On Fri, Jul 20, 2018 at 7:47 PM, John W. Linville <linvi...@redhat.com> wrote: > A mostly "behind the scenes" aspect of the Linux wireless LAN > (IEEE802.11) subsystem has changed upstream with some system-level > effects. Replacing an existing package with a cleaner but functionally > equivalent package is an attractive option, but there are questions > regarding the update path, how signing the the regulatory database > should be handled, and how to communicate these changes outward to the > greater community. > > BACKGROUND > > Usage of Radio Frequency (RF) spectrum is generally regulated around > the world. Even unlicensed use of RF spectrum (as with wireless LANs) > is still subject to some level of local regulation. In order to support > lawful use of wireless LAN technology with Linux around the world, the > Linux kernel uses an externally maintained wireless regulatory database > that encodes relevant information about known regulatory (i.e. > legal/governmental) domains. The kernel makes use of this information > in order to enforce wireless regulations for devices under its control, > although some devices further apply their own limitations at the driver > or firmware level as well. > > For a long time, the Linux kernel relied on a piece of software known > as the Central Regulatory Domain Agent (CRDA) to feed the kernel with > regulatory information in response to UDEV events that indicate what > regulatory domain's rules are to be enforced. Also, within Fedora, the > setregdomain script runs when a wireless netdev is instantiated. By > default this script attempts to relate a system’s TZ (i.e. Time Zone) > setting to a corresponding country code, which is then used to set the > wireless regulatory domain. The setregdomain script can also use > alternate means to determine which regulatory domain to request, as > detailed in the setregdomain.1 man page. > > CRDA is maintained upstream in the crda git repository. The wireless > regulatory database is maintained upstream in the wireless-regdb git > repository. In Fedora, snapshots of both of the above repositories are > built within the single crda RPM. This is done to enable build-time > generation of a key used to sign the wireless-regdb database which is > then embedded in the concurrently built crda binary in order to > validate the integrity of the database at runtime. > > As of about the upstream 4.15 version of Linux, the kernel's wireless > developers had determined that future updates to the wireless > regulatory database may require format changes that would be > incompatible with the existing CRDA implementation. Further, Linux > kernel changes over the years now enabled the kernel to load firmware > images without requiring a userland helper like CRDA. So, the Linux > kernel wireless developers implemented an in-kernel replacement for > CRDA. This includes code to validate the wireless regulatory database > against a signature that is built into the kernel itself. > > SITUATION > > At present, Fedora kernels are configured to use the in-kernel wireless > regulatory database loader. If the database is not found or is not > signed with the key built into the kernel, then boot time messages > appear, creating the appearance of a problem. For now, there is no > actual problem. Fedora systems continue to ship with CRDA, which > continues to perform its original duties with no effective problem as > of yet. The harmless "error" messages during boot are the only symptom. > (NOTE: the latest Fedora CRDA package works around this issue, > eliminating the spurious "error" messages at boot time.) > > For now, there is no emergency. We could continue to ship CRDA as-is. > My concern is that eventually, CRDA may no longer be a viable option > for future regulatory database updates. In that case, it would seem to > be better to change now than to do so in a rush later. > > Also, replacing the combined CRDA/wireless-regdb "crda" package with a > new "wireless-regdb" package removes a wart from our package collection > and replaces it with a package that is simpler and easier to understand > how it is built. That will be one less "why did they do that?" item > hanging around in Fedora. > > ACTION IN PROGRESS > > I have created a wireless-regdb package for Fedora. It simply installs > the wireless regulatory database binary into /usr/lib/firmware, > alongside it's pre-existing detached signature. This package includes a > version of the setregdomain script that is virtually identical to what > is in the existing crda package. This package has been approved for > Fedora, but I recently imported the SRPM to the Fedora repository. (htt > ps://bugzilla.redhat.com/show_bug.cgi?id=1598921) > > I have been testing with the above package for a couple of weeks, and > the user experience is identical to using the crda solution. Given this > success, I believe that the best option will be to push this package > with Obsoletes and Provides for crda. This will facilitate a painless > and nearly silent upgrade for existing Fedora users. I will follow-up > with retiring the crda package itself a bit later. Other required > actions will include replacing any packaging references to the crda > package with an equivalent wireless-regdb reference (or simply > eliminating such a reference) and making the requisite changes to > comps.xml or any future equivalent. > > QUESTIONS > > Are there reasons to oppose the Obsolete/Provides upgrade path from > crda to wireless-regdb? Would it be desirable to require users to > manually intervene by installing wireless-regdb by hand? FWIW, I do not > see any benefit from requiring any manual steps.
Querying dnf it looks like the only package that explicitly depends on the package or any of the binaries is wireless-tools so I think that should be straight forward with Obsolete/Provides. It also means the old crda is removed on upgrade so it won't hang around to conflict/cause issues later. > Is it acceptable to trust the upstream signature of the wireless > regulatory database? Or do we need to use some sort of Fedora > signature? If the latter, can it be a (semi-)permanent (e.g. per > release) signature, which could be maintained in the kernel sources? Or > must it be a per build signature? How can that be accommodated, other > than through another combined package? (My personal preference is to > simply trust the upstream signature that is already being built into > the kernel.) I don't see an issue with the upstream signing, we do have the infrastructure in Fedora for signing kernels (for secure boot) and kernel modules with Fedora keys, presumably being loaded into the kernel the signing mechanism would be the same as for modules (I'm making wild assumptions here) so it's possible to have Fedora infra sign it too if it's desirable. We just need to make sure you're in the right ACLs. > Does the crda to wireless-regdb upgrade effort need a Feature page (or > similar) in Fedora? Personally I don't think it does, it's nothing of real note that we need to notify people about as it should basically be seamless to the end user or there's no great added wireless functionality the user would care about. > Since crda is a "critical path" component, what special considerations > need to be accommodated in order to replace it with wireless-regdb? The "critical path" packages are generated based on dependencies of blocking components so the new package should replace the old one once it's landed and in a compose replacing it so there should be nothing manual needed here. _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/message/3GSLXFU3XRMJNQE3ODFBFVODNBJUFEBE/
