// Copyright 2014 Google Inc. All Rights Reserved. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. // You may obtain a copy of the License at // // http://www.apache.org/licenses/LICENSE-2.0 // // Unless required by applicable law or agreed to in writing, software // distributed under the License is distributed on an "AS IS" BASIS, // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. // +build linux // Provides Filesystem Stats package fs import ( "bufio" "fmt" "io/ioutil" "os" "os/exec" "path" "path/filepath" "regexp" "strconv" "strings" "syscall" "github.com/docker/docker/pkg/mount" "github.com/google/cadvisor/devicemapper" "github.com/google/cadvisor/utils" dockerutil "github.com/google/cadvisor/utils/docker" zfs "github.com/mistifyio/go-zfs" "k8s.io/klog" ) const ( LabelSystemRoot = "root" LabelDockerImages = "docker-images" LabelRktImages = "rkt-images" LabelCrioImages = "crio-images" ) const ( // The block size in bytes. statBlockSize uint64 = 512 // The maximum number of `disk usage` tasks that can be running at once. maxConcurrentOps = 20 ) // A pool for restricting the number of consecutive `du` and `find` tasks running. var pool = make(chan struct{}, maxConcurrentOps) func init() { for i := 0; i < maxConcurrentOps; i++ { releaseToken() } } func claimToken() { <-pool } func releaseToken() { pool <- struct{}{} } type partition struct { mountpoint string major uint minor uint fsType string blockSize uint } type RealFsInfo struct { // Map from block device path to partition information. partitions map[string]partition // Map from label to block device path. // Labels are intent-specific tags that are auto-detected. labels map[string]string // Map from mountpoint to mount information. mounts map[string]*mount.Info // devicemapper client dmsetup devicemapper.DmsetupClient // fsUUIDToDeviceName is a map from the filesystem UUID to its device name. fsUUIDToDeviceName map[string]string } type Context struct { // docker root directory. Docker DockerContext RktPath string Crio CrioContext } type DockerContext struct { Root string Driver string DriverStatus map[string]string } type CrioContext struct { Root string } func NewFsInfo(context Context) (FsInfo, error) { mounts, err := mount.GetMounts(nil) if err != nil { return nil, err } fsUUIDToDeviceName, err := getFsUUIDToDeviceNameMap() if err != nil { // UUID is not always available across different OS distributions. // Do not fail if there is an error. klog.Warningf("Failed to get disk UUID mapping, getting disk info by uuid will not work: %v", err) } // Avoid devicemapper container mounts - these are tracked by the ThinPoolWatcher excluded := []string{fmt.Sprintf("%s/devicemapper/mnt", context.Docker.Root)} fsInfo := &RealFsInfo{ partitions: processMounts(mounts, excluded), labels: make(map[string]string, 0), mounts: make(map[string]*mount.Info, 0), dmsetup: devicemapper.NewDmsetupClient(), fsUUIDToDeviceName: fsUUIDToDeviceName, } for _, mount := range mounts { fsInfo.mounts[mount.Mountpoint] = mount } fsInfo.addRktImagesLabel(context, mounts) // need to call this before the log line below printing out the partitions, as this function may // add a "partition" for devicemapper to fsInfo.partitions fsInfo.addDockerImagesLabel(context, mounts) fsInfo.addCrioImagesLabel(context, mounts) klog.V(1).Infof("Filesystem UUIDs: %+v", fsInfo.fsUUIDToDeviceName) klog.V(1).Infof("Filesystem partitions: %+v", fsInfo.partitions) fsInfo.addSystemRootLabel(mounts) return fsInfo, nil } // getFsUUIDToDeviceNameMap creates the filesystem uuid to device name map // using the information in /dev/disk/by-uuid. If the directory does not exist, // this function will return an empty map. func getFsUUIDToDeviceNameMap() (map[string]string, error) { const dir = "/dev/disk/by-uuid" if _, err := os.Stat(dir); os.IsNotExist(err) { return make(map[string]string), nil } files, err := ioutil.ReadDir(dir) if err != nil { return nil, err } fsUUIDToDeviceName := make(map[string]string) for _, file := range files { path := filepath.Join(dir, file.Name()) target, err := os.Readlink(path) if err != nil { klog.Warningf("Failed to resolve symlink for %q", path) continue } device, err := filepath.Abs(filepath.Join(dir, target)) if err != nil { return nil, fmt.Errorf("failed to resolve the absolute path of %q", filepath.Join(dir, target)) } fsUUIDToDeviceName[file.Name()] = device } return fsUUIDToDeviceName, nil } func processMounts(mounts []*mount.Info, excludedMountpointPrefixes []string) map[string]partition { partitions := make(map[string]partition, 0) supportedFsType := map[string]bool{ // all ext systems are checked through prefix. "btrfs": true, "tmpfs": true, "xfs": true, "zfs": true, } for _, mount := range mounts { if !strings.HasPrefix(mount.Fstype, "ext") && !supportedFsType[mount.Fstype] { continue } // Avoid bind mounts. if _, ok := partitions[mount.Source]; ok { continue } hasPrefix := false for _, prefix := range excludedMountpointPrefixes { if strings.HasPrefix(mount.Mountpoint, prefix) { hasPrefix = true break } } if hasPrefix { continue } // btrfs fix: following workaround fixes wrong btrfs Major and Minor Ids reported in /proc/self/mountinfo. // instead of using values from /proc/self/mountinfo we use stat to get Ids from btrfs mount point if mount.Fstype == "btrfs" && mount.Major == 0 && strings.HasPrefix(mount.Source, "/dev/") { major, minor, err := getBtrfsMajorMinorIds(mount) if err != nil { klog.Warningf("%s", err) } else { mount.Major = major mount.Minor = minor } } partitions[mount.Source] = partition{ fsType: mount.Fstype, mountpoint: mount.Mountpoint, major: uint(mount.Major), minor: uint(mount.Minor), } } return partitions } // getDockerDeviceMapperInfo returns information about the devicemapper device and "partition" if // docker is using devicemapper for its storage driver. If a loopback device is being used, don't // return any information or error, as we want to report based on the actual partition where the // loopback file resides, inside of the loopback file itself. func (self *RealFsInfo) getDockerDeviceMapperInfo(context DockerContext) (string, *partition, error) { if context.Driver != DeviceMapper.String() { return "", nil, nil } dataLoopFile := context.DriverStatus[dockerutil.DriverStatusDataLoopFile] if len(dataLoopFile) > 0 { return "", nil, nil } dev, major, minor, blockSize, err := dockerDMDevice(context.DriverStatus, self.dmsetup) if err != nil { return "", nil, err } return dev, &partition{ fsType: DeviceMapper.String(), major: major, minor: minor, blockSize: blockSize, }, nil } // addSystemRootLabel attempts to determine which device contains the mount for /. func (self *RealFsInfo) addSystemRootLabel(mounts []*mount.Info) { for _, m := range mounts { if m.Mountpoint == "/" { self.partitions[m.Source] = partition{ fsType: m.Fstype, mountpoint: m.Mountpoint, major: uint(m.Major), minor: uint(m.Minor), } self.labels[LabelSystemRoot] = m.Source return } } } // addDockerImagesLabel attempts to determine which device contains the mount for docker images. func (self *RealFsInfo) addDockerImagesLabel(context Context, mounts []*mount.Info) { dockerDev, dockerPartition, err := self.getDockerDeviceMapperInfo(context.Docker) if err != nil { klog.Warningf("Could not get Docker devicemapper device: %v", err) } if len(dockerDev) > 0 && dockerPartition != nil { self.partitions[dockerDev] = *dockerPartition self.labels[LabelDockerImages] = dockerDev } else { self.updateContainerImagesPath(LabelDockerImages, mounts, getDockerImagePaths(context)) } } func (self *RealFsInfo) addCrioImagesLabel(context Context, mounts []*mount.Info) { if context.Crio.Root != "" { crioPath := context.Crio.Root crioImagePaths := map[string]struct{}{ "/": {}, } for _, dir := range []string{"overlay", "overlay2"} { crioImagePaths[path.Join(crioPath, dir+"-images")] = struct{}{} } for crioPath != "/" && crioPath != "." { crioImagePaths[crioPath] = struct{}{} crioPath = filepath.Dir(crioPath) } self.updateContainerImagesPath(LabelCrioImages, mounts, crioImagePaths) } } func (self *RealFsInfo) addRktImagesLabel(context Context, mounts []*mount.Info) { if context.RktPath != "" { rktPath := context.RktPath rktImagesPaths := map[string]struct{}{ "/": {}, } for rktPath != "/" && rktPath != "." { rktImagesPaths[rktPath] = struct{}{} rktPath = filepath.Dir(rktPath) } self.updateContainerImagesPath(LabelRktImages, mounts, rktImagesPaths) } } // Generate a list of possible mount points for docker image management from the docker root directory. // Right now, we look for each type of supported graph driver directories, but we can do better by parsing // some of the context from `docker info`. func getDockerImagePaths(context Context) map[string]struct{} { dockerImagePaths := map[string]struct{}{ "/": {}, } // TODO(rjnagal): Detect docker root and graphdriver directories from docker info. dockerRoot := context.Docker.Root for _, dir := range []string{"devicemapper", "btrfs", "aufs", "overlay", "overlay2", "zfs"} { dockerImagePaths[path.Join(dockerRoot, dir)] = struct{}{} } for dockerRoot != "/" && dockerRoot != "." { dockerImagePaths[dockerRoot] = struct{}{} dockerRoot = filepath.Dir(dockerRoot) } return dockerImagePaths } // This method compares the mountpoints with possible container image mount points. If a match is found, // the label is added to the partition. func (self *RealFsInfo) updateContainerImagesPath(label string, mounts []*mount.Info, containerImagePaths map[string]struct{}) { var useMount *mount.Info for _, m := range mounts { if _, ok := containerImagePaths[m.Mountpoint]; ok { if useMount == nil || (len(useMount.Mountpoint) < len(m.Mountpoint)) { useMount = m } } } if useMount != nil { self.partitions[useMount.Source] = partition{ fsType: useMount.Fstype, mountpoint: useMount.Mountpoint, major: uint(useMount.Major), minor: uint(useMount.Minor), } self.labels[label] = useMount.Source } } func (self *RealFsInfo) GetDeviceForLabel(label string) (string, error) { dev, ok := self.labels[label] if !ok { return "", fmt.Errorf("non-existent label %q", label) } return dev, nil } func (self *RealFsInfo) GetLabelsForDevice(device string) ([]string, error) { labels := []string{} for label, dev := range self.labels { if dev == device { labels = append(labels, label) } } return labels, nil } func (self *RealFsInfo) GetMountpointForDevice(dev string) (string, error) { p, ok := self.partitions[dev] if !ok { return "", fmt.Errorf("no partition info for device %q", dev) } return p.mountpoint, nil } func (self *RealFsInfo) GetFsInfoForPath(mountSet map[string]struct{}) ([]Fs, error) { filesystems := make([]Fs, 0) deviceSet := make(map[string]struct{}) diskStatsMap, err := getDiskStatsMap("/proc/diskstats") if err != nil { return nil, err } for device, partition := range self.partitions { _, hasMount := mountSet[partition.mountpoint] _, hasDevice := deviceSet[device] if mountSet == nil || (hasMount && !hasDevice) { var ( err error fs Fs ) switch partition.fsType { case DeviceMapper.String(): fs.Capacity, fs.Free, fs.Available, err = getDMStats(device, partition.blockSize) klog.V(5).Infof("got devicemapper fs capacity stats: capacity: %v free: %v available: %v:", fs.Capacity, fs.Free, fs.Available) fs.Type = DeviceMapper case ZFS.String(): fs.Capacity, fs.Free, fs.Available, err = getZfstats(device) fs.Type = ZFS default: var inodes, inodesFree uint64 if utils.FileExists(partition.mountpoint) { fs.Capacity, fs.Free, fs.Available, inodes, inodesFree, err = getVfsStats(partition.mountpoint) fs.Inodes = &inodes fs.InodesFree = &inodesFree fs.Type = VFS } else { klog.V(4).Infof("unable to determine file system type, partition mountpoint does not exist: %v", partition.mountpoint) } } if err != nil { klog.V(4).Infof("Stat fs failed. Error: %v", err) } else { deviceSet[device] = struct{}{} fs.DeviceInfo = DeviceInfo{ Device: device, Major: uint(partition.major), Minor: uint(partition.minor), } fs.DiskStats = diskStatsMap[device] filesystems = append(filesystems, fs) } } } return filesystems, nil } var partitionRegex = regexp.MustCompile(`^(?:(?:s|v|xv)d[a-z]+\d*|dm-\d+)$`) func getDiskStatsMap(diskStatsFile string) (map[string]DiskStats, error) { diskStatsMap := make(map[string]DiskStats) file, err := os.Open(diskStatsFile) if err != nil { if os.IsNotExist(err) { klog.Warningf("Not collecting filesystem statistics because file %q was not found", diskStatsFile) return diskStatsMap, nil } return nil, err } defer file.Close() scanner := bufio.NewScanner(file) for scanner.Scan() { line := scanner.Text() words := strings.Fields(line) if !partitionRegex.MatchString(words[2]) { continue } // 8 50 sdd2 40 0 280 223 7 0 22 108 0 330 330 deviceName := path.Join("/dev", words[2]) wordLength := len(words) offset := 3 var stats = make([]uint64, wordLength-offset) if len(stats) < 11 { return nil, fmt.Errorf("could not parse all 11 columns of /proc/diskstats") } var error error for i := offset; i < wordLength; i++ { stats[i-offset], error = strconv.ParseUint(words[i], 10, 64) if error != nil { return nil, error } } diskStats := DiskStats{ ReadsCompleted: stats[0], ReadsMerged: stats[1], SectorsRead: stats[2], ReadTime: stats[3], WritesCompleted: stats[4], WritesMerged: stats[5], SectorsWritten: stats[6], WriteTime: stats[7], IoInProgress: stats[8], IoTime: stats[9], WeightedIoTime: stats[10], } diskStatsMap[deviceName] = diskStats } return diskStatsMap, nil } func (self *RealFsInfo) GetGlobalFsInfo() ([]Fs, error) { return self.GetFsInfoForPath(nil) } func major(devNumber uint64) uint { return uint((devNumber >> 8) & 0xfff) } func minor(devNumber uint64) uint { return uint((devNumber & 0xff) | ((devNumber >> 12) & 0xfff00)) } func (self *RealFsInfo) GetDeviceInfoByFsUUID(uuid string) (*DeviceInfo, error) { deviceName, found := self.fsUUIDToDeviceName[uuid] if !found { return nil, ErrNoSuchDevice } p, found := self.partitions[deviceName] if !found { return nil, fmt.Errorf("cannot find device %q in partitions", deviceName) } return &DeviceInfo{deviceName, p.major, p.minor}, nil } func (self *RealFsInfo) GetDirFsDevice(dir string) (*DeviceInfo, error) { buf := new(syscall.Stat_t) err := syscall.Stat(dir, buf) if err != nil { return nil, fmt.Errorf("stat failed on %s with error: %s", dir, err) } major := major(buf.Dev) minor := minor(buf.Dev) for device, partition := range self.partitions { if partition.major == major && partition.minor == minor { return &DeviceInfo{device, major, minor}, nil } } mount, found := self.mounts[dir] // try the parent dir if not found until we reach the root dir // this is an issue on btrfs systems where the directory is not // the subvolume for !found { pathdir, _ := filepath.Split(dir) // break when we reach root if pathdir == "/" { break } // trim "/" from the new parent path otherwise the next possible // filepath.Split in the loop will not split the string any further dir = strings.TrimSuffix(pathdir, "/") mount, found = self.mounts[dir] } if found && mount.Fstype == "btrfs" && mount.Major == 0 && strings.HasPrefix(mount.Source, "/dev/") { major, minor, err := getBtrfsMajorMinorIds(mount) if err != nil { klog.Warningf("%s", err) } else { return &DeviceInfo{mount.Source, uint(major), uint(minor)}, nil } } return nil, fmt.Errorf("could not find device with major: %d, minor: %d in cached partitions map", major, minor) } func GetDirUsage(dir string) (UsageInfo, error) { var usage UsageInfo if dir == "" { return usage, fmt.Errorf("invalid directory") } rootInfo, err := os.Stat(dir) if err != nil { return usage, fmt.Errorf("could not stat %q to get inode usage: %v", dir, err) } rootStat, ok := rootInfo.Sys().(*syscall.Stat_t) if !ok { return usage, fmt.Errorf("unsuported fileinfo for getting inode usage of %q", dir) } rootDevId := rootStat.Dev // dedupedInode stores inodes that could be duplicates (nlink > 1) dedupedInodes := make(map[uint64]struct{}) err = filepath.Walk(dir, func(path string, info os.FileInfo, err error) error { if os.IsNotExist(err) { // expected if files appear/vanish return nil } if err != nil { return fmt.Errorf("unable to count inodes for part of dir %s: %s", dir, err) } // according to the docs, Sys can be nil if info.Sys() == nil { return fmt.Errorf("fileinfo Sys is nil") } s, ok := info.Sys().(*syscall.Stat_t) if !ok { return fmt.Errorf("unsupported fileinfo; could not convert to stat_t") } if s.Dev != rootDevId { // don't descend into directories on other devices return filepath.SkipDir } if s.Nlink > 1 { if _, ok := dedupedInodes[s.Ino]; !ok { // Dedupe things that could be hardlinks dedupedInodes[s.Ino] = struct{}{} usage.Bytes += uint64(s.Blocks) * statBlockSize usage.Inodes++ } } else { usage.Bytes += uint64(s.Blocks) * statBlockSize usage.Inodes++ } return nil }) return usage, nil } func (self *RealFsInfo) GetDirUsage(dir string) (UsageInfo, error) { claimToken() defer releaseToken() return GetDirUsage(dir) } func getVfsStats(path string) (total uint64, free uint64, avail uint64, inodes uint64, inodesFree uint64, err error) { var s syscall.Statfs_t if err = syscall.Statfs(path, &s); err != nil { return 0, 0, 0, 0, 0, err } total = uint64(s.Frsize) * s.Blocks free = uint64(s.Frsize) * s.Bfree avail = uint64(s.Frsize) * s.Bavail inodes = uint64(s.Files) inodesFree = uint64(s.Ffree) return total, free, avail, inodes, inodesFree, nil } // Devicemapper thin provisioning is detailed at // https://www.kernel.org/doc/Documentation/device-mapper/thin-provisioning.txt func dockerDMDevice(driverStatus map[string]string, dmsetup devicemapper.DmsetupClient) (string, uint, uint, uint, error) { poolName, ok := driverStatus[dockerutil.DriverStatusPoolName] if !ok || len(poolName) == 0 { return "", 0, 0, 0, fmt.Errorf("Could not get dm pool name") } out, err := dmsetup.Table(poolName) if err != nil { return "", 0, 0, 0, err } major, minor, dataBlkSize, err := parseDMTable(string(out)) if err != nil { return "", 0, 0, 0, err } return poolName, major, minor, dataBlkSize, nil } // parseDMTable parses a single line of `dmsetup table` output and returns the // major device, minor device, block size, and an error. func parseDMTable(dmTable string) (uint, uint, uint, error) { dmTable = strings.Replace(dmTable, ":", " ", -1) dmFields := strings.Fields(dmTable) if len(dmFields) < 8 { return 0, 0, 0, fmt.Errorf("Invalid dmsetup status output: %s", dmTable) } major, err := strconv.ParseUint(dmFields[5], 10, 32) if err != nil { return 0, 0, 0, err } minor, err := strconv.ParseUint(dmFields[6], 10, 32) if err != nil { return 0, 0, 0, err } dataBlkSize, err := strconv.ParseUint(dmFields[7], 10, 32) if err != nil { return 0, 0, 0, err } return uint(major), uint(minor), uint(dataBlkSize), nil } func getDMStats(poolName string, dataBlkSize uint) (uint64, uint64, uint64, error) { out, err := exec.Command("dmsetup", "status", poolName).Output() if err != nil { return 0, 0, 0, err } used, total, err := parseDMStatus(string(out)) if err != nil { return 0, 0, 0, err } used *= 512 * uint64(dataBlkSize) total *= 512 * uint64(dataBlkSize) free := total - used return total, free, free, nil } func parseDMStatus(dmStatus string) (uint64, uint64, error) { dmStatus = strings.Replace(dmStatus, "/", " ", -1) dmFields := strings.Fields(dmStatus) if len(dmFields) < 8 { return 0, 0, fmt.Errorf("Invalid dmsetup status output: %s", dmStatus) } used, err := strconv.ParseUint(dmFields[6], 10, 64) if err != nil { return 0, 0, err } total, err := strconv.ParseUint(dmFields[7], 10, 64) if err != nil { return 0, 0, err } return used, total, nil } // getZfstats returns ZFS mount stats using zfsutils func getZfstats(poolName string) (uint64, uint64, uint64, error) { dataset, err := zfs.GetDataset(poolName) if err != nil { return 0, 0, 0, err } total := dataset.Used + dataset.Avail + dataset.Usedbydataset return total, dataset.Avail, dataset.Avail, nil } // Simple io.Writer implementation that counts how many bytes were written. type byteCounter struct{ bytesWritten uint64 } func (b *byteCounter) Write(p []byte) (int, error) { b.bytesWritten += uint64(len(p)) return len(p), nil } // Get major and minor Ids for a mount point using btrfs as filesystem. func getBtrfsMajorMinorIds(mount *mount.Info) (int, int, error) { // btrfs fix: following workaround fixes wrong btrfs Major and Minor Ids reported in /proc/self/mountinfo. // instead of using values from /proc/self/mountinfo we use stat to get Ids from btrfs mount point buf := new(syscall.Stat_t) err := syscall.Stat(mount.Source, buf) if err != nil { err = fmt.Errorf("stat failed on %s with error: %s", mount.Source, err) return 0, 0, err } klog.V(4).Infof("btrfs mount %#v", mount) if buf.Mode&syscall.S_IFMT == syscall.S_IFBLK { err := syscall.Stat(mount.Mountpoint, buf) if err != nil { err = fmt.Errorf("stat failed on %s with error: %s", mount.Mountpoint, err) return 0, 0, err } klog.V(4).Infof("btrfs dev major:minor %d:%d\n", int(major(buf.Dev)), int(minor(buf.Dev))) klog.V(4).Infof("btrfs rdev major:minor %d:%d\n", int(major(buf.Rdev)), int(minor(buf.Rdev))) return int(major(buf.Dev)), int(minor(buf.Dev)), nil } else { return 0, 0, fmt.Errorf("%s is not a block device", mount.Source) } }