[BUG] Clean resolver storage rejections terminate Launchd #20
Labels
No labels
alpha-1
bug
ci
docs
duplicate
enhancement
help wanted
invalid
performance
phase-6
question
refactor
security
wontfix
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
erix/launchd#20
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
Launchd classifies a redirection-resolver storage rejection as an internal
failure even after the original launch rollback and transaction abort have
completed. A normal resource rejection can therefore terminate the service
instead of returning an ordinary launch failure.
Reproduction
On Launchd
3eb4b959755438fba403146139f4bf5b39464bea, let Vfsd reject aresolver registration with its storage failure class and one of its actual
producer results:
INVALID_ARGUMENT,BUSY, orRESOURCE_EXHAUSTED.Allow all original rollback and abort steps to finish successfully, then map
the retained resolver error through
runtime::error_map::launch_error.The same classification problem affects local resolver-slot exhaustion after
the original local capability cleanup succeeds. The repair's focused mapper
and resolver-wire controls exercise these cases without running a guest.
Expected and actual behavior
Expected: preserve the exact recognized Vfsd result with the launch storage
failure class. Report local exhaustion as
RESOURCE_EXHAUSTEDwith that class.Actual: the missing resolver case produces
INTERNALwith the materializationfailure class, selecting the existing service fail-stop path.
Unknown results, mismatched classes, malformed replies, transport failures,
and incomplete or uncertain cleanup must keep their existing internal,
pending, or fail-stop behavior. This repair adds no authority and does not
change cleanup order.
Environment and identity
feature/posix-compat.fe7b078a13126691d59179889b6c3993e9056dc5.x86_64-unknown-none.Evidence and prior checks
The repair passed four mapper controls, seven resolver-wire/lifetime controls,
and the full 445 library plus 216 binary tests. Formatting, strict Clippy,
private Rustdoc, both native compile checks, Markdown, whitespace, and the
actual candidate-tree source-size check passed.
These establish the scoped classification repair. They do not attribute an
existing guest or CI failure to a particular inner resolver error. The
published fix's current-source CI result remains to be observed.
Tracking
Related work: Launchd PR #2.
and the completely cleaned local exhaustion result.
original refusal paths, with meaningful mapper and lifetime controls.
5d8cddb0preserves the recognized storage rejection after successful rollback and transaction abort. Unknown replies and uncertain cleanup retain their original refusal paths. The runtime component catalog still selects the previous Launchd revision, so this does not yet establish a guest fix; current-source CI is pending.CI 398 is green for
5d8cddb0, including all 661 tests and the strict compile checks. All three acceptance items are complete, so I’m closing this classification issue. Selecting the updated Launchd in the runtime catalog remains separate.