> Result<()> {
    let report = format_crash_report(crash_info)?;

    // Always output human-readable format to stderr
    eprintln!("\n{}", "=".repeat(80));
    eprintln!("CALLTRACE CRASH REPORT");
    eprintln!("{}", "=".repeat(80));
    eprintln!("{}", report);
    eprintln!("{}", "=".repeat(80));

    // Write JSON crash report to file
    if let Some(base_filename) = crate::get_base_output_filename() {
        let crash_file = format!("{}.json", base_filename);

        // Convert to JSON format
        let crash_json = convert_crash_info_to_json(crash_info)?;

        // Serialize to JSON
        let json_content = match serde_json::to_string_pretty(&crash_json) {
            Ok(json) => json,
            Err(e) => {
                eprintln!("CallTrace: Failed to serialize crash info to JSON: {}", e);
                return Ok(());
            }
        };

        if let Err(e) = std::fs::write(&crash_file, &json_content) {
            eprintln!(
                "CallTrace: Failed to write crash report to {}: {}",
                crash_file, e
            );
        } else {
            eprintln!("CallTrace: Crash report written to {}", crash_file);
        }
    }

    Ok(())
}

/// Format crash report as human-readable text
fn format_crash_report(crash_info: &CrashInfo) -> Result<String> {
    let mut report = String::new();

    report.push_str(&format!(
        "Signal: {} ({})\n",
        crash_info.signal, crash_info.signal_name
    ));
    report.push_str(&format!("Thread ID: {}\n", crash_info.thread_id));
    report.push_str(&format!(
        "Time: {}\n",
        format_system_time(&crash_info.crash_time)
    ));

    if let Some(fault_addr) = crash_info.fault_address {
        report.push_str(&format!("Fault Address: 0x{:016x}\n", fault_addr));
    }

    if let Some(ip) = crash_info.instruction_pointer {
        report.push_str(&format!("Instruction Pointer: 0x{:016x}\n", ip));
    }

    if let Some(sp) = crash_info.stack_pointer {
        report.push_str(&format!("Stack Pointer: 0x{:016x}\n", sp));
    }

    report.push_str("\nStack Trace:\n");
    for (i, frame) in crash_info.backtrace.iter().enumerate() {
        report.push_str(&format!("  #{:2}: 0x{:016x}", i, frame.address));

        if let Some(ref func) = frame.function_name {
            report.push_str(&format!(" in {}", func));
            if let Some(offset) = frame.offset {
                report.push_str(&format!("+0x{:x}", offset));
            }
        }

        if let Some(ref lib) = frame.library_name {
            report.push_str(&format!(" ({})", lib));
        }

        report.push('\n');
    }

    if let Some(ref call_tree) = crash_info.call_tree_snapshot {
        report.push_str("\nCall Tree at Crash:\n");
        report.push_str(call_tree);
    }

    Ok(report)
}

/// Convert CrashInfo to JSON-serializable format with normal trace session data
fn convert_crash_info_to_json(crash_info: &CrashInfo) -> Result<crate::json_output::TraceSession> {
    // Generate normal trace session data first
    let mut trace_session = crate::JsonOutputGenerator::new()
        .generate_output(&crate::CALL_TREE_MANAGER)
        .map_err(|e| {
            CallTraceError::InitializationError(format!(
                "Failed to generate trace session: {:?}",
                e
            ))
        })?;

    // Format timestamp
    let (crash_time_str, crash_timestamp) = format_crash_time(&crash_info.crash_time);

    // Convert backtrace
    let backtrace_json = crash_info
        .backtrace
        .iter()
        .map(|frame| crate::json_output::StackFrame {
            address: format!("0x{:016x}", frame.address),
            function_name: frame.function_name.clone(),
            library_name: frame.library_name.clone(),
            offset: frame.offset.map(|off| format!("0x{:x}", off)),
        })
        .collect();

    // Create crash info
    let crash_info_json = crate::json_output::CrashInfo {
        signal: crash_info.signal,
        signal_name: crash_info.signal_name.clone(),
        thread_id: crash_info.thread_id,
        fault_address: crash_info
            .fault_address
            .map(|addr| format!("0x{:016x}", addr)),
        instruction_pointer: crash_info
            .instruction_pointer
            .map(|ip| format!("0x{:016x}", ip)),
        stack_pointer: crash_info.stack_pointer.map(|sp| format!("0x{:016x}", sp)),
        register_context: crash_info.register_context.clone(),
        backtrace: backtrace_json,
        crash_time: crash_time_str,
        crash_timestamp,
    };

    // Add crash info to trace session
    trace_session.crash = Some(crash_info_json);

    Ok(trace_session)
}

/// Format crash time for JSON output
fn format_crash_time(system_time: &std::time::SystemTime) -> (String, f64) {
    match system_time.duration_since(std::time::UNIX_EPOCH) {
        Ok(duration) => {
            let total_seconds = duration.as_secs();
            let microseconds = duration.subsec_micros();
            let timestamp_decimal = total_seconds as f64 + (microseconds as f64 / 1_000_000.0);

            // Get human readable time
            let readable_time = get_readable_time(total_seconds);

            (readable_time, timestamp_decimal)
        }
        Err(_) => (format!("{:?}", system_time), 0.0),
    }
}

/// Format SystemTime to human-readable string
fn format_system_time(system_time: &std::time::SystemTime) -> String {
    match system_time.duration_since(std::time::UNIX_EPOCH) {
        Ok(duration) => {
            let total_seconds = duration.as_secs();
            let microseconds = duration.subsec_micros();

            // Convert to local time using UNIX utilities approach
            // Format as: YYYY-MM-DD HH:MM:SS.uuuuuu (timestamp)
            let timestamp_decimal = format!("{}.{:06}", total_seconds, microseconds);

            // Try to get human readable time via system command
            let readable_time = get_readable_time(total_seconds);

            format!("{} ({})", readable_time, timestamp_decimal)
        }
        Err(_) => {
            // Fallback to debug format if time is before UNIX epoch
            format!("{:?}", system_time)
        }
    }
}

/// Get human-readable time string from UNIX timestamp
fn get_readable_time(timestamp: u64) -> String {
    use std::process::Command;

    // Try to use system 'date' command for proper timezone handling
    if let Ok(output) = Command::new("date")
        .arg("-d")
        .arg(format!("@{}", timestamp))
        .arg("+%Y-%m-%d %H:%M:%S %Z")
        .output()
    {
        if output.status.success() {
            if let Ok(time_str) = String::from_utf8(output.stdout) {
                return time_str.trim().to_string();
            }
        }
    }

    // Fallback: Simple UTC formatting
    let total_seconds = timestamp;
    let seconds_per_day = 86400;
    let seconds_per_hour = 3600;
    let seconds_per_minute = 60;

    // Calculate days since UNIX epoch (1970-01-01)
    let days_since_epoch = total_seconds / seconds_per_day;
    let seconds_in_day = total_seconds % seconds_per_day;

    let hours = seconds_in_day / seconds_per_hour;
    let minutes = (seconds_in_day % seconds_per_hour) / seconds_per_minute;
    let seconds = seconds_in_day % seconds_per_minute;

    // Simple date calculation (approximate, doesn't handle leap years perfectly)
    let years_since_1970 = days_since_epoch / 365;
    let remaining_days = days_since_epoch % 365;
    let months = remaining_days / 30; // Rough approximation
    let days = remaining_days % 30;

    format!(
        "{:04}-{:02}-{:02} {:02}:{:02}:{:02} UTC",
        1970 + years_since_1970,
        1 + months,
        1 + days,
        hours,
        minutes,
        seconds
    )
}

/// Get human-readable signal name
fn get_signal_name(signal: i32) -> String {
    match signal {
        SIGSEGV => "SIGSEGV (Segmentation fault)".to_string(),
        SIGABRT => "SIGABRT (Abort)".to_string(),
        SIGILL => "SIGILL (Illegal instruction)".to_string(),
        SIGFPE => "SIGFPE (Floating point exception)".to_string(),
        SIGBUS => "SIGBUS (Bus error)".to_string(),
        SIGTRAP => "SIGTRAP (Trace/breakpoint trap)".to_string(),
        _ => format!("Signal {}", signal),
    }
}

/// Restore original signal handler and re-raise the signal
fn restore_and_reraise(signal: i32) {
    // Restore original handler
    if let Ok(original_handlers) = ORIGINAL_HANDLERS.lock() {
        if let Some(original_action) = original_handlers.get(&signal) {
            unsafe {
                sigaction(signal, original_action, std::ptr::null_mut());
            }
        }
    }

    // Re-raise the signal to trigger default behavior
    unsafe {
        libc::raise(signal);
    }

    // If raise didn't work, exit with error code
    unsafe {
        libc::_exit(128 + signal);
    }
}

/// Reinforce crash handlers (reinstall to override any handlers set after initialization)
pub fn reinforce_crash_handlers() -> Result<()> {
    if !CRASH_HANDLER_ENABLED.load(Ordering::Relaxed) {
        return Ok(()); // Not enabled, nothing to do
    }

    for &signal in CRASH_SIGNALS {
        // Install our crash handler (override any existing handler)
        let mut new_action: libc::sigaction = unsafe { std::mem::zeroed() };
        new_action.sa_sigaction = crash_signal_handler as *const () as usize;
        new_action.sa_flags = SA_SIGINFO | SA_RESTART | SA_NODEFER;

        // Initialize signal mask
        unsafe {
            sigemptyset(&mut new_action.sa_mask);
            for &other_signal in CRASH_SIGNALS {
                if other_signal != signal {
                    sigaddset(&mut new_action.sa_mask, other_signal);
                }
            }
        }

        let install_result = unsafe { sigaction(signal, &new_action, std::ptr::null_mut()) };

        if install_result == 0 {
            // Only print reinforcement messages during debug mode
            if std::env::var("CALLTRACE_DEBUG").is_ok() {
                eprintln!("CallTrace: Reinforced crash handler for signal {}", signal);
            }
        }
    }

    Ok(())
}

/// Check if crash handler is enabled
pub fn is_crash_handler_enabled() -> bool {
    CRASH_HANDLER_ENABLED.load(Ordering::Relaxed)
}

/// Cleanup crash handler (restore original handlers)
pub fn cleanup_crash_handler() {
    if let Ok(original_handlers) = ORIGINAL_HANDLERS.lock() {
        for (&signal, original_action) in original_handlers.iter() {
            unsafe {
                sigaction(signal, original_action, std::ptr::null_mut());
            }
        }
    }
    CRASH_HANDLER_ENABLED.store(false, Ordering::Relaxed);
}