This leaves slots available for additional uid ranges in Rosa OS.
This breaks all existing installations! Users are required to fix ownership manually.
Closes#18.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Wanted to do this for a long time, since the key is redundant. This also makes it easier to migrate to the new store interface.
Signed-off-by: Ophestra <cat@gensokyo.uk>
The SharePath is a more stable path than RunDirPath, since it is available all the time and should remain consistent. This also fits better into the intended use case of XDG_RUNTIME_DIR.
Closes#17.
Signed-off-by: Ophestra <cat@gensokyo.uk>
In some cases the signal might be delivered before the signal handler is installed, and synchronising against such a case is too expensive. Instead, use the pipe being closed as a fallback to the regular exit request. This change also moves installation of the signal handler early.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This change also spawns shim before committing system state, leaving it blocking on the setup pipe. The internal/outcome/process structure is also entirely reworked to be much more readable and less error-prone, while enabling basic performance measurements. A long-standing bug where segment lock is not held during Commit is also resolved.
Closes#19.
Signed-off-by: Ophestra <cat@gensokyo.uk>
The handle is otherwise inaccessible without the compat interface. This change also moves compatibility methods to separate adapter structs to avoid inadvertently using them.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This reduces collision with local variable names, and generally makes sense for the new store package, since it no longer specifies the state struct.
Signed-off-by: Ophestra <cat@gensokyo.uk>
The shim does not suspend output to begin with. These are leftovers from when container startup code suspends output.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This fully exposes the store internals for #19 and are final preparations for removing the legacy store interface.
This change also fixes a potential deadlock in the handle initialisation mkdir failure path. This however is never reachable in hakurei as the store is never accessed concurrently.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This replaces the Store interface with something better reflecting the underlying data format for #19. An implementation of Store is provided on top of the new code to ease transition.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This is less error-prone and fits better into internal/app which already uses check.Absolute for all pathnames.
Signed-off-by: Ophestra <cat@gensokyo.uk>
This avoids getting into an inconsistent state for simultaneous calls to List and Do on a previously unknown identity.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Some internal/app/state types were relocated to hst as part of the API. This change updates doc comments referring to them.
Signed-off-by: Ophestra <cat@gensokyo.uk>
The upper half is now a nanosecond timestamp. Lower half is still random bytes, so use lower half for short identifier.
Signed-off-by: Ophestra <cat@gensokyo.uk>
Init is not very talkative after process start even when verbose. Suspending output here is pointless and does more harm than good.
Signed-off-by: Ophestra <cat@gensokyo.uk>
The SIGTERM signal is delivered in many other cases and can lead to strange behaviour. The unconditional resume of the logger also causes strange behaviour in the cancellation forwarding path. This change also passes through additional signals.
Signed-off-by: Ophestra <cat@gensokyo.uk>
