encfs/encfs/encfs.pod

=pod

=cut

Copyright (c) 2003-2008, Valient Gough <vgough@pobox.com>
All rights reserved.

EncFS is free software; you can distribute it and/or modify it under the terms
of the GNU General Public License (GPL), as published by the Free Software
Foundation; either version 3 of the License, or (at your option) any later
version.

=head1 NAME

encfs - mounts or creates an encrypted virtual filesystem

=head1 SYNOPSIS

B<encfs> [B<--version>] [B<-s>] [B<-f>] [B<-v>|B<--verbose>] 
[B<-i MINUTES>|B<--idle=MINUTES>] [B<--extpass=program>] 
[B<-S>|B<--stdinpass>] [B<--anykey>] [B<--forcedecode>] 
[B<-d>|B<--fuse-debug>] [B<--public>] [B<--no-default-flags>]
[B<--ondemand>] [B<--delaymount>] [B<--reverse>] [B<--standard>] 
[B<-o FUSE_OPTION>]
I<rootdir> I<mountPoint> 
[B<--> [I<Fuse Mount Options>]]

=head1 DESCRIPTION

B<EncFS> creates a virtual encrypted filesystem which stores encrypted data in
the I<rootdir> directory and makes the unencrypted data visible at the
I<mountPoint> directory.  The user must supply a password which is used to
(indirectly) encrypt both filenames and file contents.

If B<EncFS> is unable to find a supported filesystem at the specified
I<rootdir>, then the user will be asked if they wish to create a new encrypted
filesystem at the specified location.  Options will be presented to the user
allowing some control over the algorithms to use.  As B<EncFS> matures, there
may be an increasing number of choices.

=head1 OPTIONS

=over 4

=item B<-i>, B<--idle=MINUTES>

Enable automatic unmount of the filesystem after a period of inactivity.  The
period is specified in minutes, so the shortest timeout period that can be
requested is one minute.  B<EncFS> will not automatically unmount if there are
files open within the filesystem, even if they are open in read-only mode.
However simply having files open does not count as activity.

=item B<-f>

The B<-f> (I<foreground>) option causes B<EncFS> to run in the foreground.
Normally B<EncFS> spawns off as a daemon and runs in the background, returning
control to the spawning shell.  With the B<-f> option, it will run in the
foreground and any warning/debug log messages will be displayed on standard
error.  In the default (background) mode, all log messages are logged via
syslog.

=item B<-v>, B<--verbose>

Causes B<EncFS> to enable logging of various debug channels within B<EncFS>.
Normally these logging messages are disabled and have no effect.  It is
recommended that you run in foreground (B<-f>) mode when running with verbose
enabled.

=item B<-s>

The B<-s> (I<single threaded>) option causes B<EncFS> to run in single threaded
mode.  By default, B<EncFS> runs in multi-threaded mode.  This option is used
during B<EncFS> development in order to simplify debugging and allow it to run
under memory checking tools.

=item B<-d>, B<--fuse-debug>

Enables debugging within the B<FUSE> library.  This should only be used if you
suspect a problem within B<FUSE> itself (not B<EncFS>), as it generates a lot
of low-level data and is not likely to be very helpful in general problem
tracking.  Try I<verbose> mode (B<-v>) first, which gives a higher level view
of what is happening within B<EncFS>.

=item B<--forcedecode>

This option only has an effect on filesystems which use MAC block headers.  By
default, if a block is decoded and the stored MAC doesn't match what is
calculated, then an IO error is returned to the application and the block is
not returned.  However, by specifying B<--forcedecode>, only an error will be
logged and the data will still be returned to the application.  This may be
useful for attempting to read corrupted files.

=item B<--public>

Attempt to make encfs behave as a typical multi-user filesystem.  By default,
all FUSE based filesystems are visible only to the user who mounted them.  No
other users (including root) can view the filesystem contents.  The B<--public>
option does two things.  It adds the FUSE flags "allow_other" and
"default_permission" when mounting the filesystem, which tells FUSE to allow
other users to access the filesystem, and to use the ownership permissions
provided by the filesystem.  Secondly, the B<--public> flag changes how encfs's
node creation functions work - as they will try and set ownership of new nodes
based on the caller identification.

B<Warning>: In order for this to work, encfs must be run as root -- otherwise
it will not have the ability to change ownership of files.  I recommend that
you instead investigate if the fuse allow_other option can be used to do what
you want before considering the use of B<--public>.

=item B<--ondemand>

Mount the filesystem on-demand.  This currently only makes sense in combination
with B<--idle> and B<--extpass> options.  When the filesystem becomes idle,
instead of exiting, B<EncFS> stops allowing access to the filesystem by
internally dropping its reference to it.  If someone attempts to access the
filesystem again, the extpass program is used to prompt the user for the
password.  If this succeeds, then the filesystem becomes available again.

=item B<--delaymount>

Do not mount the filesystem when encfs starts; instead, delay mounting until
first use. This option only makes sense with B<--ondemand>.

=item B<--require-macs>

If creating a new filesystem, this forces block authentication code headers to
be enabled.  When mounting an existing filesystem, this causes encfs to exit
if block authentication code headers are not enabled.

This can be used to improve security in case the ciphertext is vulnerable to
tampering, by preventing an attacker from disabling MACs in the config file.

=item B<--reverse>

Normally B<EncFS> provides a plaintext view of data on demand.  Normally it
stores enciphered data and displays plaintext data.  With B<--reverse> it takes
as source plaintext data and produces enciphered data on-demand.  This can be
useful for creating remote encrypted backups, where you do not wish to keep the
local files unencrypted.

For example, the following would create an encrypted view in /tmp/crypt-view.

    encfs --reverse /home/me /tmp/crypt-view

You could then copy the /tmp/crypt-view directory in order to have a copy of
the encrypted data.  You must also keep a copy of the file /home/me/.encfs6.xml
which contains the filesystem information.  Together, the two can be used to
reproduce the unencrypted data:

    ENCFS6_CONFIG=/home/me/.encfs6.xml encfs /tmp/crypt-view /tmp/plain-view

Now /tmp/plain-view contains the same data as /home/me

Note that B<--reverse> mode only works with limited configuration options, so
many settings may be disabled when used.

=item B<--nocache>

Disable the kernel's cache of file attributes.
Setting this option makes EncFS pass "attr_timeout=0" and "entry_timeout=0" to
FUSE. This makes sure that modifications to the backing files that occour
outside EncFS show up immediately in the EncFS mount. The main use case
for "--nocache" is reverse mode.

=item B<--standard>

If creating a new filesystem, this automatically selects standard configuration
options, to help with automatic filesystem creation.  This is the set of
options that should be used unless you know what you're doing and have read the
documentation.

When not creating a filesystem, this flag does nothing.

=item B<-o FUSE_ARG>

Pass through B<FUSE> args to the underlying library.  This makes it easy to
pass FUSE options when mounting B<EncFS> via mount (and /etc/fstab).  Eg:

    mount encfs#/home/me-crypt /home/me -t fuse -o kernel_cache

Note that encfs arguments cannot be set this way.  If you need to set encfs
arguments, create a wrapper, such as  encfs-reverse;

    #!/bin/sh
    encfs --reverse $*

Then mount using the script path

    mount encfs-reverse#/home/me /home/me-crypt -t fuse

=item B<-->

The B<--> option tells B<EncFS> to send any remaining arguments directly to
B<FUSE>.  In turn, B<FUSE> passes the arguments to B<fusermount>.  See
the B<fusermount> help page for information on available commands.

=item B<--no-default-flags>

B<Encfs> adds the FUSE flags "use_ino" and "default_permissions" by default, as
of version 1.2.2, because that improves compatibility with some programs.  If
for some reason you need to disable one or both of these flags, use the option
B<--no-default-flags>.

The following command lines produce the same result:

    encfs raw crypt
    encfs --no-default-flags raw crypt -- -o use_ino,default_permissions

=item B<--extpass=program>

Specify an external program to use for getting the user password.  When the
external program is spawned, the environment variable "RootDir" will be set to
contain the path to the root directory.  The program should print the password
to standard output.

B<EncFS> takes everything returned from the program to be the password, except
for a trailing newline (\n) which will be removed.

For example, specifying B<--extpass>=I</usr/lib/ssh/ssh-askpass> will cause
B<EncFS> to use ssh's password prompt program.  

B<Note>: B<EncFS> reads at most 2k of data from the password program, and it
removes any trailing newline.  Versions before 1.4.x accepted only 64 bytes of
text.

=item B<-S>, B<--stdinpass>

Read password from standard input, without prompting.  This may be useful for
scripting encfs mounts.

Note that you should make sure the filesystem and mount points exist first.
Otherwise encfs will prompt for the filesystem creation options, which may
interfere with your script.

=item B<--anykey>

Turn off key validation checking.  This allows B<EncFS> to be used with
secondary passwords.  This could be used to store a separate set of files in an
encrypted filesystem.  B<EncFS> ignores files which do not decode properly, so
files created with separate passwords will only be visible when the filesystem
is mounted with their associated password.

Note that if the primary password is changed (using B<encfsctl>), the other
passwords will not be usable unless the primary password is set back to what it
was, as the other passwords rely on an invalid decoding of the volume key,
which will not remain the same if the primary password is changed.

B<Warning>: Use this option at your own risk.

=back

=head1 ENVIRONMENT VARIABLES

=over 4

=item B<ENCFS6_CONFIG>

Which config file (typically named .encfs6.xml) to use.
By default, the config file is read from the encrypted directory.
Using this option allows to store the config file separated from the
encrypted files.

Warning: If you lose the config file, the encrypted file contents are
irrecoverably lost. It contains the master key encrypted with your
password. Without the master key, recovery is impossible, even if you
know the password.

=back

=head1 EXAMPLES

Create a new encrypted filesystem.  Store the raw (encrypted) data in
"~/.crypt" , and make the unencrypted data visible in "~/crypt".  Both
directories are in the home directory in this example.  This example shows the
full output of encfs as it asks the user if they wish to create the filesystem:

    % encfs ~/.crypt ~/crypt
    Directory "/home/me/.crypt" does not exist, create (y,n)?y
    Directory "/home/me/crypt" does not exist, create (y,n)?y
    Creating new encrypted volume.
    Please choose from one of the following options:
     enter "x" for expert configuration mode,
     enter "p" for pre-configured paranoia mode,
     anything else, or an empty line will select standard mode.
    ?> 

    Standard configuration selected.
    Using cipher Blowfish, key size 160, block size 512
    New Password: <password entered here>
    Verify: <password entered here>

The filesystem is now mounted and visible in I<~/crypt>.  If files are created
there, they can be seen in encrypted form in I<~/.crypt>.  To unmount the
filesystem, use I<fusermount> with the B<-u> (unmount) option:

    % fusermount -u ~/crypt

Another example.  To mount the same filesystem, but have fusermount name the
mount point '/dev/foo' (as shown in I<df> and other tools which read
/etc/mtab), and also request kernel-level caching of file data (which are both
special arguments to fusermount):

    % encfs ~/.crypt ~/crypt -- -n /dev/foo -c

Or, if you find strange behavior under some particular program when working in
an encrypted filesystem, it may be helpful to run in verbose mode while
reproducing the problem and send along the output with the problem report:

    % encfs -v -f ~/.crypt ~/crypt 2> encfs-report.txt

In order to avoid leaking sensitive information through the debugging channels,
all warnings and debug messages (as output in verbose mode) contain only
encrypted filenames.  You can use the I<encfsctl> program's I<decode> function
to decode filenames if desired.

=head1 CAVEATS

B<EncFS> is not a true filesystem.  It does not deal with any of the actual
storage or maintenance of files.  It simply translates requests (encrypting or
decrypting as necessary) and passes the requests through to the underlying
host filesystem.  Therefore any limitations of the host filesystem will be
inherited by B<EncFS> (or possibly be further limited).

One such limitation is filename length.  If your underlying filesystem limits
you to N characters in a filename, then B<EncFS> will limit you to approximately
3*(N-2)/4.  For example if the host filesystem limits to 256 characters, then
B<EncFS> will be limited to 190 character filenames.  This is because encrypted
filenames are always longer than plaintext filenames.

=head1 FILESYSTEM OPTIONS

When B<EncFS> is given a root directory which does not contain an existing
B<EncFS> filesystem, it will give the option to create one.  Note that options
can only be set at filesystem creation time.  There is no support for modifying
a filesystem's options in-place.  

If you want to upgrade a filesystem to use newer features, then you need to
create a new filesystem and mount both the old filesystem and new filesystem at
the same time and copy the old to the new.

Multiple instances of encfs can be run at the same time, including different
versions of encfs, as long as they are compatible with the current FUSE module
on your system.

A choice is provided for two pre-configured settings ('standard' and
'paranoia'), along with an expert configuration mode.

I<Standard> mode uses the following settings:
    Cipher: AES
    Key Size: 192 bits
    PBKDF2 with 1/2 second runtime, 160 bit salt
    Filesystem Block Size: 1024 bytes
    Filename Encoding: Block encoding with IV chaining
    Unique initialization vector file headers
    File holes passed through

I<Paranoia> mode uses the following settings:
    Cipher: AES
    Key Size: 256 bits
    PBKDF2 with 3 second runtime, 160 bit salt
    Filesystem Block Size: 1024 bytes
    Filename Encoding: Block encoding with IV chaining
    Unique initialization vector file headers
    Message Authentication Code block headers
    External IV Chaining
    File holes passed through

In the expert / manual configuration mode, each of the above options is
configurable.  Here is a list of current options with some notes about what
they mean:

=head1 Key Derivation Function

As of version 1.5, B<EncFS> now uses PBKDF2 as the default key derivation
function.  The number of iterations in the keying function is selected based on
wall clock time to generate the key.  In standard mode, a target time of 0.5
seconds is used, and in paranoia mode a target of 3.0 seconds is used.

On a 1.6Ghz AMD 64 system, roughly 64k iterations of the key derivation
function can be handled in half a second.  The exact number of iterations to
use is stored in the configuration file, as it is needed to remount the
filesystem.

If an B<EncFS> filesystem configuration from 1.4.x is modified with version 1.5
(such as when using encfsctl to change the password), then the new PBKDF2
function will be used and the filesystem will no longer be readable by older
versions.

=over 4

=item I<Cipher>

Which encryption algorithm to use.  The list is generated automatically based
on what supported algorithms B<EncFS> found in the encryption libraries.
When using a recent version of B<OpenSSL>, Blowfish and AES are the typical
options.

Blowfish is an 8 byte cipher - encoding 8 bytes at a time.  AES is a 16 byte
cipher.

=item I<Cipher Key Size>

Many, if not all, of the supported ciphers support multiple key lengths.  There
is not really much need to have enormous key lengths.  Even 160 bits (the
default) is probably overkill.

=item I<Filesystem Block Size>

This is the size (in bytes) that B<EncFS> deals with at one time.  Each block
gets its own initialization vector and is encoded in the cipher's
cipher-block-chaining mode.  A partial block at the end of a file is encoded
using a stream mode to avoid having to store the filesize somewhere.

Having larger block sizes reduces the overhead of B<EncFS> a little, but it can
also add overhead if your programs read small parts of files.  In order to read
a single byte from a file, the entire block that contains that byte must be
read and decoded, so a large block size adds overhead to small requests.  With
write calls it is even worse, as a block must be read and decoded, the change
applied and the block encoded and written back out.

The default is 512 bytes as of version 1.0.  It was hard coded to 64 bytes in
version 0.x, which was not as efficient as the current setting for general
usage.

=item I<Filename Encoding>

B<New in 1.1>. A choice is given between stream encoding of filename and block
encoding.  The advantage of stream encoding is that the encoded filenames will
be as short as possible.  If you have a filename with a single letter, it will
be very short in the encoded form, where as block encoded filenames are always
rounded up to the block size of the encryption cipher (8 bytes for Blowfish and
16 bytes for AES).

The advantage of block encoding mode is that filename lenths all come out as a
multiple of the cipher block size.  This means that someone looking at your
encrypted data can't tell as much about the length of your filenames.  It is
on by default, as it takes a similar amount of time to using the stream cipher.
However stream cipher mode may be useful if you want shorter encrypted
filenames for some reason.

Prior to version 1.1, only stream encoding was supported.

=item I<Filename Initialization Vector Chaining>

B<New in 1.1>.  In previous versions of B<EncFS>, each filename element in
a path was encoded separately.  So if "foo" encoded to "XXX", then it would
always encode that way (given the same encryption key), no matter if the path
was "a/b/foo", or "aa/foo/cc", etc.  That meant it was possible for someone
looking at the encrypted data to see if two files in different directories had
the same name, even though they wouldn't know what that name decoded to.

With initialization vector chaining, each directory gets its own initialization
vector.  So "a/foo" and "b/foo" will have completely different encoded names
for "foo".  This features has almost no performance impact (for most
operations), and so is the default in all modes.

B<Note:> One significant performance exception is directory renames.  Since the
initialization vector for filename encoding depends on the directory path, any
rename requires re-encoding every filename in the tree of the directory being
changed.  If there are thousands of files, then EncFS will have to do thousands
of renames.  It may also be possible that EncFS will come across a file that it
can't decode or doesn't have permission to move during the rename operation, in
which case it will attempt to undo any changes it made up to that point and the
rename will fail.

=item I<Per-File Initialization Vectors>

B<New in 1.1>.  In previous versions of B<EncFS>, each file was encoded in the
same way.  Each block in a file has always had its own initialization vector,
but in a deterministic way so that block N in one file is encoded in the same
was as block N in another file.  That made it possible for someone to tell if
two files were identical (or parts of the file were identical) by comparing the
encoded data.

With per-file initialization vectors, each file gets its own 64bit random
initialization vector, so that each file is encrypted in a different way.

This option is enabled by default.

=item I<External IV Chaining>

B<New in 1.1.3>.  This option is closely related to Per-File Initialization
Vectors and Filename Initialization Vector Chaining.  Basically it extends the
initialization vector chaining from filenames to the per-file initialization
vector.

When this option is enabled, the per-file initialization vector is encoded
using the initialization vector derived from the filename initialization vector
chaining code.  This means that the data in a file becomes tied to the
filename.  If an encrypted file is renamed outside of encfs, it will no longer
be decodable within encfs.  Note that unless Block MAC headers are enabled, the
decoding error will not be detected and will result in reading random looking
data.

There is a cost associated with this.  When External IV Chaining is enabled,
hard links will not be allowed within the filesystem, as there would be no way
to properly decode two different filenames pointing to the same data.

Also, renaming a file requires modifying the file header.  So renames will only
be allowed when the user has write access to the file.

Because of these limits, this option is disabled by default for standard mode
(and enabled by default for paranoia mode).

=item I<Block MAC headers>

B<New to 1.1>.  If this is enabled, every block in every file is stored along
with a cryptographic checksum (Message Authentication Code).  This makes it
virtually impossible to modify a file without the change being detected by
B<EncFS>.  B<EncFS> will refuse to read data which does not pass the checksum,
and will log the error and return an IO error to the application.

This adds substantial overhead (default being 8 bytes per filesystem block),
plus computational overhead, and is not enabled by default except in paranoia
mode.

When this is not enabled and if B<EncFS> is asked to read modified or corrupted
data, it will have no way to verify that the decoded data is what was
originally encoded.

=item I<File-hole pass-through>

Make encfs leave holes in files.  If a block is read as all zeros, it will be
assumed to be a hole and will be left as 0's when read (not deciphered).  This
is required if accessing encfs using the SMB protocol.

Enabled by default.  Can be disabled in expert mode.

=back

=head1 Attacks

The primary goal of B<EncFS> is to protect data off-line.  That is, provide a
convenient way of storing files in a way that will frustrate any attempt to
read them if the files are later intercepted.

Some algorithms in B<EncFS> are also meant to frustrate on-line attacks where
an attacker is assumed to be able to modify the files.

The most intrusive attacks, where an attacker has complete control of the
user's machine (and can therefore modify B<EncFS>, or B<FUSE>, or the kernel
itself) are not guarded against.  Do not assume that encrypted files will
protect your sensitive data if you enter your password into a compromised
computer.  How you determine that the computer is safe to use is beyond the
scope of this documentation.

That said, here are some example attacks and data gathering techniques on the
filesystem contents along with the algorithms B<EncFS> supports to thwart them:

=over 4

=item B<Attack>: modifying a few bytes of an encrypted file (without knowing
what they will decode to).

B<EncFS> does not use any form of XOR encryption which would allow
single bytes to be modified without affecting others.  Most modifications
would affect dozens or more bytes.  Additionally, MAC Block headers can be
used to identify any changes to files.

=item B<Attack>: copying a random block of one file to a random block of another file.

Each block has its own [deterministic] initialization vector.

=item B<Attack>: copying block N to block N of another file.

When the Per-File Initialization Vector support is enabled (default
in 1.1.x filesystems), a copied block will not decode properly when copied to
another file.

=item B<Attack>: copying an entire file to another file.

Can be prevented by enabling External IV Chaining mode.

=item B<Attack>: determine if two filenames are the same by looking at encrypted names.

Filename Initialization Vector chaining prevents this by giving each file a
64-bit initialization vector derived from its full path name.

=item B<Attack>: compare if two files contain the same data.

Per-File Initialization Vector support prevents this.

=back

=head1 DISCLAIMER

This library is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.  Please refer to the "COPYING" file distributed with
B<EncFS> for complete details.

=head1 AUTHORS

B<EncFS> was written by B<< Valient Gough <vgough@pobox.com> >>.

=head1 SEE ALSO

encfsctl(1)