[Bug 1695578] Re: shim-signed trigger should not fail when attempting to re-prompt noninteractively and we've prompted before

[Mathieu Trudel-Lapierre]

** Description changed:

+ [Impact]
+ Users may require updates to DKMS modules while shim validation is disabled, and expect to do these updates non-interactively, such as via unattended-upgrades.
+ 
+ [Test Case]
+ == Interactive ==
+ (on an EFI system with proper Secure Boot support)
+ 1- Install an old version of a DKMS package (such as bbswitch-dkms or r8168-dkms)
+ 2- Follow the prompts on install to disable shim validation; reboot.
+ 
+ == Uninteractive ==
+ (on an EFI system with proper Secure Boot support)
+ 1- Install an old version of a DKMS package (such as bbswitch-dkms or r8168-dkms)
+ 2- Follow the prompts on install to disable shim validation; reboot.
+ 3- Run 'sudo apt update'
+ 4- Enable unattended-upgrades (see https://help.ubuntu.com/lts/serverguide/automatic-updates.html)
+ 5- Wait/trigger unattended upgrades; observe the behavior.
+ 
+ == New DKMS package install uninteractively ==
+ 1- Enable unattended upgrades
+ 2- Prepare a new PPA package update that adds a dependency on a -dkms package.
+ 3- Wait/trigger unattended upgrades; observe the behavior.
+ 
+ The upgrades should run non-interactively without issue whenever
+ possible, but forcefully prompt the user if a new DKMS package is being
+ installed uninteractively:
+ 
+  - If the user installs a new DKMS module, we should not silently proceed.  Either the user should be prompted, or if we're noninteractive, the trigger should fail.
+  - If the user has not installed any new DKMS modules, but we have an interactive frontend, we should prompt.
+  - If the user has not installed any new DKMS modules, and we're noninteractive, the trigger should silently pass.
+ 
+ 
+ [Regression Potential]
+ Failure to complete an update in non-interactive (unattended-upgrades) when DKMS modules are in use, triggering a failure in apt's upgrade process would be a regression of this update. Also, being unable to correctly recognize the current state of Secure Boot and shim validation, or incorrectly returning before prompting for the password required to toggle shim validation when the shim validation state make sense to be changed (ie. prompting to enable when it is disabled only, prompting to disable only if it's currently enabled).
+ 
+ ---
+ 
  We currently always fail with an error when update-secureboot-policy has
  been called, we detect that secureboot needs to be disabled for a dkms
  module, and we don't have an interactive debconf frontend.  However, as
  a result this means that if the user has previously made a conscious
  decision *not* to disable secureboot, despite having dkms modules
  installed, a non-interactive package upgrade will fail.
  
  It doesn't make sense for a non-interactive package upgrade to fail
  merely because the user's secureboot setting is ill-advised.
  
  We should ensure that:
  
-  - If the user installs a new DKMS module, we should not silently proceed.  Either the user should be prompted, or if we're noninteractive, the trigger should fail.
-  - If the user has not installed any new DKMS modules, but we have an interactive frontend, we should prompt.
-  - If the user has not installed any new DKMS modules, and we're noninteractive, the trigger should silently pass.
+  - If the user installs a new DKMS module, we should not silently proceed.  Either the user should be prompted, or if we're noninteractive, the trigger should fail.
+  - If the user has not installed any new DKMS modules, but we have an interactive frontend, we should prompt.
+  - If the user has not installed any new DKMS modules, and we're noninteractive, the trigger should silently pass.
  
  To know whether new DKMS modules have been installed, we should capture
  the list from /var/lib/dkms and store it under /var/lib/shim-signed on
  each successful invocation.
  
  For upgrade purposes, the shim-signed postinst should detect that we are
  upgrading from a version of the package that did not yet record the list
  in /var/lib/shim-signed, and record any DKMS modules present, so that
  these are not considered "new".  We only want to do this on upgrade, not
  on a new install of shim-signed; on a new install, the trigger should
  already handle this for us.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to shim-signed in Ubuntu.
https://bugs.launchpad.net/bugs/1695578

Title:
  shim-signed trigger should not fail when attempting to re-prompt
  noninteractively and we've prompted before

Status in shim-signed package in Ubuntu:
  Fix Released

Bug description:
  [Impact]
  Users may require updates to DKMS modules while shim validation is disabled, and expect to do these updates non-interactively, such as via unattended-upgrades.

  [Test Case]
  == Interactive ==
  (on an EFI system with proper Secure Boot support)
  1- Install an old version of a DKMS package (such as bbswitch-dkms or r8168-dkms)
  2- Follow the prompts on install to disable shim validation; reboot.

  == Uninteractive ==
  (on an EFI system with proper Secure Boot support)
  1- Install an old version of a DKMS package (such as bbswitch-dkms or r8168-dkms)
  2- Follow the prompts on install to disable shim validation; reboot.
  3- Run 'sudo apt update'
  4- Enable unattended-upgrades (see https://help.ubuntu.com/lts/serverguide/automatic-updates.html)
  5- Wait/trigger unattended upgrades; observe the behavior.

  == New DKMS package install uninteractively ==
  1- Enable unattended upgrades
  2- Prepare a new PPA package update that adds a dependency on a -dkms package.
  3- Wait/trigger unattended upgrades; observe the behavior.

  The upgrades should run non-interactively without issue whenever
  possible, but forcefully prompt the user if a new DKMS package is
  being installed uninteractively:

   - If the user installs a new DKMS module, we should not silently proceed.  Either the user should be prompted, or if we're noninteractive, the trigger should fail.
   - If the user has not installed any new DKMS modules, but we have an interactive frontend, we should prompt.
   - If the user has not installed any new DKMS modules, and we're noninteractive, the trigger should silently pass.

  
  [Regression Potential]
  Failure to complete an update in non-interactive (unattended-upgrades) when DKMS modules are in use, triggering a failure in apt's upgrade process would be a regression of this update. Also, being unable to correctly recognize the current state of Secure Boot and shim validation, or incorrectly returning before prompting for the password required to toggle shim validation when the shim validation state make sense to be changed (ie. prompting to enable when it is disabled only, prompting to disable only if it's currently enabled).

  ---

  We currently always fail with an error when update-secureboot-policy
  has been called, we detect that secureboot needs to be disabled for a
  dkms module, and we don't have an interactive debconf frontend.
  However, as a result this means that if the user has previously made a
  conscious decision *not* to disable secureboot, despite having dkms
  modules installed, a non-interactive package upgrade will fail.

  It doesn't make sense for a non-interactive package upgrade to fail
  merely because the user's secureboot setting is ill-advised.

  We should ensure that:

   - If the user installs a new DKMS module, we should not silently proceed.  Either the user should be prompted, or if we're noninteractive, the trigger should fail.
   - If the user has not installed any new DKMS modules, but we have an interactive frontend, we should prompt.
   - If the user has not installed any new DKMS modules, and we're noninteractive, the trigger should silently pass.

  To know whether new DKMS modules have been installed, we should
  capture the list from /var/lib/dkms and store it under /var/lib/shim-
  signed on each successful invocation.

  For upgrade purposes, the shim-signed postinst should detect that we
  are upgrading from a version of the package that did not yet record
  the list in /var/lib/shim-signed, and record any DKMS modules present,
  so that these are not considered "new".  We only want to do this on
  upgrade, not on a new install of shim-signed; on a new install, the
  trigger should already handle this for us.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/shim-signed/+bug/1695578/+subscriptions