package zfs // #include // #include // #include "common.h" // #include "zpool.h" // #include "zfs.h" import "C" import ( "errors" "fmt" "path" "sort" "strings" "time" "unsafe" ) const ( msgDatasetIsNil = "Dataset handle not initialized or its closed" ) // DatasetProperties type is map of dataset or volume properties prop -> value type DatasetProperties map[Prop]string // DatasetType defines enum of dataset types type DatasetType int32 const ( // DatasetTypeFilesystem - file system dataset DatasetTypeFilesystem DatasetType = (1 << 0) // DatasetTypeSnapshot - snapshot of dataset DatasetTypeSnapshot = (1 << 1) // DatasetTypeVolume - volume (virtual block device) dataset DatasetTypeVolume = (1 << 2) // DatasetTypePool - pool dataset DatasetTypePool = (1 << 3) // DatasetTypeBookmark - bookmark dataset DatasetTypeBookmark = (1 << 4) ) // HoldTag - user holds tags type HoldTag struct { Name string Timestamp time.Time } // Dataset - ZFS dataset object type Dataset struct { list C.dataset_list_ptr Type DatasetType Properties map[Prop]Property Children []Dataset } func (d *Dataset) openChildren() (err error) { d.Children = make([]Dataset, 0, 5) list := C.dataset_list_children(d.list) for list != nil { dataset := Dataset{list: list} dataset.Type = DatasetType(C.dataset_type(d.list)) dataset.Properties = make(map[Prop]Property) err = dataset.ReloadProperties() if err != nil { return } d.Children = append(d.Children, dataset) list = C.dataset_next(list) } for ci := range d.Children { if err = d.Children[ci].openChildren(); err != nil { return } } return } // DatasetOpenAll recursive get handles to all available datasets on system // (file-systems, volumes or snapshots). func DatasetOpenAll() (datasets []Dataset, err error) { var dataset Dataset dataset.list = C.dataset_list_root() for dataset.list != nil { dataset.Type = DatasetType(C.dataset_type(dataset.list)) err = dataset.ReloadProperties() if err != nil { return } datasets = append(datasets, dataset) dataset.list = C.dataset_next(dataset.list) } for ci := range datasets { if err = datasets[ci].openChildren(); err != nil { return } } return } // DatasetCloseAll close all datasets in slice and all of its recursive // children datasets func DatasetCloseAll(datasets []Dataset) { for _, d := range datasets { d.Close() } } // DatasetOpen open dataset and all of its recursive children datasets func DatasetOpen(path string) (d Dataset, err error) { if d, err = DatasetOpenSingle(path); err != nil { return } err = d.openChildren() return } // DatasetOpenSingle open dataset without opening all of its recursive // children datasets func DatasetOpenSingle(path string) (d Dataset, err error) { csPath := C.CString(path) d.list = C.dataset_open(csPath) C.free(unsafe.Pointer(csPath)) if d.list == nil || d.list.zh == nil { err = LastError() if err == nil { err = fmt.Errorf("dataset not found") } err = fmt.Errorf("%s - %s", err.Error(), path) return } d.Type = DatasetType(C.dataset_type(d.list)) err = d.ReloadProperties() if err != nil { return } return } func datasetPropertiesTonvlist(props map[Prop]Property) ( cprops C.nvlist_ptr, err error) { // convert properties to nvlist C type cprops = C.new_property_nvlist() if cprops == nil { err = errors.New("Failed to allocate properties") return } for prop, value := range props { csValue := C.CString(value.Value) r := C.property_nvlist_add( cprops, C.zfs_prop_to_name(C.zfs_prop_t(prop)), csValue) C.free(unsafe.Pointer(csValue)) if r != 0 { err = errors.New("Failed to convert property") return } } return } // DatasetCreate create a new filesystem or volume on path representing // pool/dataset or pool/parent/dataset func DatasetCreate(path string, dtype DatasetType, props map[Prop]Property) (d Dataset, err error) { var cprops C.nvlist_ptr if cprops, err = datasetPropertiesTonvlist(props); err != nil { return } defer C.nvlist_free(cprops) csPath := C.CString(path) errcode := C.dataset_create(csPath, C.zfs_type_t(dtype), cprops) C.free(unsafe.Pointer(csPath)) if errcode != 0 { err = LastError() return } return DatasetOpen(path) } // Close close dataset and all its recursive children datasets (close handle // and cleanup dataset object/s from memory) func (d *Dataset) Close() { // path, _ := d.Path() Global.Mtx.Lock() C.dataset_list_close(d.list) d.list = nil Global.Mtx.Unlock() for _, cd := range d.Children { cd.Close() } } // reOpen - close and open dataset. Not thread safe! func (d *Dataset) reOpen() (err error) { d.Close() *d, err = DatasetOpen(d.Properties[DatasetPropName].Value) return } // Destroy destroys the dataset. The caller must make sure that the filesystem // isn't mounted, and that there are no active dependents. Set Defer argument // to true to defer destruction for when dataset is not in use. Call Close() to // cleanup memory. func (d *Dataset) Destroy(Defer bool) (err error) { if len(d.Children) > 0 { path, e := d.Path() if e != nil { return } dsType, e := d.GetProperty(DatasetPropType) if e != nil { dsType.Value = err.Error() // just put error (why it didn't fetch property type) } err = errors.New("Cannot destroy dataset " + path + ": " + dsType.Value + " has children") return } if d.list != nil { if ec := C.dataset_destroy(d.list, booleanT(Defer)); ec != 0 { err = LastError() } } else { err = errors.New(msgDatasetIsNil) } return } // IsSnapshot - retrun true if datset is snapshot func (d *Dataset) IsSnapshot() (ok bool) { path := d.Properties[DatasetPropName].Value ok = (d.Type == DatasetTypeSnapshot || strings.Contains(path, "@")) return } // DestroyRecursive recursively destroy children of dataset and dataset. func (d *Dataset) DestroyRecursive() (err error) { var path string if path, err = d.Path(); err != nil { return } if !strings.Contains(path, "@") { // not snapshot if len(d.Children) > 0 { for _, c := range d.Children { if err = c.DestroyRecursive(); err != nil { return } // close handle to destroyed child dataset c.Close() } // clear closed children array d.Children = make([]Dataset, 0) } err = d.Destroy(false) } else { var parent Dataset tmp := strings.Split(path, "@") ppath, snapname := tmp[0], tmp[1] if parent, err = DatasetOpen(ppath); err != nil { return } defer parent.Close() if len(parent.Children) > 0 { for _, c := range parent.Children { if path, err = c.Path(); err != nil { return } if strings.Contains(path, "@") { continue // skip other snapshots } if c, err = DatasetOpen(path + "@" + snapname); err != nil { continue } if err = c.DestroyRecursive(); err != nil { c.Close() return } c.Close() } } err = d.Destroy(false) } return } // Pool returns pool dataset belongs to func (d *Dataset) Pool() (p Pool, err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } p.list = C.dataset_get_pool(d.list) if p.list != nil && p.list.zph != nil { err = p.ReloadProperties() return } err = LastError() return } // PoolName - return name of the pool func (d *Dataset) PoolName() string { path := d.Properties[DatasetPropName].Value i := strings.Index(path, "/") if i < 0 { return path } return path[0:i] } // ReloadProperties re-read dataset's properties func (d *Dataset) ReloadProperties() (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } d.Properties = make(map[Prop]Property) C.zfs_refresh_properties(d.list.zh) for prop := DatasetPropType; prop < DatasetNumProps; prop++ { var plist *C.property_list_t if prop == DatasetPropMounted || prop == DatasetPropMountpoint { // prevent zfs mountpoint race conditions Global.Mtx.Lock() plist = C.read_dataset_property(d.list, C.int(prop)) Global.Mtx.Unlock() } else { plist = C.read_dataset_property(d.list, C.int(prop)) } if plist == nil { continue } d.Properties[prop] = Property{Value: C.GoString(&(*plist).value[0]), Source: C.GoString(&(*plist).source[0])} C.free_properties(plist) } return } // GetProperty reload and return single specified property. This also reloads requested // property in Properties map. func (d *Dataset) GetProperty(p Prop) (prop Property, err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } if p == DatasetPropMounted || p == DatasetPropMountpoint { // prevent zfs mountpoint race conditions Global.Mtx.Lock() defer Global.Mtx.Unlock() } plist := C.read_dataset_property(d.list, C.int(p)) if plist == nil { err = LastError() return } defer C.free_properties(plist) prop = Property{Value: C.GoString(&(*plist).value[0]), Source: C.GoString(&(*plist).source[0])} d.Properties[p] = prop return } // GetUserProperty - lookup and return user propery func (d *Dataset) GetUserProperty(p string) (prop Property, err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } csp := C.CString(p) defer C.free(unsafe.Pointer(csp)) plist := C.read_user_property(d.list, csp) if plist == nil { err = LastError() return } defer C.free_properties(plist) prop = Property{Value: C.GoString(&(*plist).value[0]), Source: C.GoString(&(*plist).source[0])} return } // SetProperty set ZFS dataset property to value. Not all properties can be set, // some can be set only at creation time and some are read only. // Always check if returned error and its description. func (d *Dataset) SetProperty(p Prop, value string) (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } csValue := C.CString(value) if p == DatasetPropMounted || p == DatasetPropMountpoint { // prevent zfs mountpoint race conditions Global.Mtx.Lock() defer Global.Mtx.Unlock() } errcode := C.dataset_prop_set(d.list, C.zfs_prop_t(p), csValue) C.free(unsafe.Pointer(csValue)) if errcode != 0 { err = LastError() return } // Update Properties member with change made plist := C.read_dataset_property(d.list, C.int(p)) if plist == nil { err = LastError() return } defer C.free_properties(plist) d.Properties[p] = Property{Value: C.GoString(&(*plist).value[0]), Source: C.GoString(&(*plist).source[0])} return } // SetUserProperty - func (d *Dataset) SetUserProperty(prop, value string) (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } csValue := C.CString(value) csProp := C.CString(prop) errcode := C.dataset_user_prop_set(d.list, csProp, csValue) C.free(unsafe.Pointer(csValue)) C.free(unsafe.Pointer(csProp)) if errcode != 0 { err = LastError() } return } // Clone - clones the dataset. The target must be of the same type as // the source. func (d *Dataset) Clone(target string, props map[Prop]Property) (rd Dataset, err error) { var cprops C.nvlist_ptr if d.list == nil { err = errors.New(msgDatasetIsNil) return } if cprops, err = datasetPropertiesTonvlist(props); err != nil { return } defer C.nvlist_free(cprops) csTarget := C.CString(target) defer C.free(unsafe.Pointer(csTarget)) if errc := C.dataset_clone(d.list, csTarget, cprops); errc != 0 { err = LastError() return } rd, err = DatasetOpen(target) return } // DatasetSnapshot create dataset snapshot. Set recur to true to snapshot child datasets. func DatasetSnapshot(path string, recur bool, props map[Prop]Property) (rd Dataset, err error) { var cprops C.nvlist_ptr if cprops, err = datasetPropertiesTonvlist(props); err != nil { return } defer C.nvlist_free(cprops) csPath := C.CString(path) defer C.free(unsafe.Pointer(csPath)) if errc := C.dataset_snapshot(csPath, booleanT(recur), cprops); errc != 0 { err = LastError() return } rd, err = DatasetOpen(path) return } // Path return zfs dataset path/name func (d *Dataset) Path() (path string, err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } name := C.dataset_get_name(d.list) path = C.GoString(name) return } // Rollback rollabck's dataset snapshot func (d *Dataset) Rollback(snap *Dataset, force bool) (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } if errc := C.dataset_rollback(d.list, snap.list, booleanT(force)); errc != 0 { err = LastError() return } d.ReloadProperties() return } // Promote promotes dataset clone func (d *Dataset) Promote() (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } if errc := C.dataset_promote(d.list); errc != 0 { err = LastError() return } d.ReloadProperties() return } // Rename dataset func (d *Dataset) Rename(newName string, recur, forceUnmount bool) (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } csNewName := C.CString(newName) defer C.free(unsafe.Pointer(csNewName)) if errc := C.dataset_rename(d.list, csNewName, booleanT(recur), booleanT(forceUnmount)); errc != 0 { err = LastError() return } d.ReloadProperties() return } // IsMounted checks to see if the mount is active. If the filesystem is mounted, // sets in 'where' argument the current mountpoint, and returns true. Otherwise, // returns false. func (d *Dataset) IsMounted() (mounted bool, where string) { if d.list == nil { return } Global.Mtx.Lock() defer Global.Mtx.Unlock() mp := C.dataset_is_mounted(d.list) // defer C.free(mp) if mounted = (mp != nil); mounted { where = C.GoString(mp) C.free(unsafe.Pointer(mp)) } return } // Mount the given filesystem. func (d *Dataset) Mount(options string, flags int) (err error) { Global.Mtx.Lock() defer Global.Mtx.Unlock() if d.list == nil { err = errors.New(msgDatasetIsNil) return } csOptions := C.CString(options) defer C.free(unsafe.Pointer(csOptions)) if ec := C.dataset_mount(d.list, csOptions, C.int(flags)); ec != 0 { err = LastError() } return } // Unmount the given filesystem. func (d *Dataset) Unmount(flags int) (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } if ec := C.dataset_unmount(d.list, C.int(flags)); ec != 0 { err = LastError() } return } // UnmountAll unmount this filesystem and any children inheriting the // mountpoint property. func (d *Dataset) UnmountAll(flags int) (err error) { if d.list == nil { err = errors.New(msgDatasetIsNil) return } // This is implemented recursive because zfs_unmountall() didn't work if len(d.Children) > 0 { for _, c := range d.Children { if err = c.UnmountAll(flags); err != nil { return } } } return d.Unmount(flags) } // Hold - Adds a single reference, named with the tag argument, to the snapshot. // Each snapshot has its own tag namespace, and tags must be unique within that space. func (d *Dataset) Hold(flag string) (err error) { var path string var pd Dataset if path, err = d.Path(); err != nil { return } if !strings.Contains(path, "@") { err = fmt.Errorf("'%s' is not a snapshot", path) return } pd, err = DatasetOpenSingle(path[:strings.Index(path, "@")]) if err != nil { return } defer pd.Close() csSnapName := C.CString(path[strings.Index(path, "@")+1:]) defer C.free(unsafe.Pointer(csSnapName)) csFlag := C.CString(flag) defer C.free(unsafe.Pointer(csFlag)) if 0 != C.zfs_hold(pd.list.zh, csSnapName, csFlag, booleanT(false), -1) { err = LastError() } return } // Release - Removes a single reference, named with the tag argument, from the specified snapshot. // The tag must already exist for each snapshot. If a hold exists on a snapshot, attempts to destroy // that snapshot by using the zfs destroy command return EBUSY. func (d *Dataset) Release(flag string) (err error) { var path string var pd Dataset if path, err = d.Path(); err != nil { return } if !strings.Contains(path, "@") { err = fmt.Errorf("'%s' is not a snapshot", path) return } pd, err = DatasetOpenSingle(path[:strings.Index(path, "@")]) if err != nil { return } defer pd.Close() csSnapName := C.CString(path[strings.Index(path, "@")+1:]) defer C.free(unsafe.Pointer(csSnapName)) csFlag := C.CString(flag) defer C.free(unsafe.Pointer(csFlag)) if 0 != C.zfs_release(pd.list.zh, csSnapName, csFlag, booleanT(false)) { err = LastError() } return } // Holds - Lists all existing user references for the given snapshot func (d *Dataset) Holds() (tags []HoldTag, err error) { var nvl *C.nvlist_t var nvp *C.nvpair_t var tu64 C.uint64_t var path string if path, err = d.Path(); err != nil { return } if !strings.Contains(path, "@") { err = fmt.Errorf("'%s' is not a snapshot", path) return } if 0 != C.zfs_get_holds(d.list.zh, &nvl) { err = LastError() return } defer C.nvlist_free(nvl) tags = make([]HoldTag, 0, 5) for nvp = C.nvlist_next_nvpair(nvl, nvp); nvp != nil; { tag := C.nvpair_name(nvp) C.nvpair_value_uint64(nvp, &tu64) tags = append(tags, HoldTag{ Name: C.GoString(tag), Timestamp: time.Unix(int64(tu64), 0), }) nvp = C.nvlist_next_nvpair(nvl, nvp) } return } // DatasetPropertyToName convert property to name // ( returns built in string representation of property name). // This is optional, you can represent each property with string // name of choice. func DatasetPropertyToName(p Prop) (name string) { if p == DatasetNumProps { return "numofprops" } prop := C.zfs_prop_t(p) name = C.GoString(C.zfs_prop_to_name(prop)) return } // DestroyPromote - Same as DestroyRecursive() except it will not destroy // any dependent clones, but promote them first. // This function will navigate any dependency chain // of cloned datasets using breadth first search to promote according and let // you remove dataset regardless of its cloned dependencies. // Note: that this function wan't work when you want to destroy snapshot this way. // However it will destroy all snaphsot of destroyed dataset without dependencies, // otherwise snapshot will move to promoted clone func (d *Dataset) DestroyPromote() (err error) { var snaps []Dataset var clones []string // We need to save list of child snapshots, to destroy them latter // since they will be moved to promoted clone var psnaps []string if clones, err = d.Clones(); err != nil { return } if len(clones) > 0 { var cds Dataset // For this to always work we need to promote youngest clone // in terms of most recent origin snapshot or creation time if // cloned from same snapshot if cds, err = DatasetOpen(clones[0]); err != nil { return } defer cds.Close() // since promote will move the snapshots to promoted dataset // we need to check and resolve possible name conflicts if snaps, err = d.Snapshots(); err != nil { return } for _, s := range snaps { spath := s.Properties[DatasetPropName].Value sname := spath[strings.Index(spath, "@"):] // conflict and resolve if ok, _ := cds.FindSnapshotName(sname); ok { // snapshot with the same name already exist volname := path.Base(spath[:strings.Index(spath, "@")]) sname = sname + "." + volname if err = s.Rename(spath+"."+volname, false, true); err != nil { return } } psnaps = append(psnaps, sname) } if err = cds.Promote(); err != nil { return } } // destroy child datasets, since this works recursive for _, cd := range d.Children { if err = cd.DestroyPromote(); err != nil { return } } d.Children = make([]Dataset, 0) if err = d.Destroy(false); err != nil { return } // Load with new promoted snapshots if len(clones) > 0 && len(psnaps) > 0 { var cds Dataset if cds, err = DatasetOpen(clones[0]); err != nil { return } defer cds.Close() // try to destroy (promoted) snapshots now for _, sname := range psnaps { if ok, snap := cds.FindSnapshotName(sname); ok { snap.Destroy(false) } } } return } // Snapshots - filter and return all snapshots of dataset func (d *Dataset) Snapshots() (snaps []Dataset, err error) { for _, ch := range d.Children { if !ch.IsSnapshot() { continue } snaps = append(snaps, ch) } return } // FindSnapshot - returns true if given path is one of dataset snaphsots func (d *Dataset) FindSnapshot(path string) (ok bool, snap Dataset) { for _, ch := range d.Children { if !ch.IsSnapshot() { continue } if ok = (path == ch.Properties[DatasetPropName].Value); ok { snap = ch break } } return } // FindSnapshotName - returns true and snapshot if given snapshot // name eg. '@snap1' is one of dataset snaphsots func (d *Dataset) FindSnapshotName(name string) (ok bool, snap Dataset) { return d.FindSnapshot(d.Properties[DatasetPropName].Value + name) } // Clones - get list of all dataset paths cloned from this // dataset or this snapshot // List is sorted descedent by origin snapshot order func (d *Dataset) Clones() (clones []string, err error) { // Clones can only live on same pool var root Dataset var sortDesc []Dataset if root, err = DatasetOpen(d.PoolName()); err != nil { return } defer root.Close() dIsSnapshot := d.IsSnapshot() // USe breadth first search to find all clones queue := make(chan Dataset, 1024) defer close(queue) // This will close and cleanup all queue <- root // start from the root element for { select { case ds := <-queue: // pull from queue (breadth first search) for _, ch := range ds.Children { origin := ch.Properties[DatasetPropOrigin].Value if len(origin) > 0 { if dIsSnapshot && origin == d.Properties[DatasetPropName].Value { // if this dataset is snaphot ch.Properties[DatasetNumProps+1000] = d.Properties[DatasetPropCreateTXG] sortDesc = append(sortDesc, ch) } else { // Check if origin of this dataset is one of snapshots ok, snap := d.FindSnapshot(origin) if !ok { continue } ch.Properties[DatasetNumProps+1000] = snap.Properties[DatasetPropCreateTXG] sortDesc = append(sortDesc, ch) } } queue <- ch } default: sort.Sort(clonesCreateDesc(sortDesc)) // This way we get clones ordered from most recent sanpshots first for _, c := range sortDesc { clones = append(clones, c.Properties[DatasetPropName].Value) } return } } return }