+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
package main
import (
"io"
"io/ioutil"
"os"
+ "os/exec"
"path/filepath"
"regexp"
"strconv"
// something to lock during IO, typically a sync.Mutex (or nil
// to skip locking)
locker sync.Locker
+
+ os osWithStats
+}
+
+// DeviceID returns a globally unique ID for the volume's root
+// directory, consisting of the filesystem's UUID and the path from
+// filesystem root to storage directory, joined by "/". For example,
+// the DeviceID for a local directory "/mnt/xvda1/keep" might be
+// "fa0b6166-3b55-4994-bd3f-92f4e00a1bb0/keep".
+func (v *UnixVolume) DeviceID() string {
+ giveup := func(f string, args ...interface{}) string {
+ log.Printf(f+"; using blank DeviceID for volume %s", append(args, v)...)
+ return ""
+ }
+ buf, err := exec.Command("findmnt", "--noheadings", "--target", v.Root).CombinedOutput()
+ if err != nil {
+ return giveup("findmnt: %s (%q)", err, buf)
+ }
+ findmnt := strings.Fields(string(buf))
+ if len(findmnt) < 2 {
+ return giveup("could not parse findmnt output: %q", buf)
+ }
+ fsRoot, dev := findmnt[0], findmnt[1]
+
+ absRoot, err := filepath.Abs(v.Root)
+ if err != nil {
+ return giveup("resolving relative path %q: %s", v.Root, err)
+ }
+ realRoot, err := filepath.EvalSymlinks(absRoot)
+ if err != nil {
+ return giveup("resolving symlinks in %q: %s", absRoot, err)
+ }
+
+ // Find path from filesystem root to realRoot
+ var fsPath string
+ if strings.HasPrefix(realRoot, fsRoot+"/") {
+ fsPath = realRoot[len(fsRoot):]
+ } else if fsRoot == "/" {
+ fsPath = realRoot
+ } else if fsRoot == realRoot {
+ fsPath = ""
+ } else {
+ return giveup("findmnt reports mount point %q which is not a prefix of volume root %q", fsRoot, realRoot)
+ }
+
+ if !strings.HasPrefix(dev, "/") {
+ return giveup("mount %q device %q is not a path", fsRoot, dev)
+ }
+
+ fi, err := os.Stat(dev)
+ if err != nil {
+ return giveup("stat %q: %s\n", dev, err)
+ }
+ ino := fi.Sys().(*syscall.Stat_t).Ino
+
+ // Find a symlink in /dev/disk/by-uuid/ whose target is (i.e.,
+ // has the same inode as) the mounted device
+ udir := "/dev/disk/by-uuid"
+ d, err := os.Open(udir)
+ if err != nil {
+ return giveup("opening %q: %s", udir, err)
+ }
+ uuids, err := d.Readdirnames(0)
+ if err != nil {
+ return giveup("reading %q: %s", udir, err)
+ }
+ for _, uuid := range uuids {
+ link := filepath.Join(udir, uuid)
+ fi, err = os.Stat(link)
+ if err != nil {
+ log.Printf("error: stat %q: %s", link, err)
+ continue
+ }
+ if fi.Sys().(*syscall.Stat_t).Ino == ino {
+ return uuid + fsPath
+ }
+ }
+ return giveup("could not find entry in %q matching %q", udir, dev)
}
// Examples implements VolumeWithExamples.
if v.DirectoryReplication == 0 {
v.DirectoryReplication = 1
}
- _, err := os.Stat(v.Root)
+ _, err := v.os.Stat(v.Root)
return err
}
return MethodDisabledError
}
p := v.blockPath(loc)
- f, err := os.OpenFile(p, os.O_RDWR|os.O_APPEND, 0644)
+ f, err := v.os.OpenFile(p, os.O_RDWR|os.O_APPEND, 0644)
if err != nil {
return err
}
defer f.Close()
- if v.locker != nil {
- v.locker.Lock()
- defer v.locker.Unlock()
+ if err := v.lock(context.TODO()); err != nil {
+ return err
}
- if e := lockfile(f); e != nil {
+ defer v.unlock()
+ if e := v.lockfile(f); e != nil {
return e
}
- defer unlockfile(f)
+ defer v.unlockfile(f)
ts := syscall.NsecToTimespec(time.Now().UnixNano())
- return syscall.UtimesNano(p, []syscall.Timespec{ts, ts})
+ v.os.stats.Tick(&v.os.stats.UtimesOps)
+ err = syscall.UtimesNano(p, []syscall.Timespec{ts, ts})
+ v.os.stats.TickErr(err)
+ return err
}
// Mtime returns the stored timestamp for the given locator.
func (v *UnixVolume) Mtime(loc string) (time.Time, error) {
p := v.blockPath(loc)
- fi, err := os.Stat(p)
+ fi, err := v.os.Stat(p)
if err != nil {
return time.Time{}, err
}
// Lock the locker (if one is in use), open the file for reading, and
// call the given function if and when the file is ready to read.
func (v *UnixVolume) getFunc(ctx context.Context, path string, fn func(io.Reader) error) error {
- if v.locker != nil {
- v.locker.Lock()
- defer v.locker.Unlock()
- }
- if ctx.Err() != nil {
- return ctx.Err()
+ if err := v.lock(ctx); err != nil {
+ return err
}
- f, err := os.Open(path)
+ defer v.unlock()
+ f, err := v.os.Open(path)
if err != nil {
return err
}
defer f.Close()
- return fn(f)
+ return fn(NewCountingReader(ioutil.NopCloser(f), v.os.stats.TickInBytes))
}
// stat is os.Stat() with some extra sanity checks.
func (v *UnixVolume) stat(path string) (os.FileInfo, error) {
- stat, err := os.Stat(path)
+ stat, err := v.os.Stat(path)
if err == nil {
if stat.Size() < 0 {
err = os.ErrInvalid
// Get retrieves a block, copies it to the given slice, and returns
// the number of bytes copied.
func (v *UnixVolume) Get(ctx context.Context, loc string, buf []byte) (int, error) {
+ return getWithPipe(ctx, loc, buf, v)
+}
+
+// ReadBlock implements BlockReader.
+func (v *UnixVolume) ReadBlock(ctx context.Context, loc string, w io.Writer) error {
path := v.blockPath(loc)
stat, err := v.stat(path)
if err != nil {
- return 0, v.translateError(err)
- }
- if stat.Size() > int64(len(buf)) {
- return 0, TooLongError
+ return v.translateError(err)
}
- var read int
- size := int(stat.Size())
- err = v.getFunc(ctx, path, func(rdr io.Reader) error {
- read, err = io.ReadFull(rdr, buf[:size])
+ return v.getFunc(ctx, path, func(rdr io.Reader) error {
+ n, err := io.Copy(w, rdr)
+ if err == nil && n != stat.Size() {
+ err = io.ErrUnexpectedEOF
+ }
return err
})
- return read, err
}
// Compare returns nil if Get(loc) would return the same content as
// returns a FullError. If the write fails due to some other error,
// that error is returned.
func (v *UnixVolume) Put(ctx context.Context, loc string, block []byte) error {
+ return putWithPipe(ctx, loc, block, v)
+}
+
+// ReadBlock implements BlockWriter.
+func (v *UnixVolume) WriteBlock(ctx context.Context, loc string, rdr io.Reader) error {
if v.ReadOnly {
return MethodDisabledError
}
return err
}
- tmpfile, tmperr := ioutil.TempFile(bdir, "tmp"+loc)
+ tmpfile, tmperr := v.os.TempFile(bdir, "tmp"+loc)
if tmperr != nil {
log.Printf("ioutil.TempFile(%s, tmp%s): %s", bdir, loc, tmperr)
return tmperr
}
+
bpath := v.blockPath(loc)
- if v.locker != nil {
- v.locker.Lock()
- defer v.locker.Unlock()
- }
- select {
- case <-ctx.Done():
- return ctx.Err()
- default:
+ if err := v.lock(ctx); err != nil {
+ return err
}
- if _, err := tmpfile.Write(block); err != nil {
+ defer v.unlock()
+ n, err := io.Copy(tmpfile, rdr)
+ v.os.stats.TickOutBytes(uint64(n))
+ if err != nil {
log.Printf("%s: writing to %s: %s\n", v, bpath, err)
tmpfile.Close()
- os.Remove(tmpfile.Name())
+ v.os.Remove(tmpfile.Name())
return err
}
if err := tmpfile.Close(); err != nil {
log.Printf("closing %s: %s\n", tmpfile.Name(), err)
- os.Remove(tmpfile.Name())
+ v.os.Remove(tmpfile.Name())
return err
}
- if err := os.Rename(tmpfile.Name(), bpath); err != nil {
+ if err := v.os.Rename(tmpfile.Name(), bpath); err != nil {
log.Printf("rename %s %s: %s\n", tmpfile.Name(), bpath, err)
- os.Remove(tmpfile.Name())
- return err
+ return v.os.Remove(tmpfile.Name())
}
return nil
}
// current state, or nil if an error occurs.
//
func (v *UnixVolume) Status() *VolumeStatus {
- var fs syscall.Statfs_t
- var devnum uint64
-
- if fi, err := os.Stat(v.Root); err == nil {
- devnum = fi.Sys().(*syscall.Stat_t).Dev
- } else {
+ fi, err := v.os.Stat(v.Root)
+ if err != nil {
log.Printf("%s: os.Stat: %s\n", v, err)
return nil
}
+ devnum := fi.Sys().(*syscall.Stat_t).Dev
- err := syscall.Statfs(v.Root, &fs)
- if err != nil {
+ var fs syscall.Statfs_t
+ if err := syscall.Statfs(v.Root, &fs); err != nil {
log.Printf("%s: statfs: %s\n", v, err)
return nil
}
//
func (v *UnixVolume) IndexTo(prefix string, w io.Writer) error {
var lastErr error
- rootdir, err := os.Open(v.Root)
+ rootdir, err := v.os.Open(v.Root)
if err != nil {
return err
}
defer rootdir.Close()
+ v.os.stats.Tick(&v.os.stats.ReaddirOps)
for {
names, err := rootdir.Readdirnames(1)
if err == io.EOF {
continue
}
blockdirpath := filepath.Join(v.Root, names[0])
- blockdir, err := os.Open(blockdirpath)
+ blockdir, err := v.os.Open(blockdirpath)
if err != nil {
log.Print("Error reading ", blockdirpath, ": ", err)
lastErr = err
continue
}
+ v.os.stats.Tick(&v.os.stats.ReaddirOps)
for {
fileInfo, err := blockdir.Readdir(1)
if err == io.EOF {
if v.ReadOnly {
return MethodDisabledError
}
- if v.locker != nil {
- v.locker.Lock()
- defer v.locker.Unlock()
+ if err := v.lock(context.TODO()); err != nil {
+ return err
}
+ defer v.unlock()
p := v.blockPath(loc)
- f, err := os.OpenFile(p, os.O_RDWR|os.O_APPEND, 0644)
+ f, err := v.os.OpenFile(p, os.O_RDWR|os.O_APPEND, 0644)
if err != nil {
return err
}
defer f.Close()
- if e := lockfile(f); e != nil {
+ if e := v.lockfile(f); e != nil {
return e
}
- defer unlockfile(f)
+ defer v.unlockfile(f)
// If the block has been PUT in the last blobSignatureTTL
// seconds, return success without removing the block. This
// protects data from garbage collection until it is no longer
// possible for clients to retrieve the unreferenced blocks
// anyway (because the permission signatures have expired).
- if fi, err := os.Stat(p); err != nil {
+ if fi, err := v.os.Stat(p); err != nil {
return err
} else if time.Since(fi.ModTime()) < time.Duration(theConfig.BlobSignatureTTL) {
return nil
}
if theConfig.TrashLifetime == 0 {
- return os.Remove(p)
+ return v.os.Remove(p)
}
- return os.Rename(p, fmt.Sprintf("%v.trash.%d", p, time.Now().Add(theConfig.TrashLifetime.Duration()).Unix()))
+ return v.os.Rename(p, fmt.Sprintf("%v.trash.%d", p, time.Now().Add(theConfig.TrashLifetime.Duration()).Unix()))
}
// Untrash moves block from trash back into store
return MethodDisabledError
}
+ v.os.stats.Tick(&v.os.stats.ReaddirOps)
files, err := ioutil.ReadDir(v.blockDir(loc))
if err != nil {
return err
for _, f := range files {
if strings.HasPrefix(f.Name(), prefix) {
foundTrash = true
- err = os.Rename(v.blockPath(f.Name()), v.blockPath(loc))
+ err = v.os.Rename(v.blockPath(f.Name()), v.blockPath(loc))
if err == nil {
break
}
return v.DirectoryReplication
}
+// InternalStats returns I/O and filesystem ops counters.
+func (v *UnixVolume) InternalStats() interface{} {
+ return &v.os.stats
+}
+
+// lock acquires the serialize lock, if one is in use. If ctx is done
+// before the lock is acquired, lock returns ctx.Err() instead of
+// acquiring the lock.
+func (v *UnixVolume) lock(ctx context.Context) error {
+ if v.locker == nil {
+ return nil
+ }
+ locked := make(chan struct{})
+ go func() {
+ v.locker.Lock()
+ close(locked)
+ }()
+ select {
+ case <-ctx.Done():
+ go func() {
+ <-locked
+ v.locker.Unlock()
+ }()
+ return ctx.Err()
+ case <-locked:
+ return nil
+ }
+}
+
+// unlock releases the serialize lock, if one is in use.
+func (v *UnixVolume) unlock() {
+ if v.locker == nil {
+ return
+ }
+ v.locker.Unlock()
+}
+
// 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)
+func (v *UnixVolume) lockfile(f *os.File) error {
+ v.os.stats.Tick(&v.os.stats.FlockOps)
+ err := syscall.Flock(int(f.Fd()), syscall.LOCK_EX)
+ v.os.stats.TickErr(err)
+ return err
}
-func unlockfile(f *os.File) error {
- return syscall.Flock(int(f.Fd()), syscall.LOCK_UN)
+func (v *UnixVolume) unlockfile(f *os.File) error {
+ err := syscall.Flock(int(f.Fd()), syscall.LOCK_UN)
+ v.os.stats.TickErr(err)
+ return err
}
// Where appropriate, translate a more specific filesystem error to an
if deadline > time.Now().Unix() {
return nil
}
- err = os.Remove(path)
+ err = v.os.Remove(path)
if err != nil {
log.Printf("EmptyTrash: Remove %v: %v", path, err)
return nil
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)
}
+
+type unixStats struct {
+ statsTicker
+ OpenOps uint64
+ StatOps uint64
+ FlockOps uint64
+ UtimesOps uint64
+ CreateOps uint64
+ RenameOps uint64
+ UnlinkOps uint64
+ ReaddirOps uint64
+}
+
+func (s *unixStats) TickErr(err error) {
+ if err == nil {
+ return
+ }
+ s.statsTicker.TickErr(err, fmt.Sprintf("%T", err))
+}
+
+type osWithStats struct {
+ stats unixStats
+}
+
+func (o *osWithStats) Open(name string) (*os.File, error) {
+ o.stats.Tick(&o.stats.OpenOps)
+ f, err := os.Open(name)
+ o.stats.TickErr(err)
+ return f, err
+}
+
+func (o *osWithStats) OpenFile(name string, flag int, perm os.FileMode) (*os.File, error) {
+ o.stats.Tick(&o.stats.OpenOps)
+ f, err := os.OpenFile(name, flag, perm)
+ o.stats.TickErr(err)
+ return f, err
+}
+
+func (o *osWithStats) Remove(path string) error {
+ o.stats.Tick(&o.stats.UnlinkOps)
+ err := os.Remove(path)
+ o.stats.TickErr(err)
+ return err
+}
+
+func (o *osWithStats) Rename(a, b string) error {
+ o.stats.Tick(&o.stats.RenameOps)
+ err := os.Rename(a, b)
+ o.stats.TickErr(err)
+ return err
+}
+
+func (o *osWithStats) Stat(path string) (os.FileInfo, error) {
+ o.stats.Tick(&o.stats.StatOps)
+ fi, err := os.Stat(path)
+ o.stats.TickErr(err)
+ return fi, err
+}
+
+func (o *osWithStats) TempFile(dir, base string) (*os.File, error) {
+ o.stats.Tick(&o.stats.CreateOps)
+ f, err := ioutil.TempFile(dir, base)
+ o.stats.TickErr(err)
+ return f, err
+}