feat(agent): report btrfs filesystems as storage pools (#2315)

Co-authored-by: henrygd <hank@henrygd.me>
This commit is contained in:
Ani Betts
2026-09-10 01:53:39 +02:00
committed by GitHub
parent 98687be2f2
commit 8d6a5d5f6e
36 changed files with 1990 additions and 681 deletions

26
agent/btrfs/btrfs.go Normal file
View File

@@ -0,0 +1,26 @@
// Package btrfs reads btrfs filesystem state from sysfs.
package btrfs
// Filesystem is a mounted btrfs filesystem read from /sys/fs/btrfs/<uuid>.
type Filesystem struct {
UUID string // stable filesystem UUID from sysfs
MountID string // kernel filesystem identity for matching monitored mounts
IODevice string // sole member block-device name, empty for multi-device/unknown pools
Name string // label, else first mountpoint, else UUID
Size uint64 // effective usable capacity, or raw member capacity when Raw
Raw bool // capacity and usage are physical bytes, unsuitable for disk alerts
Alloc uint64 // raw bytes allocated to data, metadata and system chunks
Health string // ONLINE, or DEGRADED when a device is missing
NRead uint64 // cumulative bytes read across member devices
NWrite uint64 // cumulative bytes written across member devices
Devices []Device
}
// Device is one member device (devinfo/<devid>) with its error counters.
type Device struct {
Name string // "devid N"; sysfs does not expose the block device path
State string // ONLINE or MISSING
ReadErrs uint64
WriteErrs uint64
CorruptionErrs uint64
}

285
agent/btrfs/btrfs_linux.go Normal file
View File

@@ -0,0 +1,285 @@
//go:build linux
package btrfs
import (
"errors"
"fmt"
"os"
"path/filepath"
"strconv"
"strings"
"unsafe"
"github.com/henrygd/beszel/agent/utils"
"golang.org/x/sys/unix"
)
var (
sysfsPath = "/sys/fs/btrfs"
mountsPath = "/proc/self/mounts"
mountinfoPath = "/proc/self/mountinfo"
mountUUID = MountID
deviceSize = ioctlDeviceSize
filesystemUsage = statfsUsage
)
// Filesystems returns all mounted btrfs filesystems, or nil when there are none.
func Filesystems() ([]Filesystem, error) {
entries, err := os.ReadDir(sysfsPath)
if errors.Is(err, os.ErrNotExist) {
return nil, nil
}
if err != nil {
return nil, err
}
mounts := mountpointsByDevice()
var filesystems []Filesystem
for _, entry := range entries {
if !entry.IsDir() || entry.Name() == "features" {
continue
}
fs, err := readFilesystem(filepath.Join(sysfsPath, entry.Name()), mounts)
if err != nil {
return nil, fmt.Errorf("btrfs %s: %w", entry.Name(), err)
}
filesystems = append(filesystems, fs)
}
return filesystems, nil
}
func readFilesystem(dir string, mounts map[string]string) (Filesystem, error) {
fs := Filesystem{UUID: filepath.Base(dir), Name: utils.ReadStringFile(filepath.Join(dir, "label")), Health: "UNKNOWN"}
for _, kind := range []string{"data", "metadata", "system"} {
if value, ok := utils.ReadUintFile(filepath.Join(dir, "allocation", kind, "disk_used")); ok {
fs.Alloc += value
}
}
// devices/<name> links to the block device's sysfs directory.
devices, err := os.ReadDir(filepath.Join(dir, "devices"))
if err != nil && !errors.Is(err, os.ErrNotExist) {
return fs, err
}
mountpoint := mounts["uuid:"+fs.UUID]
if fs.Name == "" {
fs.Name = mountpoint
}
var backingSize uint64
for _, dev := range devices {
if mountpoint == "" {
mountpoint = mounts[dev.Name()]
}
if fs.Name == "" {
fs.Name = mountpoint
}
devDir := filepath.Join(dir, "devices", dev.Name())
if size, ok := utils.ReadUintFile(filepath.Join(devDir, "size")); ok {
backingSize += size * 512
}
if stat := strings.Fields(utils.ReadStringFile(filepath.Join(devDir, "stat"))); len(stat) >= 7 {
fs.NRead += parseUint(stat[2]) * 512
fs.NWrite += parseUint(stat[6]) * 512
}
}
devids, err := os.ReadDir(filepath.Join(dir, "devinfo"))
if err != nil && !errors.Is(err, os.ErrNotExist) {
return fs, err
}
capacityAvailable := len(devids) > 0
healthKnown := len(devids) > 0
for _, devid := range devids {
devDir := filepath.Join(dir, "devinfo", devid.Name())
// Replacement targets do not add filesystem capacity.
replaceTarget, _ := utils.ReadUintFile(filepath.Join(devDir, "replace_target"))
if replaceTarget != 1 {
devid, err := strconv.ParseUint(devid.Name(), 10, 64)
if err != nil {
return fs, err
}
size, err := deviceSize(mountpoint, devid)
if err != nil {
capacityAvailable = false
}
fs.Size += size
}
dev := Device{Name: "devid " + devid.Name(), State: "ONLINE"}
missing := utils.ReadStringFile(filepath.Join(devDir, "missing"))
if missing != "0" && missing != "1" {
healthKnown = false
dev.State = "UNKNOWN"
}
if missing == "1" {
dev.State = "MISSING"
fs.Health = "DEGRADED"
}
for line := range strings.Lines(utils.ReadStringFile(filepath.Join(devDir, "error_stats"))) {
if fields := strings.Fields(line); len(fields) == 2 {
switch fields[0] {
case "read_errs":
dev.ReadErrs = parseUint(fields[1])
case "write_errs":
dev.WriteErrs = parseUint(fields[1])
case "corruption_errs":
dev.CorruptionErrs = parseUint(fields[1])
}
}
}
fs.Devices = append(fs.Devices, dev)
}
// Use one capacity source for the whole filesystem: device IDs cannot be
// reliably matched to block-device names in sysfs. A partial ioctl result
// must not be added to the complete backing-device total.
if !capacityAvailable {
fs.Size = backingSize
}
if fs.Health != "DEGRADED" && healthKnown {
fs.Health = "ONLINE"
}
fs.MountID = mountUUID(mountpoint)
if len(devices) == 1 && len(devids) == 1 && fs.Health == "ONLINE" {
fs.IODevice = devices[0].Name()
}
fs.Raw = true
if used, available, err := filesystemUsage(mountpoint); err == nil {
// Effective capacity excludes reserved/unavailable space, so Size-Alloc
// is available to applications and the usage ratio matches df.
fs.Size, fs.Alloc, fs.Raw = used+available, used, false
}
if fs.Name == "" {
fs.Name = filepath.Base(dir)
}
return fs, nil
}
// mountpointsByDevice prefers UUID matches from mountinfo and retains source
// device names as a fallback for environments where FS_INFO is unavailable.
func mountpointsByDevice() map[string]string {
mounts := mountpointsByUUID(utils.ReadStringFile(mountinfoPath), mountUUID)
for line := range strings.Lines(utils.ReadStringFile(mountsPath)) {
fields := strings.Fields(line)
if len(fields) < 3 || fields[2] != "btrfs" {
continue
}
device := fields[0]
if resolved, err := filepath.EvalSymlinks(device); err == nil {
device = resolved
}
if _, seen := mounts[filepath.Base(device)]; !seen {
mounts[filepath.Base(device)] = unescapeMountPath(fields[1])
}
}
return mounts
}
func parseUint(s string) uint64 {
n, _ := strconv.ParseUint(s, 10, 64)
return n
}
// ioctlDeviceSize reads Btrfs's recorded device size, which can be smaller
// than the block device after a filesystem resize. BTRFS_IOC_DEV_INFO is
// _IOWR(0x94, 30, struct btrfs_ioctl_dev_info_args), a 4096-byte ABI structure.
func ioctlDeviceSize(mountpoint string, devid uint64) (uint64, error) {
if mountpoint == "" {
return 0, errors.New("no accessible mountpoint")
}
f, err := os.Open(mountpoint)
if err != nil {
return 0, err
}
defer f.Close()
args := struct {
Devid uint64
UUID [16]byte
BytesUsed uint64
TotalBytes uint64
Reserved [4096 - 40]byte
}{Devid: devid}
_, _, errno := unix.Syscall(unix.SYS_IOCTL, f.Fd(), 0xd000941e, uintptr(unsafe.Pointer(&args)))
if errno != 0 {
return 0, errno
}
return args.TotalBytes, nil
}
// The filesystem magic is unsigned even when Statfs_t.Type is int32.
func isBtrfs(stat *unix.Statfs_t) bool {
return uint32(stat.Type) == unix.BTRFS_SUPER_MAGIC
}
func statfsUsage(path string) (used, available uint64, err error) {
if path == "" {
return 0, 0, errors.New("no accessible mountpoint")
}
var stat unix.Statfs_t
if err = unix.Statfs(path, &stat); err != nil {
return
}
if !isBtrfs(&stat) {
return 0, 0, errors.New("mountpoint is not Btrfs")
}
blockSize := uint64(stat.Bsize)
return (stat.Blocks - min(stat.Blocks, stat.Bfree)) * blockSize, min(stat.Blocks, stat.Bavail) * blockSize, nil
}
// MountID returns the filesystem UUID via BTRFS_IOC_FS_INFO. Unlike statfs
// f_fsid, this identity is shared by all subvolumes and bind mounts.
func MountID(path string) string {
if path == "" {
return ""
}
var stat unix.Statfs_t
if unix.Statfs(path, &stat) != nil || !isBtrfs(&stat) {
return ""
}
f, err := os.Open(path)
if err != nil {
return ""
}
defer f.Close()
args := struct {
MaxID uint64
NumDevices uint64
FSID [16]byte
Reserved [992]byte
}{}
// _IOR(0x94, 31, 1024). Reuse the platform's read-direction bits;
// MIPS/PowerPC use a different encoding than asm-generic.
request := uintptr(unix.FS_IOC_GETFLAGS&0xe0000000) | 0x0400941f
_, _, errno := unix.Syscall(unix.SYS_IOCTL, f.Fd(), request, uintptr(unsafe.Pointer(&args)))
if errno != 0 {
return ""
}
id := args.FSID
return fmt.Sprintf("%x-%x-%x-%x-%x", id[:4], id[4:6], id[6:8], id[8:10], id[10:])
}
// Btrfs mountinfo device numbers can be virtual (0:N), so query the UUID
// through the mount instead of comparing those numbers with sysfs block devs.
// Retry another path when a bind mount is inaccessible. Once resolved, reuse
// the result for that mount device to avoid opening every Docker bind mount.
func mountpointsByUUID(mountinfo string, identify func(string) string) map[string]string {
mounts := make(map[string]string)
resolved := make(map[string]bool)
for line := range strings.Lines(mountinfo) {
before, after, ok := strings.Cut(line, " - ")
fields, fs := strings.Fields(before), strings.Fields(after)
if !ok || len(fields) < 6 || len(fs) < 3 || fs[0] != "btrfs" || resolved[fields[2]] {
continue
}
path := unescapeMountPath(fields[4])
uuid := identify(path)
if uuid == "" {
continue
}
resolved[fields[2]] = true
if mounts["uuid:"+uuid] == "" {
mounts["uuid:"+uuid] = path
}
}
return mounts
}
func unescapeMountPath(path string) string {
return strings.NewReplacer(`\040`, " ", `\011`, "\t", `\012`, "\n", `\134`, `\`).Replace(path)
}

View File

@@ -0,0 +1,274 @@
//go:build testing && linux
package btrfs
import (
"os"
"path/filepath"
"strconv"
"testing"
"github.com/henrygd/beszel/agent/utils"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"golang.org/x/sys/unix"
)
func TestFilesystems(t *testing.T) {
root := t.TempDir()
oldSysfs, oldMounts := sysfsPath, mountsPath
sysfsPath, mountsPath = root, filepath.Join(root, "mounts")
t.Cleanup(func() { sysfsPath, mountsPath = oldSysfs, oldMounts })
fsDir := filepath.Join(root, "1b2c3d4e-0000-0000-0000-000000000000")
write := func(rel, content string) {
path := filepath.Join(fsDir, rel)
require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o755))
require.NoError(t, os.WriteFile(path, []byte(content), 0o644))
}
require.NoError(t, os.MkdirAll(filepath.Join(root, "features"), 0o755))
oldUsage := filesystemUsage
filesystemUsage = func(string) (uint64, uint64, error) { return 0, 0, os.ErrNotExist }
t.Cleanup(func() { filesystemUsage = oldUsage })
oldDeviceSize := deviceSize
t.Cleanup(func() { deviceSize = oldDeviceSize })
deviceSize = func(_ string, devid uint64) (uint64, error) {
value, _ := utils.ReadUintFile(filepath.Join(fsDir, "recorded-size", strconv.FormatUint(devid, 10)))
return value, nil
}
// Recorded member capacities differ from the unchanged backing devices.
write("recorded-size/1", "256000\n")
write("recorded-size/2", "128000\n")
write("label", "tank\n")
write("allocation/data/disk_used", "4096\n")
write("allocation/metadata/disk_used", "2048\n")
write("allocation/system/disk_used", "1024\n")
write("devices/sda/size", "1000\n")
write("devices/sda/stat", "10 0 200 0 20 0 400 0 0 0 0\n")
write("devices/sdb/size", "1000\n")
write("devices/sdb/stat", "10 0 100 0 20 0 100 0 0 0 0\n")
write("devinfo/1/missing", "0\n")
write("devinfo/1/error_stats", "write_errs 1\nread_errs 2\nflush_errs 0\ncorruption_errs 3\ngeneration_errs 0\n")
write("devinfo/2/missing", "1\n")
filesystems, err := Filesystems()
require.NoError(t, err)
require.Len(t, filesystems, 1)
assert.Equal(t, Filesystem{
UUID: "1b2c3d4e-0000-0000-0000-000000000000", Raw: true, Name: "tank", Size: 384000, Alloc: 7168, Health: "DEGRADED", NRead: 153600, NWrite: 256000,
Devices: []Device{
{Name: "devid 1", State: "ONLINE", ReadErrs: 2, WriteErrs: 1, CorruptionErrs: 3},
{Name: "devid 2", State: "MISSING"},
},
}, filesystems[0])
// Unlabeled filesystems fall back to the first mountpoint, then the UUID.
write("label", "\n")
require.NoError(t, os.WriteFile(mountsPath, []byte(
"/dev/sdz1 /other btrfs rw 0 0\n/dev/sdb /mnt/storage btrfs rw 0 0\n/dev/sdb /mnt/storage/sub btrfs rw,subvol=/sub 0 0\n",
), 0o644))
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, "/mnt/storage", filesystems[0].Name)
require.NoError(t, os.Remove(mountsPath))
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, "1b2c3d4e-0000-0000-0000-000000000000", filesystems[0].Name)
write("devinfo/3/replace_target", "1\n")
write("recorded-size/3", "512000\n")
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, uint64(384000), filesystems[0].Size, "replacement target must not inflate capacity")
deviceSize = func(string, uint64) (uint64, error) { return 0, os.ErrPermission }
filesystems, err = Filesystems()
require.NoError(t, err)
require.Len(t, filesystems, 1)
assert.Equal(t, uint64(1024000), filesystems[0].Size)
assert.Equal(t, "DEGRADED", filesystems[0].Health)
assert.Equal(t, uint64(153600), filesystems[0].NRead)
// A partial ioctl result must not be mixed with the backing-device total.
deviceSize = func(_ string, devid uint64) (uint64, error) {
if devid == 2 {
return 0, os.ErrPermission
}
return 256000, nil
}
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, uint64(1024000), filesystems[0].Size)
// With no mount visible (e.g. Docker), the real lookup falls back too.
deviceSize = ioctlDeviceSize
filesystems, err = Filesystems()
require.NoError(t, err)
require.Len(t, filesystems, 1)
assert.Equal(t, uint64(1024000), filesystems[0].Size)
filesystemUsage = func(string) (uint64, uint64, error) { return 100, 900, nil }
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, uint64(1000), filesystems[0].Size)
assert.Equal(t, uint64(100), filesystems[0].Alloc)
assert.False(t, filesystems[0].Raw)
}
func TestFilesystemsNoBtrfs(t *testing.T) {
oldPath := sysfsPath
sysfsPath = filepath.Join(t.TempDir(), "missing")
t.Cleanup(func() { sysfsPath = oldPath })
filesystems, err := Filesystems()
require.NoError(t, err)
assert.Nil(t, filesystems)
}
func TestIoctlDeviceSizeFailure(t *testing.T) {
_, err := ioctlDeviceSize("", 1)
require.Error(t, err)
_, err = ioctlDeviceSize(t.TempDir(), 1)
require.Error(t, err)
assert.ErrorIs(t, err, unix.ENOTTY)
}
func TestMountpointsDecodeEscapes(t *testing.T) {
oldMounts := mountsPath
mountsPath = filepath.Join(t.TempDir(), "mounts")
t.Cleanup(func() { mountsPath = oldMounts })
require.NoError(t, os.WriteFile(mountsPath, []byte("/dev/test-btrfs /mnt/my\\040data btrfs rw 0 0\n"), 0o644))
assert.Equal(t, "/mnt/my data", mountpointsByDevice()["test-btrfs"])
}
func TestFilesystemWithoutDevinfo(t *testing.T) {
root := t.TempDir()
require.NoError(t, os.MkdirAll(filepath.Join(root, "devices", "sda"), 0755))
require.NoError(t, os.WriteFile(filepath.Join(root, "devices", "sda", "size"), []byte("1000"), 0644))
fs, err := readFilesystem(root, nil)
require.NoError(t, err)
assert.Equal(t, uint64(512000), fs.Size)
assert.True(t, fs.Raw)
assert.Equal(t, "UNKNOWN", fs.Health)
assert.Empty(t, fs.Devices)
require.NoError(t, os.MkdirAll(filepath.Join(root, "devinfo", "1"), 0755))
fs, err = readFilesystem(root, nil)
require.NoError(t, err)
assert.Equal(t, "UNKNOWN", fs.Health)
require.Len(t, fs.Devices, 1)
assert.Equal(t, "UNKNOWN", fs.Devices[0].State)
// Some older interfaces lack the devices directory too.
fs, err = readFilesystem(t.TempDir(), nil)
require.NoError(t, err)
assert.Equal(t, "UNKNOWN", fs.Health)
}
func TestLocalBtrfsUsage(t *testing.T) {
path := os.Getenv("BESZEL_TEST_BTRFS_MOUNT")
if path == "" {
t.Skip("set BESZEL_TEST_BTRFS_MOUNT for read-only live validation")
}
used, available, err := statfsUsage(path)
require.NoError(t, err)
filesystems, err := Filesystems()
require.NoError(t, err)
for _, fs := range filesystems {
if !fs.Raw && fs.Alloc == used && fs.Size == used+available {
t.Logf("pool=%s used=%d available=%d effective_capacity=%d", fs.Name, used, available, fs.Size)
return
}
}
t.Fatal("collector did not report the mounted filesystem's usable capacity")
}
func TestMountID(t *testing.T) {
assert.Empty(t, MountID(""))
assert.Empty(t, MountID(filepath.Join(t.TempDir(), "missing")))
path := os.Getenv("BESZEL_TEST_BTRFS_MOUNT")
if path == "" {
t.Skip("set BESZEL_TEST_BTRFS_MOUNT for live identity validation")
}
id := MountID(path)
require.NotEmpty(t, id)
assert.Equal(t, id, MountID(filepath.Join(path, ".")))
}
func TestMountinfoUUIDLookup(t *testing.T) {
info := `1 0 0:40 /@ /inaccessible ro shared:1 - btrfs /dev/mapper/unavailable rw
2 0 0:40 /@/docker/hosts /etc/hosts ro - btrfs /dev/mapper/unavailable rw
3 0 0:40 /@/docker/hostname /etc/hostname ro - btrfs /dev/mapper/unavailable rw
4 0 0:41 /subvol /extra-filesystems/my\040disk ro master:2 - btrfs /dev/missing rw
5 0 0:42 / /ext4 ro - ext4 /dev/mapper/unavailable rw
malformed
6 0 0:43 / /bad ro - btrfs
`
var calls []string
mounts := mountpointsByUUID(info, func(path string) string {
calls = append(calls, path)
switch path {
case "/etc/hosts":
return "root-uuid"
case "/extra-filesystems/my disk":
return "extra-uuid"
}
return ""
})
assert.Equal(t, map[string]string{"uuid:root-uuid": "/etc/hosts", "uuid:extra-uuid": "/extra-filesystems/my disk"}, mounts)
assert.Equal(t, []string{"/inaccessible", "/etc/hosts", "/extra-filesystems/my disk"}, calls)
}
func TestDockerFilesystemWithoutDeviceNodes(t *testing.T) {
root := t.TempDir()
oldSysfs, oldMounts, oldInfo, oldUUID, oldUsage := sysfsPath, mountsPath, mountinfoPath, mountUUID, filesystemUsage
t.Cleanup(func() {
sysfsPath, mountsPath, mountinfoPath, mountUUID, filesystemUsage = oldSysfs, oldMounts, oldInfo, oldUUID, oldUsage
})
sysfsPath = filepath.Join(root, "sysfs")
mountsPath = filepath.Join(root, "missing-mounts")
mountinfoPath = filepath.Join(root, "mountinfo")
uuid := "11111111-1111-4111-8111-111111111111"
dir := filepath.Join(sysfsPath, uuid)
for path, content := range map[string]string{"devices/dm-0/size": "1000", "devinfo/1/missing": "0"} {
target := filepath.Join(dir, path)
require.NoError(t, os.MkdirAll(filepath.Dir(target), 0755))
require.NoError(t, os.WriteFile(target, []byte(content), 0644))
}
require.NoError(t, os.WriteFile(mountinfoPath, []byte("2 1 0:40 /@/docker/hosts /etc/hosts ro - btrfs /dev/mapper/not-in-container rw\n"), 0644))
mountUUID = func(path string) string {
if path == "/etc/hosts" {
return uuid
}
return ""
}
filesystemUsage = func(path string) (uint64, uint64, error) { require.Equal(t, "/etc/hosts", path); return 100, 900, nil }
fs, err := Filesystems()
require.NoError(t, err)
require.Len(t, fs, 1)
assert.Equal(t, uuid, fs[0].MountID)
assert.Equal(t, "dm-0", fs[0].IODevice)
assert.False(t, fs[0].Raw)
assert.Equal(t, uint64(1000), fs[0].Size)
}
func TestLivePoolMountIdentity(t *testing.T) {
path := os.Getenv("BESZEL_TEST_BTRFS_MOUNT")
if path == "" {
t.Skip("set BESZEL_TEST_BTRFS_MOUNT for live validation")
}
id := MountID(path)
require.NotEmpty(t, id)
pools, err := Filesystems()
require.NoError(t, err)
for _, pool := range pools {
if pool.UUID != id {
continue
}
assert.Equal(t, id, pool.MountID)
assert.False(t, pool.Raw)
t.Logf("uuid=%s mount_identity=%s io_device=%s raw=%v", pool.UUID, pool.MountID, pool.IODevice, pool.Raw)
return
}
t.Fatal("mounted Btrfs filesystem was not discovered")
}

View File

@@ -0,0 +1,11 @@
//go:build !linux
package btrfs
import "errors"
func Filesystems() ([]Filesystem, error) {
return nil, errors.ErrUnsupported
}
func MountID(string) string { return "" }