coreutils_rs/dd/

core.rs

use std::fs::{File, OpenOptions};
use std::io::{self, Read, Seek, SeekFrom, Write};
use std::time::Instant;

/// Status output level for dd.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub enum StatusLevel {
    /// Print transfer stats at end (default).
    #[default]
    Default,
    /// No informational messages to stderr.
    None,
    /// Print periodic transfer stats (like GNU dd `status=progress`).
    Progress,
    /// Like default but also suppress error messages.
    NoError,
    /// Print record counts but suppress transfer speed/bytes line.
    NoXfer,
}

/// Conversion flags for dd (`conv=` option).
#[derive(Debug, Clone, Default)]
pub struct DdConv {
    /// Convert to lowercase.
    pub lcase: bool,
    /// Convert to uppercase.
    pub ucase: bool,
    /// Swap every pair of input bytes.
    pub swab: bool,
    /// Continue after read errors.
    pub noerror: bool,
    /// Do not truncate the output file.
    pub notrunc: bool,
    /// Pad every input block with NULs to ibs-size.
    pub sync: bool,
    /// Call fdatasync on output before finishing.
    pub fdatasync: bool,
    /// Call fsync on output before finishing.
    pub fsync: bool,
    /// Fail if the output file already exists.
    pub excl: bool,
    /// Do not create the output file.
    pub nocreat: bool,
    /// Convert fixed-length records to newline-terminated (unblock).
    pub unblock: bool,
    /// Convert newline-terminated records to fixed-length (block).
    pub block: bool,
}

/// Input/output flags for dd (`iflag=`/`oflag=` options).
#[derive(Debug, Clone, Default)]
pub struct DdFlags {
    pub append: bool,
    pub direct: bool,
    pub directory: bool,
    pub dsync: bool,
    pub sync: bool,
    pub fullblock: bool,
    pub nonblock: bool,
    pub noatime: bool,
    pub nocache: bool,
    pub noctty: bool,
    pub nofollow: bool,
    pub count_bytes: bool,
    pub skip_bytes: bool,
}

/// Configuration for a dd operation.
#[derive(Debug, Clone)]
pub struct DdConfig {
    /// Input file path (None = stdin).
    pub input: Option<String>,
    /// Output file path (None = stdout).
    pub output: Option<String>,
    /// Input block size in bytes.
    pub ibs: usize,
    /// Output block size in bytes.
    pub obs: usize,
    /// Conversion block size (for block/unblock).
    pub cbs: usize,
    /// Copy only this many input blocks (None = unlimited).
    pub count: Option<u64>,
    /// Skip this many ibs-sized blocks at start of input.
    pub skip: u64,
    /// Skip this many obs-sized blocks at start of output.
    pub seek: u64,
    /// Conversion options.
    pub conv: DdConv,
    /// Status output level.
    pub status: StatusLevel,
    /// Input flags.
    pub iflag: DdFlags,
    /// Output flags.
    pub oflag: DdFlags,
}

impl Default for DdConfig {
    fn default() -> Self {
        DdConfig {
            input: None,
            output: None,
            ibs: 512,
            obs: 512,
            cbs: 0,
            count: None,
            skip: 0,
            seek: 0,
            conv: DdConv::default(),
            status: StatusLevel::default(),
            iflag: DdFlags::default(),
            oflag: DdFlags::default(),
        }
    }
}

/// Statistics from a dd copy operation.
#[derive(Debug, Clone, Default)]
pub struct DdStats {
    /// Number of full input blocks read.
    pub records_in_full: u64,
    /// Number of partial input blocks read.
    pub records_in_partial: u64,
    /// Number of full output blocks written.
    pub records_out_full: u64,
    /// Number of partial output blocks written.
    pub records_out_partial: u64,
    /// Total bytes copied.
    pub bytes_copied: u64,
}

/// Parse a GNU dd SIZE string with optional suffix and `x` multiplier.
///
/// Suffix conventions (matching GNU dd):
///   - Single letter = binary (powers of 1024): k/K, M, G, T, P, E
///   - `xB` suffix = decimal (powers of 1000): kB, KB, MB, GB, TB, PB, EB
///   - `xIB` suffix = explicit binary: KiB, MiB, GiB, TiB, PiB, EiB
///   - Special: c (1), w (2), b (512)
///
/// The `x` operator multiplies terms and chains recursively,
/// so `1x2x4` = 1 * (2 * 4) = 8.
pub fn parse_size(s: &str) -> Result<u64, String> {
    let s = s.trim();
    if s.is_empty() {
        return Err("empty size string".to_string());
    }

    // GNU dd supports 'x' as multiplication: e.g. "2x512", "1Mx2", "1x2x4"
    // Split on first 'x' and recurse on the right side for chaining.
    if let Some(pos) = s.find('x') {
        let left = parse_size_single(&s[..pos])?;
        let right = parse_size(&s[pos + 1..])?;
        return left
            .checked_mul(right)
            .ok_or_else(|| format!("size overflow: {} * {}", left, right));
    }

    parse_size_single(s)
}

fn parse_size_single(s: &str) -> Result<u64, String> {
    if s.is_empty() {
        return Err("empty size string".to_string());
    }

    // Find where the numeric part ends
    let num_end = s.find(|c: char| !c.is_ascii_digit()).unwrap_or(s.len());

    if num_end == 0 {
        return Err(format!("invalid number: '{}'", s));
    }

    let num: u64 = s[..num_end]
        .parse()
        .map_err(|e| format!("invalid number '{}': {}", &s[..num_end], e))?;

    let suffix = &s[num_end..];
    // GNU dd suffix convention: single letter = binary (powers of 1024),
    // xB suffix = decimal (powers of 1000), xIB suffix = binary (explicit).
    let multiplier: u64 = match suffix {
        "" => 1,
        "c" => 1,
        "w" => 2,
        "b" => 512,
        "k" | "K" => 1024,
        "kB" | "KB" => 1000,
        "KiB" => 1024,
        "M" => 1_048_576,
        "MB" => 1_000_000,
        "MiB" => 1_048_576,
        "G" => 1_073_741_824,
        "GB" => 1_000_000_000,
        "GiB" => 1_073_741_824,
        "T" => 1_099_511_627_776,
        "TB" => 1_000_000_000_000,
        "TiB" => 1_099_511_627_776,
        "P" => 1_125_899_906_842_624,
        "PB" => 1_000_000_000_000_000,
        "PiB" => 1_125_899_906_842_624,
        "E" => 1_152_921_504_606_846_976,
        "EB" => 1_000_000_000_000_000_000,
        "EiB" => 1_152_921_504_606_846_976,
        _ => return Err(format!("invalid suffix: '{}'", suffix)),
    };

    num.checked_mul(multiplier)
        .ok_or_else(|| format!("size overflow: {} * {}", num, multiplier))
}

/// Parse dd command-line arguments (key=value pairs).
pub fn parse_dd_args(args: &[String]) -> Result<DdConfig, String> {
    let mut config = DdConfig::default();
    let mut bs_set = false;

    for arg in args {
        if let Some((key, value)) = arg.split_once('=') {
            match key {
                "if" => config.input = Some(value.to_string()),
                "of" => config.output = Some(value.to_string()),
                "bs" => {
                    let size = parse_size(value)? as usize;
                    config.ibs = size;
                    config.obs = size;
                    bs_set = true;
                }
                "ibs" => {
                    if !bs_set {
                        config.ibs = parse_size(value)? as usize;
                    }
                }
                "obs" => {
                    if !bs_set {
                        config.obs = parse_size(value)? as usize;
                    }
                }
                "cbs" => config.cbs = parse_size(value)? as usize,
                "count" => config.count = Some(parse_size(value)?),
                "skip" => config.skip = parse_size(value)?,
                "seek" => config.seek = parse_size(value)?,
                "conv" => {
                    for flag in value.split(',') {
                        match flag {
                            "lcase" => config.conv.lcase = true,
                            "ucase" => config.conv.ucase = true,
                            "swab" => config.conv.swab = true,
                            "noerror" => config.conv.noerror = true,
                            "notrunc" => config.conv.notrunc = true,
                            "sync" => config.conv.sync = true,
                            "fdatasync" => config.conv.fdatasync = true,
                            "fsync" => config.conv.fsync = true,
                            "excl" => config.conv.excl = true,
                            "nocreat" => config.conv.nocreat = true,
                            "block" => config.conv.block = true,
                            "unblock" => config.conv.unblock = true,
                            "" => {}
                            _ => return Err(format!("invalid conversion: '{}'", flag)),
                        }
                    }
                }
                "iflag" => {
                    for flag in value.split(',') {
                        parse_flag(flag, &mut config.iflag)?;
                    }
                }
                "oflag" => {
                    for flag in value.split(',') {
                        parse_flag(flag, &mut config.oflag)?;
                    }
                }
                "status" => {
                    config.status = match value {
                        "none" => StatusLevel::None,
                        "noxfer" => StatusLevel::NoXfer,
                        "noerror" => StatusLevel::NoError,
                        "progress" => StatusLevel::Progress,
                        _ => return Err(format!("invalid status level: '{}'", value)),
                    };
                }
                _ => return Err(format!("unrecognized operand: '{}'", arg)),
            }
        } else {
            return Err(format!("unrecognized operand: '{}'", arg));
        }
    }

    // Validate conflicting options
    if config.conv.lcase && config.conv.ucase {
        return Err("conv=lcase and conv=ucase are mutually exclusive".to_string());
    }
    if config.conv.excl && config.conv.nocreat {
        return Err("conv=excl and conv=nocreat are mutually exclusive".to_string());
    }

    Ok(config)
}

/// Parse a single iflag/oflag value into the DdFlags struct.
fn parse_flag(flag: &str, flags: &mut DdFlags) -> Result<(), String> {
    match flag {
        "append" => flags.append = true,
        "direct" => flags.direct = true,
        "directory" => flags.directory = true,
        "dsync" => flags.dsync = true,
        "sync" => flags.sync = true,
        "fullblock" => flags.fullblock = true,
        "nonblock" => flags.nonblock = true,
        "noatime" => flags.noatime = true,
        "nocache" => flags.nocache = true,
        "noctty" => flags.noctty = true,
        "nofollow" => flags.nofollow = true,
        "count_bytes" => flags.count_bytes = true,
        "skip_bytes" => flags.skip_bytes = true,
        "" => {}
        _ => return Err(format!("invalid flag: '{}'", flag)),
    }
    Ok(())
}

/// Read a full block from the reader, retrying on partial reads.
/// Returns the number of bytes actually read (0 means EOF).
fn read_full_block(reader: &mut dyn Read, buf: &mut [u8]) -> io::Result<usize> {
    let mut total = 0;
    while total < buf.len() {
        match reader.read(&mut buf[total..]) {
            Ok(0) => break,
            Ok(n) => total += n,
            Err(ref e) if e.kind() == io::ErrorKind::Interrupted => continue,
            Err(e) => return Err(e),
        }
    }
    Ok(total)
}

/// Apply conversion options to a data block in-place.
pub fn apply_conversions(data: &mut [u8], conv: &DdConv) {
    if conv.swab {
        // Swap every pair of bytes using u64 word-at-a-time processing.
        // Process 8 bytes at a time: rotate each u16 pair within the u64.
        let (prefix, chunks, suffix) = unsafe { data.align_to_mut::<u64>() };
        // Handle unaligned prefix bytes
        let pairs_pre = prefix.len() / 2;
        for i in 0..pairs_pre {
            prefix.swap(i * 2, i * 2 + 1);
        }
        // Process aligned u64 chunks: swap adjacent bytes in each pair
        // For each u64 AABBCCDD_EEFFGGHH, we want BBAADDCC_FFEEHHGG
        // This is: ((x & 0xFF00FF00FF00FF00) >> 8) | ((x & 0x00FF00FF00FF00FF) << 8)
        for w in chunks.iter_mut() {
            let x = *w;
            *w = ((x & 0xFF00FF00FF00FF00) >> 8) | ((x & 0x00FF00FF00FF00FF) << 8);
        }
        // Handle remaining suffix bytes
        let pairs_suf = suffix.len() / 2;
        for i in 0..pairs_suf {
            suffix.swap(i * 2, i * 2 + 1);
        }
    }

    if conv.lcase {
        for b in data.iter_mut() {
            b.make_ascii_lowercase();
        }
    } else if conv.ucase {
        for b in data.iter_mut() {
            b.make_ascii_uppercase();
        }
    }
}

/// Skip input blocks by reading and discarding them.
fn skip_input(reader: &mut dyn Read, blocks: u64, block_size: usize) -> io::Result<()> {
    let mut discard_buf = vec![0u8; block_size];
    for _ in 0..blocks {
        let n = read_full_block(reader, &mut discard_buf)?;
        if n == 0 {
            break;
        }
    }
    Ok(())
}

/// Skip input by reading and discarding exactly `bytes` bytes.
fn skip_input_bytes(reader: &mut dyn Read, bytes: u64) -> io::Result<()> {
    let mut remaining = bytes;
    let mut discard_buf = [0u8; 8192];
    while remaining > 0 {
        let chunk = std::cmp::min(remaining, discard_buf.len() as u64) as usize;
        let n = reader.read(&mut discard_buf[..chunk])?;
        if n == 0 {
            break;
        }
        remaining -= n as u64;
    }
    Ok(())
}

/// Skip input blocks by seeking (for seekable file inputs).
fn skip_input_seek(file: &mut File, blocks: u64, block_size: usize) -> io::Result<()> {
    let offset = blocks * block_size as u64;
    file.seek(SeekFrom::Start(offset))?;
    Ok(())
}

/// Seek output by writing zero blocks (for non-seekable outputs) or using seek.
fn seek_output(writer: &mut Box<dyn Write>, seek_blocks: u64, block_size: usize) -> io::Result<()> {
    // Try to seek if the writer supports it. Since we use Box<dyn Write>,
    // we write zero blocks for the general case.
    let zero_block = vec![0u8; block_size];
    for _ in 0..seek_blocks {
        writer.write_all(&zero_block)?;
    }
    Ok(())
}

/// Seek output on a file using actual file seeking.
fn seek_output_file(file: &mut File, seek_blocks: u64, block_size: usize) -> io::Result<()> {
    let offset = seek_blocks * block_size as u64;
    file.seek(SeekFrom::Start(offset))?;
    Ok(())
}

/// Check if any data conversion options are enabled.
#[cfg(target_os = "linux")]
fn has_conversions(conv: &DdConv) -> bool {
    conv.lcase || conv.ucase || conv.swab || conv.sync || conv.block || conv.unblock
}

/// Check if any iflag/oflag fields require the generic path.
/// Note: noatime is excluded because the raw path already uses O_NOATIME.
/// fullblock is excluded because the raw read loop already reads full blocks.
#[cfg(target_os = "linux")]
fn has_flags(flags: &DdFlags) -> bool {
    flags.append
        || flags.direct
        || flags.directory
        || flags.dsync
        || flags.sync
        || flags.nonblock
        || flags.nocache
        || flags.noctty
        || flags.nofollow
        || flags.count_bytes
        || flags.skip_bytes
}

/// Raw-syscall fast path: when both input and output are file paths,
/// ibs == obs, no conversions, and no iflag/oflag are set, bypass
/// Box<dyn Read/Write> and use libc::read/write directly. Handles
/// char devices (e.g. /dev/zero) that copy_file_range can't handle.
#[cfg(target_os = "linux")]
fn try_raw_dd(config: &DdConfig) -> Option<io::Result<DdStats>> {
    if config.input.is_none() || config.output.is_none() {
        return None;
    }
    if has_conversions(&config.conv) || config.ibs != config.obs {
        return None;
    }
    // Bail out if any iflag/oflag is set — we don't apply open() flags here
    if has_flags(&config.iflag) || has_flags(&config.oflag) {
        return None;
    }

    let start_time = Instant::now();
    let in_path = config.input.as_ref().unwrap();
    let out_path = config.output.as_ref().unwrap();

    // Build CStrings before opening any FDs to avoid leaks on interior NUL
    let in_cstr = match std::ffi::CString::new(in_path.as_str()) {
        Ok(c) => c,
        Err(_) => {
            return Some(Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                format!("input path contains NUL byte: '{}'", in_path),
            )));
        }
    };
    let out_cstr = match std::ffi::CString::new(out_path.as_str()) {
        Ok(c) => c,
        Err(_) => {
            return Some(Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                format!("output path contains NUL byte: '{}'", out_path),
            )));
        }
    };

    // Open input (O_CLOEXEC prevents FD inheritance in child processes)
    let in_fd = unsafe {
        libc::open(
            in_cstr.as_ptr(),
            libc::O_RDONLY | libc::O_CLOEXEC | libc::O_NOATIME,
        )
    };
    let in_fd = if in_fd < 0 {
        let first_err = io::Error::last_os_error();
        if first_err.raw_os_error() == Some(libc::EPERM) {
            // Retry without O_NOATIME — only EPERM means "file not owned by us"
            let fd = unsafe { libc::open(in_cstr.as_ptr(), libc::O_RDONLY | libc::O_CLOEXEC) };
            if fd < 0 {
                return Some(Err(io::Error::last_os_error()));
            }
            fd
        } else {
            return Some(Err(first_err));
        }
    } else {
        in_fd
    };

    // Open output (O_CLOEXEC prevents FD inheritance)
    let mut oflags = libc::O_WRONLY | libc::O_CLOEXEC;
    if config.conv.excl {
        oflags |= libc::O_CREAT | libc::O_EXCL;
    } else if config.conv.nocreat {
        // don't create
    } else {
        oflags |= libc::O_CREAT;
    }
    if !config.conv.notrunc && !config.conv.excl {
        oflags |= libc::O_TRUNC;
    }

    let out_fd = unsafe { libc::open(out_cstr.as_ptr(), oflags, 0o666 as libc::mode_t) };
    if out_fd < 0 {
        unsafe { libc::close(in_fd) };
        return Some(Err(io::Error::last_os_error()));
    }

    // Hint kernel for sequential readahead on input
    unsafe {
        libc::posix_fadvise(in_fd, 0, 0, libc::POSIX_FADV_SEQUENTIAL);
    }

    // Handle skip (seek input) — use checked_mul to prevent overflow
    if config.skip > 0 {
        let offset = match (config.skip as u64).checked_mul(config.ibs as u64) {
            Some(o) if o <= i64::MAX as u64 => o as i64,
            _ => {
                unsafe {
                    libc::close(in_fd);
                    libc::close(out_fd);
                }
                return Some(Err(io::Error::new(
                    io::ErrorKind::InvalidInput,
                    "skip offset overflow",
                )));
            }
        };
        if unsafe { libc::lseek(in_fd, offset, libc::SEEK_SET) } < 0 {
            // lseek failed (e.g. char device) — read and discard full blocks
            let mut discard = vec![0u8; config.ibs];
            'skip: for _ in 0..config.skip {
                let mut skipped = 0usize;
                while skipped < config.ibs {
                    let n = unsafe {
                        libc::read(
                            in_fd,
                            discard[skipped..].as_mut_ptr() as *mut _,
                            config.ibs - skipped,
                        )
                    };
                    if n > 0 {
                        skipped += n as usize;
                    } else if n == 0 {
                        break 'skip; // EOF
                    } else {
                        let err = io::Error::last_os_error();
                        if err.kind() == io::ErrorKind::Interrupted {
                            continue;
                        }
                        // Non-EINTR error during skip — log and abort skip phase
                        eprintln!("dd: error skipping input: {}", err);
                        break 'skip;
                    }
                }
            }
        }
    }

    // Handle seek (seek output) — use checked_mul to prevent overflow
    if config.seek > 0 {
        let offset = match (config.seek as u64).checked_mul(config.obs as u64) {
            Some(o) if o <= i64::MAX as u64 => o as i64,
            _ => {
                unsafe {
                    libc::close(in_fd);
                    libc::close(out_fd);
                }
                return Some(Err(io::Error::new(
                    io::ErrorKind::InvalidInput,
                    "seek offset overflow",
                )));
            }
        };
        if unsafe { libc::lseek(out_fd, offset, libc::SEEK_SET) } < 0 {
            let err = io::Error::last_os_error();
            unsafe {
                libc::close(in_fd);
                libc::close(out_fd);
            }
            return Some(Err(err));
        }
    }

    let mut stats = DdStats::default();
    let bs = config.ibs;
    let mut ibuf = vec![0u8; bs];
    let count_limit = config.count;

    loop {
        if let Some(limit) = count_limit {
            if stats.records_in_full + stats.records_in_partial >= limit {
                break;
            }
        }

        // Raw read — retry on EINTR, loop for full block
        let mut total_read = 0usize;
        let mut read_error = false;
        while total_read < bs {
            let ret = unsafe {
                libc::read(
                    in_fd,
                    ibuf[total_read..].as_mut_ptr() as *mut _,
                    bs - total_read,
                )
            };
            if ret > 0 {
                total_read += ret as usize;
            } else if ret == 0 {
                break; // EOF
            } else {
                let err = io::Error::last_os_error();
                if err.kind() == io::ErrorKind::Interrupted {
                    continue;
                }
                if config.conv.noerror {
                    eprintln!("dd: error reading '{}': {}", in_path, err);
                    read_error = true;
                    break;
                }
                unsafe {
                    libc::close(in_fd);
                    libc::close(out_fd);
                }
                return Some(Err(err));
            }
        }

        // conv=noerror: skip entire bad block (GNU behavior)
        if read_error {
            stats.records_in_partial += 1;
            continue;
        }

        if total_read == 0 {
            break;
        }

        if total_read == bs {
            stats.records_in_full += 1;
        } else {
            stats.records_in_partial += 1;
        }

        // Raw write — retry on EINTR, treat write(0) as error
        let mut written = 0usize;
        while written < total_read {
            let ret = unsafe {
                libc::write(
                    out_fd,
                    ibuf[written..].as_ptr() as *const _,
                    total_read - written,
                )
            };
            if ret > 0 {
                written += ret as usize;
            } else if ret == 0 {
                // write() returning 0 is abnormal — treat as error
                unsafe {
                    libc::close(in_fd);
                    libc::close(out_fd);
                }
                return Some(Err(io::Error::new(
                    io::ErrorKind::WriteZero,
                    "write returned 0",
                )));
            } else {
                let err = io::Error::last_os_error();
                if err.kind() == io::ErrorKind::Interrupted {
                    continue;
                }
                unsafe {
                    libc::close(in_fd);
                    libc::close(out_fd);
                }
                return Some(Err(err));
            }
        }

        stats.bytes_copied += written as u64;
        if written == bs {
            stats.records_out_full += 1;
        } else {
            stats.records_out_partial += 1;
        }
    }

    // fsync / fdatasync — propagate errors
    if config.conv.fsync {
        if unsafe { libc::fsync(out_fd) } < 0 {
            let err = io::Error::last_os_error();
            unsafe {
                libc::close(in_fd);
                libc::close(out_fd);
            }
            return Some(Err(err));
        }
    } else if config.conv.fdatasync {
        if unsafe { libc::fdatasync(out_fd) } < 0 {
            let err = io::Error::last_os_error();
            unsafe {
                libc::close(in_fd);
                libc::close(out_fd);
            }
            return Some(Err(err));
        }
    }

    unsafe { libc::close(in_fd) };
    // Check close(out_fd) — on NFS, close can report deferred write errors
    if unsafe { libc::close(out_fd) } < 0 {
        return Some(Err(io::Error::last_os_error()));
    }

    if config.status != StatusLevel::None {
        print_stats(&stats, start_time.elapsed(), config.status);
    }

    Some(Ok(stats))
}

/// Fast path: use copy_file_range when both input and output are files
/// and no conversions are needed. This is zero-copy in the kernel.
#[cfg(target_os = "linux")]
fn try_copy_file_range_dd(config: &DdConfig) -> Option<io::Result<DdStats>> {
    // Only usable when both are files, no conversions, and ibs == obs
    if config.input.is_none() || config.output.is_none() {
        return None;
    }
    if has_conversions(&config.conv) || config.ibs != config.obs {
        return None;
    }

    let start_time = Instant::now();
    let in_path = config.input.as_ref().unwrap();
    let out_path = config.output.as_ref().unwrap();

    let in_file = match File::open(in_path) {
        Ok(f) => f,
        Err(e) => return Some(Err(e)),
    };

    // Hint kernel for sequential readahead
    {
        use std::os::unix::io::AsRawFd;
        unsafe {
            libc::posix_fadvise(in_file.as_raw_fd(), 0, 0, libc::POSIX_FADV_SEQUENTIAL);
        }
    }

    let mut out_opts = OpenOptions::new();
    out_opts.write(true);
    if config.conv.excl {
        out_opts.create_new(true);
    } else if !config.conv.nocreat {
        out_opts.create(true);
    }
    if !config.conv.notrunc && !config.conv.excl {
        out_opts.truncate(true);
    }

    let out_file = match out_opts.open(out_path) {
        Ok(f) => f,
        Err(e) => return Some(Err(e)),
    };

    use std::os::unix::io::AsRawFd;
    let in_fd = in_file.as_raw_fd();
    let out_fd = out_file.as_raw_fd();

    // Handle skip
    let skip_bytes = config.skip * config.ibs as u64;
    let seek_bytes = config.seek * config.obs as u64;
    let mut in_off: i64 = skip_bytes as i64;
    let mut out_off: i64 = seek_bytes as i64;

    let mut stats = DdStats::default();
    let block_size = config.ibs;

    // Determine total bytes to copy
    let total_to_copy = config.count.map(|count| count * block_size as u64);

    let mut bytes_remaining = total_to_copy;
    loop {
        let chunk = match bytes_remaining {
            Some(0) => break,
            Some(r) => r.min(block_size as u64 * 1024) as usize, // copy in large chunks
            None => block_size * 1024,
        };

        // SAFETY: in_fd and out_fd are valid file descriptors (files are open for the
        // lifetime of this function). in_off and out_off are valid, aligned i64 pointers
        // with no aliasing. The kernel updates offsets atomically. Return value is checked:
        // negative = error, 0 = EOF, positive = bytes copied.
        let ret = unsafe {
            libc::syscall(
                libc::SYS_copy_file_range,
                in_fd,
                &mut in_off as *mut i64,
                out_fd,
                &mut out_off as *mut i64,
                chunk,
                0u32,
            )
        };

        if ret < 0 {
            let err = io::Error::last_os_error();
            if err.raw_os_error() == Some(libc::EINVAL)
                || err.raw_os_error() == Some(libc::ENOSYS)
                || err.raw_os_error() == Some(libc::EXDEV)
            {
                return None; // Fall back to regular copy
            }
            return Some(Err(err));
        }
        if ret == 0 {
            break;
        }

        let copied = ret as u64;
        stats.bytes_copied += copied;

        // Track block stats
        let full_blocks = copied / block_size as u64;
        let partial = copied % block_size as u64;
        stats.records_in_full += full_blocks;
        stats.records_out_full += full_blocks;
        if partial > 0 {
            stats.records_in_partial += 1;
            stats.records_out_partial += 1;
        }

        if let Some(ref mut r) = bytes_remaining {
            *r = r.saturating_sub(copied);
        }
    }

    // fsync / fdatasync
    if config.conv.fsync {
        if let Err(e) = out_file.sync_all() {
            return Some(Err(e));
        }
    } else if config.conv.fdatasync {
        if let Err(e) = out_file.sync_data() {
            return Some(Err(e));
        }
    }

    if config.status != StatusLevel::None {
        print_stats(&stats, start_time.elapsed(), config.status);
    }

    Some(Ok(stats))
}

/// Perform the dd copy operation.
pub fn dd_copy(config: &DdConfig) -> io::Result<DdStats> {
    // Try zero-copy fast path on Linux (file-to-file)
    #[cfg(target_os = "linux")]
    {
        if let Some(result) = try_copy_file_range_dd(config) {
            return result;
        }
    }
    // Raw syscall fast path: handles devices like /dev/zero where copy_file_range fails
    #[cfg(target_os = "linux")]
    {
        if let Some(result) = try_raw_dd(config) {
            return result;
        }
    }
    let start_time = Instant::now();

    // Only clone file handles when skip/seek are needed (avoids dup() syscalls otherwise)
    let needs_input_seek = config.skip > 0;
    let needs_output_seek = config.seek > 0;

    let mut input_file: Option<File> = None;
    let mut input: Box<dyn Read> = if let Some(ref path) = config.input {
        let file = File::open(path)
            .map_err(|e| io::Error::new(e.kind(), format!("failed to open '{}': {}", path, e)))?;
        // Hint kernel for sequential readahead
        #[cfg(target_os = "linux")]
        {
            use std::os::unix::io::AsRawFd;
            unsafe {
                libc::posix_fadvise(file.as_raw_fd(), 0, 0, libc::POSIX_FADV_SEQUENTIAL);
            }
        }
        if needs_input_seek {
            input_file = Some(file.try_clone()?);
        }
        Box::new(file)
    } else {
        Box::new(io::stdin())
    };

    // Handle output file creation/opening
    let mut output_file: Option<File> = None;
    let mut output: Box<dyn Write> = if let Some(ref path) = config.output {
        let mut opts = OpenOptions::new();
        opts.write(true);

        if config.conv.excl {
            // excl: fail if file exists (create_new implies create)
            opts.create_new(true);
        } else if config.conv.nocreat {
            // nocreat: do not create, file must exist
            // Don't set create at all
        } else {
            opts.create(true);
        }

        if config.conv.notrunc {
            opts.truncate(false);
        } else if !config.conv.excl {
            // Default: truncate (but not with excl since create_new starts fresh)
            opts.truncate(true);
        }

        let file = opts
            .open(path)
            .map_err(|e| io::Error::new(e.kind(), format!("failed to open '{}': {}", path, e)))?;
        if needs_output_seek || config.conv.fsync || config.conv.fdatasync {
            // Clone for: (1) seek positioning (Box<dyn Write> can't seek directly),
            // and (2) sync_all/sync_data at end. Safe because dup()-cloned fds
            // share the same open file description.
            output_file = Some(file.try_clone()?);
        }
        Box::new(file)
    } else {
        Box::new(io::stdout())
    };

    // Skip input — use seek() for file inputs to avoid reading and discarding data
    if config.skip > 0 {
        if config.iflag.skip_bytes {
            // skip_bytes: skip N bytes, not N blocks
            if let Some(ref mut f) = input_file {
                f.seek(SeekFrom::Start(config.skip))?;
                let seeked = f.try_clone()?;
                input = Box::new(seeked);
            } else {
                skip_input_bytes(&mut input, config.skip)?;
            }
        } else if let Some(ref mut f) = input_file {
            skip_input_seek(f, config.skip, config.ibs)?;
            // Rebuild the input Box with a clone at the seeked position
            let seeked = f.try_clone()?;
            input = Box::new(seeked);
        } else {
            skip_input(&mut input, config.skip, config.ibs)?;
        }
    }

    // Seek output blocks
    if config.seek > 0 {
        if let Some(ref mut f) = output_file {
            seek_output_file(f, config.seek, config.obs)?;
            // Rebuild the output Box with a new clone at the seeked position
            let seeked = f.try_clone()?;
            output = Box::new(seeked);
        } else {
            seek_output(&mut output, config.seek, config.obs)?;
        }
    }

    let mut stats = DdStats::default();
    let mut ibuf = vec![0u8; config.ibs];
    let mut obuf: Vec<u8> = Vec::with_capacity(config.obs);
    let mut unblock_buf: Vec<u8> = Vec::new();
    // For count_bytes mode, track total bytes read
    let mut bytes_read_total: u64 = 0;

    loop {
        // Check count limit
        if let Some(count) = config.count {
            if config.iflag.count_bytes {
                if bytes_read_total >= count {
                    break;
                }
            } else if stats.records_in_full + stats.records_in_partial >= count {
                break;
            }
        }

        // When count_bytes is active, limit the read to the remaining bytes
        let read_size = if config.iflag.count_bytes {
            if let Some(count) = config.count {
                let remaining = count.saturating_sub(bytes_read_total);
                std::cmp::min(config.ibs, remaining as usize)
            } else {
                config.ibs
            }
        } else {
            config.ibs
        };
        if read_size == 0 {
            break;
        }

        // Read one input block
        let n = match read_full_block(&mut input, &mut ibuf[..read_size]) {
            Ok(n) => n,
            Err(e) => {
                if config.conv.noerror {
                    if config.status != StatusLevel::None {
                        eprintln!("dd: error reading input: {}", e);
                    }
                    // On noerror with sync, fill the entire block with NULs
                    if config.conv.sync {
                        ibuf.fill(0);
                        config.ibs
                    } else {
                        continue;
                    }
                } else {
                    return Err(e);
                }
            }
        };

        if n == 0 {
            break;
        }

        bytes_read_total += n as u64;

        // Track full vs partial blocks
        if n == config.ibs {
            stats.records_in_full += 1;
        } else {
            stats.records_in_partial += 1;
            // Pad if conv=sync: spaces for block/unblock, NULs otherwise
            if config.conv.sync {
                let pad_byte = if config.conv.block || config.conv.unblock {
                    b' '
                } else {
                    0u8
                };
                ibuf[n..config.ibs].fill(pad_byte);
            }
        }

        // Determine the data slice to use and apply conversions in-place
        let effective_len = if config.conv.sync { config.ibs } else { n };
        apply_conversions(&mut ibuf[..effective_len], &config.conv);

        // Apply unblock conversion: split fixed-length records into
        // newline-terminated records with trailing spaces stripped
        let write_data: &[u8] = if config.conv.unblock && config.cbs > 0 {
            unblock_buf.clear();
            let data = &ibuf[..effective_len];
            let mut pos = 0;
            while pos < data.len() {
                let end = std::cmp::min(pos + config.cbs, data.len());
                let record = &data[pos..end];
                // Strip trailing spaces
                let trimmed_len = record
                    .iter()
                    .rposition(|&b| b != b' ')
                    .map(|p| p + 1)
                    .unwrap_or(0);
                unblock_buf.extend_from_slice(&record[..trimmed_len]);
                unblock_buf.push(b'\n');
                pos = end;
            }
            &unblock_buf
        } else {
            &ibuf[..effective_len]
        };

        // Buffer output and flush when we have enough for a full output block.
        // Use efficient buffer management: write directly from ibuf when possible,
        // only buffer when ibs != obs.
        let wd_len = write_data.len();
        if config.ibs == config.obs && obuf.is_empty() && !config.conv.unblock {
            // Fast path: ibs == obs, write directly
            output.write_all(write_data)?;
            if wd_len == config.obs {
                stats.records_out_full += 1;
            } else {
                stats.records_out_partial += 1;
            }
            stats.bytes_copied += wd_len as u64;
            // Skip the drain loop below since we wrote directly
            continue;
        }

        // Append write_data to obuf and drain full output blocks.
        // We write directly from write_data when possible to avoid copying
        // through obuf. Only buffer the remainder that doesn't fill a block.
        let obs = config.obs;
        let mut wd_off = 0;

        // If obuf has leftover bytes, try to complete a full block
        if !obuf.is_empty() {
            let need = obs - obuf.len();
            if write_data.len() >= need {
                obuf.extend_from_slice(&write_data[..need]);
                output.write_all(&obuf)?;
                stats.records_out_full += 1;
                stats.bytes_copied += obs as u64;
                obuf.clear();
                wd_off = need;
            } else {
                obuf.extend_from_slice(write_data);
                wd_off = write_data.len();
            }
        }

        // Write full blocks directly from write_data (zero-copy)
        let remaining_wd = &write_data[wd_off..];
        let full_blocks = remaining_wd.len() / obs;
        if full_blocks > 0 {
            let full_len = full_blocks * obs;
            output.write_all(&remaining_wd[..full_len])?;
            stats.records_out_full += full_blocks as u64;
            stats.bytes_copied += full_len as u64;
            wd_off += full_len;
        }

        // Buffer any remaining partial block
        let leftover = &write_data[wd_off..];
        if !leftover.is_empty() {
            obuf.extend_from_slice(leftover);
        }
    }

    // Flush remaining partial output block
    if !obuf.is_empty() {
        output.write_all(&obuf)?;
        stats.records_out_partial += 1;
        stats.bytes_copied += obuf.len() as u64;
    }

    // Flush output
    output.flush()?;

    // fsync / fdatasync (output_file is Some when seek or sync was requested)
    if let Some(ref f) = output_file {
        if config.conv.fsync {
            f.sync_all()?;
        } else if config.conv.fdatasync {
            f.sync_data()?;
        }
    }

    let elapsed = start_time.elapsed();

    // Print status
    if config.status != StatusLevel::None {
        print_stats(&stats, elapsed, config.status);
    }

    Ok(stats)
}

/// Print dd transfer statistics to stderr.
fn print_stats(stats: &DdStats, elapsed: std::time::Duration, status: StatusLevel) {
    eprintln!(
        "{}+{} records in",
        stats.records_in_full, stats.records_in_partial
    );
    eprintln!(
        "{}+{} records out",
        stats.records_out_full, stats.records_out_partial
    );

    if status == StatusLevel::NoXfer {
        return;
    }

    let secs = elapsed.as_secs_f64();
    if secs > 0.0 {
        let rate = stats.bytes_copied as f64 / secs;
        eprintln!(
            "{} bytes copied, {:.6} s, {}/s",
            stats.bytes_copied,
            secs,
            human_size(rate as u64)
        );
    } else {
        eprintln!("{} bytes copied", stats.bytes_copied);
    }
}

/// Format a byte count as a human-readable string (e.g., "1.5 MB").
fn human_size(bytes: u64) -> String {
    const UNITS: &[&str] = &["B", "kB", "MB", "GB", "TB", "PB", "EB"];
    let mut size = bytes as f64;
    for &unit in UNITS {
        if size < 1000.0 {
            if size == size.floor() {
                return format!("{} {}", size as u64, unit);
            }
            return format!("{:.1} {}", size, unit);
        }
        size /= 1000.0;
    }
    format!("{:.1} EB", size * 1000.0)
}

/// Print help message for dd.
pub fn print_help() {
    eprint!(
        "\
Usage: dd [OPERAND]...
  or:  dd OPTION
Copy a file, converting and formatting according to the operands.

  bs=BYTES        read and write up to BYTES bytes at a time (default: 512)
  cbs=BYTES       convert BYTES bytes at a time
  conv=CONVS      convert the file as per the comma separated symbol list
  count=N         copy only N input blocks
  ibs=BYTES       read up to BYTES bytes at a time (default: 512)
  if=FILE         read from FILE instead of stdin
  iflag=FLAGS     read as per the comma separated symbol list
  obs=BYTES       write BYTES bytes at a time (default: 512)
  of=FILE         write to FILE instead of stdout
  oflag=FLAGS     write as per the comma separated symbol list
  seek=N          skip N obs-sized blocks at start of output
  skip=N          skip N ibs-sized blocks at start of input
  status=LEVEL    LEVEL of information to print to stderr;
                  'none' suppresses everything but error messages,
                  'noerror' suppresses the final transfer statistics,
                  'progress' shows periodic transfer statistics

  BLOCKS and BYTES may be followed by the following multiplicative suffixes:
  c=1, w=2, b=512, kB=1000, K=1024, MB=1000*1000, M=1024*1024,
  GB=1000*1000*1000, GiB=1024*1024*1024, and so on for T, P, E.

Each CONV symbol may be:

  lcase     change upper case to lower case
  ucase     change lower case to upper case
  swab      swap every pair of input bytes
  sync      pad every input block with NULs to ibs-size
  noerror   continue after read errors
  notrunc   do not truncate the output file
  fdatasync physically write output file data before finishing
  fsync     likewise, but also write metadata
  excl      fail if the output file already exists
  nocreat   do not create the output file

Each FLAG symbol may be:

  append    append mode (makes sense only for output; conv=notrunc suggested)
  direct    use direct I/O for data
  directory fail unless a directory
  dsync     use synchronized I/O for data
  sync      likewise, but also for metadata
  fullblock accumulate full blocks of input (iflag only)
  nonblock  use non-blocking I/O
  noatime   do not update access time
  nocache   Request to drop cache
  noctty    do not assign controlling terminal from file
  nofollow  do not follow symlinks
  count_bytes  treat 'count=N' as a byte count (iflag only)
  skip_bytes   treat 'skip=N' as a byte count (iflag only)

  --help     display this help and exit
  --version  output version information and exit
"
    );
}

/// Print version information for dd.
pub fn print_version() {
    eprintln!("dd (fcoreutils) {}", env!("CARGO_PKG_VERSION"));
}