seaweedfs/weed/mount/weedfs_dir_read.go
Chris Lu 952afd810c
Fix dead lock (#5815)
* reduce locks to avoid dead lock

Flush->FlushData->uplloadPipeline.FluahAll
uploaderCount>0

goroutine 1 [sync.Cond.Wait, 71 minutes]:
sync.runtime_notifyListWait(0xc0007ae4d0, 0x0)
	/usr/local/go/src/runtime/sema.go:569 +0x159
sync.(*Cond).Wait(0xc001a59290?)
	/usr/local/go/src/sync/cond.go:70 +0x85
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*UploadPipeline).waitForCurrentWritersToComplete(0xc0002ee4d0)
	/github/workspace/weed/mount/page_writer/upload_pipeline_lock.go:58 +0x32
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*UploadPipeline).FlushAll(0xc0002ee4d0)
	/github/workspace/weed/mount/page_writer/upload_pipeline.go:151 +0x25
github.com/seaweedfs/seaweedfs/weed/mount.(*ChunkedDirtyPages).FlushData(0xc00087e840)
	/github/workspace/weed/mount/dirty_pages_chunked.go:54 +0x29
github.com/seaweedfs/seaweedfs/weed/mount.(*PageWriter).FlushData(...)
	/github/workspace/weed/mount/page_writer.go:50
github.com/seaweedfs/seaweedfs/weed/mount.(*WFS).doFlush(0xc0006ad600, 0xc00030d380, 0x0, 0x0)
	/github/workspace/weed/mount/weedfs_file_sync.go:101 +0x169
github.com/seaweedfs/seaweedfs/weed/mount.(*WFS).Flush(0xc0006ad600, 0xc001a594a8?, 0xc0004c1ca0)
	/github/workspace/weed/mount/weedfs_file_sync.go:59 +0x48
github.com/hanwen/go-fuse/v2/fuse.doFlush(0xc0000da870?, 0xc0004c1b08)

SaveContent -> MemChunk.RLock ->
	ChunkedDirtyPages.saveChunkedFileIntervalToStorage
	pages.fh.AddChunks([]*filer_pb.FileChunk{chunk})
		fh.entryLock.Lock()

sync.(*RWMutex).Lock(0x0?)
	/usr/local/go/src/sync/rwmutex.go:146 +0x31
github.com/seaweedfs/seaweedfs/weed/mount.(*FileHandle).AddChunks(0xc00030d380, {0xc00028bdc8, 0x1, 0x1})
	/github/workspace/weed/mount/filehandle.go:93 +0x45
github.com/seaweedfs/seaweedfs/weed/mount.(*ChunkedDirtyPages).saveChunkedFileIntervalToStorage(0xc00087e840, {0x2be7ac0, 0xc00018d9e0}, 0x0, 0x121, 0x17e3c624565ace45, 0x1?)
	/github/workspace/weed/mount/dirty_pages_chunked.go:80 +0x2d4
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*MemChunk).SaveContent(0xc0008d9130, 0xc0008093e0)
	/github/workspace/weed/mount/page_writer/page_chunk_mem.go:115 +0x112
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*UploadPipeline).moveToSealed.func1()
	/github/workspace/weed/mount/page_writer/upload_pipeline.go:187 +0x55
github.com/seaweedfs/seaweedfs/weed/util.(*LimitedConcurrentExecutor).Execute.func1()
	/github/workspace/weed/util/limited_executor.go:38 +0x62
created by github.com/seaweedfs/seaweedfs/weed/util.(*LimitedConcurrentExecutor).Execute in goroutine 1
	/github/workspace/weed/util/limited_executor.go:33 +0x97

On metadata update
fh.entryLock.Lock()
	fh.dirtyPages.Destroy()
		up.chunksLock.Lock => each sealed chunk.FreeReference => MemChunk.Lock

goroutine 134 [sync.RWMutex.Lock, 71 minutes]:
sync.runtime_SemacquireRWMutex(0xc0007c3558?, 0xea?, 0x3fb0800?)
	/usr/local/go/src/runtime/sema.go:87 +0x25
sync.(*RWMutex).Lock(0xc0007c35a8?)
	/usr/local/go/src/sync/rwmutex.go:151 +0x6a
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*MemChunk).FreeResource(0xc0008d9130)
	/github/workspace/weed/mount/page_writer/page_chunk_mem.go:38 +0x2a
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*SealedChunk).FreeReference(0xc00071cdb0, {0xc0006ba1a0, 0x20})
	/github/workspace/weed/mount/page_writer/upload_pipeline.go:38 +0xb7
github.com/seaweedfs/seaweedfs/weed/mount/page_writer.(*UploadPipeline).Shutdown(0xc0002ee4d0)
	/github/workspace/weed/mount/page_writer/upload_pipeline.go:220 +0x185
github.com/seaweedfs/seaweedfs/weed/mount.(*ChunkedDirtyPages).Destroy(0xc0008cea40?)
	/github/workspace/weed/mount/dirty_pages_chunked.go:87 +0x17
github.com/seaweedfs/seaweedfs/weed/mount.(*PageWriter).Destroy(...)
	/github/workspace/weed/mount/page_writer.go:78
github.com/seaweedfs/seaweedfs/weed/mount.NewSeaweedFileSystem.func3({0xc00069a6c0, 0x30}, 0x6?)
	/github/workspace/weed/mount/weedfs.go:119 +0x17a
github.com/seaweedfs/seaweedfs/weed/mount/meta_cache.NewMetaCache.func1({0xc00069a6c0?, 0xc00069a480?}, 0x4015b40?)
	/github/workspace/weed/mount/meta_cache/meta_cache.go:37 +0x1c
github.com/seaweedfs/seaweedfs/weed/mount/meta_cache.SubscribeMetaEvents.func1(0xc000661810)
	/github/workspace/weed/mount/meta_cache/meta_cache_subscribe.go:43 +0x570

* use locked entry everywhere

* modifiable remote entry

* skip locking after getting lock from fhLockTable
2024-07-24 23:46:40 -07:00

241 lines
6.8 KiB
Go

package mount
import (
"context"
"github.com/hanwen/go-fuse/v2/fuse"
"github.com/seaweedfs/seaweedfs/weed/filer"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/mount/meta_cache"
"math"
"sync"
)
type DirectoryHandleId uint64
const (
directoryStreamBaseOffset = 2 // . & ..
)
type DirectoryHandle struct {
isFinished bool
entryStream []*filer.Entry
entryStreamOffset uint64
}
func (dh *DirectoryHandle) reset() {
*dh = DirectoryHandle{
isFinished: false,
entryStream: []*filer.Entry{},
entryStreamOffset: directoryStreamBaseOffset,
}
}
type DirectoryHandleToInode struct {
// shares the file handle id sequencer with FileHandleToInode{nextFh}
sync.Mutex
dir2inode map[DirectoryHandleId]*DirectoryHandle
}
func NewDirectoryHandleToInode() *DirectoryHandleToInode {
return &DirectoryHandleToInode{
dir2inode: make(map[DirectoryHandleId]*DirectoryHandle),
}
}
func (wfs *WFS) AcquireDirectoryHandle() (DirectoryHandleId, *DirectoryHandle) {
wfs.fhmap.Lock()
fh := wfs.fhmap.nextFh
wfs.fhmap.nextFh++
wfs.fhmap.Unlock()
wfs.dhmap.Lock()
defer wfs.dhmap.Unlock()
dh := new(DirectoryHandle)
dh.reset()
wfs.dhmap.dir2inode[DirectoryHandleId(fh)] = dh
return DirectoryHandleId(fh), dh
}
func (wfs *WFS) GetDirectoryHandle(dhid DirectoryHandleId) *DirectoryHandle {
wfs.dhmap.Lock()
defer wfs.dhmap.Unlock()
if dh, found := wfs.dhmap.dir2inode[dhid]; found {
return dh
}
dh := new(DirectoryHandle)
dh.reset()
wfs.dhmap.dir2inode[dhid] = dh
return dh
}
func (wfs *WFS) ReleaseDirectoryHandle(dhid DirectoryHandleId) {
wfs.dhmap.Lock()
defer wfs.dhmap.Unlock()
delete(wfs.dhmap.dir2inode, dhid)
}
// Directory handling
/** Open directory
*
* Unless the 'default_permissions' mount option is given,
* this method should check if opendir is permitted for this
* directory. Optionally opendir may also return an arbitrary
* filehandle in the fuse_file_info structure, which will be
* passed to readdir, releasedir and fsyncdir.
*/
func (wfs *WFS) OpenDir(cancel <-chan struct{}, input *fuse.OpenIn, out *fuse.OpenOut) (code fuse.Status) {
if !wfs.inodeToPath.HasInode(input.NodeId) {
return fuse.ENOENT
}
dhid, _ := wfs.AcquireDirectoryHandle()
out.Fh = uint64(dhid)
return fuse.OK
}
/** Release directory
*
* If the directory has been removed after the call to opendir, the
* path parameter will be NULL.
*/
func (wfs *WFS) ReleaseDir(input *fuse.ReleaseIn) {
wfs.ReleaseDirectoryHandle(DirectoryHandleId(input.Fh))
}
/** Synchronize directory contents
*
* If the directory has been removed after the call to opendir, the
* path parameter will be NULL.
*
* If the datasync parameter is non-zero, then only the user data
* should be flushed, not the meta data
*/
func (wfs *WFS) FsyncDir(cancel <-chan struct{}, input *fuse.FsyncIn) (code fuse.Status) {
return fuse.OK
}
/** Read directory
*
* The filesystem may choose between two modes of operation:
*
* 1) The readdir implementation ignores the offset parameter, and
* passes zero to the filler function's offset. The filler
* function will not return '1' (unless an error happens), so the
* whole directory is read in a single readdir operation.
*
* 2) The readdir implementation keeps track of the offsets of the
* directory entries. It uses the offset parameter and always
* passes non-zero offset to the filler function. When the buffer
* is full (or an error happens) the filler function will return
* '1'.
*/
func (wfs *WFS) ReadDir(cancel <-chan struct{}, input *fuse.ReadIn, out *fuse.DirEntryList) (code fuse.Status) {
return wfs.doReadDirectory(input, out, false)
}
func (wfs *WFS) ReadDirPlus(cancel <-chan struct{}, input *fuse.ReadIn, out *fuse.DirEntryList) (code fuse.Status) {
return wfs.doReadDirectory(input, out, true)
}
func (wfs *WFS) doReadDirectory(input *fuse.ReadIn, out *fuse.DirEntryList, isPlusMode bool) fuse.Status {
dh := wfs.GetDirectoryHandle(DirectoryHandleId(input.Fh))
if input.Offset == 0 {
dh.reset()
} else if dh.isFinished && input.Offset >= dh.entryStreamOffset {
entryCurrentIndex := input.Offset - dh.entryStreamOffset
if uint64(len(dh.entryStream)) <= entryCurrentIndex {
return fuse.OK
}
}
isEarlyTerminated := false
dirPath, code := wfs.inodeToPath.GetPath(input.NodeId)
if code != fuse.OK {
return code
}
var dirEntry fuse.DirEntry
processEachEntryFn := func(entry *filer.Entry) bool {
dirEntry.Name = entry.Name()
dirEntry.Mode = toSyscallMode(entry.Mode)
inode := wfs.inodeToPath.Lookup(dirPath.Child(dirEntry.Name), entry.Crtime.Unix(), entry.IsDirectory(), len(entry.HardLinkId) > 0, entry.Inode, isPlusMode)
dirEntry.Ino = inode
if !isPlusMode {
if !out.AddDirEntry(dirEntry) {
isEarlyTerminated = true
return false
}
} else {
entryOut := out.AddDirLookupEntry(dirEntry)
if entryOut == nil {
isEarlyTerminated = true
return false
}
if fh, found := wfs.fhmap.FindFileHandle(inode); found {
glog.V(4).Infof("readdir opened file %s", dirPath.Child(dirEntry.Name))
entry = filer.FromPbEntry(string(dirPath), fh.GetEntry().GetEntry())
}
wfs.outputFilerEntry(entryOut, inode, entry)
}
return true
}
if input.Offset < directoryStreamBaseOffset {
if !isPlusMode {
if input.Offset == 0 {
out.AddDirEntry(fuse.DirEntry{Mode: fuse.S_IFDIR, Name: "."})
}
out.AddDirEntry(fuse.DirEntry{Mode: fuse.S_IFDIR, Name: ".."})
} else {
if input.Offset == 0 {
out.AddDirLookupEntry(fuse.DirEntry{Mode: fuse.S_IFDIR, Name: "."})
}
out.AddDirLookupEntry(fuse.DirEntry{Mode: fuse.S_IFDIR, Name: ".."})
}
input.Offset = directoryStreamBaseOffset
}
var lastEntryName string
if input.Offset >= dh.entryStreamOffset {
if input.Offset > dh.entryStreamOffset {
entryPreviousIndex := (input.Offset - dh.entryStreamOffset) - 1
if uint64(len(dh.entryStream)) > entryPreviousIndex {
lastEntryName = dh.entryStream[entryPreviousIndex].Name()
dh.entryStream = dh.entryStream[entryPreviousIndex:]
dh.entryStreamOffset = input.Offset - 1
}
}
entryCurrentIndex := input.Offset - dh.entryStreamOffset
for uint64(len(dh.entryStream)) > entryCurrentIndex {
entry := dh.entryStream[entryCurrentIndex]
if processEachEntryFn(entry) {
lastEntryName = entry.Name()
entryCurrentIndex++
} else {
// early terminated
return fuse.OK
}
}
}
var err error
if err = meta_cache.EnsureVisited(wfs.metaCache, wfs, dirPath); err != nil {
glog.Errorf("dir ReadDirAll %s: %v", dirPath, err)
return fuse.EIO
}
listErr := wfs.metaCache.ListDirectoryEntries(context.Background(), dirPath, lastEntryName, false, int64(math.MaxInt32), func(entry *filer.Entry) bool {
dh.entryStream = append(dh.entryStream, entry)
return processEachEntryFn(entry)
})
if listErr != nil {
glog.Errorf("list meta cache: %v", listErr)
return fuse.EIO
}
if !isEarlyTerminated {
dh.isFinished = true
}
return fuse.OK
}