zosstorage/src/fs/plan.rs

// REGION: API
// api: fs::FsKind { Vfat, Btrfs, Bcachefs }
// api: fs::FsSpec { kind: FsKind, devices: Vec<String>, label: String }
// api: fs::FsPlan { specs: Vec<FsSpec> }
// api: fs::FsResult { kind: FsKind, devices: Vec<String>, uuid: String, label: String }
// api: fs::plan_filesystems(parts: &[crate::partition::PartitionResult], cfg: &crate::config::types::Config) -> crate::Result<FsPlan>
// api: fs::make_filesystems(plan: &FsPlan, cfg: &crate::types::Config) -> crate::Result<Vec<FsResult>>
// REGION: API-END
//
// REGION: RESPONSIBILITIES
// - Map partition roles to concrete filesystems (vfat for ESP, btrfs for data, bcachefs for SSD+HDD).
// - Execute mkfs operations via external tooling wrappers and capture resulting UUIDs/labels.
// Non-goals: partition layout decisions, mount orchestration, device discovery.
// REGION: RESPONSIBILITIES-END
//
// REGION: EXTENSION_POINTS
// ext: support additional filesystems or tuning flags through Config (e.g., more btrfs/bcachefs options).
// ext: dry-run mode to emit mkfs commands without executing (future).
// REGION: EXTENSION_POINTS-END
//
// REGION: SAFETY
// safety: must not run mkfs on non-empty or unexpected partitions; assume prior validation enforced.
// safety: ensure labels follow reserved semantics (ZOSBOOT for ESP, ZOSDATA for all data FS).
// safety: mkfs.btrfs uses -f in apply path immediately after partitioning to handle leftover signatures.
// REGION: SAFETY-END
//
// REGION: ERROR_MAPPING
// errmap: external mkfs/blkid failures -> crate::Error::Tool with captured stderr.
// errmap: planning mismatches -> crate::Error::Filesystem with context.
// REGION: ERROR_MAPPING-END
//
// REGION: TODO
// todo: bcachefs tuning flags mapping from config (compression/checksum/cache_mode) deferred
// todo: add UUID consistency checks across multi-device filesystems
// REGION: TODO-END
//! Filesystem planning and creation for zosstorage.
//!
//! Maps partition results to concrete filesystems (vfat, btrfs, bcachefs)
//! and executes mkfs operations via external tooling wrappers.
//!
//! See [fn plan_filesystems](plan.rs:1) and
//! [fn make_filesystems](plan.rs:1).

use crate::{
    Error, Result,
    partition::{PartRole, PartitionResult},
    types::{Config, Topology},
    util::{run_cmd, run_cmd_capture, which_tool},
};
use std::fs;
use tracing::{debug, warn};

/// Filesystem kinds supported by zosstorage.
#[derive(Debug, Clone, Copy)]
pub enum FsKind {
    /// FAT32 for the EFI System Partition.
    Vfat,
    /// Btrfs for data filesystems in single/dual topologies.
    Btrfs,
    /// Bcachefs for SSD+HDD topology (SSD as cache/promote, HDD backing).
    Bcachefs,
}

/// Declarative specification for creating a filesystem.
#[derive(Debug, Clone)]
pub struct FsSpec {
    /// Filesystem kind.
    pub kind: FsKind,
    /// Source device(s):
    /// - single path for vfat and btrfs
    /// - two paths for bcachefs (cache + backing)
    pub devices: Vec<String>,
    /// Filesystem label:
    /// - "ZOSBOOT" for ESP
    /// - "ZOSDATA" for all data filesystems
    pub label: String,
}

/// Plan of filesystem creations.
#[derive(Debug, Clone)]
pub struct FsPlan {
    /// All filesystem creation specs.
    pub specs: Vec<FsSpec>,
}

/// Result of creating a filesystem.
#[derive(Debug, Clone)]
pub struct FsResult {
    /// Filesystem kind.
    pub kind: FsKind,
    /// Devices the filesystem was created on.
    pub devices: Vec<String>,
    /// Filesystem UUID (string as reported by blkid or related).
    pub uuid: String,
    /// Filesystem label ("ZOSBOOT" or "ZOSDATA").
    pub label: String,
}

/// Determine which partitions get which filesystem based on topology.
///
/// Rules:
/// - ESP partitions => Vfat with label from cfg.filesystem.vfat.label (reserved "ZOSBOOT")
/// - Data partitions => Btrfs with label cfg.filesystem.btrfs.label ("ZOSDATA"), unless topology SsdHddBcachefs
/// - SsdHddBcachefs => pair one Cache partition (SSD) with one Data partition (HDD) into one Bcachefs FsSpec with devices [cache, data] and label cfg.filesystem.bcachefs.label ("ZOSDATA")
/// - DualIndependent/BtrfsRaid1 => map each Data partition to its own Btrfs FsSpec (raid profile concerns are handled later during mkfs)
pub fn plan_filesystems(parts: &[PartitionResult], cfg: &Config) -> Result<FsPlan> {
    let mut specs: Vec<FsSpec> = Vec::new();

    // Always map ESP partitions
    for p in parts.iter().filter(|p| matches!(p.role, PartRole::Esp)) {
        specs.push(FsSpec {
            kind: FsKind::Vfat,
            devices: vec![p.device_path.clone()],
            label: cfg.filesystem.vfat.label.clone(),
        });
    }

    match cfg.topology {
        Topology::SsdHddBcachefs => {
            // Expect exactly one cache (SSD) and at least one data (HDD). Use the first data for pairing.
            let cache = parts
                .iter()
                .find(|p| matches!(p.role, PartRole::Cache))
                .ok_or_else(|| {
                    Error::Filesystem(
                        "expected a Cache partition for SsdHddBcachefs topology".to_string(),
                    )
                })?;
            let data = parts
                .iter()
                .find(|p| matches!(p.role, PartRole::Data))
                .ok_or_else(|| {
                    Error::Filesystem(
                        "expected a Data partition for SsdHddBcachefs topology".to_string(),
                    )
                })?;

            specs.push(FsSpec {
                kind: FsKind::Bcachefs,
                devices: vec![cache.device_path.clone(), data.device_path.clone()],
                label: cfg.filesystem.bcachefs.label.clone(),
            });
        }
        Topology::BtrfsRaid1 => {
            // Group all Data partitions into a single Btrfs filesystem across multiple devices.
            let data_devs: Vec<String> = parts
                .iter()
                .filter(|p| matches!(p.role, PartRole::Data))
                .map(|p| p.device_path.clone())
                .collect();

            if data_devs.len() < 2 {
                return Err(Error::Filesystem(
                    "BtrfsRaid1 topology requires at least 2 data partitions".to_string(),
                ));
            }

            specs.push(FsSpec {
                kind: FsKind::Btrfs,
                devices: data_devs,
                label: cfg.filesystem.btrfs.label.clone(),
            });
        }
        Topology::Bcachefs2Copy => {
            // Group all Data partitions into a single Bcachefs filesystem across multiple devices (2-copy semantics).
            let data_devs: Vec<String> = parts
                .iter()
                .filter(|p| matches!(p.role, PartRole::Data))
                .map(|p| p.device_path.clone())
                .collect();
            if data_devs.len() < 2 {
                return Err(Error::Filesystem(
                    "Bcachefs2Copy topology requires at least 2 data partitions".to_string(),
                ));
            }
            specs.push(FsSpec {
                kind: FsKind::Bcachefs,
                devices: data_devs,
                label: cfg.filesystem.bcachefs.label.clone(),
            });
        }
        Topology::BcachefsSingle => {
            // Single-device bcachefs on the sole Data partition.
            let data = parts
                .iter()
                .find(|p| matches!(p.role, PartRole::Data))
                .ok_or_else(|| {
                    Error::Filesystem(
                        "expected a Data partition for BcachefsSingle topology".to_string(),
                    )
                })?;
            specs.push(FsSpec {
                kind: FsKind::Bcachefs,
                devices: vec![data.device_path.clone()],
                label: cfg.filesystem.bcachefs.label.clone(),
            });
        }
        Topology::BtrfsSingle | Topology::DualIndependent => {
            // Map Data partition(s) to Btrfs (single device per partition for DualIndependent).
            for p in parts.iter().filter(|p| matches!(p.role, PartRole::Data)) {
                specs.push(FsSpec {
                    kind: FsKind::Btrfs,
                    devices: vec![p.device_path.clone()],
                    label: cfg.filesystem.btrfs.label.clone(),
                });
            }
        }
    }

    if specs.is_empty() {
        return Err(Error::Filesystem(
            "no filesystems to create from provided partitions".to_string(),
        ));
    }

    Ok(FsPlan { specs })
}

/// Create the filesystems and return identity info (UUIDs, labels).
///
//// Uses external tooling via util wrappers (mkfs.vfat, mkfs.btrfs, bcachefs format).
/// Notes:
/// - This initial implementation applies labels and creates filesystems with minimal flags.
/// - Btrfs RAID profile (e.g., raid1) will be applied in a follow-up by mapping config to mkfs flags.
/// - UUID is captured via blkid -o export on the first device of each spec.
pub fn make_filesystems(plan: &FsPlan, cfg: &Config) -> Result<Vec<FsResult>> {
    // Discover required tools up-front
    let vfat_tool = which_tool("mkfs.vfat")?;
    let btrfs_tool = which_tool("mkfs.btrfs")?;
    let bcachefs_tool = which_tool("bcachefs")?;
    let blkid_tool = which_tool("blkid")?;

    if blkid_tool.is_none() {
        return Err(Error::Filesystem(
            "blkid not found in PATH; cannot capture filesystem UUIDs".into(),
        ));
    }
    let blkid = blkid_tool.unwrap();

    let mut results: Vec<FsResult> = Vec::new();

    for spec in &plan.specs {
        match spec.kind {
            FsKind::Vfat => {
                let Some(ref mkfs) = vfat_tool else {
                    return Err(Error::Filesystem("mkfs.vfat not found in PATH".into()));
                };
                if spec.devices.len() != 1 {
                    return Err(Error::Filesystem("vfat requires exactly one device".into()));
                }
                let dev = &spec.devices[0];
                // mkfs.vfat -n LABEL /dev/...
                run_cmd(&[mkfs.as_str(), "-n", spec.label.as_str(), dev.as_str()])?;

                // Capture UUID
                let uuid = capture_uuid(&blkid, dev)?;
                results.push(FsResult {
                    kind: FsKind::Vfat,
                    devices: vec![dev.clone()],
                    uuid,
                    label: spec.label.clone(),
                });
            }
            FsKind::Btrfs => {
                let Some(ref mkfs) = btrfs_tool else {
                    return Err(Error::Filesystem("mkfs.btrfs not found in PATH".into()));
                };
                if spec.devices.is_empty() {
                    return Err(Error::Filesystem(
                        "btrfs requires at least one device".into(),
                    ));
                }
                // mkfs.btrfs -L LABEL [ -m raid1 -d raid1 (when multi-device/raid1) ] dev1 [dev2 ...]
                let mut args: Vec<String> = vec![mkfs.clone(), "-L".into(), spec.label.clone()];

                // If this Btrfs is multi-device (as planned in BtrfsRaid1 topology),
                // set metadata/data profiles to raid1. This keeps plan/apply consistent.
                if spec.devices.len() >= 2 {
                    args.push("-m".into());
                    args.push("raid1".into());
                    args.push("-d".into());
                    args.push("raid1".into());
                }

                // Note: compression is a mount-time option for btrfs; we will apply it in mount phase.
                // Leaving mkfs-time compression unset by design.

                // Force formatting in apply path to avoid leftover signatures on freshly created partitions.
                // Safe because we just created these partitions in this run.
                args.push("-f".into());

                args.extend(spec.devices.iter().cloned());
                let args_ref: Vec<&str> = args.iter().map(|s| s.as_str()).collect();
                run_cmd(&args_ref)?;

                // Capture UUID from the first device
                let dev0 = &spec.devices[0];
                let uuid = capture_uuid(&blkid, dev0)?;
                results.push(FsResult {
                    kind: FsKind::Btrfs,
                    devices: spec.devices.clone(),
                    uuid,
                    label: spec.label.clone(),
                });
            }
            FsKind::Bcachefs => {
                let Some(ref mkfs) = bcachefs_tool else {
                    return Err(Error::Filesystem("bcachefs not found in PATH".into()));
                };
                if spec.devices.is_empty() {
                    return Err(Error::Filesystem(
                        "bcachefs requires at least one device".into(),
                    ));
                }
                // bcachefs format --label LABEL [--replicas=2] dev1 [dev2 ...]
                // Apply replicas policy for Bcachefs2Copy topology (data+metadata replicas = 2)
                let mut args: Vec<String> = vec![
                    mkfs.clone(),
                    "format".into(),
                    "--label".into(),
                    spec.label.clone(),
                ];
                if matches!(cfg.topology, Topology::Bcachefs2Copy) {
                    args.push("--replicas=2".into());
                }
                args.extend(spec.devices.iter().cloned());
                let args_ref: Vec<&str> = args.iter().map(|s| s.as_str()).collect();
                run_cmd(&args_ref)?;

                // Capture UUID from the first device
                let dev0 = &spec.devices[0];
                let uuid = capture_uuid(&blkid, dev0)?;
                results.push(FsResult {
                    kind: FsKind::Bcachefs,
                    devices: spec.devices.clone(),
                    uuid,
                    label: spec.label.clone(),
                });
            }
        }
    }

    debug!("make_filesystems: created {} filesystems", results.len());
    Ok(results)
}

fn capture_uuid(blkid: &str, dev: &str) -> Result<String> {
    // blkid -o export /dev/...
    let out = run_cmd_capture(&[blkid, "-o", "export", dev])?;
    let map = parse_blkid_export(&out.stdout);
    // Prefer ID_FS_UUID if present, fall back to UUID
    if let Some(u) = map.get("ID_FS_UUID") {
        return Ok(u.clone());
    }
    if let Some(u) = map.get("UUID") {
        return Ok(u.clone());
    }
    warn!("blkid did not report UUID for {}", dev);
    Err(Error::Filesystem(format!(
        "missing UUID in blkid output for {}",
        dev
    )))
}

/// Minimal parser for blkid -o export KEY=VAL lines.
fn parse_blkid_export(s: &str) -> std::collections::HashMap<String, String> {
    let mut map = std::collections::HashMap::new();
    for line in s.lines() {
        if let Some((k, v)) = line.split_once('=') {
            map.insert(k.trim().to_string(), v.trim().to_string());
        }
    }
    map
}

/// Probe existing filesystems on the system and return their identities (kind, uuid, label).
///
/// This inspects /proc/partitions and uses `blkid -o export` on each device to detect:
/// - Data filesystems: Btrfs or Bcachefs with label "ZOSDATA"
/// - ESP filesystems: Vfat with label "ZOSBOOT"
/// Multi-device filesystems (e.g., btrfs) are de-duplicated by UUID.
///
/// Returns:
/// - Vec<FsResult> with at most one entry per filesystem UUID.
pub fn probe_existing_filesystems() -> Result<Vec<FsResult>> {
    let Some(blkid) = which_tool("blkid")? else {
        return Err(Error::Filesystem(
            "blkid not found in PATH; cannot probe existing filesystems".into(),
        ));
    };

    let content = fs::read_to_string("/proc/partitions")
        .map_err(|e| Error::Filesystem(format!("/proc/partitions read error: {}", e)))?;

    let mut results_by_uuid: std::collections::HashMap<String, FsResult> =
        std::collections::HashMap::new();

    for line in content.lines() {
        let line = line.trim();
        if line.is_empty() || line.starts_with("major") {
            continue;
        }
        // Format: major minor #blocks name
        let parts: Vec<&str> = line.split_whitespace().collect();
        if parts.len() < 4 {
            continue;
        }
        let name = parts[3];
        // Skip pseudo devices commonly not relevant (loop, ram, zram, fd)
        if name.starts_with("loop")
            || name.starts_with("ram")
            || name.starts_with("zram")
            || name.starts_with("fd")
        {
            continue;
        }

        let dev_path = format!("/dev/{}", name);
        // Probe with blkid -o export; ignore non-zero statuses meaning "nothing found"
        let out = match run_cmd_capture(&[blkid.as_str(), "-o", "export", dev_path.as_str()]) {
            Ok(o) => o,
            Err(Error::Tool { status, .. }) if status != 0 => {
                // No recognizable signature; skip
                continue;
            }
            Err(_) => {
                // Unexpected failure; skip this device
                continue;
            }
        };

        let map = parse_blkid_export(&out.stdout);
        let ty = map.get("TYPE").cloned().unwrap_or_default();
        let label = map
            .get("ID_FS_LABEL")
            .cloned()
            .or_else(|| map.get("LABEL").cloned())
            .unwrap_or_default();
        let uuid = map
            .get("ID_FS_UUID")
            .cloned()
            .or_else(|| map.get("UUID").cloned());

        let (kind_opt, expected_label) = match ty.as_str() {
            "btrfs" => (Some(FsKind::Btrfs), "ZOSDATA"),
            "bcachefs" => (Some(FsKind::Bcachefs), "ZOSDATA"),
            "vfat" => (Some(FsKind::Vfat), "ZOSBOOT"),
            _ => (None, ""),
        };

        if let (Some(kind), Some(u)) = (kind_opt, uuid) {
            // Enforce reserved label semantics
            if !expected_label.is_empty() && label != expected_label {
                continue;
            }

            // Deduplicate multi-device filesystems by UUID; record first-seen device
            results_by_uuid.entry(u.clone()).or_insert(FsResult {
                kind,
                devices: vec![dev_path.clone()],
                uuid: u,
                label: label.clone(),
            });
        }
    }

    Ok(results_by_uuid.into_values().collect())
}

#[cfg(test)]
mod tests_parse {
    use super::parse_blkid_export;

    #[test]
    fn parse_export_ok() {
        let s = "ID_FS_UUID=abcd-1234\nUUID=abcd-1234\nTYPE=btrfs\n";
        let m = parse_blkid_export(s);
        assert_eq!(m.get("ID_FS_UUID").unwrap(), "abcd-1234");
        assert_eq!(m.get("TYPE").unwrap(), "btrfs");
    }
}