vfs: revise locking in file and dir to fix race conditions

s3-about
Nick Craig-Wood 2019-11-06 21:48:43 +00:00
parent e0d9314059
commit 51efb349ac
2 changed files with 120 additions and 67 deletions

View File

@ -20,14 +20,15 @@ import (
// Dir represents a directory entry
type Dir struct {
vfs *VFS
inode uint64 // inode number
f fs.Fs
parent *Dir // parent, nil for root
vfs *VFS // read only
inode uint64 // read only: inode number
f fs.Fs // read only
mu sync.RWMutex // protects the following
parent *Dir // parent, nil for root
path string
modTime time.Time
entry fs.Directory
mu sync.Mutex // protects the following
read time.Time // time directory entry last read
items map[string]Node // directory entries - can be empty but not nil
}
@ -50,6 +51,8 @@ func (d *Dir) String() string {
if d == nil {
return "<nil *Dir>"
}
d.mu.RLock()
defer d.mu.RUnlock()
return d.path + "/"
}
@ -70,7 +73,9 @@ func (d *Dir) Mode() (mode os.FileMode) {
// Name (base) of the directory - satisfies Node interface
func (d *Dir) Name() (name string) {
d.mu.RLock()
name = path.Base(d.path)
d.mu.RUnlock()
if name == "." {
name = "/"
}
@ -79,6 +84,8 @@ func (d *Dir) Name() (name string) {
// Path of the directory - satisfies Node interface
func (d *Dir) Path() (name string) {
d.mu.RLock()
defer d.mu.RUnlock()
return d.path
}
@ -105,6 +112,7 @@ func (d *Dir) Node() Node {
func (d *Dir) forgetDirPath(relativePath string) {
if dir := d.cachedDir(relativePath); dir != nil {
dir.walk(func(dir *Dir) {
// this is called with the mutex held
fs.Debugf(dir.path, "forgetting directory cache")
dir.read = time.Time{}
dir.items = make(map[string]Node)
@ -139,7 +147,9 @@ func (d *Dir) invalidateDir(absPath string) {
// if entryType is a directory it invalidates the parent of the directory too.
func (d *Dir) changeNotify(relativePath string, entryType fs.EntryType) {
defer log.Trace(d.path, "relativePath=%q, type=%v", relativePath, entryType)("")
d.mu.RLock()
absPath := path.Join(d.path, relativePath)
d.mu.RUnlock()
d.invalidateDir(findParent(absPath))
if entryType == fs.EntryDirectory {
d.invalidateDir(absPath)
@ -154,7 +164,10 @@ func (d *Dir) changeNotify(relativePath string, entryType fs.EntryType) {
// you cannot clear the cache for the Dir's ancestors or siblings.
func (d *Dir) ForgetPath(relativePath string, entryType fs.EntryType) {
defer log.Trace(d.path, "relativePath=%q, type=%v", relativePath, entryType)("")
if absPath := path.Join(d.path, relativePath); absPath != "" {
d.mu.RLock()
absPath := path.Join(d.path, relativePath)
d.mu.RUnlock()
if absPath != "" {
d.invalidateDir(findParent(absPath))
}
if entryType == fs.EntryDirectory {
@ -164,6 +177,8 @@ func (d *Dir) ForgetPath(relativePath string, entryType fs.EntryType) {
// walk runs a function on all cached directories. It will be called
// on a directory's children first.
//
// The mutex will be held for the directory when fun is called
func (d *Dir) walk(fun func(*Dir)) {
d.mu.Lock()
defer d.mu.Unlock()
@ -176,18 +191,11 @@ func (d *Dir) walk(fun func(*Dir)) {
fun(d)
}
// stale returns true if the directory contents will be read the next
// time it is accessed. stale must be called with d.mu held.
func (d *Dir) stale(when time.Time) bool {
_, stale := d.age(when)
return stale
}
// age returns the duration since the last time the directory contents
// was read and the content is cosidered stale. age will be 0 and
// stale true if the last read time is empty.
// age must be called with d.mu held.
func (d *Dir) age(when time.Time) (age time.Duration, stale bool) {
func (d *Dir) _age(when time.Time) (age time.Duration, stale bool) {
if d.read.IsZero() {
return age, true
}
@ -202,11 +210,13 @@ func (d *Dir) age(when time.Time) (age time.Duration, stale bool) {
// reading everything again
func (d *Dir) rename(newParent *Dir, fsDir fs.Directory) {
d.ForgetAll()
d.mu.Lock()
d.parent = newParent
d.entry = fsDir
d.path = fsDir.Remote()
d.modTime = fsDir.ModTime(context.TODO())
d.read = time.Time{}
d.mu.Unlock()
}
// addObject adds a new object or directory to the directory
@ -228,7 +238,7 @@ func (d *Dir) delObject(leaf string) {
// read the directory and sets d.items - must be called with the lock held
func (d *Dir) _readDir() error {
when := time.Now()
if age, stale := d.age(when); stale {
if age, stale := d._age(when); stale {
if age != 0 {
fs.Debugf(d.path, "Re-reading directory (%v old)", age)
}
@ -317,9 +327,9 @@ func (d *Dir) _readDirFromEntries(entries fs.DirEntries, dirTree dirtree.DirTree
// readDirTree forces a refresh of the complete directory tree
func (d *Dir) readDirTree() error {
d.mu.Lock()
d.mu.RLock()
f, path := d.f, d.path
d.mu.Unlock()
d.mu.RUnlock()
when := time.Now()
fs.Debugf(path, "Reading directory tree")
dt, err := walk.NewDirTree(context.TODO(), f, path, false, -1)
@ -394,6 +404,8 @@ func (d *Dir) isEmpty() (bool, error) {
// ModTime returns the modification time of the directory
func (d *Dir) ModTime() time.Time {
d.mu.RLock()
defer d.mu.RUnlock()
// fs.Debugf(d.path, "Dir.ModTime %v", d.modTime)
return d.modTime
}
@ -409,8 +421,8 @@ func (d *Dir) SetModTime(modTime time.Time) error {
return EROFS
}
d.mu.Lock()
defer d.mu.Unlock()
d.modTime = modTime
d.mu.Unlock()
return nil
}

View File

@ -16,11 +16,11 @@ import (
// File represents a file
type File struct {
inode uint64 // inode number
inode uint64 // inode number - read only
size int64 // size of file - read and written with atomic int64 - must be 64 bit aligned
d *Dir // parent directory - read only
mu sync.Mutex // protects the following
mu sync.RWMutex // protects the following
d *Dir // parent directory
o fs.Object // NB o may be nil if file is being written
leaf string // leaf name of the object
rwOpenCount int // number of open files on this handle
@ -66,6 +66,8 @@ func (f *File) IsDir() bool {
// Mode bits of the file or directory - satisfies Node interface
func (f *File) Mode() (mode os.FileMode) {
f.mu.RLock()
defer f.mu.RUnlock()
mode = f.d.vfs.Opt.FilePerms
if f.appendMode {
mode |= os.ModeAppend
@ -75,11 +77,15 @@ func (f *File) Mode() (mode os.FileMode) {
// Name (base) of the directory - satisfies Node interface
func (f *File) Name() (name string) {
f.mu.RLock()
defer f.mu.RUnlock()
return f.leaf
}
// Path returns the full path of the file
func (f *File) Path() string {
f.mu.RLock()
defer f.mu.RUnlock()
return path.Join(f.d.path, f.leaf)
}
@ -101,8 +107,11 @@ func (f *File) Node() Node {
// applyPendingRename runs a previously set rename operation if there are no
// more remaining writers. Call without lock held.
func (f *File) applyPendingRename() {
f.mu.RLock()
fun := f.pendingRenameFun
if fun == nil || f.writingInProgress() {
writing := f._writingInProgress()
f.mu.RUnlock()
if fun == nil || writing {
return
}
fs.Debugf(f.o, "Running delayed rename now")
@ -115,16 +124,19 @@ func (f *File) applyPendingRename() {
// Otherwise it will queue the rename operation on the remote until no writers
// remain.
func (f *File) rename(ctx context.Context, destDir *Dir, newName string) error {
if features := f.d.f.Features(); features.Move == nil && features.Copy == nil {
err := errors.Errorf("Fs %q can't rename files (no server side Move or Copy)", f.d.f)
f.mu.RLock()
d := f.d
f.mu.RUnlock()
if features := d.f.Features(); features.Move == nil && features.Copy == nil {
err := errors.Errorf("Fs %q can't rename files (no server side Move or Copy)", d.f)
fs.Errorf(f.Path(), "Dir.Rename error: %v", err)
return err
}
renameCall := func(ctx context.Context) error {
newPath := path.Join(destDir.path, newName)
dstOverwritten, _ := f.d.f.NewObject(ctx, newPath)
newObject, err := operations.Move(ctx, f.d.f, dstOverwritten, newPath, f.o)
dstOverwritten, _ := d.f.NewObject(ctx, newPath)
newObject, err := operations.Move(ctx, d.f, dstOverwritten, newPath, f.o)
if err != nil {
fs.Errorf(f.Path(), "File.Rename error: %v", err)
return err
@ -155,7 +167,10 @@ func (f *File) rename(ctx context.Context, destDir *Dir, newName string) error {
return nil
}
if f.writingInProgress() {
f.mu.RLock()
writing := f._writingInProgress()
f.mu.RUnlock()
if writing {
fs.Debugf(f.o, "File is currently open, delaying rename %p", f)
f.mu.Lock()
f.d = destDir
@ -182,6 +197,7 @@ func (f *File) addWriter(h Handle) {
// delWriter removes a write handle from the file
func (f *File) delWriter(h Handle, modifiedCacheFile bool) (lastWriterAndModified bool) {
f.mu.Lock()
defer f.applyPendingRename()
defer f.mu.Unlock()
var found = -1
for i := range f.writers {
@ -207,7 +223,6 @@ func (f *File) delWriter(h Handle, modifiedCacheFile bool) (lastWriterAndModifie
if lastWriterAndModified {
f.modified = false
}
defer f.applyPendingRename()
return
}
@ -231,8 +246,8 @@ func (f *File) delRWOpen() {
// Note that file handles which are in pending open state aren't
// counted.
func (f *File) rwOpens() int {
f.mu.Lock()
defer f.mu.Unlock()
f.mu.RLock()
defer f.mu.RUnlock()
return f.rwOpenCount
}
@ -256,8 +271,8 @@ func (f *File) activeWriters() int {
//
// if NoModTime is set then it returns the mod time of the directory
func (f *File) ModTime() (modTime time.Time) {
f.mu.Lock()
defer f.mu.Unlock()
f.mu.RLock()
defer f.mu.RUnlock()
if !f.d.vfs.Opt.NoModTime {
// if o is nil it isn't valid yet or there are writers, so return the size so far
@ -283,11 +298,11 @@ func nonNegative(i int64) int64 {
// Size of the file
func (f *File) Size() int64 {
f.mu.Lock()
defer f.mu.Unlock()
f.mu.RLock()
defer f.mu.RUnlock()
// if o is nil it isn't valid yet or there are writers, so return the size so far
if f.writingInProgress() {
if f._writingInProgress() {
return atomic.LoadInt64(&f.size)
}
return nonNegative(f.o.Size())
@ -304,22 +319,23 @@ func (f *File) SetModTime(modTime time.Time) error {
f.pendingModTime = modTime
// Only update the ModTime when there are no writers, setObject will do it
if !f.writingInProgress() {
return f.applyPendingModTime()
if !f._writingInProgress() {
return f._applyPendingModTime()
}
// queue up for later, hoping f.o becomes available
return nil
}
// call with the mutex held
func (f *File) applyPendingModTime() error {
defer func() { f.pendingModTime = time.Time{} }()
// Apply a pending mod time
// Call with the write mutex held
func (f *File) _applyPendingModTime() error {
if f.pendingModTime.IsZero() {
return nil
}
defer func() { f.pendingModTime = time.Time{} }()
if f.o == nil {
return errors.New("Cannot apply ModTime, file object is not available")
}
@ -327,19 +343,20 @@ func (f *File) applyPendingModTime() error {
err := f.o.SetModTime(context.TODO(), f.pendingModTime)
switch err {
case nil:
fs.Debugf(f.o, "File.applyPendingModTime OK")
fs.Debugf(f.o, "File._applyPendingModTime OK")
case fs.ErrorCantSetModTime, fs.ErrorCantSetModTimeWithoutDelete:
// do nothing, in order to not break "touch somefile" if it exists already
default:
fs.Errorf(f, "File.applyPendingModTime error: %v", err)
fs.Errorf(f, "File._applyPendingModTime error: %v", err)
return err
}
return nil
}
// writingInProgress returns true of there are any open writers
func (f *File) writingInProgress() bool {
// _writingInProgress returns true of there are any open writers
// Call with read lock held
func (f *File) _writingInProgress() bool {
return f.o == nil || len(f.writers) != 0 || f.readWriterClosing
}
@ -352,7 +369,7 @@ func (f *File) setSize(n int64) {
func (f *File) setObject(o fs.Object) {
f.mu.Lock()
f.o = o
_ = f.applyPendingModTime()
_ = f._applyPendingModTime()
f.mu.Unlock()
f.d.addObject(f)
@ -362,21 +379,21 @@ func (f *File) setObject(o fs.Object) {
// the directory cache
func (f *File) setObjectNoUpdate(o fs.Object) {
f.mu.Lock()
defer f.mu.Unlock()
f.o = o
f.mu.Unlock()
}
// Get the current fs.Object - may be nil
func (f *File) getObject() fs.Object {
f.mu.Lock()
defer f.mu.Unlock()
f.mu.RLock()
defer f.mu.RUnlock()
return f.o
}
// exists returns whether the file exists already
func (f *File) exists() bool {
f.mu.Lock()
defer f.mu.Unlock()
f.mu.RLock()
defer f.mu.RUnlock()
return f.o != nil
}
@ -386,11 +403,11 @@ func (f *File) exists() bool {
// Call without the mutex held
func (f *File) waitForValidObject() (o fs.Object, err error) {
for i := 0; i < 50; i++ {
f.mu.Lock()
f.mu.RLock()
o = f.o
nwriters := len(f.writers)
wclosing := f.readWriterClosing
f.mu.Unlock()
f.mu.RUnlock()
if o != nil {
return o, nil
}
@ -421,12 +438,16 @@ func (f *File) openRead() (fh *ReadFileHandle, err error) {
// openWrite open the file for write
func (f *File) openWrite(flags int) (fh *WriteFileHandle, err error) {
if f.d.vfs.Opt.ReadOnly {
f.mu.RLock()
d := f.d
f.mu.RUnlock()
if d.vfs.Opt.ReadOnly {
return nil, EROFS
}
// fs.Debugf(o, "File.openWrite")
fh, err = newWriteFileHandle(f.d, f, f.Path(), flags)
fh, err = newWriteFileHandle(d, f, f.Path(), flags)
if err != nil {
fs.Errorf(f, "File.openWrite failed: %v", err)
return nil, err
@ -438,13 +459,17 @@ func (f *File) openWrite(flags int) (fh *WriteFileHandle, err error) {
//
// It uses the open flags passed in.
func (f *File) openRW(flags int) (fh *RWFileHandle, err error) {
f.mu.RLock()
d := f.d
f.mu.RUnlock()
// FIXME chunked
if flags&accessModeMask != os.O_RDONLY && f.d.vfs.Opt.ReadOnly {
if flags&accessModeMask != os.O_RDONLY && d.vfs.Opt.ReadOnly {
return nil, EROFS
}
// fs.Debugf(o, "File.openRW")
fh, err = newRWFileHandle(f.d, f, f.Path(), flags)
fh, err = newRWFileHandle(d, f, f.Path(), flags)
if err != nil {
fs.Errorf(f, "File.openRW failed: %v", err)
return nil, err
@ -461,7 +486,11 @@ func (f *File) Sync() error {
// Remove the file
func (f *File) Remove() error {
if f.d.vfs.Opt.ReadOnly {
f.mu.RLock()
d := f.d
f.mu.RUnlock()
if d.vfs.Opt.ReadOnly {
return EROFS
}
f.muRW.Lock() // muRW must be locked before mu to avoid
@ -479,10 +508,10 @@ func (f *File) Remove() error {
f.muRW.Unlock()
// Remove the item from the directory listing
f.d.delObject(f.Name())
d.delObject(f.Name())
// Remove the object from the cache
if f.d.vfs.Opt.CacheMode >= CacheModeMinimal {
f.d.vfs.cache.remove(f.Path())
if d.vfs.Opt.CacheMode >= CacheModeMinimal {
d.vfs.cache.remove(f.Path())
}
return nil
}
@ -494,16 +523,22 @@ func (f *File) RemoveAll() error {
// DirEntry returns the underlying fs.DirEntry - may be nil
func (f *File) DirEntry() (entry fs.DirEntry) {
f.mu.RLock()
defer f.mu.RUnlock()
return f.o
}
// Dir returns the directory this file is in
func (f *File) Dir() *Dir {
f.mu.RLock()
defer f.mu.RUnlock()
return f.d
}
// VFS returns the instance of the VFS
func (f *File) VFS() *VFS {
f.mu.RLock()
defer f.mu.RUnlock()
return f.d.vfs
}
@ -552,7 +587,9 @@ func (f *File) Open(flags int) (fd Handle, err error) {
// If append is set then set read to force openRW
if flags&os.O_APPEND != 0 {
read = true
f.mu.Lock()
f.appendMode = true
f.mu.Unlock()
}
// If truncate is set then set write to force openRW
@ -561,8 +598,11 @@ func (f *File) Open(flags int) (fd Handle, err error) {
}
// Open the correct sort of handle
CacheMode := f.d.vfs.Opt.CacheMode
if CacheMode >= CacheModeMinimal && (f.d.vfs.cache.opens(f.Path()) > 0 || f.d.vfs.cache.exists(f.Path())) {
f.mu.RLock()
d := f.d
f.mu.RUnlock()
CacheMode := d.vfs.Opt.CacheMode
if CacheMode >= CacheModeMinimal && (d.vfs.cache.opens(f.Path()) > 0 || d.vfs.cache.exists(f.Path())) {
fd, err = f.openRW(flags)
} else if read && write {
if CacheMode >= CacheModeMinimal {
@ -591,7 +631,7 @@ func (f *File) Open(flags int) (fd Handle, err error) {
}
// if creating a file, add the file to the directory
if err == nil && flags&os.O_CREATE != 0 {
f.d.addObject(f)
d.addObject(f)
}
return fd, err
}
@ -603,13 +643,14 @@ func (f *File) Truncate(size int64) (err error) {
f.mu.Lock()
writers := make([]Handle, len(f.writers))
copy(writers, f.writers)
o := f.o
f.mu.Unlock()
// FIXME: handle closing writer
// If have writers then call truncate for each writer
if len(writers) != 0 {
fs.Debugf(f.o, "Truncating %d file handles", len(writers))
fs.Debugf(f, "Truncating %d file handles", len(writers))
for _, h := range writers {
truncateErr := h.Truncate(size)
if truncateErr != nil {
@ -620,11 +661,11 @@ func (f *File) Truncate(size int64) (err error) {
}
// If no writers, and size is already correct then all done
if f.o.Size() == size {
if o.Size() == size {
return nil
}
fs.Debugf(f.o, "Truncating file")
fs.Debugf(f, "Truncating file")
// Otherwise if no writers then truncate the file by opening
// the file and truncating it.