libgocryptfs/mount.go
Jakob Unterwurzacher e36a0ebf18 main: add "-sharedstorage" flag
At the moment, it does two things:

1. Disable stat() caching so changes to the backing storage show up
   immediately.
2. Disable hard link tracking, as the inode numbers on the backing
   storage are not stable when files are deleted and re-created behind
   our back. This would otherwise produce strange "file does not exist"
   and other errors.

Mitigates https://github.com/rfjakob/gocryptfs/issues/156
2017-11-12 20:06:13 +01:00

375 lines
13 KiB
Go

package main
import (
"encoding/json"
"fmt"
"log/syslog"
"net"
"os"
"os/exec"
"os/signal"
"path"
"path/filepath"
"runtime"
"runtime/debug"
"strings"
"syscall"
"time"
"github.com/hanwen/go-fuse/fuse"
"github.com/hanwen/go-fuse/fuse/nodefs"
"github.com/hanwen/go-fuse/fuse/pathfs"
"github.com/rfjakob/gocryptfs/internal/configfile"
"github.com/rfjakob/gocryptfs/internal/cryptocore"
"github.com/rfjakob/gocryptfs/internal/ctlsock"
"github.com/rfjakob/gocryptfs/internal/exitcodes"
"github.com/rfjakob/gocryptfs/internal/fusefrontend"
"github.com/rfjakob/gocryptfs/internal/fusefrontend_reverse"
"github.com/rfjakob/gocryptfs/internal/readpassword"
"github.com/rfjakob/gocryptfs/internal/tlog"
)
// doMount mounts an encrypted directory.
// Called from main.
func doMount(args *argContainer) int {
// Check mountpoint
var err error
args.mountpoint, err = filepath.Abs(flagSet.Arg(1))
if err != nil {
tlog.Fatal.Printf("Invalid mountpoint: %v", err)
os.Exit(exitcodes.MountPoint)
}
// We cannot mount "/home/user/.cipher" at "/home/user" because the mount
// will hide ".cipher" also for us.
if args.cipherdir == args.mountpoint || strings.HasPrefix(args.cipherdir, args.mountpoint+"/") {
tlog.Fatal.Printf("Mountpoint %q would shadow cipherdir %q, this is not supported",
args.mountpoint, args.cipherdir)
os.Exit(exitcodes.MountPoint)
}
// Reverse-mounting "/foo" at "/foo/mnt" means we would be recursively
// encrypting ourselves.
if strings.HasPrefix(args.mountpoint, args.cipherdir+"/") {
tlog.Fatal.Printf("Mountpoint %q is contained in cipherdir %q, this is not supported",
args.mountpoint, args.cipherdir)
os.Exit(exitcodes.MountPoint)
}
if args.nonempty {
err = checkDir(args.mountpoint)
} else {
err = checkDirEmpty(args.mountpoint)
}
if err != nil {
tlog.Fatal.Printf("Invalid mountpoint: %v", err)
os.Exit(exitcodes.MountPoint)
}
// Open control socket early so we can error out before asking the user
// for the password
if args.ctlsock != "" {
// We must use an absolute path because we cd to / when daemonizing.
// This messes up the delete-on-close logic in the unix socket object.
args.ctlsock, _ = filepath.Abs(args.ctlsock)
var sock net.Listener
sock, err = net.Listen("unix", args.ctlsock)
if err != nil {
tlog.Fatal.Printf("ctlsock: %v", err)
os.Exit(exitcodes.CtlSock)
}
args._ctlsockFd = sock
// Close also deletes the socket file
defer func() {
err = sock.Close()
if err != nil {
tlog.Warn.Print(err)
}
}()
}
// Get master key (may prompt for the password)
var masterkey []byte
var confFile *configfile.ConfFile
if args.masterkey != "" {
// "-masterkey"
masterkey = parseMasterKey(args.masterkey)
} else if args.zerokey {
// "-zerokey"
tlog.Info.Printf("Using all-zero dummy master key.")
tlog.Info.Printf(tlog.ColorYellow +
"ZEROKEY MODE PROVIDES NO SECURITY AT ALL AND SHOULD ONLY BE USED FOR TESTING." +
tlog.ColorReset)
masterkey = make([]byte, cryptocore.KeyLen)
} else {
// Load master key from config file
// Prompts the user for the password
masterkey, confFile, err = loadConfig(args)
if err != nil {
if args._ctlsockFd != nil {
// Close the socket file (which also deletes it)
args._ctlsockFd.Close()
}
exitcodes.Exit(err)
}
readpassword.CheckTrailingGarbage()
printMasterKey(masterkey)
}
// We cannot use JSON for pretty-printing as the fields are unexported
tlog.Debug.Printf("cli args: %#v", args)
// Initialize FUSE server
srv := initFuseFrontend(masterkey, args, confFile)
tlog.Info.Println(tlog.ColorGreen + "Filesystem mounted and ready." + tlog.ColorReset)
// We have been forked into the background, as evidenced by the set
// "notifypid".
if args.notifypid > 0 {
// Chdir to the root directory so we don't block unmounting the CWD
os.Chdir("/")
// Switch to syslog
if !args.nosyslog {
// Switch all of our logs and the generic logger to syslog
tlog.Info.SwitchToSyslog(syslog.LOG_USER | syslog.LOG_INFO)
tlog.Debug.SwitchToSyslog(syslog.LOG_USER | syslog.LOG_DEBUG)
tlog.Warn.SwitchToSyslog(syslog.LOG_USER | syslog.LOG_WARNING)
tlog.SwitchLoggerToSyslog(syslog.LOG_USER | syslog.LOG_WARNING)
// Daemons should redirect stdin, stdout and stderr
redirectStdFds()
}
// Disconnect from the controlling terminal by creating a new session.
// This prevents us from getting SIGINT when the user presses Ctrl-C
// to exit a running script that has called gocryptfs.
_, err = syscall.Setsid()
if err != nil {
tlog.Warn.Printf("Setsid: %v", err)
}
// Send SIGUSR1 to our parent
sendUsr1(args.notifypid)
}
// Increase the open file limit to 4096. This is not essential, so do it after
// we have switched to syslog and don't bother the user with warnings.
setOpenFileLimit()
// Wait for SIGINT in the background and unmount ourselves if we get it.
// This prevents a dangling "Transport endpoint is not connected"
// mountpoint if the user hits CTRL-C.
handleSigint(srv, args.mountpoint)
// Return memory that was allocated for scrypt (64M by default!) and other
// stuff that is no longer needed to the OS
debug.FreeOSMemory()
// Jump into server loop. Returns when it gets an umount request from the kernel.
srv.Serve()
return 0
}
// setOpenFileLimit tries to increase the open file limit to 4096 (the default hard
// limit on Linux).
func setOpenFileLimit() {
var lim syscall.Rlimit
err := syscall.Getrlimit(syscall.RLIMIT_NOFILE, &lim)
if err != nil {
tlog.Warn.Printf("Getting RLIMIT_NOFILE failed: %v", err)
return
}
if lim.Cur >= 4096 {
return
}
lim.Cur = 4096
err = syscall.Setrlimit(syscall.RLIMIT_NOFILE, &lim)
if err != nil {
tlog.Warn.Printf("Setting RLIMIT_NOFILE to %+v failed: %v", lim, err)
// %+v output: "{Cur:4097 Max:4096}" ^
}
}
// initFuseFrontend - initialize gocryptfs/fusefrontend
// Calls os.Exit on errors
func initFuseFrontend(masterkey []byte, args *argContainer, confFile *configfile.ConfFile) *fuse.Server {
// Reconciliate CLI and config file arguments into a fusefrontend.Args struct
// that is passed to the filesystem implementation
cryptoBackend := cryptocore.BackendGoGCM
if args.openssl {
cryptoBackend = cryptocore.BackendOpenSSL
}
if args.aessiv {
cryptoBackend = cryptocore.BackendAESSIV
}
// forceOwner implies allow_other, as documented.
// Set this early, so args.allow_other can be relied on below this point.
if args._forceOwner != nil {
args.allow_other = true
}
frontendArgs := fusefrontend.Args{
Cipherdir: args.cipherdir,
PlaintextNames: args.plaintextnames,
LongNames: args.longnames,
CryptoBackend: cryptoBackend,
ConfigCustom: args._configCustom,
Raw64: args.raw64,
NoPrealloc: args.noprealloc,
HKDF: args.hkdf,
SerializeReads: args.serialize_reads,
ForceDecode: args.forcedecode,
ForceOwner: args._forceOwner,
}
// confFile is nil when "-zerokey" or "-masterkey" was used
if confFile != nil {
// Settings from the config file override command line args
frontendArgs.PlaintextNames = confFile.IsFeatureFlagSet(configfile.FlagPlaintextNames)
frontendArgs.Raw64 = confFile.IsFeatureFlagSet(configfile.FlagRaw64)
frontendArgs.HKDF = confFile.IsFeatureFlagSet(configfile.FlagHKDF)
if confFile.IsFeatureFlagSet(configfile.FlagAESSIV) {
frontendArgs.CryptoBackend = cryptocore.BackendAESSIV
} else if args.reverse {
tlog.Fatal.Printf("AES-SIV is required by reverse mode, but not enabled in the config file")
os.Exit(exitcodes.Usage)
}
}
// If allow_other is set and we run as root, try to give newly created files to
// the right user.
if args.allow_other && os.Getuid() == 0 {
frontendArgs.PreserveOwner = true
}
jsonBytes, _ := json.MarshalIndent(frontendArgs, "", "\t")
tlog.Debug.Printf("frontendArgs: %s", string(jsonBytes))
var finalFs pathfs.FileSystem
var ctlSockBackend ctlsock.Interface
// pathFsOpts are passed into go-fuse/pathfs
pathFsOpts := &pathfs.PathNodeFsOptions{ClientInodes: true}
if args.sharedstorage {
// shared storage mode disables hard link tracking as the backing inode
// numbers may change behind our back:
// https://github.com/rfjakob/gocryptfs/issues/156
pathFsOpts.ClientInodes = false
}
if args.reverse {
// The dance with the intermediate variables is because we need to
// cast the FS into pathfs.FileSystem *and* ctlsock.Interface. This
// avoids using interface{}.
fs := fusefrontend_reverse.NewFS(masterkey, frontendArgs)
finalFs = fs
ctlSockBackend = fs
// Reverse mode is read-only, so we don't need a working link().
// Disable hard link tracking to avoid strange breakage on duplicate
// inode numbers ( https://github.com/rfjakob/gocryptfs/issues/149 ).
pathFsOpts.ClientInodes = false
} else {
fs := fusefrontend.NewFS(masterkey, frontendArgs)
finalFs = fs
ctlSockBackend = fs
}
// fusefrontend / fusefrontend_reverse have initialized their crypto with
// derived keys (HKDF), we can purge the master key from memory.
for i := range masterkey {
masterkey[i] = 0
}
// We have opened the socket early so that we cannot fail here after
// asking the user for the password
if args._ctlsockFd != nil {
go ctlsock.Serve(args._ctlsockFd, ctlSockBackend)
}
pathFs := pathfs.NewPathNodeFs(finalFs, pathFsOpts)
var fuseOpts *nodefs.Options
if args.sharedstorage {
// sharedstorage mode sets all cache timeouts to zero so changes to the
// backing shared storage show up immediately.
fuseOpts = &nodefs.Options{}
} else {
fuseOpts = &nodefs.Options{
// These options are to be compatible with libfuse defaults,
// making benchmarking easier.
NegativeTimeout: time.Second,
AttrTimeout: time.Second,
EntryTimeout: time.Second,
}
}
conn := nodefs.NewFileSystemConnector(pathFs.Root(), fuseOpts)
mOpts := fuse.MountOptions{
// Writes and reads are usually capped at 128kiB on Linux through
// the FUSE_MAX_PAGES_PER_REQ kernel constant in fuse_i.h. Our
// sync.Pool buffer pools are sized acc. to the default. Users may set
// the kernel constant higher, and Synology NAS kernels are known to
// have it >128kiB. We cannot handle more than 128kiB, so we tell
// the kernel to limit the size explicitely.
MaxWrite: fuse.MAX_KERNEL_WRITE,
Options: []string{fmt.Sprintf("max_read=%d", fuse.MAX_KERNEL_WRITE)},
}
if args.allow_other {
tlog.Info.Printf(tlog.ColorYellow + "The option \"-allow_other\" is set. Make sure the file " +
"permissions protect your data from unwanted access." + tlog.ColorReset)
mOpts.AllowOther = true
// Make the kernel check the file permissions for us
mOpts.Options = append(mOpts.Options, "default_permissions")
}
if args.forcedecode {
tlog.Info.Printf(tlog.ColorYellow + "THE OPTION \"-forcedecode\" IS ACTIVE. GOCRYPTFS WILL RETURN CORRUPT DATA!" +
tlog.ColorReset)
}
if args.nonempty {
mOpts.Options = append(mOpts.Options, "nonempty")
}
// Set values shown in "df -T" and friends
// First column, "Filesystem"
fsname := args.cipherdir
if args.fsname != "" {
fsname = args.fsname
}
mOpts.Options = append(mOpts.Options, "fsname="+fsname)
// Second column, "Type", will be shown as "fuse." + Name
mOpts.Name = "gocryptfs"
if args.reverse {
mOpts.Name += "-reverse"
}
// Add a volume name if running osxfuse. Otherwise the Finder will show it as
// something like "osxfuse Volume 0 (gocryptfs)".
if runtime.GOOS == "darwin" {
mOpts.Options = append(mOpts.Options, "volname="+path.Base(args.mountpoint))
}
// The kernel enforces read-only operation, we just have to pass "ro".
// Reverse mounts are always read-only.
if args.ro || args.reverse {
mOpts.Options = append(mOpts.Options, "ro")
}
// Add additional mount options (if any) after the stock ones, so the user has
// a chance to override them.
if args.ko != "" {
parts := strings.Split(args.ko, ",")
tlog.Debug.Printf("Adding -ko mount options: %v", parts)
mOpts.Options = append(mOpts.Options, parts...)
}
srv, err := fuse.NewServer(conn.RawFS(), args.mountpoint, &mOpts)
if err != nil {
tlog.Fatal.Printf("fuse.NewServer failed: %v", err)
if runtime.GOOS == "darwin" {
tlog.Info.Printf("Maybe you should run: /Library/Filesystems/osxfuse.fs/Contents/Resources/load_osxfuse")
}
os.Exit(exitcodes.FuseNewServer)
}
srv.SetDebug(args.fusedebug)
// All FUSE file and directory create calls carry explicit permission
// information. We need an unrestricted umask to create the files and
// directories with the requested permissions.
syscall.Umask(0000)
return srv
}
func handleSigint(srv *fuse.Server, mountpoint string) {
ch := make(chan os.Signal, 1)
signal.Notify(ch, os.Interrupt)
signal.Notify(ch, syscall.SIGTERM)
go func() {
<-ch
err := srv.Unmount()
if err != nil {
tlog.Warn.Print(err)
if runtime.GOOS == "linux" {
// MacOSX does not support lazy unmount
tlog.Info.Printf("Trying lazy unmount")
cmd := exec.Command("fusermount", "-u", "-z", mountpoint)
cmd.Stdout = os.Stdout
cmd.Stderr = os.Stderr
cmd.Run()
}
}
os.Exit(exitcodes.SigInt)
}()
}