capture-boot

#!/bin/bash
# capture-boot: Capture dmesg immediately on boot to catch shutdown traces
# This runs very early in boot before dmesg is cleared

set -u  # Only fail on undefined variables, not on command failures

# Pattern for critical kernel errors (excluding firmware bugs)
# Match various BUG formats and warnings
critical_pattern="(Call Trace:|Oops:|Kernel panic|kernel BUG|BUG at|BUG:|^[[:space:]]*BUG:|WARNING:|WARN_ON|WARN\(|RIP:|stack backtrace|general protection fault|page fault|NULL pointer dereference)"

# Directory for saving traces
trace_dir="/var/log/shutdown-traces"
if ! mkdir -p "$trace_dir" 2>/dev/null; then
	# Can't create directory, exit silently
	exit 0
fi

# Capture dmesg immediately (before it's cleared)
# Try multiple methods to capture kernel messages
dmesg_output=$(dmesg -T 2>/dev/null || dmesg 2>/dev/null)

# Also try reading from /dev/kmsg directly (kernel ring buffer)
# This might contain messages that aren't in dmesg yet
if [[ -r /dev/kmsg ]]; then
	kmsg_output=$(timeout 1 cat /dev/kmsg 2>/dev/null | head -1000 || true)
	if [[ -n "$kmsg_output" ]]; then
		# Merge with dmesg output
		dmesg_output="${dmesg_output}${dmesg_output:+$'\n'}${kmsg_output}"
	fi
fi

if [[ -z "$dmesg_output" ]]; then
	exit 0
fi

# Check for traces - save if ANY traces are found (not just near reboot messages)
# This captures BUG traces that happen during shutdown/reboot
has_traces=$(echo "$dmesg_output" | grep -iE "$critical_pattern" | grep -vE "\[Firmware Bug\]|Firmware Bug:" | head -1)
has_reboot=$(echo "$dmesg_output" | grep -E "reboot: Restarting|reboot: machine restart|reboot:" | head -1)

if [[ -n "$has_traces" ]]; then
	# Found traces - save them
	timestamp=$(date '+%Y%m%d-%H%M%S')
	trace_file="$trace_dir/boot-capture-$timestamp.log"
	
	{
		echo "=== Boot-time dmesg capture (traces detected) ==="
		echo "Timestamp: $(date)"
		echo "Boot ID: $(cat /proc/sys/kernel/random/boot_id 2>/dev/null || echo 'unknown')"
		echo ""
		
		# Extract traces with context
		echo "=== Critical traces detected ==="
		echo "$dmesg_output" | awk -v pattern="$critical_pattern" '
			{
				lines[NR] = $0
			}
			END {
				max_lines = NR
				last_printed = 0
				for (i = 1; i <= max_lines; i++) {
					if (lines[i] ~ pattern && lines[i] !~ /\[Firmware Bug\]|Firmware Bug:/ && i > last_printed) {
						# Show 100 lines of context around traces
						start = (i - 50 > 1) ? i - 50 : 1
						end = (i + 100 < max_lines) ? i + 100 : max_lines
						for (j = start; j <= end; j++) {
							print lines[j]
						}
						print "---"
						last_printed = end
					}
				}
			}
		'
		
		echo ""
		echo "=== Full dmesg output ==="
		echo "$dmesg_output"
	} > "$trace_file" 2>&1
	
	# Make it readable
	chmod 644 "$trace_file" 2>/dev/null || true
fi

exit 0