157 lines
7.8 KiB
Markdown
157 lines
7.8 KiB
Markdown
|
# Software Bill of Materials (SBOM)
|
||
|
|
||
|
SBOM is a collection of information of each software component
|
||
|
you are supplying/building. Similar to a package manager on Linux
|
||
|
based systems, it holds information of as many software parts as
|
||
|
possible. This information can be a version, name of the software, URL,
|
||
|
license information and more. A SBOM can be saved in various formats.
|
||
|
In coreboot it's saved as "uSWID" file. uSWID is not a standard or
|
||
|
specification but it doesn't need to be, since it's basically just an
|
||
|
array/list of CoSWID (Concise Software Identification) files which in
|
||
|
turn are specified by a RFC specification. CoSWID files are saved in a
|
||
|
CBOR format. CBOR is like JSON if JSON were a binary format. Similar
|
||
|
to a package manager the CoSWID format can link multiple softwares
|
||
|
together. For example on most modern Intel systems FSP is included as
|
||
|
a dependency of coreboot. That kind of relationship between software
|
||
|
components (among others) can be expressed in an uSWID file. That makes
|
||
|
firmware/software much more transparent. One could for example create a
|
||
|
software that takes a coreboot firmware image as input and
|
||
|
automatically creates a graph with all software components the coreboot
|
||
|
image contains and their relationship to each other.
|
||
|
|
||
|
|
||
|
## SWID/CoSWID
|
||
|
|
||
|
SWID is a standard hidden behind an ISO paywall.
|
||
|
It generally identifies/describes Software components. Since SWID files
|
||
|
are written in XML, they can get too large for devices with network and
|
||
|
storage constraints. CoSWID is basically SWID but in CBOR binary
|
||
|
format, which makes it far smaller compared to its big brother. Also,
|
||
|
CoSWID is a RFC specification (so publicly accessible). Therefore
|
||
|
CoSWID is the standard used in coreboot SBOM. But one CoSWID file/tag
|
||
|
can only describe one single software, but since software is usually
|
||
|
composed of multiple parts (especially in firmware with many binary
|
||
|
blobs) uSWID was born as a container format to hold multiple CoSWID
|
||
|
files. It also has a magic value, that makes software capable of
|
||
|
extracting uSWID/CoSWID data without the need to understand the
|
||
|
underlying format of the binary (in coreboot it's the CBFS and in EDK2
|
||
|
it's the COFF). To get a simple overview of how a SWID/CoSWID file
|
||
|
looks like, just take a look at the various "templates" in src/sbom/.
|
||
|
There are of course other SBOM specifications out there, but most of
|
||
|
them are rather blown up and don't support a binary format at all.
|
||
|
|
||
|
|
||
|
## coreboot implementation
|
||
|
|
||
|
Quick overview of how things are generated:
|
||
|
|
||
|
![Generation of an SBOM File in coreboot][sbom_generation]
|
||
|
|
||
|
[sbom_generation]: sbom_generation.svg
|
||
|
|
||
|
After all SBOM data has been fetched from all the software components,
|
||
|
the 'goswid' tool links them all together into one sbom.uswid file.
|
||
|
Therefore the goswid tool is basically a linker that takes multiple
|
||
|
CoSWID/SWID files and converts them into one uSWID file. Although the
|
||
|
image shows only Files in JSON format it is also possible to supply
|
||
|
them in XML or CBOR format.
|
||
|
|
||
|
The final SBOM file is located inside the CBFS.
|
||
|
For each software component in coreboot SBOM, there is an option in
|
||
|
Kconfig (usually called `CONFIG_INCLUDE_[software-name]_SBOM`) to either
|
||
|
include or not include SBOM metadata for the specified software.
|
||
|
Furthermore there is a `CONFIG_SBOM_[software-name]_PATH` option which
|
||
|
contains a path to a SWID/CoSWID file in a format of choice
|
||
|
(being either JSON, XML or CBOR). `CONFIG_SBOM_[software-name]_PATH`
|
||
|
option usually defaults to a very generic CoSWID file in JSON format
|
||
|
(which are stored in src/sbom/). That at least gives minimal
|
||
|
information like the name of the software and maybe a version.
|
||
|
But it is always preferred, that the `CONFIG_SBOM_[software-name]_PATH`
|
||
|
is set to a custom CoSWID/SWID file that contains much more information
|
||
|
(like version/commit-hash, license, URL, dependencies, ...).
|
||
|
Therefore using the defaults is by any means to be avoided, since they
|
||
|
hold very little information or even worse wrong information.
|
||
|
Furthermore some of these Kconfig options have a suboption
|
||
|
(usually called `CONFIG_SBOM_[software-name]_GENERATE`) to generate
|
||
|
some basic SBOM data for the specified software component, in order to
|
||
|
get at least some bit of information about it by analyzing the binary
|
||
|
(for binary blobs) or querying information via git (for open source
|
||
|
projects). This is for example currently done for all payloads. For
|
||
|
each payload the commit hash used in the build is taken and put into
|
||
|
the SBOM file. For open-source projects (like all payloads) crucial
|
||
|
information like the current commit-hash of the payload can easily be
|
||
|
put into the SBOM file. Extracting information out of binary blobs is a
|
||
|
bit trickier for obvious reasons. For closed source binary blobs it is
|
||
|
therefore recommended that vendors and software-engineers create a SBOM
|
||
|
file as part of their build process and add a path to that SBOM file
|
||
|
via Kconfig options in coreboot (`CONFIG_SBOM_[software-name]_PATH`).
|
||
|
That way the final SBOM has much more useful and correct data.
|
||
|
|
||
|
|
||
|
## Build coreboot with SBOM
|
||
|
|
||
|
Directly under the 'General setup' Kconfig menu is a
|
||
|
'Software Bill of Materials (SBOM)' submenu where all options are to
|
||
|
enable/disable SBOM integration in to the corebeoot build.
|
||
|
Therefore one can just enable/disable them via `make menuconfig`.
|
||
|
|
||
|
|
||
|
## What to do as Developer of a binary blob (which is used in coreboot)
|
||
|
|
||
|
1. Generate a SWID/CoSWID/uSWID File in either JSON, XML or CBOR Format
|
||
|
as part of your software build process
|
||
|
|
||
|
2. Supply that generated File along with your binary blob (preferably
|
||
|
not inside the blob)
|
||
|
|
||
|
3. To build coreboot: Add `CONFIG_SBOM_[software-name]_PATH` to your
|
||
|
defconfig pointing to your [software-name] generated File.
|
||
|
|
||
|
|
||
|
## What to do as Developer of an open source project (which is used in coreboot)
|
||
|
|
||
|
1. Generate a SWID/CoSWID/uSWID file in either JSON, XML or CBOR format
|
||
|
as part of your software's build process. For example in form of a
|
||
|
Makefile target.
|
||
|
|
||
|
2. Change src/sbom/Makefile.inc (in order to know where to find the
|
||
|
CoSWID/SWID/uSWID file) as well as the Makefile in coreboot which
|
||
|
builds said software. For example for GRUB2 that could mean to add a
|
||
|
Makefile target in payloads/external/GRUB2/Makefile.
|
||
|
|
||
|
|
||
|
## Problems
|
||
|
|
||
|
What to do if the binary blob that is included in coreboot's build
|
||
|
already has a SBOM file embedded in the binary? One could supply the
|
||
|
path of the software binary itself (e.g. me.bin) as SBOM file path for
|
||
|
the software in question. Which would basically mean to set
|
||
|
`CONFIG_SBOM_[software-name]_PATH=/path/to/me.bin`. This is possible
|
||
|
since the 'goswid' tooling is able to extract uSWID information out of
|
||
|
an unknown binary format because of uSWIDs magic value. But even if
|
||
|
coreboot can extract the uSWID data there is still the question of what
|
||
|
to do next. One can do one of the following:
|
||
|
|
||
|
- Do not include the Software's SBOM data in the final SBOM of
|
||
|
coreboot. Data would not be duplicated, but therefore not included
|
||
|
in coreboot SBOM file.
|
||
|
|
||
|
- Extract the uSWID/CoSWID information from the binary and also
|
||
|
include it in the coreboot SBOM. That would mean, that SBOM data
|
||
|
is duplicated.
|
||
|
|
||
|
The first solution should in general be preferred, since its no
|
||
|
problem if SBOM data is located at multiple locations/binaries if they
|
||
|
don't have a direct dependency on each other. It would be good if
|
||
|
software that cannot run on its own only supplies the SBOM data along
|
||
|
with it as kind of extra file instead of embedded in an unknown binary
|
||
|
blob. coreboot can then just take it and include it in its own SBOM
|
||
|
file. If on the other hand the binary can function on its own (e.g. EC
|
||
|
or BMC binary), it is generally preferred that the software supplies
|
||
|
its own SBOM data and coreboot just simply doesn't include it in its
|
||
|
own SBOM file. That would make a more or less clear distinction and
|
||
|
avoids duplication in case the BMC or EC is updated (without updating
|
||
|
coreboot). The distinction is not always easy and this problem is
|
||
|
currently not considered in the implementation, since none of the
|
||
|
software components currently create a SBOM file on their own.
|