dwipe 1.0.7__py3-none-any.whl → 2.0.0__py3-none-any.whl

dwipe/WipeJob.py

@@ -0,0 +1,1243 @@
+"""
+WipeJob class for handling disk/partition wiping operations
+
+"""
+# pylint: disable=broad-exception-raised,broad-exception-caught
+import os
+import json
+import time
+import threading
+import random
+import traceback
+import subprocess
+import mmap
+from types import SimpleNamespace
+
+from .Utils import Utils
+
+
+class WipeJob:
+    """Handles disk/partition wiping operations with progress tracking"""
+
+    # O_DIRECT requires aligned buffers and write sizes
+    BLOCK_SIZE = 4096  # Alignment requirement for O_DIRECT
+    WRITE_SIZE = 1 * 1024 * 1024  # 1MB (must be multiple of BLOCK_SIZE)
+    BUFFER_SIZE = WRITE_SIZE  # Same size for O_DIRECT
+
+    # Marker constants (separate from O_DIRECT writes)
+    MARKER_SIZE = 16 * 1024  # 16KB for marker
+    STATE_OFFSET = 15 * 1024  # where json is written (for marker buffer)
+
+    # Aligned buffers allocated with mmap (initialized at module load)
+    buffer = None  # Random data buffer (memoryview)
+    buffer_mem = None  # Underlying mmap object
+    zero_buffer = None  # Zero buffer (memoryview)
+    zero_buffer_mem = None  # Underlying mmap object
+
+    @staticmethod
+    def _get_dirty_kb():
+        """Read Dirty pages from /proc/meminfo (in KB)"""
+        try:
+            with open('/proc/meminfo', 'r', encoding='utf-8') as f:
+                for line in f:
+                    if line.startswith('Dirty:'):
+                        return int(line.split()[1])
+        except Exception:
+            pass
+        return 0
+
+    @staticmethod
+    def _get_total_memory_mb():
+        """Read total system memory from /proc/meminfo (in MB)"""
+        try:
+            with open('/proc/meminfo', 'r', encoding='utf-8') as f:
+                for line in f:
+                    if line.startswith('MemTotal:'):
+                        # Value is in KB, convert to MB
+                        return int(line.split()[1]) // 1024
+        except Exception:
+            pass
+        return 8192  # Default to 8GB if we can't read
+
+    @staticmethod
+    def _rebalance_buffer(buffer):
+        """Rebalance byte distribution to be more uniform
+
+        Takes random data and redistributes bytes to make the distribution
+        more uniform, which helps with chi-squared verification while maintaining
+        cryptographic randomness.
+        """
+        # Count byte frequencies
+        byte_counts = [0] * 256
+        for byte in buffer:
+            byte_counts[byte] += 1
+
+        # Calculate expected frequency (uniform distribution)
+        expected = len(buffer) / 256
+
+        # Find overused and underused bytes
+        overused = []  # (byte_value, excess_count)
+        underused = []  # (byte_value, deficit_count)
+
+        for byte_val in range(256):
+            count = byte_counts[byte_val]
+            diff = count - expected
+            if diff > expected * 0.05:  # More than 5% over expected
+                overused.append((byte_val, int(diff * 0.7)))  # Replace 70% of excess
+            elif diff < -expected * 0.05:  # More than 5% under expected
+                underused.append((byte_val, int(-diff * 0.7)))  # Add to 70% of deficit
+
+        if not overused or not underused:
+            return buffer  # Already well-balanced
+
+        # Create a mutable copy
+        result = bytearray(buffer)
+
+        # Build replacement plan
+        replacements_made = 0
+        target_replacements = min(sum(count for _, count in overused),
+                                  sum(count for _, count in underused))
+
+        # Randomly replace overused bytes with underused bytes
+        underused_idx = 0
+        for overused_byte, excess in overused:
+            if replacements_made >= target_replacements:
+                break
+
+            # Find positions of this overused byte
+            positions = [i for i, b in enumerate(result) if b == overused_byte]
+
+            # Replace randomly selected positions
+            replace_count = min(excess, target_replacements - replacements_made)
+            positions_to_replace = random.sample(positions, min(replace_count, len(positions)))
+
+            for pos in positions_to_replace:
+                if underused_idx >= len(underused):
+                    break
+                underused_byte, deficit = underused[underused_idx]
+                result[pos] = underused_byte
+                replacements_made += 1
+
+                # Move to next underused byte when we've used up this one's deficit
+                if replacements_made % max(1, deficit) == 0:
+                    underused_idx += 1
+
+        return bytes(result)
+
+    def __init__(self, device_path, total_size, opts=None, resume_from=0, resume_mode=None):
+        self.opts = opts if opts else SimpleNamespace(dry_run=False)
+        self.device_path = device_path
+        self.total_size = total_size
+        self.do_abort = False
+        self.thread = None
+
+        self.start_mono = time.monotonic()  # Track the start time
+        self.total_written = resume_from  # Start from resumed offset if resuming
+        self.resume_from = resume_from  # Track resume offset
+        self.resume_mode = resume_mode  # Original mode if resuming (overrides opts.wipe_mode)
+        self.wr_hists = []  # list of (mono, written)
+        self.done = False
+        self.exception = None  # in case of issues
+
+        # Multi-pass tracking
+        self.passes = getattr(opts, 'passes', 1)  # Total number of passes to perform
+        self.current_pass = 0  # Current pass number (0-indexed)
+
+        # Verification tracking
+        self.verify_phase = False  # True when verifying
+        self.verify_start_mono = None  # Start time of verify phase
+        self.verify_progress = 0  # Bytes verified so far
+        self.verify_pct = 0  # Percentage of disk being verified (e.g., 2 for 2%)
+        self.verify_result = None  # "zeroed", "random", "not-wiped", "mixed"
+
+        self.expected_pattern = None  # "zero" or "random" for what we wrote
+        self.is_verify_only = False  # True if this is a standalone verify job
+
+        # Periodic marker updates for crash recovery
+        self.last_marker_update_mono = time.monotonic() - 25  # Last time we wrote progress marker
+        self.marker_update_interval = 30  # Update marker every 30 seconds
+
+        ## SLOWDOWN / STALL DETECTION/ABORT FEATURE
+        ## 
+        self.slowdown_stop = getattr(opts, 'slowdown_stop', 16)
+        self.stall_timeout = getattr(opts, 'stall_timeout', 60)
+        self.max_slowdown_ratio = 0
+        self.max_stall_secs = 0
+            # Initialize tracking variables
+        self.baseline_speed = None  # Bytes per second
+        self.baseline_end_mono = None  # When baseline measurement ended
+            # Stall tracking
+        self.last_progress_mono = time.monotonic()  # Last time we made progress
+        self.last_progress_written = resume_from  # Bytes written at last progress check
+            # For periodic slowdown checks (every 10 seconds)
+        self.last_slowdown_check = 0
+            # Initialize write history for speed calculation
+        self.wr_hists.append(SimpleNamespace(mono=self.start_mono, written=resume_from))
+        
+        # ERROR ABORT FEATURE
+        self.max_consecutive_errors = 3 # a control
+        self.max_total_errors = 100 # a control
+        self.reopen_on_error = True # a control
+        self.reopen_count = 0 # cumulative (info only)
+        self.total_errors = 0 # cumulative
+
+
+
+    @staticmethod
+    def start_job(device_path, total_size, opts):
+        """Start a wipe job in a background thread
+
+        If an existing marker with stopped state is found, resumes from that offset.
+        Uses current passes setting to determine target bytes (may differ from original).
+
+        Smart resume logic for multi-pass:
+        - If resuming in the final pass, checks pattern on disk
+        - If pattern matches expected for final pass, resumes from current position
+        - If pattern doesn't match, restarts final pass from beginning
+        This prevents mixed patterns (e.g., random then zeros) in the final result.
+        """
+        resume_from = 0
+        resume_mode = None
+
+        # Check for existing marker to resume from
+        device_name = device_path.replace('/dev/', '')
+        existing_marker = WipeJob.read_marker_buffer(device_name)
+        if existing_marker and existing_marker.scrubbed_bytes > 0:
+            scrubbed = existing_marker.scrubbed_bytes
+            resume_mode = existing_marker.mode  # Use original mode ('Rand' or 'Zero')
+            passes = getattr(opts, 'passes', 1)
+
+            # Check if the marker indicates a completed wipe
+            # Old markers without 'passes' field: assume 1 pass
+            marker_passes = getattr(existing_marker, 'passes', 1)
+            marker_target = total_size * marker_passes
+
+            if scrubbed >= marker_target:
+                # Wipe was completed with marker_passes - user wants to re-wipe
+                # Start fresh from 0
+                resume_from = 0
+                resume_mode = None
+            else:
+                   # Partial/stopped wipe - resume from where it left off
+                resume_from = scrubbed
+                # Ensure we don't resume in the marker area
+                if resume_from < WipeJob.MARKER_SIZE:
+                    resume_from = WipeJob.MARKER_SIZE
+                # Also ensure not past the end (sanity check)
+                if resume_from > total_size * getattr(opts, 'passes', 1):
+                    resume_from = 0  # Start over if marker corrupted
+
+                # Partial/stopped wipe - resume from where it left off
+                # Smart resume: check if we're in the final pass
+                current_pass = scrubbed // total_size
+                last_pass_num = passes - 1
+
+                if current_pass >= last_pass_num:
+                    # Resuming in final pass - check pattern on disk matches expected
+                    # Create temporary job to use get_pass_pattern method
+                    temp_job = WipeJob(device_path, total_size, opts)
+                    temp_job.passes = passes
+                    expected_is_random = temp_job.get_pass_pattern(last_pass_num, resume_mode)
+
+                    # Detect actual pattern on disk
+                    actual_is_random = WipeJob.detect_pattern_on_disk(device_path)
+
+                    if expected_is_random != actual_is_random:
+                        # Pattern mismatch - restart final pass from beginning
+                        resume_from = last_pass_num * total_size
+                    else:
+                        # Pattern matches - resume from current position
+                        resume_from = scrubbed
+
+        job = WipeJob(device_path=device_path, total_size=total_size, opts=opts,
+                     resume_from=resume_from, resume_mode=resume_mode)
+        job.thread = threading.Thread(target=job.write_partition)
+        job.thread.start()
+        return job
+
+    @staticmethod
+    def start_verify_job(device_path, total_size, opts, expected_pattern=None):
+        """Start a verification-only job in a background thread
+
+        Args:
+            device_path: Path to device (e.g., '/dev/sda1')
+            total_size: Total size in bytes
+            opts: Options namespace with verify_pct
+            expected_pattern: "zeroed", "random", or None (auto-detect)
+        """
+        job = WipeJob(device_path=device_path, total_size=total_size, opts=opts)
+        job.is_verify_only = True  # Mark as standalone verification job
+        job.expected_pattern = expected_pattern
+        verify_pct = getattr(opts, 'verify_pct', 0)
+        if verify_pct == 0:
+            verify_pct = 2  # Default to 2% if not set
+
+        # Initialize verify state BEFORE starting thread to avoid showing "0%"
+        job.verify_pct = verify_pct
+        job.verify_start_mono = time.monotonic()
+        job.verify_progress = 0
+        job.wr_hists = [SimpleNamespace(mono=job.verify_start_mono, written=0)]
+        job.verify_phase = True  # Set before thread starts
+
+        def verify_runner():
+            try:
+                # Read existing marker to determine the mode and expected pattern
+                device_name = os.path.basename(device_path)
+                existing_marker = WipeJob.read_marker_buffer(device_name)
+                if existing_marker:
+                    # Infer expected pattern from marker if not already set
+                    if job.expected_pattern is None:
+                        job.expected_pattern = "random" if existing_marker.mode == 'Rand' else "zeroed"
+
+                job.verify_partition(verify_pct)
+
+                # Write marker with verification status
+                if existing_marker:
+                    is_random = existing_marker.mode == 'Rand'
+                    job._write_marker_with_verify_status(is_random)
+                    # Note: _write_marker_with_verify_status sets job.done in its finally block
+                else:
+                    # No marker - just mark as done
+                    job.done = True
+            except Exception:
+                job.exception = traceback.format_exc()
+            finally:
+                # ALWAYS ensure job is marked as done, even if exception or early return
+                if not job.done:
+                    job.done = True
+
+        job.thread = threading.Thread(target=verify_runner)
+        job.thread.start()
+        return job
+
+    def _check_for_stall(self, current_monotonic):
+        """Check for stall (no progress) - called frequently"""
+        if self.stall_timeout <= 0:
+            return False
+        
+        time_since_progress = current_monotonic - self.last_progress_mono
+        self.max_stall_secs = max(time_since_progress, self.max_stall_secs)
+        if time_since_progress >= self.stall_timeout:
+            self.do_abort = True
+            self.exception = f"Stall detected: No progress for {time_since_progress:.1f} seconds"
+            return True
+        
+        return False
+
+    def _check_for_slowdown(self, current_monotonic):
+        """Check for slowdown - called every 10 seconds"""
+        if self.slowdown_stop <= 0 or self.baseline_speed is None or self.baseline_speed <= 0:
+            return False
+        
+        # Calculate current speed over last 30 seconds
+        floor = current_monotonic - 30
+        recent_history = [h for h in self.wr_hists if h.mono >= floor]
+        
+        if len(recent_history) >= 2:
+            recent_start = recent_history[0]
+            recent_written = self.total_written - recent_start.written
+            recent_elapsed = current_monotonic - recent_start.mono
+            
+            if recent_elapsed > 1.0:
+                current_speed = recent_written / recent_elapsed
+                self.baseline_speed = max(self.baseline_speed, current_speed)
+                slowdown_ratio = self.baseline_speed / max(current_speed, 1)
+                slowdown_ratio = int(round(slowdown_ratio, 0))
+                self.max_slowdown_ratio = max(self.max_slowdown_ratio, slowdown_ratio)
+                
+                if slowdown_ratio > self.slowdown_stop:
+                    self.do_abort = True
+                    self.exception = (f"Slowdown abort: ({Utils.human(current_speed)}B/s)"
+                                     f" is 1/{slowdown_ratio} baseline")
+                    return True
+        
+        return False
+
+    def _update_baseline_if_needed(self, current_monotonic):
+        """Update baseline speed measurement if still in first 60 seconds"""
+        if self.baseline_speed is not None:
+            return  # Baseline already established
+        
+        if (current_monotonic - self.start_mono) >= 60:
+            total_written_60s = self.total_written - self.resume_from
+            elapsed_60s = current_monotonic - self.start_mono
+            if elapsed_60s > 0:
+                self.baseline_speed = total_written_60s / elapsed_60s
+                self.baseline_end_mono = current_monotonic
+                self.last_slowdown_check = current_monotonic  # Start slowdown checking
+
+
+
+    def get_status_str(self):
+        """Get human-readable status string"""
+        elapsed_time = time.monotonic() - self.start_mono
+        write_rate = self.total_written / elapsed_time if elapsed_time > 0 else 0
+        percent_complete = (self.total_written / self.total_size) * 100
+        return (f"Write rate: {write_rate / (1024 * 1024):.2f} MB/s, "
+                f"Completed: {percent_complete:.2f}%")
+
+    def get_status(self):
+        """Get status tuple: (elapsed, percent, rate, eta)
+
+        Returns stats for current phase only:
+        - Write phase (0-100%): elapsed/rate/eta for writing
+        - Flushing phase: 100% FLUSH while kernel syncs to device
+        - Verify phase (v0-v100%): elapsed/rate/eta for verification only
+        """
+        pct_str, rate_str, when_str = '', '', ''
+        mono = time.monotonic()
+
+        if self.verify_phase:
+            # Verification phase: v0% to v100%
+            if self.verify_start_mono is None:
+                self.verify_start_mono = mono
+
+            elapsed_time = mono - self.verify_start_mono
+            progress = self.verify_progress
+
+            # Calculate total bytes to verify (verify_pct% of total_size)
+            if self.verify_pct > 0:
+                total_to_verify = self.total_size * self.verify_pct / 100
+            else:
+                total_to_verify = self.total_size
+
+            # Calculate verification percentage (0-100)
+            pct = int((progress / total_to_verify) * 100) if total_to_verify > 0 else 0
+            pct_str = f'v{pct}%'
+
+            if self.do_abort:
+                pct_str = 'STOP'
+
+            # Track verification progress for rate calculation
+            self.wr_hists.append(SimpleNamespace(mono=mono, written=progress))
+            floor = mono - 30
+            while len(self.wr_hists) >= 3 and self.wr_hists[1].mono >= floor:
+                del self.wr_hists[0]
+            delta_mono = mono - self.wr_hists[0].mono
+            physical_rate = (progress - self.wr_hists[0].written) / delta_mono if delta_mono > 1.0 else 0
+            # Scale rate to show "effective" verification rate (as if verifying 100% of disk)
+            effective_rate = physical_rate * (100 / self.verify_pct) if self.verify_pct > 0 else physical_rate
+            rate_str = f'{Utils.human(int(round(effective_rate, 0)))}/s'
+
+            if physical_rate > 0:
+                remaining = total_to_verify - progress
+                when = int(round(remaining / physical_rate))
+                when_str = Utils.ago_str(when)
+            else:
+                when_str = '0'
+
+            return Utils.ago_str(int(round(elapsed_time))), pct_str, rate_str, when_str
+        else:
+            # Write phase: 0-100% (across all passes)
+            written = self.total_written
+            elapsed_time = mono - self.start_mono
+
+            # Calculate progress across all passes
+            # total_written represents cumulative bytes across all passes
+            total_work = self.total_size * self.passes
+            pct = (self.total_written / total_work) * 100 if total_work > 0 else 0
+            # Cap at 100% (can exceed if passes changed during resume)
+            pct = min(pct, 100)
+            pct_str = f'{int(round(pct))}%'
+            if self.do_abort:
+                pct_str = 'STOP'
+
+            # Calculate rate using sliding window to avoid RAM buffering inflation
+            # Track write progress history
+            self.wr_hists.append(SimpleNamespace(mono=mono, written=written))
+            floor = mono - 30  # 30 second window
+            while len(self.wr_hists) >= 3 and self.wr_hists[1].mono >= floor:
+                del self.wr_hists[0]
+
+            # Calculate rate from sliding window
+            delta_mono = mono - self.wr_hists[0].mono
+            rate = (written - self.wr_hists[0].written) / delta_mono if delta_mono > 1.0 else 0
+
+            rate_str = f'{Utils.human(int(round(rate, 0)))}/s'
+
+            if rate > 0:
+                # ETA based on total remaining work across all passes
+                remaining_work = total_work - self.total_written
+                when = int(round(remaining_work / rate))
+                when_str = Utils.ago_str(when)
+            else:
+                when_str = '0'
+
+            return Utils.ago_str(int(round(elapsed_time))), pct_str, rate_str, when_str
+
+    def prep_marker_buffer(self, is_random, verify_status=None):
+        """Get the 1st 16KB to write:
+        - 15K zeros
+        - JSON status + zero fill to 1KB
+
+        Marker format (JSON):
+            - unixtime: Unix timestamp when marker was written
+            - scrubbed_bytes: Total bytes written (including all passes)
+            - size_bytes: Total device size in bytes
+            - passes: Number of passes intended/completed
+            - mode: 'Rand' or 'Zero' (final desired pattern)
+            - verify_status: 'pass', 'fail', or omitted (not verified)
+
+        Args:
+            is_random: bool, whether random data was written
+            verify_status: str, "pass", "fail", or None (not verified)
+        """
+        data = {"unixtime": int(time.time()),
+                "scrubbed_bytes": self.total_written,
+                "size_bytes": self.total_size,
+                "passes": self.passes,
+                "mode": 'Rand' if is_random else 'Zero'
+                }
+        if verify_status is not None:
+            data["verify_status"] = verify_status
+        json_data = json.dumps(data).encode('utf-8')
+        buffer = bytearray(self.MARKER_SIZE)  # Only 16KB, not 1MB
+        buffer[:self.STATE_OFFSET] = b'\x00' * self.STATE_OFFSET
+        buffer[self.STATE_OFFSET:self.STATE_OFFSET + len(json_data)] = json_data
+        remaining_size = self.MARKER_SIZE - (self.STATE_OFFSET + len(json_data))
+        buffer[self.STATE_OFFSET + len(json_data):] = b'\x00' * remaining_size
+        return buffer
+
+    def get_pass_pattern(self, pass_number, desired_mode):
+        """Determine what pattern to write for a given pass
+
+        For multi-pass (>1), alternates patterns ending on desired:
+        - 4-pass Rand: Zero, Rand, Zero, Rand
+        - 4-pass Zero: Rand, Zero, Rand, Zero
+        - 2-pass Rand: Zero, Rand
+        - 2-pass Zero: Rand, Zero
+
+        Args:
+            pass_number: 0-indexed pass number
+            desired_mode: 'Rand' or 'Zero' - the final desired pattern
+
+        Returns:
+            bool: True for random, False for zeros
+        """
+        if self.passes == 1:
+            # Single pass: just write desired pattern
+            return desired_mode == 'Rand'
+
+        # Multi-pass: alternate patterns, ending on desired
+        # Final pass is always desired pattern
+        if pass_number == self.passes - 1:
+            return desired_mode == 'Rand'
+
+        # Earlier passes: alternate, starting with opposite
+        # If desired is Rand: pass 0=Zero, 1=Rand, 2=Zero, 3=Rand
+        # If desired is Zero: pass 0=Rand, 1=Zero, 2=Rand, 3=Zero
+        if desired_mode == 'Rand':
+            # Even passes (0, 2, ...) = Zero, odd (1, 3, ...) = Rand
+            return pass_number % 2 == 1
+        else:
+            # Even passes (0, 2, ...) = Rand, odd (1, 3, ...) = Zero
+            return pass_number % 2 == 0
+
+    def maybe_update_marker(self, is_random):
+        """Periodically update marker to enable crash recovery
+
+        Updates marker every marker_update_interval seconds (default 30s).
+        This allows resume to work even after crashes, power loss, or kill -9.
+
+        Args:
+            is_random: bool, whether random data is being written
+
+        Returns:
+            None
+        """
+        now_mono = time.monotonic()
+        if now_mono - self.last_marker_update_mono < self.marker_update_interval:
+            return  # Not time yet
+
+        # Marker writes use separate file handle (buffered I/O, not O_DIRECT)
+        # because marker buffer is not aligned
+        try:
+            with open(self.device_path, 'r+b') as marker_file:
+                marker_file.seek(0)
+                marker_file.write(self.prep_marker_buffer(is_random))
+            self.last_marker_update_mono = now_mono
+        except Exception:
+            # If marker update fails, just continue - we'll try again in 30s
+            pass
+
+    @staticmethod
+    def detect_pattern_on_disk(device_path, sample_size=16*1024):
+        """Read a sample from disk after header to detect zeros vs random
+
+        Reads sample_size bytes starting at WRITE_SIZE (after the marker area).
+        Used for smart resume to determine if we're in the middle of a Zero or Rand pass.
+
+        Args:
+            device_path: Path to device (e.g., '/dev/sda1')
+            sample_size: Number of bytes to sample (default 16KB)
+
+        Returns:
+            bool: True if random pattern detected, False if zeros detected
+        """
+        try:
+            with open(device_path, 'rb') as device:
+                # Skip past the marker area (first 16KB)
+                device.seek(WipeJob.WRITE_SIZE)
+                data = device.read(sample_size)
+
+                if not data:
+                    return False  # Can't read, assume zeros
+
+                # Check if all zeros
+                non_zero_count = sum(1 for byte in data if byte != 0)
+                # If less than 1% non-zero bytes, consider it zeros
+                if non_zero_count < len(data) * 0.01:
+                    return False  # Zeros detected
+
+                # Otherwise, assume random
+                return True
+        except Exception:
+            return False  # Error reading, assume zeros
+
+    def _get_device_major_minor(self):
+        """Get major:minor device numbers for the device
+
+        Returns:
+            str: "major:minor" or None if unable to determine
+        """
+        try:
+            stat_info = os.stat(self.device_path)
+            major = os.major(stat_info.st_rdev)
+            minor = os.minor(stat_info.st_rdev)
+            return f"{major}:{minor}"
+        except Exception:
+            return None
+
+    def _setup_ionice(self):
+        """Setup I/O priority to best-effort class, lowest priority"""
+        try:
+            # Class 2 = best-effort, priority 7 = lowest (0 is highest, 7 is lowest)
+            subprocess.run(["ionice", "-c", "2", "-n", "7", "-p", str(os.getpid())],
+                          capture_output=True, check=False)
+        except Exception:
+            pass
+
+    @staticmethod
+    def read_marker_buffer(device_name):
+        """Open the device and read the first 16 KB"""
+        try:
+            with open(f'/dev/{device_name}', 'rb') as device:
+                device.seek(0)
+                buffer = device.read(WipeJob.MARKER_SIZE)
+        except Exception:
+            return None  # cannot find info
+
+        if buffer[:WipeJob.STATE_OFFSET] != b'\x00' * (WipeJob.STATE_OFFSET):
+            return None  # First 15 KB are not zeros
+
+        # Extract JSON data from the next 1 KB Strip trailing zeros
+        json_data_bytes = buffer[WipeJob.STATE_OFFSET:WipeJob.MARKER_SIZE].rstrip(b'\x00')
+
+        if not json_data_bytes:
+            return None  # No JSON data found
+
+        # Deserialize the JSON data
+        try:
+            data = json.loads(json_data_bytes.decode('utf-8'))
+        except (json.JSONDecodeError, Exception):
+            return None  # Invalid JSON data!
+
+        rv = {}
+        for key, value in data.items():
+            if key in ('unixtime', 'scrubbed_bytes', 'size_bytes', 'passes') and isinstance(value, int):
+                rv[key] = value
+            elif key in ('mode', 'verify_status') and isinstance(value, str):
+                rv[key] = value
+            else:
+                return None  # bogus data
+        # Old markers: 4 fields (no passes, no verify_status)
+        # New markers: 5 fields minimum (with passes), 6 with verify_status
+        if len(rv) < 4 or len(rv) > 6:
+            return None  # bogus data
+        return SimpleNamespace(**rv)
+
+    def write_partition(self):
+        """Writes random chunks to a device and updates the progress status.
+
+        Performs multiple passes if self.passes > 1 with alternating patterns:
+        - Multi-pass Rand: Zero, Rand, Zero, Rand (ends on Rand)
+        - Multi-pass Zero: Rand, Zero, Rand, Zero (ends on Zero)
+        If verify_pct > 0, automatically starts verification after write completes.
+        Supports resuming from a previous stopped wipe with current passes setting.
+        """
+        # Use resume_mode if set (from existing marker), otherwise current mode
+        mode_to_use = self.resume_mode if self.resume_mode else self.opts.wipe_mode.replace('+V', '')
+        desired_mode = mode_to_use  # 'Rand' or 'Zero' - the final desired pattern
+        self.expected_pattern = "random" if desired_mode == 'Rand' else "zeroed"
+
+        # Calculate target bytes based on current passes setting
+        target_bytes = self.passes * self.total_size
+
+        # If already at or beyond target (user reduced passes), mark as done
+        if self.total_written >= target_bytes:
+            self.done = True
+            return
+
+        try:
+            # Set low I/O priority to be nice to other system processes
+            self._setup_ionice()
+
+            # Open device with O_DIRECT for unbuffered I/O (bypasses page cache)
+            # O_DIRECT gives maximum performance with zero dirty pages
+            if not self.opts.dry_run:
+                fd = os.open(self.device_path, os.O_WRONLY | os.O_DIRECT)
+            else:
+                fd = None
+
+            try:
+                # Continue writing until we reach target_bytes
+                while self.total_written < target_bytes and not self.do_abort:
+                    # Calculate current pass and offset within pass
+                    self.current_pass = self.total_written // self.total_size
+                    offset_in_pass = self.total_written % self.total_size
+                    
+                    # SKIP MARKER AREA - don't overwrite it!
+                    if offset_in_pass < self.MARKER_SIZE:
+                        self.total_written += self.MARKER_SIZE - offset_in_pass
+                        offset_in_pass = self.MARKER_SIZE
+
+                    # Determine pattern for this pass (alternating for multi-pass)
+                    is_random_pass = self.get_pass_pattern(self.current_pass, desired_mode)
+
+                    # Seek to current position (O_DIRECT requires block-aligned seeks)
+                    if not self.opts.dry_run:
+                        os.lseek(fd, offset_in_pass, os.SEEK_SET)
+
+                    # Write until end of current pass or target_bytes, whichever comes first
+                    pass_remaining = self.total_size - offset_in_pass
+                    total_remaining = target_bytes - self.total_written
+                    bytes_to_write_this_pass = min(pass_remaining, total_remaining)
+
+                    pass_bytes_written = 0
+
+                    while pass_bytes_written < bytes_to_write_this_pass and not self.do_abort:
+                        current_mono = time.monotonic()
+
+                        # Update baseline if needed (first 60 seconds)
+                        self._update_baseline_if_needed(current_mono)
+                        
+                        # Check for stall (frequently)
+                        if self._check_for_stall(current_mono):
+                            break
+                        
+                        # Check for slowdown (every 10 seconds)
+                        if self.baseline_speed is not None:
+                            time_since_last_check = current_mono - self.last_slowdown_check
+                            if time_since_last_check >= 10:
+                                if self._check_for_slowdown(current_mono):
+                                    break
+                                self.last_slowdown_check = current_mono
+                        
+                        # Update progress tracking
+                        if self.total_written > self.last_progress_written:
+                            self.last_progress_mono = current_mono
+
+      
+                        # Calculate chunk size (must be block-aligned for O_DIRECT)
+                        remaining = bytes_to_write_this_pass - pass_bytes_written
+                        chunk_size = min(WipeJob.WRITE_SIZE, remaining)
+                        # Round down to block boundary
+                        chunk_size = (chunk_size // WipeJob.BLOCK_SIZE) * WipeJob.BLOCK_SIZE
+                        if chunk_size == 0:
+                            break
+
+                        # Select buffer based on pass type
+                        if is_random_pass:
+                            # Use slice of random buffer (still aligned via memoryview)
+                            chunk = WipeJob.buffer[:chunk_size]
+                        else:
+                            # Use zero buffer
+                            chunk = WipeJob.zero_buffer[:chunk_size]
+
+                        if self.opts.dry_run:
+                            bytes_written = chunk_size
+                            time.sleep(0.001)
+                        else:
+                            try:
+                                # Write with O_DIRECT (bypasses page cache)
+                                bytes_written, fd = self.safe_write(fd, chunk)
+                            except Exception as e:
+                                # Save exception for debugging
+                                self.exception = str(e)
+                                self.do_abort = True
+                                bytes_written = 0
+
+                        self.total_written += bytes_written
+                        pass_bytes_written += bytes_written
+
+                        # Periodically update marker for crash recovery (every 30s)
+                        # Note: marker writes use separate buffered file handle
+                        if not self.opts.dry_run and self.total_written > self.MARKER_SIZE:
+                            marker_is_random = (desired_mode == 'Rand')
+                            self.maybe_update_marker(marker_is_random)
+
+                        # Check for errors or incomplete writes
+                        if bytes_written < chunk_size:
+                            break
+
+                # O_DIRECT has no dirty pages - close is instant
+            finally:
+                # Close device file descriptor
+                if fd is not None:
+                    os.close(fd)
+
+            # Write final marker buffer at beginning after ALL passes complete
+            # Skip marker write on abort to avoid blocking on problematic devices
+            # Use separate buffered file handle (marker is not O_DIRECT aligned)
+            if not self.opts.dry_run and self.total_written > 0 and not self.do_abort:
+                try:
+                    final_is_random = (desired_mode == 'Rand')
+                    with open(self.device_path, 'r+b') as marker_file:
+                        marker_file.seek(0)
+                        marker_file.write(self.prep_marker_buffer(final_is_random))
+                except Exception:
+                    pass  # Marker write failure shouldn't fail the whole job
+
+            # Auto-start verification if enabled and write completed successfully
+            verify_pct = getattr(self.opts, 'verify_pct', 0)
+            auto_verify = getattr(self.opts, 'wipe_mode', "").endswith('+V')
+            if auto_verify and verify_pct > 0 and not self.do_abort and not self.exception:
+                self.verify_partition(verify_pct)
+                # Write marker with verification status after verification completes
+                # Use desired_mode to determine if random or zero
+                is_random = (desired_mode == 'Rand')
+                self._write_marker_with_verify_status(is_random)
+            else:
+                self.done = True
+        except Exception:
+            self.exception = traceback.format_exc()
+        finally:
+            # ALWAYS ensure job is marked as done, even if exception or early return
+            if not self.done:
+                self.done = True
+
+    def safe_write(self, fd, chunk):
+        """Safe write with error recovery.
+        
+        Returns:
+            tuple: (bytes_written, fd) - bytes_written is either:
+                   - Actual bytes written (success)
+                   - len(chunk) (failed but non-fatal - skip entire chunk)
+                   fd might be new if reopened
+        
+        Raises:
+            Exception: If should abort (too many consecutive errors)
+        """
+        consecutive_errors = 0
+        while True:  # Keep trying until success, skip, or abort
+            try:
+                bytes_written = os.write(fd, chunk)
+                self.reopen_count = 0
+                return bytes_written, fd # success
+                
+            except Exception as e:
+                consecutive_errors += 1
+                self.total_errors += 1
+
+                # Check if we should abort
+                if consecutive_errors >= self.max_consecutive_errors:
+                    raise Exception(f"{consecutive_errors} consecutive write errors") from e
+
+                if self.total_errors >= self.max_total_errors:
+                    raise Exception(f"{self.total_errors} total write errors") from e
+
+                # Not fatal yet - try reopening if enabled
+                if self.reopen_on_error:
+                    try:
+                        current_pos = self.total_written
+                        # Open new fd first
+                        new_fd = os.open(self.device_path, os.O_WRONLY | os.O_DIRECT)
+                        try:
+                            # Seek to correct position on new fd
+                            os.lseek(new_fd, current_pos, os.SEEK_SET)
+                            # Only close old fd after new one is ready
+                            old_fd = fd
+                            fd = new_fd
+                            try:
+                                os.close(old_fd)
+                            except Exception:
+                                pass  # Old fd close failed, but new fd is good
+                            self.reopen_count += 1
+                        except Exception:
+                            # New fd setup failed, close it and keep using old fd
+                            os.close(new_fd)
+                            raise
+                    except Exception:
+                        # Reopen failed - count as another error and retry with old fd
+                        self.total_errors += 1
+
+                # Retry the write (continue loop)
+
+    def verify_partition(self, verify_pct):
+        """Verify partition with section-by-section analysis"""
+        # Initialize verify state
+        if not self.verify_phase:
+            self.verify_pct = verify_pct
+            self.verify_start_mono = time.monotonic()
+            self.verify_progress = 0
+            self.wr_hists = []
+            self.wr_hists.append(SimpleNamespace(mono=self.verify_start_mono, written=0))
+            self.verify_phase = True
+
+        if verify_pct == 0:
+            self.verify_result = "skipped"
+            return
+
+        # Fast-fail for zeros
+        fast_fail_zeros = (self.expected_pattern == "zeroed")
+        
+        # For unmarked disks: track if ALL bytes are zero
+        all_zeros = (self.expected_pattern is None)
+        
+        # Track section results for debugging
+        self.section_results = []  # Store (section_idx, result, stats)
+
+        try:
+            # Open with regular buffered I/O
+            if not self.opts.dry_run:
+                fd = os.open(self.device_path, os.O_RDONLY)
+            else:
+                fd = None
+
+            read_chunk_size = 64 * 1024  # 64KB chunks
+            SAMPLE_STEP = 23  # Sample every 23rd byte (~4% of data) - prime for even distribution
+
+            # Skip marker area
+            marker_skip = WipeJob.BUFFER_SIZE
+            usable_size = self.total_size - marker_skip
+            
+            # Divide disk into 100 sections for sampling
+            num_sections = 100
+            section_size = usable_size // num_sections
+
+            # Pre-allocated zero pattern for fast comparison
+            ZERO_PATTERN_64K = b'\x00' * (64 * 1024)
+
+            # Track if any section failed
+            overall_failed = False
+            failure_reason = ""
+
+            for section_idx in range(num_sections):
+                if self.do_abort or overall_failed:
+                    break
+
+                # Reset analysis for THIS SECTION
+                section_byte_counts = [0] * 256
+                section_samples = 0
+                section_found_nonzero = False
+
+                # Calculate bytes to verify in this section
+                bytes_in_section = min(section_size, usable_size - section_idx * section_size)
+                bytes_to_verify = int(bytes_in_section * verify_pct / 100)
+
+                if bytes_to_verify == 0:
+                    self.section_results.append((section_idx, "skipped", {}))
+                    continue
+
+                # Random offset within section
+                if bytes_to_verify < bytes_in_section:
+                    offset_in_section = random.randint(0, bytes_in_section - bytes_to_verify)
+                else:
+                    offset_in_section = 0
+
+                read_pos = marker_skip + (section_idx * section_size) + offset_in_section
+                verified_in_section = 0
+
+                # Seek to position in this section
+                if not self.opts.dry_run:
+                    os.lseek(fd, read_pos, os.SEEK_SET)
+
+                # Read and analyze THIS SECTION
+                while verified_in_section < bytes_to_verify:
+                    if self.do_abort:
+                        break
+
+                    chunk_size = min(read_chunk_size, bytes_to_verify - verified_in_section)
+
+                    if self.opts.dry_run:
+                        time.sleep(0.01)
+                        data = b'\x00' * chunk_size
+                    else:
+                        data = os.read(fd, chunk_size)
+                        if not data:
+                            break
+
+                    # --------------------------------------------------
+                    # SECTION ANALYSIS
+                    # --------------------------------------------------
+                    
+                    # FAST zero check for zeroed pattern
+                    if fast_fail_zeros:
+                        # Ultra-fast: compare against pre-allocated zero pattern
+                        if memoryview(data) != ZERO_PATTERN_64K[:len(data)]:
+                            failed_offset = read_pos + verified_in_section
+                            overall_failed = True
+                            failure_reason = f"non-zero at {Utils.human(failed_offset)}"
+                            break
+
+                    # FAST check for unmarked disks (looking for all zeros)
+                    if all_zeros and not section_found_nonzero:
+                        # Fast check: use bytes.count() which is C-optimized
+                        if data.count(0) != len(data):
+                            section_found_nonzero = True
+
+                    # RANDOM pattern analysis (always collect data for analysis)
+                    # Use memoryview for fast slicing
+                    mv = memoryview(data)
+                    data_len = len(data)
+                    
+                    # Sample every SAMPLE_STEP-th byte
+                    for i in range(0, data_len, SAMPLE_STEP):
+                        section_byte_counts[mv[i]] += 1
+                        section_samples += 1
+                    
+                    # --------------------------------------------------
+                    # END SECTION ANALYSIS
+                    # --------------------------------------------------
+
+                    verified_in_section += len(data)
+                    self.verify_progress += len(data)  # Track actual bytes read for progress
+
+                # After reading section, analyze it
+                if overall_failed:
+                    break
+
+                # Determine section result
+                if fast_fail_zeros:
+                    # Already passed zero check if we got here
+                    section_result = "zeroed"
+                    section_stats = {}
+                    
+                elif all_zeros:
+                    if not section_found_nonzero:
+                        section_result = "zeroed"
+                        section_stats = {}
+                    else:
+                        # Need to check if it's random
+                        section_result, section_stats = self._analyze_section_randomness(
+                            section_byte_counts, section_samples
+                        )
+                        
+                else:  # Expected random
+                    section_result, section_stats = self._analyze_section_randomness(
+                        section_byte_counts, section_samples
+                    )
+                
+                # Store section result
+                self.section_results.append((section_idx, section_result, section_stats))
+                
+                # Check if section failed
+                if (self.expected_pattern == "random" and section_result != "random") or \
+                   (self.expected_pattern == "zeroed" and section_result != "zeroed") or \
+                   (self.expected_pattern is None and section_result == "not-wiped"):
+
+                    overall_failed = True
+                    failure_reason = f"section {section_idx}: {section_result}"
+                    break
+
+            # Close file descriptor
+            if fd is not None:
+                os.close(fd)
+
+            # Determine overall result
+            if overall_failed:
+                if self.expected_pattern == "zeroed":
+                    self.verify_result = f"not-wiped ({failure_reason})"
+                elif self.expected_pattern == "random":
+                    self.verify_result = f"not-wiped ({failure_reason})"
+                else:  # unmarked
+                    # Count section results
+                    zeroed_sections = sum(1 for _, result, _ in self.section_results if result == "zeroed")
+                    random_sections = sum(1 for _, result, _ in self.section_results if result == "random")
+                    total_checked = len([r for _, r, _ in self.section_results if r != "skipped"])
+                    
+                    if zeroed_sections == total_checked:
+                        self.verify_result = "zeroed"
+                        self.expected_pattern = "zeroed"
+                    elif random_sections == total_checked:
+                        self.verify_result = "random"
+                        self.expected_pattern = "random"
+                    else:
+                        self.verify_result = f"mixed ({failure_reason})"
+            else:
+                # All sections passed
+                if self.expected_pattern == "zeroed":
+                    self.verify_result = "zeroed"
+                elif self.expected_pattern == "random":
+                    self.verify_result = "random"
+                else:  # unmarked
+                    # Determine from section consensus
+                    zeroed_sections = sum(1 for _, result, _ in self.section_results if result == "zeroed")
+                    random_sections = sum(1 for _, result, _ in self.section_results if result == "random")
+                    
+                    if zeroed_sections > random_sections:
+                        self.verify_result = "zeroed"
+                        self.expected_pattern = "zeroed"
+                    else:
+                        self.verify_result = "random"
+                        self.expected_pattern = "random"
+
+        except Exception:
+            self.exception = traceback.format_exc()
+            self.verify_result = "error"
+
+
+
+
+
+
+    def _analyze_section_randomness(self, byte_counts, total_samples):
+        """Analyze if a section appears random"""
+        if total_samples < 100:
+            return "insufficient-data", {"samples": total_samples}
+        
+        # Calculate statistics
+        max_count = max(byte_counts)
+        max_freq = max_count / total_samples
+
+        # Count unique bytes seen
+        unique_bytes = sum(1 for count in byte_counts if count > 0)
+
+        # Count completely unused bytes
+        unused_bytes = sum(1 for count in byte_counts if count == 0)
+        
+        # Calculate expected frequency and variance
+        expected = total_samples / 256
+        if expected > 0:
+            # Coefficient of variation (measure of dispersion)
+            variance = sum((count - expected) ** 2 for count in byte_counts) / 256
+            std_dev = variance ** 0.5
+            cv = std_dev / expected
+        else:
+            cv = float('inf')
+        
+        # Decision logic for "random"
+        # Good random data should:
+        # 1. Use most byte values (>200 unique)
+        # 2. No single byte dominates (<2% frequency)
+        # 3. Relatively even distribution (CV < 2.0)
+        # 4. Not too many zeros (if it's supposed to be random, not zeroed)
+        
+        is_random = (unique_bytes > 200 and      # >78% of bytes used
+                     max_freq < 0.02 and         # No byte > 2%
+                     cv < 2.0 and               # Not too lumpy
+                     byte_counts[0] / total_samples < 0.5)  # Not mostly zeros
+        
+        stats = {
+            "samples": total_samples,
+            "max_freq": max_freq,
+            "unique_bytes": unique_bytes,
+            "unused_bytes": unused_bytes,
+            "cv": cv,
+            "zero_freq": byte_counts[0] / total_samples if total_samples > 0 else 0
+        }
+        
+        if is_random:
+            return "random", stats
+        else:
+            # Check if it's zeros
+            if byte_counts[0] / total_samples > 0.95:
+                return "zeroed", stats
+            else:
+                return "not-wiped", stats
+
+
+
+
+
+
+
+
+    def _write_marker_with_verify_status(self, is_random):
+        """Write marker buffer with verification status if verification was performed
+
+        Writes marker if:
+        1. Verification was performed (verify_result is set)
+        2. Verification status would change from current marker, OR
+        3. No marker exists but verify passed (unmarked disk that's all zeros/random)
+        4. Verification was not aborted
+
+        Args:
+            is_random: bool, whether random data was written (or None to infer from verify_result)
+        """
+        try:
+            if not self.verify_result or self.verify_result == "skipped" or self.do_abort:
+                return
+
+            # Determine verify_status based on verify_result (may include debug info)
+            verify_result_base = self.verify_result.split(' ')[0] if ' ' in self.verify_result else self.verify_result
+
+            if self.expected_pattern == "random" and verify_result_base == "random":
+                new_verify_status = "pass"
+            elif self.expected_pattern == "zeroed" and verify_result_base == "zeroed":
+                new_verify_status = "pass"
+            elif verify_result_base in ("not-wiped", "mixed", "error"):
+                new_verify_status = "fail"
+            else:
+                # Mismatch: expected one pattern but got another
+                new_verify_status = "fail"
+
+            # Read existing marker to check if verify_status would change
+            device_name = os.path.basename(self.device_path)
+            existing_marker = WipeJob.read_marker_buffer(device_name)
+            existing_verify_status = (getattr(existing_marker, 'verify_status', None)
+                                      if existing_marker else None)
+
+            # For unmarked disks that verified successfully, infer the mode from verify_result
+            if not existing_marker and new_verify_status == "pass":
+                if verify_result_base == "random":
+                    is_random = True
+                    self.total_written = self.total_size  # Mark as fully wiped
+                elif verify_result_base == "zeroed":
+                    is_random = False
+                    self.total_written = self.total_size  # Mark as fully wiped
+                # Write marker for this previously unmarked disk
+            elif existing_marker:
+                # Only write if verify status changed
+                if existing_verify_status == new_verify_status:
+                    return
+                # Preserve original scrubbed_bytes if this is a verify-only job
+                if self.total_written == 0:
+                    self.total_written = existing_marker.scrubbed_bytes
+            else:
+                # No marker and verify failed - don't write marker
+                return
+
+            # Write marker with verification status
+            if not self.opts.dry_run:
+                with open(self.device_path, 'r+b') as device:
+                    device.seek(0)
+                    marker_buffer = self.prep_marker_buffer(is_random,
+                                                            verify_status=new_verify_status)
+                    device.write(marker_buffer)
+
+        except Exception:
+            # Catch ANY exception in this method to ensure self.done is always set
+            self.exception = traceback.format_exc()
+        finally:
+            # ALWAYS set done, even if there was an exception
+            self.done = True
+
+
+# Initialize the class-level buffers with mmap for O_DIRECT alignment
+if WipeJob.buffer is None:
+    # Allocate random buffer with mmap (page-aligned for O_DIRECT)
+    WipeJob.buffer_mem = mmap.mmap(-1, WipeJob.BUFFER_SIZE,
+                                   flags=mmap.MAP_PRIVATE | mmap.MAP_ANONYMOUS)
+    raw_buffer = os.urandom(WipeJob.BUFFER_SIZE)
+    rebalanced = WipeJob._rebalance_buffer(raw_buffer)
+    WipeJob.buffer_mem.write(rebalanced)
+    WipeJob.buffer_mem.seek(0)
+    WipeJob.buffer = memoryview(WipeJob.buffer_mem)
+
+    # Allocate zero buffer with mmap
+    WipeJob.zero_buffer_mem = mmap.mmap(-1, WipeJob.WRITE_SIZE,
+                                        flags=mmap.MAP_PRIVATE | mmap.MAP_ANONYMOUS)
+    WipeJob.zero_buffer_mem.write(b'\x00' * WipeJob.WRITE_SIZE)
+    WipeJob.zero_buffer_mem.seek(0)
+    WipeJob.zero_buffer = memoryview(WipeJob.zero_buffer_mem)