path: root/glep-0074.rst

---
GLEP: 74
Title: Full-tree verification using Manifest files
Author: Michał Górny <mgorny@gentoo.org>,
        Robin Hugh Johnson <robbat2@gentoo.org>,
        Ulrich Müller <ulm@gentoo.org>
Type: Standards Track
Status: Final
Version: 1.1
Created: 2017-10-21
Last-Modified: 2018-03-11
Post-History: 2017-10-26, 2017-11-16, 2018-02-08
Content-Type: text/x-rst
Requires: 59, 61
Replaces: 44, 58, 60
---

Abstract
========

This GLEP extends the Manifest file format to cover full-tree file
integrity and authenticity checks. The format aims to be future-proof,
efficient and provide means of backwards compatibility.


Motivation
==========

The Manifest files as defined by GLEP 44 [#GLEP44]_ provide the current
means of verifying the integrity of distfiles and package files
in Gentoo. Combined with OpenPGP signatures, they provide means to
ensure the authenticity of the covered files. However, as noted
in GLEP 57 [#GLEP57]_ they lack the ability to provide full-tree
authenticity verification as they do not cover any files outside
the package directory. In particular, they provide multiple ways
for a third party to inject malicious code into the ebuild environment.

Historically, the topic of providing authenticity coverage for the whole
repository has been mentioned multiple times. The most noteworthy effort
are GLEPs 58 [#GLEP58]_ and 60 [#GLEP60]_ by Robin H. Johnson from 2008.
They were accepted by the Council in 2010 but have never been
implemented. When potential implementation work started in 2017, a new
discussion about the specification arose. It prompted the creation
of a competing GLEP that would provide a redesigned alternative to
the old GLEPs.

This specification is designed with the following goals in mind:

1. It should provide means to ensure the authenticity of the complete
   repository, including preventing the injection of additional files.

2. The format should be universal enough to work both for the Gentoo
   repository and third-party repositories of different characteristics.

3. The Manifest files should be verifiable stand-alone, that is without
   knowing any details about the underlying repository format.


Specification
=============

Manifest file format
--------------------

This specification reuses and extends the Manifest file format defined
in GLEP 44 [#GLEP44]_. For the purpose of it, the *file type* field is
repurposed as a generic *tag* that could also indicate additional
(non-checksum) metadata. Appropriately, those tags can be followed by
other space-separated values.

Unless specified otherwise, the paths used in the Manifest files
are relative to the directory containing the Manifest file. The paths
must not reference the parent directory (``..``). Forward slash (``/``)
is used as path component separator.

The Manifest files use UTF-8 encoding.


Manifest file locations and nesting
-----------------------------------

The ``Manifest`` file located in the root directory of the repository
is called top-level Manifest, and it is used to perform the full-tree
verification. In order to verify the authenticity, it must be signed
using OpenPGP, using the armored cleartext format.

The top-level Manifest may reference sub-Manifests contained
in subdirectories of the repository. The sub-Manifests are traditionally
named ``Manifest``; however, the implementation must support arbitrary
names, including the possibility of multiple (split) Manifests
for a single directory. The sub-Manifest can only cover the files inside
the directory tree where it resides.

The sub-Manifest can also be signed using OpenPGP armored cleartext
format. However, the signature verification can be omitted since it
already is covered by the signed top-level Manifest.


Directory tree coverage
-----------------------

The specification provides three ways of skipping Manifest verification
of specific files and directories (recursively):

1. explicit ``IGNORE`` entries in Manifest files,

2. injected ignore paths via package manager configuration,

3. using names starting with a dot (``.``) which are always skipped.

All files that are not ignored must be covered by at least one
of the Manifests.

A single file may be matched by multiple identical or equivalent
Manifest entries, if and only if the entries have the same semantics,
specify the same size and the checksums common to both entries match.
It is an error for a single file to be matched by multiple entries
of different semantics, file size or checksum values. It is an error
to specify another entry for a file that matches ``IGNORE``, or that
is located inside an ignored directory.

The file entries (except for ``IGNORE``) can be specified for regular
files only. Symbolic links are followed when opening files
and traversing directories. It is an error to specify an entry for
a different file type. If the tree contain files of other types
that are not otherwise ignored, they need to be covered by an explicit
``IGNORE``.


Path and filename encoding
--------------------------

The path fields in the Manifest file must consist of Unicode characters
excluding the backwards slash (``\``) and characters classified
as control characters or as whitespace in the current version
of the Unicode standard [#UNICODE]_.

The implementation can optionally support extended filename encoding
to support those paths. If encoding is not supported, the implementation
must reject directories containing any files using non-compliant names,
as well as Manifest files whose filename field contains such filenames.

If encoding is supported, then all of the excluded characters that
are present in paths must be encoded using one of the following escape
sequences:

- characters in the ``U+0000`` to ``U+007F`` range can be encoded
  as ``\xHH`` where ``HH`` specifies the zero-padded, hexadecimal
  character code,

- characters in the ``U+0000`` to ``U+FFFF`` range can be encoded
  as ``\uHHHH`` where ``HHHH`` specifies the zero-padded, hexadecimal
  character code,

- characters in the UCS-4 range can be encoded as ``\UHHHHHHHH``
  where ``HHHHHHHH`` specifies the zero-padded, hexadecimal character
  code.

It is invalid for the backwards slash to be used in any other context,
and a backwards slash present in filename must be encoded. A backwards
slash used as a path component separator should be replaced by a forward
slash instead.

The encoding can be used for other characters as well. In particular,
escaping non-printable characters might be desirable.


File verification
-----------------

When verifying a file against the Manifest, the following rules are
used:

1. If the file is covered directly or indirectly by an entry
   of the ``IGNORE`` type, the verification always succeeds.

2. If the file is covered by an entry of the ``MANIFEST``, ``DATA``,
   ``MISC``, ``EBUILD`` or ``AUX`` type:

   a. if the file is not present, then the verification fails,

   b. if the file is present but has a different size or one
      of the checksums does not match, the verification fails,

   c. otherwise, the verification succeeds.

3. If the file is present but not listed in Manifest, the verification
   fails.

Unless specified otherwise, the package manager must not allow using
any files for which the verification failed. The package manager may
reject any package or even the whole repository if it may refer to files
for which the verification failed.


Timestamp verification
----------------------

The top-level Manifest file can contain a ``TIMESTAMP`` entry to account
for attacks against tree update distribution. If such an entry
is present, it should be updated every time at least one
of the Manifests changes. Every unique timestamp value must correspond
to a single tree state.

During the verification process, the client should compare the timestamp
against the update time obtained from a local clock or a trusted time
source. If the comparison result indicates that the Manifest at the time
of receiving was already significantly outdated, the client should
either fail the verification or require manual confirmation from
the user.

Furthermore, the Manifest provider may employ additional methods
of distributing the timestamps of recently generated Manifests
using a secure channel from a trusted source for exact comparison.
The exact details of such a solution are outside the scope of this
specification.

``TIMESTAMP`` entries may also be present in sub-Manifests. Those
timestamps must not be newer than the timestamp of the top-level
Manifest (if present). This specification does not define any specific
use for them.


Modern Manifest tags
--------------------

The Manifest files can specify the following tags:

``TIMESTAMP <iso8601>``
  Specifies a timestamp of when the Manifest file was last updated.
  The timestamp must be a valid second-precision ISO 8601 extended
  format combined date and time in UTC timezone, i.e. using
  the following ``strftime()`` format string: ``%Y-%m-%dT%H:%M:%SZ``.
  Optional. The package manager can use it to detect an outdated
  repository checkout as described in `Timestamp verification`_.

``MANIFEST <path> <size> <checksums>...``
  Specifies a sub-Manifest. The sub-Manifest must be verified like
  a regular file. If the verification succeeds, the entries from
  the sub-Manifest are included for verification as described
  in `Manifest file locations and nesting`_.

``IGNORE <path>``
  Ignores a subdirectory or file from Manifest checks. If the specified
  path is present, it and its contents are omitted from the Manifest
  verification (always pass). *Path* must be a plain file or directory
  path without a trailing slash. Wildcards are not supported
  and wildcard characters are interpreted literally.

``DATA <path> <size> <checksums>...``
  Specifies a regular file subject to Manifest verification. The file
  is required to pass verification. Used for all files that do not match
  any other type.

``DIST <filename> <size> <checksums>...``
  Specifies a distfile entry used to verify files fetched as part
  of ``SRC_URI``. The filename must match the filename used to store
  the fetched file as specified in the PMS [#PMS-FETCH]_. The package
  manager must reject the fetched file if it fails verification.
  ``DIST`` entries apply to all packages below the Manifest file
  specifying them.


Deprecated Manifest tags
------------------------

For backwards compatibility, the following tags are additionally
allowed at the package directory level:

``EBUILD <filename> <size> <checksums>...``
  Equivalent to the ``DATA`` type.

``MISC <path> <size> <checksums>...``
  Equivalent to the ``DATA`` type. Historically indicated that
  the package manager may ignore a verification failure if operating
  in non-strict mode. However, that behavior is deprecated.

``AUX <filename> <size> <checksums>...``
  Equivalent to the ``DATA`` type, except that the filename is relative
  to the ``files/`` subdirectory.


Algorithm for full-tree verification
------------------------------------

In order to perform full-tree verification, the following algorithm
can be used:

1. Collect all files present in the repository into *present* set.

2. Start at the top-level Manifest file. Verify its OpenPGP signature.
   Optionally verify the ``TIMESTAMP`` entry if present as specified
   in `timestamp verification`. Remove the top-level Manifest
   from the *present* set.

3. Process all ``MANIFEST`` entries, recursively. Verify the Manifest
   files according to the `file verification`_ section, and include
   their entries in the current Manifest entry list (using paths
   relative to directories containing the Manifests).

4. Process all ``IGNORE`` entries. Remove any paths matching them
   from the *present* set.

5. Collect all files covered by ``DATA``, ``MISC``, ``EBUILD``
   and ``AUX`` entries into the *covered* set.

6. Verify the entries in the *covered* set for incompatible duplicates
   and collisions with ignored files as explained in `Manifest file
   locations and nesting`_.

7. Verify all the files in the union of the *present* and *covered*
   sets, according to the `file verification`_ section.


Algorithm for finding parent Manifests
--------------------------------------

In order to find the top-level Manifest from the current directory
the following algorithm can be used:

1. Store the current directory as *original*,

2. If the current directory contains a ``Manifest`` file:

   a. If an ``IGNORE`` entry in the ``Manifest`` file covers
      the *original* directory (or one of the parent directories), stop.

   b. Otherwise, store the current directory as *last_found*.

3. If the current directory is the root system directory (``/``), stop.

4. Otherwise, enter the parent directory and jump to step 2.

Once the algorithm stops, *last_found* will contain the relevant
top-level Manifest. If *last_found* is null, then the directory tree
does not contain any valid top-level Manifest candidates and one should
be created in the *original* directory.

Once the top-level Manifest is found, its ``MANIFEST`` entries should
be used to find any sub-Manifests below the top-level Manifest,
up to and including the *original* directory. Note that those
sub-Manifests can use different filenames than ``Manifest``.


Checksum algorithms (informational)
-----------------------------------

This section is informational only. Specifying the exact set
of supported algorithms is outside the scope of this specification.

The algorithm names reserved at the time of writing are:

- ``MD5`` [#MD5]_,
- ``RMD160`` -- RIPEMD-160 [#RIPEMD160]_,
- ``SHA1`` [#SHS]_,
- ``SHA256`` and ``SHA512`` -- SHA-2 family of hashes [#SHS]_,
- ``WHIRLPOOL`` [#WHIRLPOOL]_,
- ``BLAKE2B`` and ``BLAKE2S`` -- BLAKE2 family of hashes [#BLAKE2]_,
- ``SHA3_256`` and ``SHA3_512`` -- SHA-3 family of hashes [#SHA3]_,
- ``STREEBOG256`` and ``STREEBOG512`` -- Streebog family of hashes
  [#STREEBOG]_.

The method of introducing new hashes is defined by GLEP 59 [#GLEP59]_.
It is recommended that any new hashes are named after the Python
``hashlib`` module algorithm names, transformed into uppercase.


Manifest compression
--------------------

The topic of Manifest file compression is covered by GLEP 61 [#GLEP61]_.
This section merely addresses interoperability issues between Manifest
compression and this specification.

The compressed Manifest files are required to be suffixed for their
compression algorithm. This suffix should be used to recognize
the compression and decompress Manifests transparently. The exact list
of algorithms and their corresponding suffixes are outside the scope
of this specification.

The top-level Manifest file must not be compressed. Since the OpenPGP
signature covers the uncompressed text and is compressed itself,
the data would have to be decompressed without any prior verification.
This could expose users e.g. to zip bombs or exploits on decompressor
vulnerabilities.

Whenever this specification refers to sub-Manifests, they can use any
names but are also required to use a specific compression suffix.
The ``MANIFEST`` entries are required to specify the full name including
compression suffix, and the verification is performed on the compressed
file.

The specification permits uncompressed Manifests to exist alongside
their compressed counterparts, and multiple compressed formats
to coexist. If that is the case, the files must have the same
uncompressed content and the specification is free to choose either
of the files using the same base name.


Combining multiple Manifest trees (informational)
-------------------------------------------------

This specification permits nesting multiple hierarchical Manifest trees.
In this layout, the specific directories of the Manifest tree can
be verified both as a part of another top-level Manifest,
and as an independent Manifest tree (when obtained without the parent
directory).

For this to work, the sub-Manifest file in the directory must also
satisfy the requirements for the top-level Manifest file. That is:

- it must be named ``Manifest`` and not compressed,

- it must cover all the files in this directory and its subdirectories
  (i.e. no files from the directory tree can be covered by parent
  Manifest),

- if authenticity verification is desired, it must be OpenPGP-signed.

It should be noted that if such a directory is a subdirectory of a valid
Manifest tree, the sub-Manifest needs to be valid according
to the top-level Manifest and the OpenPGP signature is disregarded
as detailed in `Manifest file locations and nesting`_. The top-level
behavior is exhibited only when the directory is obtained without parent
directories.


Package manager integration (informational)
-------------------------------------------

A package manager supporting full-tree Manifest verification should
enable it by default when using the Gentoo repository via rsync,
and require every location affecting its operation to verify
successfully before using it.

Full-tree verification can only be disabled explicitly by the user
(e.g. using configuration files). For security reasons, the package
manager must not ever attempt to disable it based on any data from
the repository. In particular, it is wrong to control it via
``metadata/layout.conf`` or based on the presence of top-level Manifest,
as it allows a malicious third-party to easily bypass verification.

Furthermore, none of the files present in the repository can be
processed before being verified against the Manifest files. This
includes ``metadata/layout.conf`` and ``profiles/repo_name`` files.
If the top-level Manifest is not present or those files do not pass
verification, the package manager with full-tree verification enabled
must reject the repository immediately.


An example Manifest file (informational)
----------------------------------------

An example top-level Manifest file for the Gentoo repository would have
the following content::

    TIMESTAMP 2017-10-30T10:11:12Z
    IGNORE distfiles
    IGNORE local
    IGNORE lost+found
    IGNORE packages
    MANIFEST app-accessibility/Manifest 14821 SHA256 1b5f.. SHA512 f7eb..
    ...
    MANIFEST eclass/Manifest.gz 50812 SHA256 8c55.. SHA512 2915..
    ...

An example modern Manifest (disregarding backwards compatibility)
for a package directory would have the following content::

    DATA SphinxTrain-0.9.1-r1.ebuild 932 SHA256 3d3b.. SHA512 be4d..
    DATA SphinxTrain-1.0.8.ebuild 912 SHA256 f681.. SHA512 0749..
    DATA metadata.xml 664 SHA256 97c6.. SHA512 1175..
    DATA files/gcc.patch 816 SHA256 b56e.. SHA512 2468..
    DATA files/gcc34.patch 333 SHA256 c107.. SHA512 9919..
    DIST SphinxTrain-0.9.1-beta.tar.gz 469617 SHA256 c1a4.. SHA512 1b33..
    DIST sphinxtrain-1.0.8.tar.gz 8925803 SHA256 548e.. SHA512 465d..


Rationale
=========

Stand-alone format
------------------

The first question that needed to be asked before proceeding with
the design was whether the Manifest file format was supposed to be
stand-alone, or tightly bound to the repository format.

The stand-alone format has been selected because of its three
advantages:

1. It is more future-proof. If an incompatible change to the repository
   format is introduced, only developers need to upgrade the tools
   they use to generate the Manifests. The tools used to verify
   the updated Manifests will continue to work.

2. It is more flexible and universal. With a dedicated tool,
   the Manifest files can be used to sign and verify arbitrary file
   sets.

3. It keeps the verification tool simpler. In particular, we can easily
   write an independent verification tool that could work on any
   distribution without needing to depend on a package manager
   implementation or rewrite parts of it.

Designing a stand-alone format requires that the Manifest carries enough
information to perform the verification following all the rules specific
to the Gentoo repository.


Tree design
-----------

The second important point of the design was determining whether
the Manifest files should be structured hierarchically, or independent.
Both options have their advantages.

In the hierarchical model, each sub-Manifest file is covered by a higher
level Manifest. As a result, only the top-level Manifest has to be
OpenPGP-signed, and subsequent Manifests need to be only verified by
checksum stored in the parent Manifest. This has the following
implications:

- Verifying any set of files in the repository requires using checksums
  from the most relevant Manifests and the parent Manifests.

- The OpenPGP signature of the top-level Manifest needs to be verified
  only once per process.

- Altering any set of files requires updating the relevant Manifests,
  and their parent Manifests up to the top-level Manifest, and signing
  the last one.

- As a result, the top-level Manifest changes on every commit,
  and various middle-level Manifests change (and need to be transferred)
  frequently.

In the independent model, each sub-Manifest file is independent
of the parent Manifests. As a result, each of them needs to be signed
and verified independently. However, the parent Manifests still need
to list sub-Manifests (albeit without verification data) in order
to detect removal or replacement of subdirectories. This has
the following implications:

- Verifying any set of files in the repository requires using checksums
  and verifying signatures of the most relevant Manifest files.

- Altering any set of files requires updating the relevant Manifests
  and signing them again.

- Parent Manifests are updated only when Manifests are added or removed
  from subdirectories. As a result, they change infrequently.

While both models have their advantages, the hierarchical model was
selected because it reduces the number of OpenPGP operations
(which are comparatively costly) to the minimum.


Tree layout restrictions
------------------------

The algorithm is meant to work primarily with ebuild repositories which
normally contain only files and directories. Directories provide
no useful metadata for verification, and specifying special entries
for additional file types is purposeless. Therefore, the specification
is restricted to dealing with regular files.

The Gentoo repository does not use symbolic links. Some Gentoo
repositories do, however. To provide a simple solution for dealing with
symlinks without having to take care to implement special handling for
them, the common behavior of implicitly resolving them is used.
Therefore, symbolic links to files are stored as if they were regular
files, and symbolic links to directories are followed as if they were
regular directories.

Dotfiles are implicitly ignored as that is a common notion used
in software written for POSIX systems. All other filenames require
explicit ``IGNORE`` lines.

An ability to inject additional ignore entries is provided to account
for site configuration affecting the repository tree -- placing
additional files in it, skipping some of the categories from syncing.
This configuration can extend beyond the limits of this GLEP,
e.g. by allowing wildcards or regular expressions.


Cross-filesystem Manifests
--------------------------

The first version of this specification had an additional requirement
that all files covered by the Manifest tree must reside on a single
filesystem. This requirement has been removed in version 1.1 for
the reasons outlined in this section.

The original rationale stated that this restriction aims to prevent
crossing filesystem boundaries in the top-level Manifest lookup
algorithm. While that seemed a good idea at the time, there is no real
reason to prevent that and this particular method worked correctly only
if the files were placed in a dedicated filesystem.

Worse than that, the original rationale did not anticipate the use
of overlayfs which combines multiple filesystems while preserving their
original metadata, including device and inode numbers. As a result,
if the repository was checked out to an overlayfs, it was quite possible
that different files had different device numbers, and the Manifest
checks failed due to crossing filesystem boundaries.

Given no clear solution to that and no good reason to reject use
of overlayfs, the restriction was lifted.

The only potential drawback of this is that the implementation may now
follow maliciously placed symbolic links pointing outside the tree.
If a regular file was replaced by such a symlink, the user could
be tricked into reporting the verification failure with the report
containing the checksums of the target file. However, for this to happen
the client would have to use rsync with ``--links`` option but without
``--safe-links`` which is neither the default behavior of rsync nor
the default configuration used by Portage.


Filename character set restriction
----------------------------------

The valid set of filename characters for the Gentoo repository
is restricted by the devmanual 'File Naming Rules' section
[#FILE-NAMING-RULES]_, and enforced via a git hook. The valid distfile
names are not restricted explicitly -- however, the PMS dependency
specification syntax [#PMS-FETCH]_ implicitly makes it impossible to use
filenames containing whitespace.

This specification aims to avoid arbitrary restrictions. For this
reason, filename characters are only restricted by excluding three
technically problematic groups:

1. The backwards slash character (``\``) is used as path separator
   on Windows systems, so it's extremely unlikely to be used in real
   filenames. For this reason it is used to implement character
   encoding with minimal risk of breaking backwards compatibility.

2. The control characters can trigger special behavior in various
   programs and confuse them from recognizing text files. In particular,
   the NULL character (``U+0000``) is normally used to indicate the end
   of a null-terminated string. Its use could therefore break
   implementations written in the C language. Other control characters
   could trigger various formatting routines, garbling text output.

3. Whitespace characters are used to separate Manifest fields
   and entries. While technically it would be enough to restrict space
   (``U+0020``) character that is normally used as the separator
   and newline (``U+000A``) character that is used to separate lines,
   all whitespace characters are forbidden to avoid confusion
   and implementation errors.

Historically, Portage attempted to overcome the whitespace limitation
by attempting to locate the size field and take everything before it
as filename. This was terribly fragile and even if it worked, it would
solve the problem only partially.

To preserve compatibility with the current implementations and given
that all of the listed characters are not allowed for the foreseeable
Gentoo uses, extended encoding support is optional. If such support
is not provided, the implementation must unconditionally reject any
such files. Ignoring them implicitly would be confusing, and it is
not possible to use them in explicit ``IGNORE`` entries.

The character encoding method provides means to overcome the character
restrictions to extend the tool usability beyond immediate Gentoo uses.
The backslash escape form based on Python unicode strings is used
since it can encode all characters within the Unicode range, the syntax
is familiar to many programmers and the backwards slash character
is extremely unlikely to appear in real filenames.

Syntax is limited to the minimum necessary to implement the encoding.
Shorthand forms (e.g. ``\t`` or ``\\``) are omitted to avoid unnecessary
complexity, and to reduce the risk of shell users using backslash
to escape space directly. The ``\x`` form is limited to ``\x00..\x7F``
range to avoid ambiguity of higher values which might be interpreted
either as UCS-2 code points or part of a UTF-8 encoded character.

Encoding stores UCS-2/UCS-4 characters directly rather than hex-encoded
UTF-8 string to simplify the implementation. In particular, it makes it
possible to process the Manifest file as UTF-8 encoded text without
having to perform additional UTF-8 decoding (and verification)
of the escaped data.

URL-encoding was considered as an alternative. However, it could collide
with ``DIST`` entries that are implicitly named after the URL filename
part where URL-encoding is pretty common.


File verification model
-----------------------

The verification model aims to provide full coverage against different
forms of attack. In particular, three different kinds of manipulation
are considered:

1. Alteration of the file content.

2. Removal of a file.

3. Addition of a new file.

In order to prevent against all three, the system requires that all
files in the repository are listed in Manifests and verified against
them.

As a special case, ignores are allowed to account for directories
that are not part of the repository but were traditionally placed inside
it. Those directories were ``distfiles``, ``local`` and ``packages``. It
could be also used to ignore VCS directories such as ``CVS``.


Non-strict Manifest verification
--------------------------------

Originally the Manifest2 format provided a special ``MISC`` tag that
was used for ``metadata.xml`` and ``ChangeLog`` files. This tag
indicated that the Manifest verification failures could be ignored for
those files unless the package manager was working in strict mode.

The first versions of this specification continued the use of this tag.
However, after a long debate it was decided to deprecate it along with
the non-strict behavior, and require all files to strictly match.

Two arguments were mentioned for the usefulness of a ``MISC`` type:

1. being able to reduce the checkout size by stripping unnecessary
   files out, and

2. being able to update automatically generated files locally
   without causing unnecessary verification failures.

However, the usefulness of ``MISC`` in both cases is doubtful.

The cases for stripping unnecessary files mostly focused around space
savings. For this purpose, stripping ``metadata.xml`` and similar files
has little value. It is much more common for users to strip whole
packages or categories. The ``MISC`` type is not suitable for that,
and so a dedicated package manager mechanism needs to be developed
instead. The same mechanism can also handle files that historically used
the ``MISC`` type. As an example, the package manager may choose
to generate both the rsync exclusion list and Manifest ignore list
using a single source list.

The cases for autogenerated files involve such cache files
as ``use.local.desc``. However, we can not include ``md5-cache`` there
due to security concerns which results in inconsistent cache handling.
Furthermore, the tools were historically modified to provide stable
output which means that their content can not change without
a non-``MISC`` content being changed first. This practically defeats
the purpose of using ``MISC``.

Finally, the non-strict mode could be used as means to an attack.
The allowance of missing or modified documentation file could be used
to spread misinformation, resulting in bad decisions made by the user.
A modified file could also be used, e.g. to exploit vulnerabilities
of an XML parser.


Timestamp field
---------------

The top-level Manifest optionally allows using a ``TIMESTAMP`` tag
to include a generation timestamp in the Manifest. A similar feature
was originally proposed in GLEP 58 [#GLEP58]_.

A malicious third-party may use the principles of exclusion or replay
[#C08]_ to deny an update to clients, while at the same time recording
the identity of clients to attack. The timestamp field can be used to
detect that.

In order to provide more complete protection, the Gentoo Infrastructure
should provide an ability to obtain the timestamps of all Manifests
from a recent timeframe over a secure channel from a trusted source
for comparison.

Strictly speaking, this information is provided by the various
``metadata/timestamp*`` files that are already present. However,
including the value in the Manifest itself has a little cost
and provides the ability to perform the verification stand-alone.

Furthermore, some of the timestamp files are added very late
in the distribution process, past the Manifest generation phase. Those
files will most likely receive ``IGNORE`` entries and therefore
be unsafe to use.

The specification permits additional timestamps in sub-Manifest files
for local use. A generic testing tool should ignore them.


New vs deprecated tags
----------------------

Out of the four types defined by Manifest2, only one is reused
and the remaining three are replaced by a single, universal ``DATA``
type.

The ``DIST`` tag is reused since the specification does not change
anything with regard to distfile handling.

The ``EBUILD`` tag could potentially be reused for generic file
verification data. However, it would be confusing if all the different
data files were marked as ``EBUILD``. Therefore, an equivalent ``DATA``
type was introduced as a replacement.

The ``MISC`` tag and the relevant non-strict mode has been removed
as being of little value, as detailed in the `Non-strict Manifest
verification`_ section.

The ``AUX`` tag is deprecated as it is redundant to ``DATA``, and has
the limiting property of implicit ``files/`` path prefix.


Finding top-level Manifest
--------------------------

The development of a reference implementation for this GLEP has brought
the following problem: how to find all the relevant Manifests when
the Manifest tool is run inside a subdirectory of the repository?

One of the options would be to provide a bi-directional linking
of Manifests via a ``PARENT`` tag. However, that would not solve
the problem when a new Manifest file is being created.

Instead, an algorithm for iterating over parent directories is proposed.
Since there is no obligatory explicit indicator for the top-level
Manifest, the algorithm assumes that the top-level Manifest
is the highest ``Manifest`` in the directory hierarchy that can cover
the current directory. This generally makes sense since the Manifest
files are required to provide coverage for all subdirectories, so all
Manifests starting from that one need to be updated.

If independent Manifest trees are nested in the directory structure,
then an ``IGNORE`` entry needs to be used to separate them.

Since sub-Manifests can use any filenames, the Manifest finding
algorithm must not short-cut the procedure by storing all ``Manifest``
files along the parent directories. Instead, it needs to retrace
the relevant sub-Manifest files along ``MANIFEST`` entries
in the top-level Manifest.


Injecting ChangeLogs into the checkout
--------------------------------------

One of the problems considered in the new Manifest format was injecting
historical and autogenerated ChangeLog into the repository. We normally
don't include those files, to reduce the checkout size. However, some
users have shown interest in them and Infra is working on providing them
via an additional rsync module.

If such files were injected into the repository, they would cause
verification failures of Manifests. To account for this, Infra could
provide ``IGNORE`` entries to allow them to exist.


Splitting distfile checksums from file checksums
------------------------------------------------

Another problem with the current Manifest format is that the checksums
for fetched files are combined with checksums for local files
in a single file inside the package directory. It has been specifically
pointed out that:

- since distfiles are sometimes reused across different packages,
  the repeating checksums are redundant [#DIST]_.
  
- mirror admins were interested in the possibility of verifying all
  the distfiles with a single tool.

This specification does not provide a clean solution to this problem.
It technically permits moving ``DIST`` entries to higher-level Manifests
but the usefulness of such a solution is doubtful.

However, for the second problem we will probably deliver a dedicated
tool working with this Manifest format.


Hash algorithms
---------------

While maintaining a consistent supported hash set is important
for interoperability, it is not a good fit for the generic layout
of this GLEP. Furthermore, it would require updating the GLEP
in the future every time the used algorithms change.

Instead, the specification focuses on listing the currently used
algorithm names for interoperability, and sets a recommendation
for consistent naming of algorithms in the future. The Python
``hashlib`` module is used as a reference since it is used
as the provider of hash functions for most of the Python software,
including Portage and PkgCore.

The basic rules for changing hash algorithms are defined in GLEP 59
[#GLEP59]_. The implementations can focus only on those algorithms
that are actually used or planned on being used. It may be feasible
to devise a new GLEP that specifies the currently used hashes (or update
GLEP 59 accordingly).


Manifest compression
--------------------

The support for Manifest compression is introduced with minimal changes
to the file format. The ``MANIFEST`` entries are required to provide
the real (compressed) file path for compatibility with other file
entries and to avoid confusion.

The compression of top-level Manifest file has been prohibited
as the specification currently does not provide any means of verifying
the file prior to decompression. If the top-level Manifest is
compressed, tooling will have to unpack the file before being able
to verify the contents. This makes it possible for a malicious third
party to attack the system by providing a compressed Manifest that
exposes decompressor vulnerabilities, or a zip bomb.

The OpenPGP cleartext signature covers the contents of the Manifest,
and is therefore compressed along with them. The possibility of using
a detached signature has been considered but it was rejected as
unnecessary complexity for minor gain.

Technically, a similar result could be effected via moving all the data
into a compressed sub-Manifest in the top directory (e.g.
``Manifest.sub.gz``), and including a ``MANIFEST`` entry for this file
in a signed, uncompressed top-level Manifest.

The existence of additional entries for uncompressed Manifest checksums
was debated. However, plain entries for the uncompressed file would
be confusing if only the compressed file existed, and conflicting
if both uncompressed and compressed variants existed. Furthermore,
it has been pointed out that ``DIST`` entries do not have
an uncompressed variant either.


Performance considerations
--------------------------

Performing a full-tree verification on every sync raises some
performance concerns for end-user systems. The initial testing has shown
that a cold-cache verification on a btrfs file system can take up around
4 minutes, with the process being mostly I/O bound. On the other hand,
it can be expected that the verification will be performed directly
after syncing, taking advantage of a warm filesystem cache.

To improve speed on I/O and/or CPU-restrained systems even further,
the algorithms can be easily extended to perform incremental
verification. Given that rsync does not preserve mtimes by default,
the tool can take advantage of mtime and Manifest comparisons to recheck
only the parts of the repository that have changed.

Furthermore, the package manager implementations can restrict checking
only to the parts of the repository that are actually being used.


Backwards Compatibility
=======================

This GLEP provides optional means of preserving backwards compatibility.
To preserve the backwards compatibility, the following needs to hold
for the ``Manifest`` file in every package directory:

- all files must be covered by the single ``Manifest`` file,

- all distfiles used by the package must be included,

- all files inside the ``files/`` subdirectory need to use
  the ``AUX`` tag (rather than ``DATA``),

- all ``.ebuild`` files need to use the ``EBUILD`` tag,

- the ``metadata.xml`` and ``ChangeLog`` files need to use
  the ``MISC`` tag,

- the Manifest can be signed to provide authenticity verification,

- an uncompressed Manifest must always exist, and a compressed Manifest
  of identical content may be present.

Once the backwards compatibility is no longer a concern, the above
no longer needs to hold and the deprecated tags can be removed.


Reference Implementation
========================

The reference implementation for this GLEP is being developed
as the gemato project [#GEMATO]_.


Credits
=======

Thanks to all the people whose contributions were invaluable
to the creation of this GLEP. This includes but is not limited to:

- Robin Hugh Johnson,
- Ulrich Müller.

Additionally, thanks to Robin Hugh Johnson for the original
MetaManifest GLEP series which served both as inspiration and source
of many concepts used in this GLEP. Recursively, also thanks to all
the people who contributed to the original GLEPs.


References
==========

.. [#GLEP44] GLEP 44: Manifest2 format
   (https://www.gentoo.org/glep/glep-0044.html)

.. [#GLEP57] GLEP 57: Security of distribution of Gentoo software
   - Overview
   (https://www.gentoo.org/glep/glep-0057.html)

.. [#GLEP58] GLEP 58: Security of distribution of Gentoo software
   - Infrastructure to User distribution - MetaManifest
   (https://www.gentoo.org/glep/glep-0058.html)

.. [#GLEP59] GLEP 59: Manifest2 hash policies and security implications
   (https://www.gentoo.org/glep/glep-0059.html)

.. [#GLEP60] GLEP 60: Manifest2 filetypes
   (https://www.gentoo.org/glep/glep-0060.html)

.. [#GLEP61] GLEP 61: Manifest2 compression
   (https://www.gentoo.org/glep/glep-0061.html)

.. [#UNICODE] The Unicode standard
   (https://unicode.org/versions/latest/)

.. [#PMS-FETCH] Package Manager Specification: Dependency Specification
   Format - SRC_URI
   (https://projects.gentoo.org/pms/6/pms.html#x1-940008.2.10)

.. [#FILE-NAMING-RULES] Ebuild File Format -- Gentoo Development Guide
   (https://devmanual.gentoo.org/ebuild-writing/file-format/#file-naming-rules)

.. [#MD5] RFC1321: The MD5 Message-Digest Algorithm
   (https://www.ietf.org/rfc/rfc1321.txt)

.. [#RIPEMD160] The hash function RIPEMD-160
   (https://homes.esat.kuleuven.be/~bosselae/ripemd160.html)

.. [#SHS] FIPS PUB 180-4: Secure Hash Standard (SHS)
   (http://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.180-4.pdf)

.. [#WHIRLPOOL] The WHIRLPOOL Hash Function
   (http://www.larc.usp.br/~pbarreto/WhirlpoolPage.html)

.. [#BLAKE2] BLAKE2 -- fast secure hashing
   (https://blake2.net/)

.. [#SHA3] FIPS PUB 202: SHA-3 Standard: Permutation-Based Hash
   and Extendable-Output Functions
   (http://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf)

.. [#STREEBOG] GOST R 34.11-2012: Streebog Hash Function
   (https://www.streebog.net/)

.. [#C08] Cappos, J et al. (2008). "Attacks on Package Managers"
   (https://www2.cs.arizona.edu/stork/packagemanagersecurity/attacks-on-package-managers.html)

.. [#DIST] According to Robin H. Johnson, 8.4% of all DIST entries
   at the time of writing are duplicate, representing 2 MiB
   out of 25 MiB of DIST entries altogether.

.. [#GEMATO] gemato: Gentoo Manifest Tool
   (https://github.com/mgorny/gemato/)


Copyright
=========
This work is licensed under the Creative Commons Attribution-ShareAlike 3.0
Unported License. To view a copy of this license, visit
https://creativecommons.org/licenses/by-sa/3.0/.