Skip to content

fix(snapshot): let a second delete job leave a reused data inode alone - #103

Merged
mattmattox merged 2 commits into
mainfrom
fix/snapshot-delete-guard
Nov 13, 2026
Merged

mattmattox merged 2 commits into
mainfrom
fix/snapshot-delete-guard

Conversation

@mattmattox

Copy link
Copy Markdown
Contributor

Two delete jobs for one snapshot can run at the same time. Mount recovery queues a job for every deleting snapshot, including one another node is deleting right now, and a double delete queues two. A job that read the descriptor before the other job freed the data inode then emptied whatever took that slot. On main, TestSecondDeleteJobLeavesReusedDataInode shows a new snapshot in that slot losing its blocks (refcount 2 → 1), and the second job fails in both cases.

The guard checks the descriptor rather than a generation. The descriptor goes before the inode is freed, so "still named, checked under the inode lock" proves the slot was not reused. A generation read after the descriptor could already belong to the new inode.

🤖 Generated with Claude Code

mattmattox and others added 2 commits November 13, 2026 14:50
A job that read the descriptor before another job of the same snapshot
freed the data inode could then empty whatever took that slot, such as a
new snapshot's data inode. The job now empties the data inode under its
lock and only while the descriptor still names it, runs one at a time per
node, and succeeds when it finds the snapshot already removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@mattmattox
mattmattox merged commit 64aaa27 into main Nov 13, 2026
13 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant