feat(sched): made tasks reference counted and closed use-after-free races in the wake and kill paths - #178
Merged
Merged
Conversation
…futex and wait queues
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
A group-id send collected matching groups into a fixed 64-entry batch and silently skipped the rest while still reporting success, so a kill of a large process group left some of its processes running. Each registry pass now sends to the group with the smallest leader pid above a cursor, visiting every matching group exactly once however many exist. Leader pids are allocated monotonically and never reused, which makes the cursor a stable iteration order.
…n unstarted task Closing a process handle released the child's proc resource reference before claiming the task, while a concurrent group teardown claiming the same CREATED task released it again after winning the state CAS, dropping the reference twice. The CAS is the mutual exclusion for unstarted teardown, so the release now lives behind it and only the winning path touches the reference. Every other destroy_unstarted_task caller runs before a proc resource exists, so the move changes nothing for them.
…r its lock The wake pin contract forbids holding an irqsave spinlock across a wake because of the remote off-CPU spin, yet the timer expiry walk does exactly that and is correct. Sleepers always re-take the queue lock in cancel_sleep before they can exit, and a task sleeps on the CPU whose queue holds it, so an expiry wake never spins. Recording the argument keeps the walk from being either miscorrected or copied into a context where it does not hold.
…erences The leader teardown batch carried raw proc_resource pointers with hand-rolled release and destroy calls, the one remaining wake batch not expressed through strong_ref. Adopting the thread's reference into the batch lets scope handle the release and keeps every deferred-wake batch on the same ownership idiom.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
High Risk
Touches core scheduler teardown, signal delivery, and synchronization primitives; refcount or lock-order mistakes could cause use-after-free, missed wakeups, or stuck tasks.
Overview
Makes
sched::taskrefcounted so teardown goes throughrelease()→task::ref_destroy()→ the reaper, with the scheduler dropping the initial reference only after off-CPU publication infinalize_pending_off_cpu.Adds
task_ref/task_ref_by_tidand updates wake, kill, signals, futex, wait queues, and proc close/kill to holdstrong_refpins and callsched::wake/force_wake_for_killonly after spinlocks drop, avoiding use-after-free when a remote wake spins for off-CPU.Leader exit batches thread teardown under
tg->lock, deferswake_allon proc wait queues and kill wakes until the lock is released, and uses refcount drops instead of immediatereaper::deferon every thread.destroy_unstarted_taskclaimsTASK_STATE_CREATED, drops proc refs, and joins the same staged reaper path instead of freeing inline.On context switch, fatal signals are not applied while
TASK_FLAG_IN_SYSCALLis set (syscall exit still handles death).send_to_group_idno longer stops at 64 groups—it iterates by smallest leader pid above a cursor until every matching process group is signaled. Tests add task pins for kill paths and a 67-group group-id send regression.Reviewed by Cursor Bugbot for commit 499ab98. Bugbot is set up for automated code reviews on this repo. Configure here.