felhom-agent/internal/storage/smartdev.go

package storage

import (
	"context"
	"os"
	"path/filepath"
	"regexp"
	"strings"

	"gitea.dooplex.hu/admin/felhom-agent/internal/hub"
)

// sysBlockRoot is the sysfs block directory. A package var so tests can point it at a fixture tree.
var sysBlockRoot = "/sys/block"

// dmWholeDisk resolves a device-mapper / LVM device to its SINGLE backing whole disk, recursing
// through stacked dm layers via /sys/block/<dm>/slaves (Fix A, SPIKE-smart-coverage-2026-07-25).
// Returns ok=false when the device is not dm, sysfs is missing, there are no slaves, or the slaves
// span MORE THAN ONE physical disk — in that last case we deliberately skip rather than guess which
// of two disks to SMART (e.g. a mirrored LV).
func dmWholeDisk(device string) (string, bool) {
	name := dmName(device)
	if name == "" {
		return "", false
	}
	disks := map[string]bool{}
	if !collectSlaveDisks(name, disks, 0) {
		return "", false
	}
	if len(disks) != 1 {
		return "", false // no disk, or an ambiguous multi-disk dm — never guess
	}
	for d := range disks {
		return "/dev/" + d, true
	}
	return "", false
}

// dmName maps a dm device path to its sysfs name (dm-N). Handles /dev/dm-N (and a bare dm-N)
// directly, and /dev/mapper/<name> by matching /sys/block/dm-*/dm/name.
func dmName(device string) string {
	base := filepath.Base(device)
	if strings.HasPrefix(base, "dm-") {
		return base
	}
	if strings.HasPrefix(device, "/dev/mapper/") {
		entries, err := os.ReadDir(sysBlockRoot)
		if err != nil {
			return ""
		}
		for _, e := range entries {
			if !strings.HasPrefix(e.Name(), "dm-") {
				continue
			}
			b, err := os.ReadFile(filepath.Join(sysBlockRoot, e.Name(), "dm", "name"))
			if err == nil && strings.TrimSpace(string(b)) == base {
				return e.Name()
			}
		}
	}
	return ""
}

// collectSlaveDisks fills `disks` with the whole-disk names backing dm `name`, recursing through
// nested dm. Returns false on missing sysfs, no slaves, or excessive nesting (loop guard).
func collectSlaveDisks(name string, disks map[string]bool, depth int) bool {
	if depth > 8 {
		return false
	}
	entries, err := os.ReadDir(filepath.Join(sysBlockRoot, name, "slaves"))
	if err != nil {
		return false
	}
	if len(entries) == 0 {
		return false
	}
	for _, e := range entries {
		s := e.Name()
		if strings.HasPrefix(s, "dm-") {
			if !collectSlaveDisks(s, disks, depth+1) {
				return false
			}
			continue
		}
		disks[wholeDiskName(s)] = true
	}
	return true
}

// wholeDiskName strips a partition suffix to the whole-disk name (sda3→sda, nvme0n1p3→nvme0n1).
func wholeDiskName(part string) string {
	if m := reNVMePartName.FindStringSubmatch(part); m != nil {
		return m[1]
	}
	if m := reSDPartName.FindStringSubmatch(part); m != nil {
		return m[1]
	}
	return part
}

var (
	reNVMePartName = regexp.MustCompile(`^(nvme[0-9]+n[0-9]+)p[0-9]+$`)
	reSDPartName   = regexp.MustCompile(`^((?:sd|hd|vd)[a-z]+)[0-9]+$`)
)

// containingMountDevice returns the device of the mount whose mountpoint is the LONGEST prefix of
// path — the filesystem that actually holds `path`. Used ONLY to pick a whole-disk device for a
// SMART read of a dir-storage that lives inside a shared filesystem (the builtin `local` on the LVM
// root); it never feeds durable_id / backing_device (which stay empty for such targets by design —
// the removable-safety guard in build()).
func containingMountDevice(mounts []Mount, path string) (string, bool) {
	clean := cleanMountPath(path)
	best, bestLen := "", -1
	for _, m := range mounts {
		if m.Device == "" {
			continue
		}
		mp := cleanMountPath(m.MountPoint)
		if mp == clean || mp == "/" || strings.HasPrefix(clean, strings.TrimRight(mp, "/")+"/") {
			if len(mp) > bestLen {
				best, bestLen = m.Device, len(mp)
			}
		}
	}
	return best, best != ""
}

// SmartReader reads per-disk SMART for a backing device, resolving partition / dm / LVM down to the
// whole disk. It exists so the localapi /disks UNION path (registry/USB drives that skip Observe's
// enrich) gets the SAME SMART read the dir targets get, without duplicating smartDeviceFor (Fix B).
// A zero-value summary (Health "") means "could not read" (nil ops, unresolvable device, or a read
// error) — distinct from a read that returned UNKNOWN — so the caller can omit it exactly like the
// dir path does.
type SmartReader struct{ ops HostOps }

// NewSmartReader wraps a HostOps for the localapi union path.
func NewSmartReader(ops HostOps) *SmartReader { return &SmartReader{ops: ops} }

// SMARTForBacking reads SMART for backingDevice (partition/dm/whole-disk). Never returns an error;
// on any failure the summary's Health is "".
func (r *SmartReader) SMARTForBacking(ctx context.Context, backingDevice string) hub.SmartSummary {
	if r == nil || r.ops == nil {
		return hub.SmartSummary{}
	}
	dev, ok := smartDeviceFor(backingDevice)
	if !ok {
		return hub.SmartSummary{}
	}
	sm, err := r.ops.SMART(ctx, dev)
	if err != nil {
		return hub.SmartSummary{}
	}
	return sm
}