X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/109cd685ecbfb5b685347731340c6dd69e630617..094024f2476d84a5b9a453ede79b243e1d282bab:/services/keepstore/volume_unix.go diff --git a/services/keepstore/volume_unix.go b/services/keepstore/volume_unix.go index 5d09e84f52..5982fb0484 100644 --- a/services/keepstore/volume_unix.go +++ b/services/keepstore/volume_unix.go @@ -2,7 +2,6 @@ package main import ( "bufio" - "bytes" "errors" "flag" "fmt" @@ -139,9 +138,8 @@ func (v *UnixVolume) Touch(loc string) error { return e } defer unlockfile(f) - now := time.Now().Unix() - utime := syscall.Utimbuf{now, now} - return syscall.Utime(p, &utime) + ts := syscall.NsecToTimespec(time.Now().UnixNano()) + return syscall.UtimesNano(p, []syscall.Timespec{ts, ts}) } // Mtime returns the stored timestamp for the given locator. @@ -182,26 +180,24 @@ func (v *UnixVolume) stat(path string) (os.FileInfo, error) { return stat, err } -// Get retrieves a block identified by the locator string "loc", and -// returns its contents as a byte slice. -// -// Get returns a nil buffer IFF it returns a non-nil error. -func (v *UnixVolume) Get(loc string) ([]byte, error) { +// Get retrieves a block, copies it to the given slice, and returns +// the number of bytes copied. +func (v *UnixVolume) Get(loc string, buf []byte) (int, error) { path := v.blockPath(loc) stat, err := v.stat(path) if err != nil { - return nil, err + return 0, v.translateError(err) + } + if stat.Size() > int64(len(buf)) { + return 0, TooLongError } - buf := bufs.Get(int(stat.Size())) + var read int + size := int(stat.Size()) err = v.getFunc(path, func(rdr io.Reader) error { - _, err = io.ReadFull(rdr, buf) + read, err = io.ReadFull(rdr, buf[:size]) return err }) - if err != nil { - bufs.Put(buf) - return nil, err - } - return buf, nil + return read, err } // Compare returns nil if Get(loc) would return the same content as @@ -209,43 +205,11 @@ func (v *UnixVolume) Get(loc string) ([]byte, error) { // bytes.Compare(), but uses less memory. func (v *UnixVolume) Compare(loc string, expect []byte) error { path := v.blockPath(loc) - stat, err := v.stat(path) - if err != nil { - return err + if _, err := v.stat(path); err != nil { + return v.translateError(err) } - bufLen := 1 << 20 - if int64(bufLen) > stat.Size() { - bufLen = int(stat.Size()) - if bufLen < 1 { - // len(buf)==0 would prevent us from handling - // empty files the same way as non-empty - // files, because reading 0 bytes at a time - // never reaches EOF. - bufLen = 1 - } - } - cmp := expect - buf := make([]byte, bufLen) return v.getFunc(path, func(rdr io.Reader) error { - // Loop invariants: all data read so far matched what - // we expected, and the first N bytes of cmp are - // expected to equal the next N bytes read from - // reader. - for { - n, err := rdr.Read(buf) - if n > len(cmp) || bytes.Compare(cmp[:n], buf[:n]) != 0 { - return collisionOrCorrupt(loc[:32], expect[:len(expect)-len(cmp)], buf[:n], rdr) - } - cmp = cmp[n:] - if err == io.EOF { - if len(cmp) != 0 { - return collisionOrCorrupt(loc[:32], expect[:len(expect)-len(cmp)], nil, nil) - } - return nil - } else if err != nil { - return err - } - } + return compareReaderWithBuf(rdr, expect, loc[:32]) }) } @@ -325,6 +289,7 @@ func (v *UnixVolume) Status() *VolumeStatus { } var blockDirRe = regexp.MustCompile(`^[0-9a-f]+$`) +var blockFileRe = regexp.MustCompile(`^[0-9a-f]{32}$`) // IndexTo writes (to the given Writer) a list of blocks found on this // volume which begin with the specified prefix. If the prefix is an @@ -341,7 +306,7 @@ var blockDirRe = regexp.MustCompile(`^[0-9a-f]+$`) // e4de7a2810f5554cd39b36d8ddb132ff+67108864 1388701136 // func (v *UnixVolume) IndexTo(prefix string, w io.Writer) error { - var lastErr error = nil + var lastErr error rootdir, err := os.Open(v.root) if err != nil { return err @@ -381,25 +346,31 @@ func (v *UnixVolume) IndexTo(prefix string, w io.Writer) error { if !strings.HasPrefix(name, prefix) { continue } + if !blockFileRe.MatchString(name) { + continue + } _, err = fmt.Fprint(w, name, "+", fileInfo[0].Size(), - " ", fileInfo[0].ModTime().Unix(), + " ", fileInfo[0].ModTime().UnixNano(), "\n") } blockdir.Close() } } -// Delete deletes the block data from the unix storage -func (v *UnixVolume) Delete(loc string) error { +// Trash trashes the block data from the unix storage +// If trashLifetime == 0, the block is deleted +// Else, the block is renamed as path/{loc}.trash.{deadline}, +// where deadline = now + trashLifetime +func (v *UnixVolume) Trash(loc string) error { // Touch() must be called before calling Write() on a block. Touch() // also uses lockfile(). This avoids a race condition between Write() - // and Delete() because either (a) the file will be deleted and Touch() + // and Trash() because either (a) the file will be trashed and Touch() // will signal to the caller that the file is not present (and needs to // be re-written), or (b) Touch() will update the file's timestamp and - // Delete() will read the correct up-to-date timestamp and choose not to - // delete the file. + // Trash() will read the correct up-to-date timestamp and choose not to + // trash the file. if v.readonly { return MethodDisabledError @@ -426,12 +397,50 @@ func (v *UnixVolume) Delete(loc string) error { // anyway (because the permission signatures have expired). if fi, err := os.Stat(p); err != nil { return err - } else { - if time.Since(fi.ModTime()) < blobSignatureTTL { - return nil + } else if time.Since(fi.ModTime()) < blobSignatureTTL { + return nil + } + + if trashLifetime == 0 { + return os.Remove(p) + } + return os.Rename(p, fmt.Sprintf("%v.trash.%d", p, time.Now().Add(trashLifetime).Unix())) +} + +// Untrash moves block from trash back into store +// Look for path/{loc}.trash.{deadline} in storage, +// and rename the first such file as path/{loc} +func (v *UnixVolume) Untrash(loc string) (err error) { + if v.readonly { + return MethodDisabledError + } + + files, err := ioutil.ReadDir(v.blockDir(loc)) + if err != nil { + return err + } + + if len(files) == 0 { + return os.ErrNotExist + } + + foundTrash := false + prefix := fmt.Sprintf("%v.trash.", loc) + for _, f := range files { + if strings.HasPrefix(f.Name(), prefix) { + foundTrash = true + err = os.Rename(v.blockPath(f.Name()), v.blockPath(loc)) + if err == nil { + break + } } } - return os.Remove(p) + + if foundTrash == false { + return os.ErrNotExist + } + + return } // blockDir returns the fully qualified directory name for the directory @@ -495,11 +504,18 @@ func (v *UnixVolume) String() string { return fmt.Sprintf("[UnixVolume %s]", v.root) } -// Writable returns false if all future Put, Mtime, and Delete calls are expected to fail. +// Writable returns false if all future Put, Mtime, and Delete calls +// are expected to fail. func (v *UnixVolume) Writable() bool { return !v.readonly } +// Replication returns the number of replicas promised by the +// underlying device (currently assumed to be 1). +func (v *UnixVolume) Replication() int { + return 1 +} + // lockfile and unlockfile use flock(2) to manage kernel file locks. func lockfile(f *os.File) error { return syscall.Flock(int(f.Fd()), syscall.LOCK_EX) @@ -508,3 +524,63 @@ func lockfile(f *os.File) error { func unlockfile(f *os.File) error { return syscall.Flock(int(f.Fd()), syscall.LOCK_UN) } + +// Where appropriate, translate a more specific filesystem error to an +// error recognized by handlers, like os.ErrNotExist. +func (v *UnixVolume) translateError(err error) error { + switch err.(type) { + case *os.PathError: + // stat() returns a PathError if the parent directory + // (not just the file itself) is missing + return os.ErrNotExist + default: + return err + } +} + +var unixTrashLocRegexp = regexp.MustCompile(`/([0-9a-f]{32})\.trash\.(\d+)$`) + +// EmptyTrash walks hierarchy looking for {hash}.trash.* +// and deletes those with deadline < now. +func (v *UnixVolume) EmptyTrash() { + var bytesDeleted, bytesInTrash int64 + var blocksDeleted, blocksInTrash int + + err := filepath.Walk(v.root, func(path string, info os.FileInfo, err error) error { + if err != nil { + log.Printf("EmptyTrash: filepath.Walk: %v: %v", path, err) + return nil + } + if info.Mode().IsDir() { + return nil + } + matches := unixTrashLocRegexp.FindStringSubmatch(path) + if len(matches) != 3 { + return nil + } + deadline, err := strconv.ParseInt(matches[2], 10, 64) + if err != nil { + log.Printf("EmptyTrash: %v: ParseInt(%v): %v", path, matches[2], err) + return nil + } + bytesInTrash += info.Size() + blocksInTrash++ + if deadline > time.Now().Unix() { + return nil + } + err = os.Remove(path) + if err != nil { + log.Printf("EmptyTrash: Remove %v: %v", path, err) + return nil + } + bytesDeleted += info.Size() + blocksDeleted++ + return nil + }) + + if err != nil { + log.Printf("EmptyTrash error for %v: %v", v.String(), err) + } + + log.Printf("EmptyTrash stats for %v: Deleted %v bytes in %v blocks. Remaining in trash: %v bytes in %v blocks.", v.String(), bytesDeleted, blocksDeleted, bytesInTrash-bytesDeleted, blocksInTrash-blocksDeleted) +}