rclone/fs/walk/walk.go
Nick Craig-Wood efd720b533 walk: Implement walk.ListR which will use ListR if at all possible
It otherwise has the nearly the same interface as walk.Walk which it
will fall back to if it can't use ListR.

Using walk.ListR will speed up file system operations by default and
use much less memory and start immediately compared to if --fast-list
had been supplied.
2019-03-16 17:41:12 +00:00

815 lines
22 KiB
Go

// Package walk walks directories
package walk
import (
"bytes"
"fmt"
"path"
"sort"
"strings"
"sync"
"time"
"github.com/ncw/rclone/fs"
"github.com/ncw/rclone/fs/filter"
"github.com/ncw/rclone/fs/list"
"github.com/pkg/errors"
)
// ErrorSkipDir is used as a return value from Walk to indicate that the
// directory named in the call is to be skipped. It is not returned as
// an error by any function.
var ErrorSkipDir = errors.New("skip this directory")
// ErrorCantListR is returned by WalkR if the underlying Fs isn't
// capable of doing a recursive listing.
var ErrorCantListR = errors.New("recursive directory listing not available")
// Func is the type of the function called for directory
// visited by Walk. The path argument contains remote path to the directory.
//
// If there was a problem walking to directory named by path, the
// incoming error will describe the problem and the function can
// decide how to handle that error (and Walk will not descend into
// that directory). If an error is returned, processing stops. The
// sole exception is when the function returns the special value
// ErrorSkipDir. If the function returns ErrorSkipDir, Walk skips the
// directory's contents entirely.
type Func func(path string, entries fs.DirEntries, err error) error
// Walk lists the directory.
//
// If includeAll is not set it will use the filters defined.
//
// If maxLevel is < 0 then it will recurse indefinitely, else it will
// only do maxLevel levels.
//
// It calls fn for each tranche of DirEntries read.
//
// Note that fn will not be called concurrently whereas the directory
// listing will proceed concurrently.
//
// Parent directories are always listed before their children
//
// This is implemented by WalkR if Config.UseUseListR is true
// and f supports it and level > 1, or WalkN otherwise.
//
// If --files-from is set then a DirTree will be constructed with just
// those files in and then walked with WalkR
//
// NB (f, path) to be replaced by fs.Dir at some point
func Walk(f fs.Fs, path string, includeAll bool, maxLevel int, fn Func) error {
if filter.Active.HaveFilesFrom() {
return walkR(f, path, includeAll, maxLevel, fn, filter.Active.MakeListR(f.NewObject))
}
// FIXME should this just be maxLevel < 0 - why the maxLevel > 1
if (maxLevel < 0 || maxLevel > 1) && fs.Config.UseListR && f.Features().ListR != nil {
return walkListR(f, path, includeAll, maxLevel, fn)
}
return walkListDirSorted(f, path, includeAll, maxLevel, fn)
}
// ListType is uses to choose which combination of files or directories is requires
type ListType byte
// Types of listing for ListR
const (
ListObjects ListType = 1 << iota // list objects only
ListDirs // list dirs only
ListAll = ListObjects | ListDirs // list files and dirs
)
// Objects returns true if the list type specifies objects
func (l ListType) Objects() bool {
return (l & ListObjects) != 0
}
// Dirs returns true if the list type specifies dirs
func (l ListType) Dirs() bool {
return (l & ListDirs) != 0
}
// Filter in (inplace) to only contain the type of list entry required
func (l ListType) Filter(in *fs.DirEntries) {
if l == ListAll {
return
}
out := (*in)[:0]
for _, entry := range *in {
switch entry.(type) {
case fs.Object:
if l.Objects() {
out = append(out, entry)
}
case fs.Directory:
if l.Dirs() {
out = append(out, entry)
}
default:
fs.Errorf(nil, "Unknown object type %T", entry)
}
}
*in = out
}
// ListR lists the directory recursively.
//
// If includeAll is not set it will use the filters defined.
//
// If maxLevel is < 0 then it will recurse indefinitely, else it will
// only do maxLevel levels.
//
// If synthesizeDirs is set then for bucket based remotes it will
// synthesize directories from the file structure. This uses extra
// memory so don't set this if you don't need directories, likewise do
// set this if you are interested in directories.
//
// It calls fn for each tranche of DirEntries read. Note that these
// don't necessarily represent a directory
//
// Note that fn will not be called concurrently whereas the directory
// listing will proceed concurrently.
//
// Directories are not listed in any particular order so you can't
// rely on parents coming before children or alphabetical ordering
//
// This is implemented by using ListR on the backend if possible and
// efficient, otherwise by Walk.
//
// NB (f, path) to be replaced by fs.Dir at some point
func ListR(f fs.Fs, path string, includeAll bool, maxLevel int, listType ListType, fn fs.ListRCallback) error {
// FIXME disable this with --no-fast-list ??? `--disable ListR` will do it...
doListR := f.Features().ListR
// Can't use ListR if...
if doListR == nil || // ...no ListR
filter.Active.HaveFilesFrom() || // ...using --files-from
maxLevel >= 0 || // ...using bounded recursion
len(filter.Active.Opt.ExcludeFile) > 0 || // ...using --exclude-file
filter.Active.BoundedRecursion() { // ...filters imply bounded recursion
return listRwalk(f, path, includeAll, maxLevel, listType, fn)
}
return listR(f, path, includeAll, listType, fn, doListR, listType.Dirs() && f.Features().BucketBased)
}
// listRwalk walks the file tree for ListR using Walk
func listRwalk(f fs.Fs, path string, includeAll bool, maxLevel int, listType ListType, fn fs.ListRCallback) error {
var listErr error
walkErr := Walk(f, path, includeAll, maxLevel, func(path string, entries fs.DirEntries, err error) error {
// Carry on listing but return the error at the end
if err != nil {
listErr = err
fs.CountError(err)
fs.Errorf(path, "error listing: %v", err)
return nil
}
listType.Filter(&entries)
return fn(entries)
})
if listErr != nil {
return listErr
}
return walkErr
}
// dirMap keeps track of directories made for bucket based remotes.
// true => directory has been sent
// false => directory has been seen but not sent
type dirMap struct {
mu sync.Mutex
m map[string]bool
root string
}
// make a new dirMap
func newDirMap(root string) *dirMap {
return &dirMap{
m: make(map[string]bool),
root: root,
}
}
// add adds a directory and parents with sent
func (dm *dirMap) add(dir string, sent bool) {
for {
if dir == dm.root || dir == "" {
return
}
currentSent, found := dm.m[dir]
if found {
// If it has been sent already then nothing more to do
if currentSent {
return
}
// If not sent already don't override
if !sent {
return
}
// currenSent == false && sent == true so needs overriding
}
dm.m[dir] = sent
// Add parents in as unsent
dir = parentDir(dir)
sent = false
}
}
// add all the directories in entries and their parents to the dirMap
func (dm *dirMap) addEntries(entries fs.DirEntries) error {
dm.mu.Lock()
defer dm.mu.Unlock()
for _, entry := range entries {
switch x := entry.(type) {
case fs.Object:
dm.add(parentDir(x.Remote()), false)
case fs.Directory:
dm.add(x.Remote(), true)
default:
return errors.Errorf("unknown object type %T", entry)
}
}
return nil
}
// send any missing parents to fn
func (dm *dirMap) sendEntries(fn fs.ListRCallback) (err error) {
// Count the strings first so we allocate the minimum memory
n := 0
for _, sent := range dm.m {
if !sent {
n++
}
}
if n == 0 {
return nil
}
dirs := make([]string, 0, n)
// Fill the dirs up then sort it
for dir, sent := range dm.m {
if !sent {
dirs = append(dirs, dir)
}
}
sort.Strings(dirs)
// Now convert to bulkier Dir in batches and send
now := time.Now()
list := NewListRHelper(fn)
for _, dir := range dirs {
err = list.Add(fs.NewDir(dir, now))
if err != nil {
return err
}
}
return list.Flush()
}
// listR walks the file tree using ListR
func listR(f fs.Fs, path string, includeAll bool, listType ListType, fn fs.ListRCallback, doListR fs.ListRFn, synthesizeDirs bool) error {
includeDirectory := filter.Active.IncludeDirectory(f)
if !includeAll {
includeAll = filter.Active.InActive()
}
var dm *dirMap
if synthesizeDirs {
dm = newDirMap(path)
}
var mu sync.Mutex
err := doListR(path, func(entries fs.DirEntries) (err error) {
if synthesizeDirs {
err = dm.addEntries(entries)
if err != nil {
return err
}
}
listType.Filter(&entries)
if !includeAll {
filteredEntries := entries[:0]
for _, entry := range entries {
var include bool
switch x := entry.(type) {
case fs.Object:
include = filter.Active.IncludeObject(x)
case fs.Directory:
include, err = includeDirectory(x.Remote())
if err != nil {
return err
}
default:
return errors.Errorf("unknown object type %T", entry)
}
if include {
filteredEntries = append(filteredEntries, entry)
} else {
fs.Debugf(entry, "Excluded from sync (and deletion)")
}
}
entries = filteredEntries
}
mu.Lock()
defer mu.Unlock()
return fn(entries)
})
if err != nil {
return err
}
if synthesizeDirs {
err = dm.sendEntries(fn)
if err != nil {
return err
}
}
return nil
}
// walkListDirSorted lists the directory.
//
// It implements Walk using non recursive directory listing.
func walkListDirSorted(f fs.Fs, path string, includeAll bool, maxLevel int, fn Func) error {
return walk(f, path, includeAll, maxLevel, fn, list.DirSorted)
}
// walkListR lists the directory.
//
// It implements Walk using recursive directory listing if
// available, or returns ErrorCantListR if not.
func walkListR(f fs.Fs, path string, includeAll bool, maxLevel int, fn Func) error {
listR := f.Features().ListR
if listR == nil {
return ErrorCantListR
}
return walkR(f, path, includeAll, maxLevel, fn, listR)
}
type listDirFunc func(fs fs.Fs, includeAll bool, dir string) (entries fs.DirEntries, err error)
func walk(f fs.Fs, path string, includeAll bool, maxLevel int, fn Func, listDir listDirFunc) error {
var (
wg sync.WaitGroup // sync closing of go routines
traversing sync.WaitGroup // running directory traversals
doClose sync.Once // close the channel once
mu sync.Mutex // stop fn being called concurrently
)
// listJob describe a directory listing that needs to be done
type listJob struct {
remote string
depth int
}
in := make(chan listJob, fs.Config.Checkers)
errs := make(chan error, 1)
quit := make(chan struct{})
closeQuit := func() {
doClose.Do(func() {
close(quit)
go func() {
for range in {
traversing.Done()
}
}()
})
}
for i := 0; i < fs.Config.Checkers; i++ {
wg.Add(1)
go func() {
defer wg.Done()
for {
select {
case job, ok := <-in:
if !ok {
return
}
entries, err := listDir(f, includeAll, job.remote)
var jobs []listJob
if err == nil && job.depth != 0 {
entries.ForDir(func(dir fs.Directory) {
// Recurse for the directory
jobs = append(jobs, listJob{
remote: dir.Remote(),
depth: job.depth - 1,
})
})
}
mu.Lock()
err = fn(job.remote, entries, err)
mu.Unlock()
// NB once we have passed entries to fn we mustn't touch it again
if err != nil && err != ErrorSkipDir {
traversing.Done()
fs.CountError(err)
fs.Errorf(job.remote, "error listing: %v", err)
closeQuit()
// Send error to error channel if space
select {
case errs <- err:
default:
}
continue
}
if err == nil && len(jobs) > 0 {
traversing.Add(len(jobs))
go func() {
// Now we have traversed this directory, send these
// jobs off for traversal in the background
for _, newJob := range jobs {
in <- newJob
}
}()
}
traversing.Done()
case <-quit:
return
}
}
}()
}
// Start the process
traversing.Add(1)
in <- listJob{
remote: path,
depth: maxLevel - 1,
}
traversing.Wait()
close(in)
wg.Wait()
close(errs)
// return the first error returned or nil
return <-errs
}
// DirTree is a map of directories to entries
type DirTree map[string]fs.DirEntries
// parentDir finds the parent directory of path
func parentDir(entryPath string) string {
dirPath := path.Dir(entryPath)
if dirPath == "." {
dirPath = ""
}
return dirPath
}
// add an entry to the tree
func (dt DirTree) add(entry fs.DirEntry) {
dirPath := parentDir(entry.Remote())
dt[dirPath] = append(dt[dirPath], entry)
}
// add a directory entry to the tree
func (dt DirTree) addDir(entry fs.DirEntry) {
dt.add(entry)
// create the directory itself if it doesn't exist already
dirPath := entry.Remote()
if _, ok := dt[dirPath]; !ok {
dt[dirPath] = nil
}
}
// Find returns the DirEntry for filePath or nil if not found
func (dt DirTree) Find(filePath string) (parentPath string, entry fs.DirEntry) {
parentPath = parentDir(filePath)
for _, entry := range dt[parentPath] {
if entry.Remote() == filePath {
return parentPath, entry
}
}
return parentPath, nil
}
// check that dirPath has a *Dir in its parent
func (dt DirTree) checkParent(root, dirPath string) {
if dirPath == root {
return
}
parentPath, entry := dt.Find(dirPath)
if entry != nil {
return
}
dt[parentPath] = append(dt[parentPath], fs.NewDir(dirPath, time.Now()))
dt.checkParent(root, parentPath)
}
// check every directory in the tree has *Dir in its parent
func (dt DirTree) checkParents(root string) {
for dirPath := range dt {
dt.checkParent(root, dirPath)
}
}
// Sort sorts all the Entries
func (dt DirTree) Sort() {
for _, entries := range dt {
sort.Stable(entries)
}
}
// Dirs returns the directories in sorted order
func (dt DirTree) Dirs() (dirNames []string) {
for dirPath := range dt {
dirNames = append(dirNames, dirPath)
}
sort.Strings(dirNames)
return dirNames
}
// Prune remove directories from a directory tree. dirNames contains
// all directories to remove as keys, with true as values. dirNames
// will be modified in the function.
func (dt DirTree) Prune(dirNames map[string]bool) error {
// We use map[string]bool to avoid recursion (and potential
// stack exhaustion).
// First we need delete directories from their parents.
for dName, remove := range dirNames {
if !remove {
// Currently all values should be
// true, therefore this should not
// happen. But this makes function
// more predictable.
fs.Infof(dName, "Directory in the map for prune, but the value is false")
continue
}
if dName == "" {
// if dName is root, do nothing (no parent exist)
continue
}
parent := parentDir(dName)
// It may happen that dt does not have a dName key,
// since directory was excluded based on a filter. In
// such case the loop will be skipped.
for i, entry := range dt[parent] {
switch x := entry.(type) {
case fs.Directory:
if x.Remote() == dName {
// the slice is not sorted yet
// to delete item
// a) replace it with the last one
dt[parent][i] = dt[parent][len(dt[parent])-1]
// b) remove last
dt[parent] = dt[parent][:len(dt[parent])-1]
// we modify a slice within a loop, but we stop
// iterating immediately
break
}
case fs.Object:
// do nothing
default:
return errors.Errorf("unknown object type %T", entry)
}
}
}
for len(dirNames) > 0 {
// According to golang specs, if new keys were added
// during range iteration, they may be skipped.
for dName, remove := range dirNames {
if !remove {
fs.Infof(dName, "Directory in the map for prune, but the value is false")
continue
}
// First, add all subdirectories to dirNames.
// It may happen that dt[dName] does not exist.
// If so, the loop will be skipped.
for _, entry := range dt[dName] {
switch x := entry.(type) {
case fs.Directory:
excludeDir := x.Remote()
dirNames[excludeDir] = true
case fs.Object:
// do nothing
default:
return errors.Errorf("unknown object type %T", entry)
}
}
// Then remove current directory from DirTree
delete(dt, dName)
// and from dirNames
delete(dirNames, dName)
}
}
return nil
}
// String emits a simple representation of the DirTree
func (dt DirTree) String() string {
out := new(bytes.Buffer)
for _, dir := range dt.Dirs() {
_, _ = fmt.Fprintf(out, "%s/\n", dir)
for _, entry := range dt[dir] {
flag := ""
if _, ok := entry.(fs.Directory); ok {
flag = "/"
}
_, _ = fmt.Fprintf(out, " %s%s\n", path.Base(entry.Remote()), flag)
}
}
return out.String()
}
func walkRDirTree(f fs.Fs, startPath string, includeAll bool, maxLevel int, listR fs.ListRFn) (DirTree, error) {
dirs := make(DirTree)
// Entries can come in arbitrary order. We use toPrune to keep
// all directories to exclude later.
toPrune := make(map[string]bool)
includeDirectory := filter.Active.IncludeDirectory(f)
var mu sync.Mutex
err := listR(startPath, func(entries fs.DirEntries) error {
mu.Lock()
defer mu.Unlock()
for _, entry := range entries {
slashes := strings.Count(entry.Remote(), "/")
switch x := entry.(type) {
case fs.Object:
// Make sure we don't delete excluded files if not required
if includeAll || filter.Active.IncludeObject(x) {
if maxLevel < 0 || slashes <= maxLevel-1 {
dirs.add(x)
} else {
// Make sure we include any parent directories of excluded objects
dirPath := x.Remote()
for ; slashes > maxLevel-1; slashes-- {
dirPath = parentDir(dirPath)
}
dirs.checkParent(startPath, dirPath)
}
} else {
fs.Debugf(x, "Excluded from sync (and deletion)")
}
// Check if we need to prune a directory later.
if !includeAll && len(filter.Active.Opt.ExcludeFile) > 0 {
basename := path.Base(x.Remote())
if basename == filter.Active.Opt.ExcludeFile {
excludeDir := parentDir(x.Remote())
toPrune[excludeDir] = true
fs.Debugf(basename, "Excluded from sync (and deletion) based on exclude file")
}
}
case fs.Directory:
inc, err := includeDirectory(x.Remote())
if err != nil {
return err
}
if includeAll || inc {
if maxLevel < 0 || slashes <= maxLevel-1 {
if slashes == maxLevel-1 {
// Just add the object if at maxLevel
dirs.add(x)
} else {
dirs.addDir(x)
}
}
} else {
fs.Debugf(x, "Excluded from sync (and deletion)")
}
default:
return errors.Errorf("unknown object type %T", entry)
}
}
return nil
})
if err != nil {
return nil, err
}
dirs.checkParents(startPath)
if len(dirs) == 0 {
dirs[startPath] = nil
}
err = dirs.Prune(toPrune)
if err != nil {
return nil, err
}
dirs.Sort()
return dirs, nil
}
// Create a DirTree using List
func walkNDirTree(f fs.Fs, path string, includeAll bool, maxLevel int, listDir listDirFunc) (DirTree, error) {
dirs := make(DirTree)
fn := func(dirPath string, entries fs.DirEntries, err error) error {
if err == nil {
dirs[dirPath] = entries
}
return err
}
err := walk(f, path, includeAll, maxLevel, fn, listDir)
if err != nil {
return nil, err
}
return dirs, nil
}
// NewDirTree returns a DirTree filled with the directory listing
// using the parameters supplied.
//
// If includeAll is not set it will use the filters defined.
//
// If maxLevel is < 0 then it will recurse indefinitely, else it will
// only do maxLevel levels.
//
// This is implemented by WalkR if f supports ListR and level > 1, or
// WalkN otherwise.
//
// If --files-from is set then a DirTree will be constructed with just
// those files in.
//
// NB (f, path) to be replaced by fs.Dir at some point
func NewDirTree(f fs.Fs, path string, includeAll bool, maxLevel int) (DirTree, error) {
if filter.Active.HaveFilesFrom() {
return walkRDirTree(f, path, includeAll, maxLevel, filter.Active.MakeListR(f.NewObject))
}
if ListR := f.Features().ListR; (maxLevel < 0 || maxLevel > 1) && ListR != nil {
return walkRDirTree(f, path, includeAll, maxLevel, ListR)
}
return walkNDirTree(f, path, includeAll, maxLevel, list.DirSorted)
}
func walkR(f fs.Fs, path string, includeAll bool, maxLevel int, fn Func, listR fs.ListRFn) error {
dirs, err := walkRDirTree(f, path, includeAll, maxLevel, listR)
if err != nil {
return err
}
skipping := false
skipPrefix := ""
emptyDir := fs.DirEntries{}
for _, dirPath := range dirs.Dirs() {
if skipping {
// Skip over directories as required
if strings.HasPrefix(dirPath, skipPrefix) {
continue
}
skipping = false
}
entries := dirs[dirPath]
if entries == nil {
entries = emptyDir
}
err = fn(dirPath, entries, nil)
if err == ErrorSkipDir {
skipping = true
skipPrefix = dirPath
if skipPrefix != "" {
skipPrefix += "/"
}
} else if err != nil {
return err
}
}
return nil
}
// GetAll runs ListR getting all the results
func GetAll(f fs.Fs, path string, includeAll bool, maxLevel int) (objs []fs.Object, dirs []fs.Directory, err error) {
err = ListR(f, path, includeAll, maxLevel, ListAll, func(entries fs.DirEntries) error {
for _, entry := range entries {
switch x := entry.(type) {
case fs.Object:
objs = append(objs, x)
case fs.Directory:
dirs = append(dirs, x)
}
}
return nil
})
return
}
// ListRHelper is used in the implementation of ListR to accumulate DirEntries
type ListRHelper struct {
callback fs.ListRCallback
entries fs.DirEntries
}
// NewListRHelper should be called from ListR with the callback passed in
func NewListRHelper(callback fs.ListRCallback) *ListRHelper {
return &ListRHelper{
callback: callback,
}
}
// send sends the stored entries to the callback if there are >= max
// entries.
func (lh *ListRHelper) send(max int) (err error) {
if len(lh.entries) >= max {
err = lh.callback(lh.entries)
lh.entries = lh.entries[:0]
}
return err
}
// Add an entry to the stored entries and send them if there are more
// than a certain amount
func (lh *ListRHelper) Add(entry fs.DirEntry) error {
if entry == nil {
return nil
}
lh.entries = append(lh.entries, entry)
return lh.send(100)
}
// Flush the stored entries (if any) sending them to the callback
func (lh *ListRHelper) Flush() error {
return lh.send(1)
}