mbox series

[RFC,0/6] KEYS: Introduce user mode key and signature parsers

Message ID 20230425173557.724688-1-roberto.sassu@huaweicloud.com
Headers show
Series KEYS: Introduce user mode key and signature parsers | expand

Message

Roberto Sassu April 25, 2023, 5:35 p.m. UTC
From: Roberto Sassu <roberto.sassu@huawei.com>

Goals
=====

Support new key and signature formats with the same kernel component.

Verify the authenticity of system data with newly supported data formats.

Mitigate the risk of parsing arbitrary data in the kernel.


Motivation
==========

Adding new functionality to the kernel comes with an increased risk of
introducing new bugs which, once exploited, can lead to a partial or full
system compromise.

Parsing arbitrary data is particularly critical, since it allows an
attacker to send a malicious sequence of bytes to exploit vulnerabilities
in the parser code. The attacker might be able to overwrite kernel memory
to bypass kernel protections, and obtain more privileges.

User Mode Drivers (UMDs) can effectively mitigate this risk. If the parser
runs in user space, even if it has a bug, it won't allow the attacker to
overwrite kernel memory.

The communication protocol between the UMD and the kernel should be simple
enough, that the kernel can immediately recognize malformed data sent by an
attacker controlling the UMD, and discard it.


Solution
========

Register a new parser of the asymmetric key type which, instead of parsing
the key blob, forwards it to a UMD, and populates the key fields from the
UMD response. That response contains the data for each field of the public
key structure, defined in the kernel, and possibly a key description.

Supporting new data formats can be achieved by simply extending the UMD. As
long as the UMD recognizes them, and provides the crypto material to the
kernel Crypto API in the expected format, the kernel does not need to be
aware of the UMD changes.

Add a new API to verify the authenticity of system data, similar to the one
for PKCS#7 signatures. As for the key parser, send the signature to a UMD,
and fill the public_key_signature structure from the UMD response.

The API still supports a very basic trust model, it accepts a key for
signature verification if it is in the supplied keyring. The API can be
extended later to support more sophisticated models.


Use cases
=========

eBPF
----

The eBPF infrastructure already offers to eBPF programs the ability to
verify PKCS#7 signatures, through the bpf_verify_pkcs7_signature() kfunc.

Add the new bpf_verify_umd_signature() kfunc, to allow eBPF programs verify
signatures in a data format that is not PKCS#7 (for example PGP).


IMA Appraisal
-------------

An alternative to appraising each file with its signature (Fedora 38) is to
build a repository of reference file digests from signed RPM headers, and
lookup the calculated digest of files being accessed in that repository
(DIGLIM[1]).

With this patch set, the kernel can verify the authenticity of RPM headers
from their PGP signature against the Linux distribution GPG keys. Once
verified, RPM headers can be parsed with a UMD to build the repository of
reference file digests.

With DIGLIM, Linux distributions are not required to change anything in
their building infrastructure (no extra data in the RPM header, no new PKI
for IMA signatures).

[1]: https://lore.kernel.org/linux-integrity/20210914163401.864635-1-roberto.sassu@huawei.com/


UMD development
===============

The header file crypto/asymmetric_keys/umd_key_sig_umh.h contains the
details of the communication protocol between the kernel and the UMD
handler.

The UMD handler should implement the commands defined, CMD_KEY and CMD_SIG,
should set the result of the processing, and fill the key and
signature-specific structures umd_key_msg_out and umd_sig_msg_out.

The UMD handler should provide the key and signature blobs in a format that
is understood by the kernel. For example, for RSA keys, it should provide
them in ASN.1 format (SEQUENCE of INTEGER).

The auth IDs of the keys and signatures should match, for signature
verification. Auth ID matching can be partial.


Patch set dependencies
======================

This patch set depends on 'usermode_driver: Add management library and
API':

https://lore.kernel.org/bpf/20230317145240.363908-1-roberto.sassu@huaweicloud.com/


Patch set content
=================

Patch 1 introduces the new parser for the asymmetric key type.

Patch 2 introduces the parser for signatures and its API.

Patch 3 introduces the system-level API for signature verification.

Patch 4 extends eBPF to use the new system-level API.

Patch 5 adds a test for UMD-parser signatures (not executed until the UMD
supports PGP).

Patch 6 introduces the skeleton of the UMD handler.


PGP
===

A work in progress implementation of the PGP format (RFC 4880 and RFC 6637)
in the UMD handler is available at:

https://github.com/robertosassu/linux/commits/pgp-signatures-umd-v1-devel-v24

It is based on a previous work of David Howells, available at:

https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-modsign.git/log/?h=pgp-parser

The patches have been adapted for use in user space.

Roberto Sassu (6):
  KEYS: asymmetric: Introduce UMD-based asymmetric key parser
  KEYS: asymmetric: Introduce UMD-based asymmetric key signature parser
  verification: Introduce verify_umd_signature() and
    verify_umd_message_sig()
  bpf: Introduce bpf_verify_umd_signature() kfunc
  selftests/bpf: Prepare a test for UMD-parsed signatures
  KEYS: asymmetric: Add UMD handler

 .gitignore                                    |   3 +
 MAINTAINERS                                   |   1 +
 certs/system_keyring.c                        | 125 ++++++
 crypto/asymmetric_keys/Kconfig                |  32 ++
 crypto/asymmetric_keys/Makefile               |  23 +
 crypto/asymmetric_keys/asymmetric_type.c      |   3 +-
 crypto/asymmetric_keys/umd_key.h              |  28 ++
 crypto/asymmetric_keys/umd_key_parser.c       | 203 +++++++++
 crypto/asymmetric_keys/umd_key_sig_loader.c   |  32 ++
 crypto/asymmetric_keys/umd_key_sig_umh.h      |  71 +++
 crypto/asymmetric_keys/umd_key_sig_umh_blob.S |   7 +
 crypto/asymmetric_keys/umd_key_sig_umh_user.c |  84 ++++
 crypto/asymmetric_keys/umd_sig_parser.c       | 416 ++++++++++++++++++
 include/crypto/umd_sig.h                      |  71 +++
 include/keys/asymmetric-type.h                |   1 +
 include/linux/verification.h                  |  48 ++
 kernel/trace/bpf_trace.c                      |  69 ++-
 ...ify_pkcs7_sig.c => verify_pkcs7_umd_sig.c} | 109 +++--
 ...kcs7_sig.c => test_verify_pkcs7_umd_sig.c} |  18 +-
 .../testing/selftests/bpf/verify_sig_setup.sh |  82 +++-
 20 files changed, 1378 insertions(+), 48 deletions(-)
 create mode 100644 crypto/asymmetric_keys/umd_key.h
 create mode 100644 crypto/asymmetric_keys/umd_key_parser.c
 create mode 100644 crypto/asymmetric_keys/umd_key_sig_loader.c
 create mode 100644 crypto/asymmetric_keys/umd_key_sig_umh.h
 create mode 100644 crypto/asymmetric_keys/umd_key_sig_umh_blob.S
 create mode 100644 crypto/asymmetric_keys/umd_key_sig_umh_user.c
 create mode 100644 crypto/asymmetric_keys/umd_sig_parser.c
 create mode 100644 include/crypto/umd_sig.h
 rename tools/testing/selftests/bpf/prog_tests/{verify_pkcs7_sig.c => verify_pkcs7_umd_sig.c} (75%)
 rename tools/testing/selftests/bpf/progs/{test_verify_pkcs7_sig.c => test_verify_pkcs7_umd_sig.c} (82%)

Comments

Jarkko Sakkinen April 26, 2023, 12:28 a.m. UTC | #1
On Tue Apr 25, 2023 at 8:35 PM EEST, Roberto Sassu wrote:
> From: Roberto Sassu <roberto.sassu@huawei.com>
>
> Introduce verify_umd_signature() and verify_umd_message_sig(), to verify
> UMD-parsed signatures from detached data. It aims to be used by kernel
> subsystems wishing to verify the authenticity of system data, with
> system-defined keyrings as trust anchor.

UMD is not generic knowledge. It is a term coined up in this patch set
so please open code it to each patch.

One discussion points should be what these handlers should be called.
Right now the patch set is misleads the reader to think as this was
some kind of "official" term and set to stone.

BR, Jarkko
Roberto Sassu April 26, 2023, 11:42 a.m. UTC | #2
On Wed, 2023-04-26 at 03:28 +0300, Jarkko Sakkinen wrote:
> On Tue Apr 25, 2023 at 8:35 PM EEST, Roberto Sassu wrote:
> > From: Roberto Sassu <roberto.sassu@huawei.com>
> > 
> > Introduce verify_umd_signature() and verify_umd_message_sig(), to verify
> > UMD-parsed signatures from detached data. It aims to be used by kernel
> > subsystems wishing to verify the authenticity of system data, with
> > system-defined keyrings as trust anchor.
> 
> UMD is not generic knowledge. It is a term coined up in this patch set
> so please open code it to each patch.

Yes, Linus also commented on this:

https://lwn.net/ml/linux-kernel/CAHk-=wihqhksXHkcjuTrYmC-vajeRcNh3s6eeoJNxS7wp77dFQ@mail.gmail.com/

I will check if the full name is mentioned at least once. So far, it
seems that using umd for function names should be ok.

> One discussion points should be what these handlers should be called.
> Right now the patch set is misleads the reader to think as this was
> some kind of "official" term and set to stone.

I proposed some naming here (dependency of this patch set):

https://lore.kernel.org/bpf/20230317145240.363908-6-roberto.sassu@huaweicloud.com/

Please let me know if it sounds reasonable to you.

Thanks

Roberto
Jarkko Sakkinen April 26, 2023, 6:25 p.m. UTC | #3
On Wed, 2023-04-26 at 13:42 +0200, Roberto Sassu wrote:
> On Wed, 2023-04-26 at 03:28 +0300, Jarkko Sakkinen wrote:
> > On Tue Apr 25, 2023 at 8:35 PM EEST, Roberto Sassu wrote:
> > > From: Roberto Sassu <roberto.sassu@huawei.com>
> > > 
> > > Introduce verify_umd_signature() and verify_umd_message_sig(), to verify
> > > UMD-parsed signatures from detached data. It aims to be used by kernel
> > > subsystems wishing to verify the authenticity of system data, with
> > > system-defined keyrings as trust anchor.
> > 
> > UMD is not generic knowledge. It is a term coined up in this patch set
> > so please open code it to each patch.
> 
> Yes, Linus also commented on this:
> 
> https://lwn.net/ml/linux-kernel/CAHk-=wihqhksXHkcjuTrYmC-vajeRcNh3s6eeoJNxS7wp77dFQ@mail.gmail.com/
> 
> I will check if the full name is mentioned at least once. So far, it
> seems that using umd for function names should be ok.

Also: "UMD-based parser for the asymmetric key type"

It is a tautology:

UMD is based on parser which based on UMD.

I.e. makes no sense.

Everyone hates three letter acronyms so I would consider not
inventing a new one out of the void.

So the corrective step would be to rename Kconfig flags as
USER_ASYMMETRIC_KEY_PARSER and USER_ASYMMETRIC_SIGNATURE_PARSER.

BR, Jarkko
Jarkko Sakkinen April 26, 2023, 6:27 p.m. UTC | #4
On Wed, 2023-04-26 at 21:25 +0300, Jarkko Sakkinen wrote:
> On Wed, 2023-04-26 at 13:42 +0200, Roberto Sassu wrote:
> > On Wed, 2023-04-26 at 03:28 +0300, Jarkko Sakkinen wrote:
> > > On Tue Apr 25, 2023 at 8:35 PM EEST, Roberto Sassu wrote:
> > > > From: Roberto Sassu <roberto.sassu@huawei.com>
> > > > 
> > > > Introduce verify_umd_signature() and verify_umd_message_sig(), to verify
> > > > UMD-parsed signatures from detached data. It aims to be used by kernel
> > > > subsystems wishing to verify the authenticity of system data, with
> > > > system-defined keyrings as trust anchor.
> > > 
> > > UMD is not generic knowledge. It is a term coined up in this patch set
> > > so please open code it to each patch.
> > 
> > Yes, Linus also commented on this:
> > 
> > https://lwn.net/ml/linux-kernel/CAHk-=wihqhksXHkcjuTrYmC-vajeRcNh3s6eeoJNxS7wp77dFQ@mail.gmail.com/
> > 
> > I will check if the full name is mentioned at least once. So far, it
> > seems that using umd for function names should be ok.
> 
> Also: "UMD-based parser for the asymmetric key type"
> 
> It is a tautology:
> 
> UMD is based on parser which based on UMD.
> 
> I.e. makes no sense.
> 
> Everyone hates three letter acronyms so I would consider not
> inventing a new one out of the void.
> 
> So the corrective step would be to rename Kconfig flags as
> USER_ASYMMETRIC_KEY_PARSER and USER_ASYMMETRIC_SIGNATURE_PARSER.

(or along the lines)

BR, Jarkko