From 48dbefc37e1c5b16ee6cd97fed272a8372b9cbdd Mon Sep 17 00:00:00 2001 From: Michael Eischer Date: Sat, 2 Nov 2024 20:27:38 +0100 Subject: [PATCH] fs / archiver: convert to handle based interface The actual implementation still relies on file paths, but with the abstraction layer in place, an FS implementation can ensure atomic file accesses in the future. --- internal/archiver/archiver.go | 86 ++++++---- internal/archiver/archiver_test.go | 207 ++++++++++++------------ internal/archiver/archiver_unix_test.go | 8 +- internal/archiver/exclude.go | 8 +- internal/archiver/file_saver.go | 13 +- internal/archiver/file_saver_test.go | 15 +- internal/fs/file.go | 2 +- internal/fs/fs_local.go | 110 +++++++++++-- internal/fs/fs_local_vss.go | 11 +- internal/fs/fs_local_vss_test.go | 2 +- internal/fs/fs_reader.go | 33 ++-- internal/fs/fs_reader_test.go | 8 +- internal/fs/fs_track.go | 6 +- internal/fs/interface.go | 25 ++- internal/fs/node_test.go | 62 ++----- internal/fs/node_unix_test.go | 16 +- internal/fs/node_windows_test.go | 8 +- internal/restic/tree_test.go | 20 +-- 18 files changed, 356 insertions(+), 284 deletions(-) diff --git a/internal/archiver/archiver.go b/internal/archiver/archiver.go index e88c15bd9..ae3edaf44 100644 --- a/internal/archiver/archiver.go +++ b/internal/archiver/archiver.go @@ -66,6 +66,11 @@ func (s *ItemStats) Add(other ItemStats) { s.TreeSizeInRepo += other.TreeSizeInRepo } +// ToNoder returns a restic.Node for a File. +type ToNoder interface { + ToNode(ignoreXattrListError bool) (*restic.Node, error) +} + type archiverRepo interface { restic.Loader restic.BlobSaver @@ -257,8 +262,8 @@ func (arch *Archiver) trackItem(item string, previous, current *restic.Node, s I } // nodeFromFileInfo returns the restic node from an os.FileInfo. -func (arch *Archiver) nodeFromFileInfo(snPath, filename string, fi os.FileInfo, ignoreXattrListError bool) (*restic.Node, error) { - node, err := arch.FS.NodeFromFileInfo(filename, fi, ignoreXattrListError) +func (arch *Archiver) nodeFromFileInfo(snPath, filename string, meta ToNoder, ignoreXattrListError bool) (*restic.Node, error) { + node, err := meta.ToNode(ignoreXattrListError) if !arch.WithAtime { node.AccessTime = node.ModTime } @@ -308,20 +313,14 @@ func (arch *Archiver) wrapLoadTreeError(id restic.ID, err error) error { // saveDir stores a directory in the repo and returns the node. snPath is the // path within the current snapshot. -func (arch *Archiver) saveDir(ctx context.Context, snPath string, dir string, fi os.FileInfo, previous *restic.Tree, complete fileCompleteFunc) (d futureNode, err error) { +func (arch *Archiver) saveDir(ctx context.Context, snPath string, dir string, meta fs.File, previous *restic.Tree, complete fileCompleteFunc) (d futureNode, err error) { debug.Log("%v %v", snPath, dir) - treeNode, err := arch.nodeFromFileInfo(snPath, dir, fi, false) + treeNode, names, err := arch.dirToNodeAndEntries(snPath, dir, meta) if err != nil { return futureNode{}, err } - names, err := fs.Readdirnames(arch.FS, dir, fs.O_NOFOLLOW) - if err != nil { - return futureNode{}, err - } - sort.Strings(names) - nodes := make([]futureNode, 0, len(names)) for _, name := range names { @@ -359,6 +358,29 @@ func (arch *Archiver) saveDir(ctx context.Context, snPath string, dir string, fi return fn, nil } +func (arch *Archiver) dirToNodeAndEntries(snPath, dir string, meta fs.File) (node *restic.Node, names []string, err error) { + err = meta.MakeReadable() + if err != nil { + return nil, nil, fmt.Errorf("openfile for readdirnames failed: %w", err) + } + + node, err = arch.nodeFromFileInfo(snPath, dir, meta, false) + if err != nil { + return nil, nil, err + } + if node.Type != restic.NodeTypeDir { + return nil, nil, fmt.Errorf("directory %v changed type, refusing to archive", snPath) + } + + names, err = meta.Readdirnames(-1) + if err != nil { + return nil, nil, fmt.Errorf("readdirnames %v failed: %w", dir, err) + } + sort.Strings(names) + + return node, names, nil +} + // futureNode holds a reference to a channel that returns a FutureNodeResult // or a reference to an already existing result. If the result is available // immediately, then storing a reference directly requires less memory than @@ -448,8 +470,23 @@ func (arch *Archiver) save(ctx context.Context, snPath, target string, previous return futureNode{}, true, nil } + meta, err := arch.FS.OpenFile(target, fs.O_NOFOLLOW, true) + if err != nil { + debug.Log("open metadata for %v returned error: %v", target, err) + return filterError(err) + } + closeFile := true + defer func() { + if closeFile { + cerr := meta.Close() + if err == nil { + err = cerr + } + } + }() + // get file info and run remaining select functions that require file information - fi, err := arch.FS.Lstat(target) + fi, err := meta.Stat() if err != nil { debug.Log("lstat() for %v returned error: %v", target, err) return filterError(err) @@ -470,7 +507,7 @@ func (arch *Archiver) save(ctx context.Context, snPath, target string, previous debug.Log("%v hasn't changed, using old list of blobs", target) arch.trackItem(snPath, previous, previous, ItemStats{}, time.Since(start)) arch.CompleteBlob(previous.Size) - node, err := arch.nodeFromFileInfo(snPath, target, fi, false) + node, err := arch.nodeFromFileInfo(snPath, target, meta, false) if err != nil { return futureNode{}, false, err } @@ -497,28 +534,28 @@ func (arch *Archiver) save(ctx context.Context, snPath, target string, previous // reopen file and do an fstat() on the open file to check it is still // a file (and has not been exchanged for e.g. a symlink) - file, err := arch.FS.OpenFile(target, fs.O_RDONLY|fs.O_NOFOLLOW) + err := meta.MakeReadable() if err != nil { - debug.Log("Openfile() for %v returned error: %v", target, err) + debug.Log("MakeReadable() for %v returned error: %v", target, err) return filterError(err) } - fi, err = file.Stat() + fi, err := meta.Stat() if err != nil { debug.Log("stat() on opened file %v returned error: %v", target, err) - _ = file.Close() return filterError(err) } // make sure it's still a file if !fi.Mode().IsRegular() { err = errors.Errorf("file %v changed type, refusing to archive", target) - _ = file.Close() return filterError(err) } + closeFile = false + // Save will close the file, we don't need to do that - fn = arch.fileSaver.Save(ctx, snPath, target, file, fi, func() { + fn = arch.fileSaver.Save(ctx, snPath, target, meta, func() { arch.StartFile(snPath) }, func() { arch.trackItem(snPath, nil, nil, ItemStats{}, 0) @@ -538,7 +575,7 @@ func (arch *Archiver) save(ctx context.Context, snPath, target string, previous return futureNode{}, false, err } - fn, err = arch.saveDir(ctx, snPath, target, fi, oldSubtree, + fn, err = arch.saveDir(ctx, snPath, target, meta, oldSubtree, func(node *restic.Node, stats ItemStats) { arch.trackItem(snItem, previous, node, stats, time.Since(start)) }) @@ -554,7 +591,7 @@ func (arch *Archiver) save(ctx context.Context, snPath, target string, previous default: debug.Log(" %v other", target) - node, err := arch.nodeFromFileInfo(snPath, target, fi, false) + node, err := arch.nodeFromFileInfo(snPath, target, meta, false) if err != nil { return futureNode{}, false, err } @@ -688,7 +725,7 @@ func (arch *Archiver) saveTree(ctx context.Context, snPath string, atree *tree, } func (arch *Archiver) dirPathToNode(snPath, target string) (node *restic.Node, err error) { - meta, err := arch.FS.OpenFile(target, fs.O_RDONLY) + meta, err := arch.FS.OpenFile(target, 0, true) if err != nil { return nil, err } @@ -700,14 +737,9 @@ func (arch *Archiver) dirPathToNode(snPath, target string) (node *restic.Node, e }() debug.Log("%v, reading dir node data from %v", snPath, target) - fi, err := meta.Stat() - if err != nil { - return nil, errors.WithStack(err) - } - // in some cases reading xattrs for directories above the backup source is not allowed // thus ignore errors for such folders. - node, err = arch.nodeFromFileInfo(snPath, target, fi, true) + node, err = arch.nodeFromFileInfo(snPath, target, meta, true) if err != nil { return nil, err } diff --git a/internal/archiver/archiver_test.go b/internal/archiver/archiver_test.go index ff072fa36..0b2957bc6 100644 --- a/internal/archiver/archiver_test.go +++ b/internal/archiver/archiver_test.go @@ -76,17 +76,12 @@ func saveFile(t testing.TB, repo archiverRepo, filename string, filesystem fs.FS startCallback = true } - file, err := arch.FS.OpenFile(filename, fs.O_RDONLY|fs.O_NOFOLLOW) + file, err := arch.FS.OpenFile(filename, fs.O_NOFOLLOW, false) if err != nil { t.Fatal(err) } - fi, err := file.Stat() - if err != nil { - t.Fatal(err) - } - - res := arch.fileSaver.Save(ctx, "/", filename, file, fi, start, completeReading, complete) + res := arch.fileSaver.Save(ctx, "/", filename, file, start, completeReading, complete) fnr := res.take(ctx) if fnr.err != nil { @@ -556,11 +551,12 @@ func rename(t testing.TB, oldname, newname string) { } } -func nodeFromFI(t testing.TB, fs fs.FS, filename string, fi os.FileInfo) *restic.Node { - node, err := fs.NodeFromFileInfo(filename, fi, false) - if err != nil { - t.Fatal(err) - } +func nodeFromFile(t testing.TB, localFs fs.FS, filename string) *restic.Node { + meta, err := localFs.OpenFile(filename, fs.O_NOFOLLOW, true) + rtest.OK(t, err) + node, err := meta.ToNode(false) + rtest.OK(t, err) + rtest.OK(t, meta.Close()) return node } @@ -688,7 +684,7 @@ func TestFileChanged(t *testing.T) { fs := &fs.Local{} fiBefore := lstat(t, filename) - node := nodeFromFI(t, fs, filename, fiBefore) + node := nodeFromFile(t, fs, filename) if fileChanged(fs, fiBefore, node, 0) { t.Fatalf("unchanged file detected as changed") @@ -729,8 +725,8 @@ func TestFilChangedSpecialCases(t *testing.T) { t.Run("type-change", func(t *testing.T) { fi := lstat(t, filename) - node := nodeFromFI(t, &fs.Local{}, filename, fi) - node.Type = "restic.NodeTypeSymlink" + node := nodeFromFile(t, &fs.Local{}, filename) + node.Type = restic.NodeTypeSymlink if !fileChanged(&fs.Local{}, fi, node, 0) { t.Fatal("node with changed type detected as unchanged") } @@ -834,7 +830,8 @@ func TestArchiverSaveDir(t *testing.T) { wg, ctx := errgroup.WithContext(context.Background()) repo.StartPackUploader(ctx, wg) - arch := New(repo, fs.Track{FS: fs.Local{}}, Options{}) + testFS := fs.Track{FS: fs.Local{}} + arch := New(repo, testFS, Options{}) arch.runWorkers(ctx, wg) arch.summary = &Summary{} @@ -846,15 +843,11 @@ func TestArchiverSaveDir(t *testing.T) { back := rtest.Chdir(t, chdir) defer back() - fi, err := os.Lstat(test.target) - if err != nil { - t.Fatal(err) - } - - ft, err := arch.saveDir(ctx, "/", test.target, fi, nil, nil) - if err != nil { - t.Fatal(err) - } + meta, err := testFS.OpenFile(test.target, fs.O_NOFOLLOW, true) + rtest.OK(t, err) + ft, err := arch.saveDir(ctx, "/", test.target, meta, nil, nil) + rtest.OK(t, err) + rtest.OK(t, meta.Close()) fnr := ft.take(ctx) node, stats := fnr.node, fnr.stats @@ -916,19 +909,16 @@ func TestArchiverSaveDirIncremental(t *testing.T) { wg, ctx := errgroup.WithContext(context.TODO()) repo.StartPackUploader(ctx, wg) - arch := New(repo, fs.Track{FS: fs.Local{}}, Options{}) + testFS := fs.Track{FS: fs.Local{}} + arch := New(repo, testFS, Options{}) arch.runWorkers(ctx, wg) arch.summary = &Summary{} - fi, err := os.Lstat(tempdir) - if err != nil { - t.Fatal(err) - } - - ft, err := arch.saveDir(ctx, "/", tempdir, fi, nil, nil) - if err != nil { - t.Fatal(err) - } + meta, err := testFS.OpenFile(tempdir, fs.O_NOFOLLOW, true) + rtest.OK(t, err) + ft, err := arch.saveDir(ctx, "/", tempdir, meta, nil, nil) + rtest.OK(t, err) + rtest.OK(t, meta.Close()) fnr := ft.take(ctx) node, stats := fnr.node, fnr.stats @@ -1665,8 +1655,8 @@ type MockFS struct { bytesRead map[string]int // tracks bytes read from all opened files } -func (m *MockFS) OpenFile(name string, flag int) (fs.File, error) { - f, err := m.FS.OpenFile(name, flag) +func (m *MockFS) OpenFile(name string, flag int, metadataOnly bool) (fs.File, error) { + f, err := m.FS.OpenFile(name, flag, metadataOnly) if err != nil { return f, err } @@ -2056,12 +2046,12 @@ type TrackFS struct { m sync.Mutex } -func (m *TrackFS) OpenFile(name string, flag int) (fs.File, error) { +func (m *TrackFS) OpenFile(name string, flag int, metadataOnly bool) (fs.File, error) { m.m.Lock() m.opened[name]++ m.m.Unlock() - return m.FS.OpenFile(name, flag) + return m.FS.OpenFile(name, flag, metadataOnly) } type failSaveRepo struct { @@ -2210,48 +2200,39 @@ func snapshot(t testing.TB, repo archiverRepo, fs fs.FS, parent *restic.Snapshot return snapshot, node } -// StatFS allows overwriting what is returned by the Lstat function. -type StatFS struct { +type overrideFS struct { fs.FS - - OverrideLstat map[string]os.FileInfo - OnlyOverrideStat bool + overrideFI os.FileInfo + overrideNode *restic.Node + overrideErr error } -func (fs *StatFS) Lstat(name string) (os.FileInfo, error) { - if !fs.OnlyOverrideStat { - if fi, ok := fs.OverrideLstat[fixpath(name)]; ok { - return fi, nil - } +func (m *overrideFS) OpenFile(name string, flag int, metadataOnly bool) (fs.File, error) { + f, err := m.FS.OpenFile(name, flag, metadataOnly) + if err != nil { + return f, err } - return fs.FS.Lstat(name) -} - -func (fs *StatFS) OpenFile(name string, flags int) (fs.File, error) { - if fi, ok := fs.OverrideLstat[fixpath(name)]; ok { - f, err := fs.FS.OpenFile(name, flags) - if err != nil { - return nil, err - } - - wrappedFile := fileStat{ - File: f, - fi: fi, - } - return wrappedFile, nil + if filepath.Base(name) == "testfile" { + return &overrideFile{f, m}, nil } - - return fs.FS.OpenFile(name, flags) + return f, nil } -type fileStat struct { +type overrideFile struct { fs.File - fi os.FileInfo + ofs *overrideFS } -func (f fileStat) Stat() (os.FileInfo, error) { - return f.fi, nil +func (f overrideFile) Stat() (os.FileInfo, error) { + return f.ofs.overrideFI, nil +} + +func (f overrideFile) ToNode(ignoreXattrListError bool) (*restic.Node, error) { + if f.ofs.overrideNode == nil { + return f.File.ToNode(ignoreXattrListError) + } + return f.ofs.overrideNode, f.ofs.overrideErr } // used by wrapFileInfo, use untyped const in order to avoid having a version @@ -2279,17 +2260,18 @@ func TestMetadataChanged(t *testing.T) { // get metadata fi := lstat(t, "testfile") localFS := &fs.Local{} - want, err := localFS.NodeFromFileInfo("testfile", fi, false) - if err != nil { - t.Fatal(err) - } + meta, err := localFS.OpenFile("testfile", fs.O_NOFOLLOW, true) + rtest.OK(t, err) + want, err := meta.ToNode(false) + rtest.OK(t, err) + rtest.OK(t, meta.Close()) - fs := &StatFS{ - FS: localFS, - OverrideLstat: map[string]os.FileInfo{ - "testfile": fi, - }, + fs := &overrideFS{ + FS: localFS, + overrideFI: fi, + overrideNode: &restic.Node{}, } + *fs.overrideNode = *want sn, node2 := snapshot(t, repo, fs, nil, "testfile") @@ -2309,7 +2291,8 @@ func TestMetadataChanged(t *testing.T) { } // modify the mode by wrapping it in a new struct, uses the consts defined above - fs.OverrideLstat["testfile"] = wrapFileInfo(fi) + fs.overrideFI = wrapFileInfo(fi) + rtest.Assert(t, !fileChanged(fs, fs.overrideFI, node2, 0), "testfile must not be considered as changed") // set the override values in the 'want' node which want.Mode = 0400 @@ -2318,16 +2301,13 @@ func TestMetadataChanged(t *testing.T) { want.UID = 51234 want.GID = 51235 } - // no user and group name - want.User = "" - want.Group = "" + // update mock node accordingly + fs.overrideNode.Mode = 0400 + fs.overrideNode.UID = want.UID + fs.overrideNode.GID = want.GID // make another snapshot _, node3 := snapshot(t, repo, fs, sn, "testfile") - // Override username and group to empty string - in case underlying system has user with UID 51234 - // See https://github.com/restic/restic/issues/2372 - node3.User = "" - node3.Group = "" // make sure that metadata was recorded successfully if !cmp.Equal(want, node3) { @@ -2342,7 +2322,7 @@ func TestMetadataChanged(t *testing.T) { func TestRacyFileSwap(t *testing.T) { files := TestDir{ - "file": TestFile{ + "testfile": TestFile{ Content: "foo bar test file", }, } @@ -2354,14 +2334,11 @@ func TestRacyFileSwap(t *testing.T) { // get metadata of current folder fi := lstat(t, ".") - tempfile := filepath.Join(tempdir, "file") + tempfile := filepath.Join(tempdir, "testfile") - statfs := &StatFS{ - FS: fs.Local{}, - OverrideLstat: map[string]os.FileInfo{ - tempfile: fi, - }, - OnlyOverrideStat: true, + statfs := &overrideFS{ + FS: fs.Local{}, + overrideFI: fi, } ctx, cancel := context.WithCancel(context.Background()) @@ -2388,14 +2365,19 @@ func TestRacyFileSwap(t *testing.T) { } } +type mockToNoder struct { + node *restic.Node + err error +} + +func (m *mockToNoder) ToNode(_ bool) (*restic.Node, error) { + return m.node, m.err +} + func TestMetadataBackupErrorFiltering(t *testing.T) { tempdir := t.TempDir() - repo := repository.TestRepository(t) - filename := filepath.Join(tempdir, "file") - rtest.OK(t, os.WriteFile(filename, []byte("example"), 0o600)) - fi, err := os.Stat(filename) - rtest.OK(t, err) + repo := repository.TestRepository(t) arch := New(repo, fs.Local{}, Options{}) @@ -2406,15 +2388,24 @@ func TestMetadataBackupErrorFiltering(t *testing.T) { return replacementErr } + nonExistNoder := &mockToNoder{ + node: &restic.Node{Type: restic.NodeTypeFile}, + err: fmt.Errorf("not found"), + } + // check that errors from reading extended metadata are properly filtered - node, err := arch.nodeFromFileInfo("file", filename+"invalid", fi, false) + node, err := arch.nodeFromFileInfo("file", filename+"invalid", nonExistNoder, false) rtest.Assert(t, node != nil, "node is missing") rtest.Assert(t, err == replacementErr, "expected %v got %v", replacementErr, err) rtest.Assert(t, filteredErr != nil, "missing inner error") // check that errors from reading irregular file are not filtered filteredErr = nil - node, err = arch.nodeFromFileInfo("file", filename, wrapIrregularFileInfo(fi), false) + nonExistNoder = &mockToNoder{ + node: &restic.Node{Type: restic.NodeTypeIrregular}, + err: fmt.Errorf(`unsupported file type "irregular"`), + } + node, err = arch.nodeFromFileInfo("file", filename, nonExistNoder, false) rtest.Assert(t, node != nil, "node is missing") rtest.Assert(t, filteredErr == nil, "error for irregular node should not have been filtered") rtest.Assert(t, strings.Contains(err.Error(), "irregular"), "unexpected error %q does not warn about irregular file mode", err) @@ -2434,17 +2425,19 @@ func TestIrregularFile(t *testing.T) { tempfile := filepath.Join(tempdir, "testfile") fi := lstat(t, "testfile") - statfs := &StatFS{ - FS: fs.Local{}, - OverrideLstat: map[string]os.FileInfo{ - tempfile: wrapIrregularFileInfo(fi), + override := &overrideFS{ + FS: fs.Local{}, + overrideFI: wrapIrregularFileInfo(fi), + overrideNode: &restic.Node{ + Type: restic.NodeTypeIrregular, }, + overrideErr: fmt.Errorf(`unsupported file type "irregular"`), } ctx, cancel := context.WithCancel(context.Background()) defer cancel() - arch := New(repo, fs.Track{FS: statfs}, Options{}) + arch := New(repo, fs.Track{FS: override}, Options{}) _, excluded, err := arch.save(ctx, "/", tempfile, nil) if err == nil { t.Fatalf("Save() should have failed") diff --git a/internal/archiver/archiver_unix_test.go b/internal/archiver/archiver_unix_test.go index 621f84826..deeab6459 100644 --- a/internal/archiver/archiver_unix_test.go +++ b/internal/archiver/archiver_unix_test.go @@ -57,12 +57,8 @@ func wrapIrregularFileInfo(fi os.FileInfo) os.FileInfo { } func statAndSnapshot(t *testing.T, repo archiverRepo, name string) (*restic.Node, *restic.Node) { - fi := lstat(t, name) - fs := &fs.Local{} - want, err := fs.NodeFromFileInfo(name, fi, false) - rtest.OK(t, err) - - _, node := snapshot(t, repo, fs, nil, name) + want := nodeFromFile(t, &fs.Local{}, name) + _, node := snapshot(t, repo, &fs.Local{}, nil, name) return want, node } diff --git a/internal/archiver/exclude.go b/internal/archiver/exclude.go index 54ced788a..418517fd9 100644 --- a/internal/archiver/exclude.go +++ b/internal/archiver/exclude.go @@ -135,9 +135,9 @@ func isExcludedByFile(filename, tagFilename, header string, rc *rejectionCache, return rejected } -func isDirExcludedByFile(dir, tagFilename, header string, fs fs.FS, warnf func(msg string, args ...interface{})) bool { - tf := fs.Join(dir, tagFilename) - _, err := fs.Lstat(tf) +func isDirExcludedByFile(dir, tagFilename, header string, fsInst fs.FS, warnf func(msg string, args ...interface{})) bool { + tf := fsInst.Join(dir, tagFilename) + _, err := fsInst.Lstat(tf) if errors.Is(err, os.ErrNotExist) { return false } @@ -153,7 +153,7 @@ func isDirExcludedByFile(dir, tagFilename, header string, fs fs.FS, warnf func(m // From this stage, errors mean tagFilename exists but it is malformed. // Warnings will be generated so that the user is informed that the // indented ignore-action is not performed. - f, err := fs.OpenFile(tf, os.O_RDONLY) + f, err := fsInst.OpenFile(tf, fs.O_RDONLY, false) if err != nil { warnf("could not open exclusion tagfile: %v", err) return false diff --git a/internal/archiver/file_saver.go b/internal/archiver/file_saver.go index dccaa9442..ca8ec2fbb 100644 --- a/internal/archiver/file_saver.go +++ b/internal/archiver/file_saver.go @@ -4,7 +4,6 @@ import ( "context" "fmt" "io" - "os" "sync" "github.com/restic/chunker" @@ -29,7 +28,7 @@ type fileSaver struct { CompleteBlob func(bytes uint64) - NodeFromFileInfo func(snPath, filename string, fi os.FileInfo, ignoreXattrListError bool) (*restic.Node, error) + NodeFromFileInfo func(snPath, filename string, meta ToNoder, ignoreXattrListError bool) (*restic.Node, error) } // newFileSaver returns a new file saver. A worker pool with fileWorkers is @@ -71,13 +70,12 @@ type fileCompleteFunc func(*restic.Node, ItemStats) // file is closed by Save. completeReading is only called if the file was read // successfully. complete is always called. If completeReading is called, then // this will always happen before calling complete. -func (s *fileSaver) Save(ctx context.Context, snPath string, target string, file fs.File, fi os.FileInfo, start func(), completeReading func(), complete fileCompleteFunc) futureNode { +func (s *fileSaver) Save(ctx context.Context, snPath string, target string, file fs.File, start func(), completeReading func(), complete fileCompleteFunc) futureNode { fn, ch := newFutureNode() job := saveFileJob{ snPath: snPath, target: target, file: file, - fi: fi, ch: ch, start: start, @@ -100,7 +98,6 @@ type saveFileJob struct { snPath string target string file fs.File - fi os.FileInfo ch chan<- futureNodeResult start func() @@ -109,7 +106,7 @@ type saveFileJob struct { } // saveFile stores the file f in the repo, then closes it. -func (s *fileSaver) saveFile(ctx context.Context, chnker *chunker.Chunker, snPath string, target string, f fs.File, fi os.FileInfo, start func(), finishReading func(), finish func(res futureNodeResult)) { +func (s *fileSaver) saveFile(ctx context.Context, chnker *chunker.Chunker, snPath string, target string, f fs.File, start func(), finishReading func(), finish func(res futureNodeResult)) { start() fnr := futureNodeResult{ @@ -156,7 +153,7 @@ func (s *fileSaver) saveFile(ctx context.Context, chnker *chunker.Chunker, snPat debug.Log("%v", snPath) - node, err := s.NodeFromFileInfo(snPath, target, fi, false) + node, err := s.NodeFromFileInfo(snPath, target, f, false) if err != nil { _ = f.Close() completeError(err) @@ -262,7 +259,7 @@ func (s *fileSaver) worker(ctx context.Context, jobs <-chan saveFileJob) { } } - s.saveFile(ctx, chnker, job.snPath, job.target, job.file, job.fi, job.start, func() { + s.saveFile(ctx, chnker, job.snPath, job.target, job.file, job.start, func() { if job.completeReading != nil { job.completeReading() } diff --git a/internal/archiver/file_saver_test.go b/internal/archiver/file_saver_test.go index 069cdc171..ce862f6fe 100644 --- a/internal/archiver/file_saver_test.go +++ b/internal/archiver/file_saver_test.go @@ -30,7 +30,7 @@ func createTestFiles(t testing.TB, num int) (files []string) { return files } -func startFileSaver(ctx context.Context, t testing.TB, fs fs.FS) (*fileSaver, context.Context, *errgroup.Group) { +func startFileSaver(ctx context.Context, t testing.TB, fsInst fs.FS) (*fileSaver, context.Context, *errgroup.Group) { wg, ctx := errgroup.WithContext(ctx) saveBlob := func(ctx context.Context, tpe restic.BlobType, buf *buffer, _ string, cb func(saveBlobResponse)) { @@ -49,8 +49,8 @@ func startFileSaver(ctx context.Context, t testing.TB, fs fs.FS) (*fileSaver, co } s := newFileSaver(ctx, wg, saveBlob, pol, workers, workers) - s.NodeFromFileInfo = func(snPath, filename string, fi os.FileInfo, ignoreXattrListError bool) (*restic.Node, error) { - return fs.NodeFromFileInfo(filename, fi, ignoreXattrListError) + s.NodeFromFileInfo = func(snPath, filename string, meta ToNoder, ignoreXattrListError bool) (*restic.Node, error) { + return meta.ToNode(ignoreXattrListError) } return s, ctx, wg @@ -72,17 +72,12 @@ func TestFileSaver(t *testing.T) { var results []futureNode for _, filename := range files { - f, err := testFs.OpenFile(filename, os.O_RDONLY) + f, err := testFs.OpenFile(filename, os.O_RDONLY, false) if err != nil { t.Fatal(err) } - fi, err := f.Stat() - if err != nil { - t.Fatal(err) - } - - ff := s.Save(ctx, filename, filename, f, fi, startFn, completeReadingFn, completeFn) + ff := s.Save(ctx, filename, filename, f, startFn, completeReadingFn, completeFn) results = append(results, ff) } diff --git a/internal/fs/file.go b/internal/fs/file.go index fa395b628..81ee4bc7a 100644 --- a/internal/fs/file.go +++ b/internal/fs/file.go @@ -67,7 +67,7 @@ func ResetPermissions(path string) error { // Readdirnames returns a list of file in a directory. Flags are passed to fs.OpenFile. // O_RDONLY and O_DIRECTORY are implied. func Readdirnames(filesystem FS, dir string, flags int) ([]string, error) { - f, err := filesystem.OpenFile(dir, O_RDONLY|O_DIRECTORY|flags) + f, err := filesystem.OpenFile(dir, O_RDONLY|O_DIRECTORY|flags, false) if err != nil { return nil, fmt.Errorf("openfile for readdirnames failed: %w", err) } diff --git a/internal/fs/fs_local.go b/internal/fs/fs_local.go index 4f8b67090..5e6c72d0a 100644 --- a/internal/fs/fs_local.go +++ b/internal/fs/fs_local.go @@ -20,18 +20,16 @@ func (fs Local) VolumeName(path string) string { return filepath.VolumeName(path) } -// OpenFile is the generalized open call; most users will use Open -// or Create instead. It opens the named file with specified flag -// (O_RDONLY etc.) and perm, (0666 etc.) if applicable. If successful, -// methods on the returned File can be used for I/O. -// If there is an error, it will be of type *PathError. -func (fs Local) OpenFile(name string, flag int) (File, error) { - f, err := os.OpenFile(fixpath(name), flag, 0) - if err != nil { - return nil, err - } - _ = setFlags(f) - return f, nil +// OpenFile opens a file or directory for reading. +// +// If metadataOnly is set, an implementation MUST return a File object for +// arbitrary file types including symlinks. The implementation may internally use +// the given file path or a file handle. In particular, an implementation may +// delay actually accessing the underlying filesystem. +// +// Only the O_NOFOLLOW and O_DIRECTORY flags are supported. +func (fs Local) OpenFile(name string, flag int, metadataOnly bool) (File, error) { + return newLocalFile(name, flag, metadataOnly) } // Lstat returns the FileInfo structure describing the named file. @@ -53,10 +51,6 @@ func (fs Local) ExtendedStat(fi os.FileInfo) ExtendedFileInfo { return ExtendedStat(fi) } -func (fs Local) NodeFromFileInfo(path string, fi os.FileInfo, ignoreXattrListError bool) (*restic.Node, error) { - return nodeFromFileInfo(path, fi, ignoreXattrListError) -} - // Join joins any number of path elements into a single path, adding a // Separator if necessary. Join calls Clean on the result; in particular, all // empty strings are ignored. On Windows, the result is a UNC path if and only @@ -97,3 +91,87 @@ func (fs Local) Base(path string) string { func (fs Local) Dir(path string) string { return filepath.Dir(path) } + +type localFile struct { + name string + flag int + f *os.File + fi os.FileInfo +} + +// See the File interface for a description of each method +var _ File = &localFile{} + +func newLocalFile(name string, flag int, metadataOnly bool) (*localFile, error) { + var f *os.File + if !metadataOnly { + var err error + f, err = os.OpenFile(fixpath(name), flag, 0) + if err != nil { + return nil, err + } + _ = setFlags(f) + } + return &localFile{ + name: name, + flag: flag, + f: f, + }, nil +} + +func (f *localFile) MakeReadable() error { + if f.f != nil { + panic("file is already readable") + } + + newF, err := newLocalFile(f.name, f.flag, false) + if err != nil { + return err + } + // replace state and also reset cached FileInfo + *f = *newF + return nil +} + +func (f *localFile) cacheFI() error { + if f.fi != nil { + return nil + } + var err error + if f.f != nil { + f.fi, err = f.f.Stat() + } else if f.flag&O_NOFOLLOW != 0 { + f.fi, err = os.Lstat(f.name) + } else { + f.fi, err = os.Stat(f.name) + } + return err +} + +func (f *localFile) Stat() (os.FileInfo, error) { + err := f.cacheFI() + // the call to cacheFI MUST happen before reading from f.fi + return f.fi, err +} + +func (f *localFile) ToNode(ignoreXattrListError bool) (*restic.Node, error) { + if err := f.cacheFI(); err != nil { + return nil, err + } + return nodeFromFileInfo(f.name, f.fi, ignoreXattrListError) +} + +func (f *localFile) Read(p []byte) (n int, err error) { + return f.f.Read(p) +} + +func (f *localFile) Readdirnames(n int) ([]string, error) { + return f.f.Readdirnames(n) +} + +func (f *localFile) Close() error { + if f.f != nil { + return f.f.Close() + } + return nil +} diff --git a/internal/fs/fs_local_vss.go b/internal/fs/fs_local_vss.go index e9bc66657..fe82b85e1 100644 --- a/internal/fs/fs_local_vss.go +++ b/internal/fs/fs_local_vss.go @@ -10,7 +10,6 @@ import ( "github.com/restic/restic/internal/errors" "github.com/restic/restic/internal/options" - "github.com/restic/restic/internal/restic" ) // VSSConfig holds extended options of windows volume shadow copy service. @@ -126,9 +125,9 @@ func (fs *LocalVss) DeleteSnapshots() { fs.snapshots = activeSnapshots } -// OpenFile wraps the Open method of the underlying file system. -func (fs *LocalVss) OpenFile(name string, flag int) (File, error) { - return fs.FS.OpenFile(fs.snapshotPath(name), flag) +// OpenFile wraps the OpenFile method of the underlying file system. +func (fs *LocalVss) OpenFile(name string, flag int, metadataOnly bool) (File, error) { + return fs.FS.OpenFile(fs.snapshotPath(name), flag, metadataOnly) } // Lstat wraps the Lstat method of the underlying file system. @@ -136,10 +135,6 @@ func (fs *LocalVss) Lstat(name string) (os.FileInfo, error) { return fs.FS.Lstat(fs.snapshotPath(name)) } -func (fs *LocalVss) NodeFromFileInfo(path string, fi os.FileInfo, ignoreXattrListError bool) (*restic.Node, error) { - return fs.FS.NodeFromFileInfo(fs.snapshotPath(path), fi, ignoreXattrListError) -} - // isMountPointIncluded is true if given mountpoint included by user. func (fs *LocalVss) isMountPointIncluded(mountPoint string) bool { if fs.excludeVolumes == nil { diff --git a/internal/fs/fs_local_vss_test.go b/internal/fs/fs_local_vss_test.go index db8d4b133..33c412fe9 100644 --- a/internal/fs/fs_local_vss_test.go +++ b/internal/fs/fs_local_vss_test.go @@ -327,7 +327,7 @@ func TestVSSFS(t *testing.T) { rtest.OK(t, err) rtest.Equals(t, origFi.Mode(), lstatFi.Mode()) - f, err := localVss.OpenFile(tempfile, os.O_RDONLY) + f, err := localVss.OpenFile(tempfile, os.O_RDONLY, false) rtest.OK(t, err) data, err := io.ReadAll(f) rtest.OK(t, err) diff --git a/internal/fs/fs_reader.go b/internal/fs/fs_reader.go index a4efa8dea..8728b274c 100644 --- a/internal/fs/fs_reader.go +++ b/internal/fs/fs_reader.go @@ -49,12 +49,7 @@ func (fs *Reader) fi() os.FileInfo { } } -// OpenFile is the generalized open call; most users will use Open -// or Create instead. It opens the named file with specified flag -// (O_RDONLY etc.) and perm, (0666 etc.) if applicable. If successful, -// methods on the returned File can be used for I/O. -// If there is an error, it will be of type *os.PathError. -func (fs *Reader) OpenFile(name string, flag int) (f File, err error) { +func (fs *Reader) OpenFile(name string, flag int, _ bool) (f File, err error) { if flag & ^(O_RDONLY|O_NOFOLLOW) != 0 { return nil, pathError("open", name, fmt.Errorf("invalid combination of flags 0x%x", flag)) @@ -127,17 +122,6 @@ func (fs *Reader) ExtendedStat(fi os.FileInfo) ExtendedFileInfo { } } -func (fs *Reader) NodeFromFileInfo(path string, fi os.FileInfo, _ bool) (*restic.Node, error) { - node := buildBasicNode(path, fi) - - // fill minimal info with current values for uid, gid - node.UID = uint32(os.Getuid()) - node.GID = uint32(os.Getgid()) - node.ChangeTime = node.ModTime - - return node, nil -} - // Join joins any number of path elements into a single path, adding a // Separator if necessary. Join calls Clean on the result; in particular, all // empty strings are ignored. On Windows, the result is a UNC path if and only @@ -235,6 +219,10 @@ type fakeFile struct { // ensure that fakeFile implements File var _ File = fakeFile{} +func (f fakeFile) MakeReadable() error { + return nil +} + func (f fakeFile) Readdirnames(_ int) ([]string, error) { return nil, pathError("readdirnames", f.name, os.ErrInvalid) } @@ -251,6 +239,17 @@ func (f fakeFile) Stat() (os.FileInfo, error) { return f.FileInfo, nil } +func (f fakeFile) ToNode(_ bool) (*restic.Node, error) { + node := buildBasicNode(f.name, f.FileInfo) + + // fill minimal info with current values for uid, gid + node.UID = uint32(os.Getuid()) + node.GID = uint32(os.Getgid()) + node.ChangeTime = node.ModTime + + return node, nil +} + // fakeDir implements Readdirnames and Readdir, everything else is delegated to fakeFile. type fakeDir struct { entries []os.FileInfo diff --git a/internal/fs/fs_reader_test.go b/internal/fs/fs_reader_test.go index e7020bc9d..7e7f6e77c 100644 --- a/internal/fs/fs_reader_test.go +++ b/internal/fs/fs_reader_test.go @@ -16,7 +16,7 @@ import ( ) func verifyFileContentOpenFile(t testing.TB, fs FS, filename string, want []byte) { - f, err := fs.OpenFile(filename, O_RDONLY) + f, err := fs.OpenFile(filename, O_RDONLY, false) if err != nil { t.Fatal(err) } @@ -37,7 +37,7 @@ func verifyFileContentOpenFile(t testing.TB, fs FS, filename string, want []byte } func verifyDirectoryContents(t testing.TB, fs FS, dir string, want []string) { - f, err := fs.OpenFile(dir, os.O_RDONLY) + f, err := fs.OpenFile(dir, O_RDONLY, false) if err != nil { t.Fatal(err) } @@ -123,7 +123,7 @@ func TestFSReader(t *testing.T) { { name: "file/Stat", f: func(t *testing.T, fs FS) { - f, err := fs.OpenFile(filename, os.O_RDONLY) + f, err := fs.OpenFile(filename, O_RDONLY, true) if err != nil { t.Fatal(err) } @@ -295,7 +295,7 @@ func TestFSReaderMinFileSize(t *testing.T) { AllowEmptyFile: test.allowEmpty, } - f, err := fs.OpenFile("testfile", os.O_RDONLY) + f, err := fs.OpenFile("testfile", O_RDONLY, false) if err != nil { t.Fatal(err) } diff --git a/internal/fs/fs_track.go b/internal/fs/fs_track.go index 9912ac45b..9ebdbb8c4 100644 --- a/internal/fs/fs_track.go +++ b/internal/fs/fs_track.go @@ -16,8 +16,8 @@ type Track struct { } // OpenFile wraps the OpenFile method of the underlying file system. -func (fs Track) OpenFile(name string, flag int) (File, error) { - f, err := fs.FS.OpenFile(fixpath(name), flag) +func (fs Track) OpenFile(name string, flag int, metadataOnly bool) (File, error) { + f, err := fs.FS.OpenFile(name, flag, metadataOnly) if err != nil { return nil, err } @@ -31,7 +31,7 @@ type trackFile struct { func newTrackFile(stack []byte, filename string, file File) *trackFile { f := &trackFile{file} - runtime.SetFinalizer(f, func(_ *trackFile) { + runtime.SetFinalizer(f, func(_ any) { fmt.Fprintf(os.Stderr, "file %s not closed\n\nStacktrack:\n%s\n", filename, stack) panic("file " + filename + " not closed") }) diff --git a/internal/fs/interface.go b/internal/fs/interface.go index 899888fb0..7ff777138 100644 --- a/internal/fs/interface.go +++ b/internal/fs/interface.go @@ -9,11 +9,18 @@ import ( // FS bundles all methods needed for a file system. type FS interface { - OpenFile(name string, flag int) (File, error) + // OpenFile opens a file or directory for reading. + // + // If metadataOnly is set, an implementation MUST return a File object for + // arbitrary file types including symlinks. The implementation may internally use + // the given file path or a file handle. In particular, an implementation may + // delay actually accessing the underlying filesystem. + // + // Only the O_NOFOLLOW and O_DIRECTORY flags are supported. + OpenFile(name string, flag int, metadataOnly bool) (File, error) Lstat(name string) (os.FileInfo, error) DeviceID(fi os.FileInfo) (deviceID uint64, err error) ExtendedStat(fi os.FileInfo) ExtendedFileInfo - NodeFromFileInfo(path string, fi os.FileInfo, ignoreXattrListError bool) (*restic.Node, error) Join(elem ...string) string Separator() string @@ -26,11 +33,23 @@ type FS interface { Base(path string) string } -// File is an open file on a file system. +// File is an open file on a file system. When opened as metadataOnly, an +// implementation may opt to perform filesystem operations using the filepath +// instead of actually opening the file. type File interface { + // MakeReadable reopens a File that was opened metadataOnly for reading. + // The method must not be called for files that are opened for reading. + // If possible, the underlying file should be reopened atomically. + // MakeReadable must work for files and directories. + MakeReadable() error + io.Reader io.Closer Readdirnames(n int) ([]string, error) Stat() (os.FileInfo, error) + // ToNode returns a restic.Node for the File. The internally used os.FileInfo + // must be consistent with that returned by Stat(). In particular, the metadata + // returned by consecutive calls to Stat() and ToNode() must match. + ToNode(ignoreXattrListError bool) (*restic.Node, error) } diff --git a/internal/fs/node_test.go b/internal/fs/node_test.go index 58facceb1..65098e304 100644 --- a/internal/fs/node_test.go +++ b/internal/fs/node_test.go @@ -17,56 +17,26 @@ import ( rtest "github.com/restic/restic/internal/test" ) -func BenchmarkNodeFillUser(t *testing.B) { - tempfile, err := os.CreateTemp("", "restic-test-temp-") - if err != nil { - t.Fatal(err) - } - - fi, err := tempfile.Stat() - if err != nil { - t.Fatal(err) - } - +func BenchmarkNodeFromFileInfo(t *testing.B) { + tempfile, err := os.CreateTemp(t.TempDir(), "restic-test-temp-") + rtest.OK(t, err) path := tempfile.Name() + rtest.OK(t, tempfile.Close()) + fs := Local{} + f, err := fs.OpenFile(path, O_NOFOLLOW, true) + rtest.OK(t, err) + _, err = f.Stat() + rtest.OK(t, err) t.ResetTimer() for i := 0; i < t.N; i++ { - _, err := fs.NodeFromFileInfo(path, fi, false) + _, err := f.ToNode(false) rtest.OK(t, err) } - rtest.OK(t, tempfile.Close()) - rtest.RemoveAll(t, tempfile.Name()) -} - -func BenchmarkNodeFromFileInfo(t *testing.B) { - tempfile, err := os.CreateTemp("", "restic-test-temp-") - if err != nil { - t.Fatal(err) - } - - fi, err := tempfile.Stat() - if err != nil { - t.Fatal(err) - } - - path := tempfile.Name() - fs := Local{} - - t.ResetTimer() - - for i := 0; i < t.N; i++ { - _, err := fs.NodeFromFileInfo(path, fi, false) - if err != nil { - t.Fatal(err) - } - } - - rtest.OK(t, tempfile.Close()) - rtest.RemoveAll(t, tempfile.Name()) + rtest.OK(t, f.Close()) } func parseTime(s string) time.Time { @@ -249,14 +219,14 @@ func TestNodeRestoreAt(t *testing.T) { rtest.OK(t, NodeCreateAt(&test, nodePath)) rtest.OK(t, NodeRestoreMetadata(&test, nodePath, func(msg string) { rtest.OK(t, fmt.Errorf("Warning triggered for path: %s: %s", nodePath, msg)) })) - fi, err := os.Lstat(nodePath) - rtest.OK(t, err) - fs := &Local{} - n2, err := fs.NodeFromFileInfo(nodePath, fi, false) + meta, err := fs.OpenFile(nodePath, O_NOFOLLOW, true) rtest.OK(t, err) - n3, err := fs.NodeFromFileInfo(nodePath, fi, true) + n2, err := meta.ToNode(false) rtest.OK(t, err) + n3, err := meta.ToNode(true) + rtest.OK(t, err) + rtest.OK(t, meta.Close()) rtest.Assert(t, n2.Equals(*n3), "unexpected node info mismatch %v", cmp.Diff(n2, n3)) rtest.Assert(t, test.Name == n2.Name, diff --git a/internal/fs/node_unix_test.go b/internal/fs/node_unix_test.go index 6b47eafba..1eb1ee506 100644 --- a/internal/fs/node_unix_test.go +++ b/internal/fs/node_unix_test.go @@ -114,16 +114,14 @@ func TestNodeFromFileInfo(t *testing.T) { return } - if fi.Sys() == nil { - t.Skip("fi.Sys() is nil") - return - } - fs := &Local{} - node, err := fs.NodeFromFileInfo(test.filename, fi, false) - if err != nil { - t.Fatal(err) - } + meta, err := fs.OpenFile(test.filename, O_NOFOLLOW, true) + rtest.OK(t, err) + node, err := meta.ToNode(false) + rtest.OK(t, err) + rtest.OK(t, meta.Close()) + + rtest.OK(t, err) switch node.Type { case restic.NodeTypeFile, restic.NodeTypeSymlink: diff --git a/internal/fs/node_windows_test.go b/internal/fs/node_windows_test.go index 1bb76b204..f75df54d3 100644 --- a/internal/fs/node_windows_test.go +++ b/internal/fs/node_windows_test.go @@ -222,11 +222,11 @@ func restoreAndGetNode(t *testing.T, tempDir string, testNode *restic.Node, warn test.OK(t, errors.Wrapf(err, "Failed to restore metadata for: %s", testPath)) fs := &Local{} - fi, err := fs.Lstat(testPath) - test.OK(t, errors.Wrapf(err, "Could not Lstat for path: %s", testPath)) - - nodeFromFileInfo, err := fs.NodeFromFileInfo(testPath, fi, false) + meta, err := fs.OpenFile(testPath, O_NOFOLLOW, true) + test.OK(t, err) + nodeFromFileInfo, err := meta.ToNode(false) test.OK(t, errors.Wrapf(err, "Could not get NodeFromFileInfo for path: %s", testPath)) + test.OK(t, meta.Close()) return testPath, nodeFromFileInfo } diff --git a/internal/restic/tree_test.go b/internal/restic/tree_test.go index f1979f135..07ca254f1 100644 --- a/internal/restic/tree_test.go +++ b/internal/restic/tree_test.go @@ -83,13 +83,17 @@ func TestNodeMarshal(t *testing.T) { } } -func TestNodeComparison(t *testing.T) { - fs := &fs.Local{} - fi, err := fs.Lstat("tree_test.go") +func nodeForFile(t *testing.T, name string) *restic.Node { + f, err := (&fs.Local{}).OpenFile(name, fs.O_NOFOLLOW, true) rtest.OK(t, err) + node, err := f.ToNode(false) + rtest.OK(t, err) + rtest.OK(t, f.Close()) + return node +} - node, err := fs.NodeFromFileInfo("tree_test.go", fi, false) - rtest.OK(t, err) +func TestNodeComparison(t *testing.T) { + node := nodeForFile(t, "tree_test.go") n2 := *node rtest.Assert(t, node.Equals(n2), "nodes aren't equal") @@ -127,11 +131,7 @@ func TestTreeEqualSerialization(t *testing.T) { builder := restic.NewTreeJSONBuilder() for _, fn := range files[:i] { - fs := &fs.Local{} - fi, err := fs.Lstat(fn) - rtest.OK(t, err) - node, err := fs.NodeFromFileInfo(fn, fi, false) - rtest.OK(t, err) + node := nodeForFile(t, fn) rtest.OK(t, tree.Insert(node)) rtest.OK(t, builder.AddNode(node))