nvx-rs: check a GPT disk's kernel partitions in the guest

クローズ
  • 何か問題がありますか？
#770 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
25/100
issue の種類
機能追加
明瞭さ
明確に書かれている
活発さ
停滞
技術スタック
linux, rust
領域
backend, testing

調査の方向性

Start with parent issue #621 and the guest GPT path in nvx::guest::gpt::read_disk; review reference implementation PR #715 on its branch for the proposed behavior. The work covers sysfs partition checks, conditional BLKRRPART rereads with deadline handling, consistent reads, returned block devices, and privileged microVM scenarios. Done means those cases are covered and kernel partitions are validated against the GPT table.

索引モデルが issue の本文から書いたものです。

説明

Deferred to: platform clones (meta B, to be filed), whose restore substitutes the distro under a running kernel. From: #621 (A3.3), which closes without it. Reference implementation: #715, not merged, on branch esaurez/a3-gpt-device-validation-20261020.

#621 specified that, after a partition-table reread, each kernel partition match its validated entry before a deadline. The first iteration of A3.3's device code leaves that out. nvx::guest::gpt::read_disk reads a GPT disk's table through the block device that the guest's kernel attached, at the device's logical block size and size, and checks it, with the expected disk GUID, through nvx::gpt. The guest then uses the partitions that the kernel created when it found the disk, such as /dev/vda2, without comparing them with the table or asking the kernel to reread it. For a disk that the host stages read-only and attaches when it creates the VM, the kernel reads the same table before the guest's init starts, and nothing changes it afterwards.

This issue tracks the rest, for when a guest needs it: for example, for a disk whose table can change under a running kernel, or for a guest whose disks other programs manage.

  • Compare the kernel's partitions of the disk, as sysfs describes them (number, start and size), with the table's, and refuse missing and extra partitions.
  • Reread the table (BLKRRPART) only when they differ, then wait, with a deadline and short pauses, until they match, and refuse them when the deadline passes. The reread needs CAP_SYS_ADMIN, and fails while a partition of the disk is open.
  • Take a consistent view: never combine partitions, or the attributes of one partition, read before and after a reread.
  • Return the partitions' block devices: the first partition's and the layers'.
  • Privileged tests as a microVM scenario, with a test-only guest program shared into the guest rather than a larger release initramfs: a crafted GPT disk; missing, extra, moved and late partitions; and an MBR that the kernel reads in place of the table.

#715 implemented all of this, and passed CI and its reviews. Built as optimized as the guest binaries, its validation of a read-only disk, as the kernel found it at boot, took 0.4 to 0.9 ms on WHP and MSHV; the reread that it skipped there took 0.6 to 1.4 ms.

主要言語
Rust
スター
758
フォーク
48
平均マージ
17時間 44分
マージ済み PR（30日）
468

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

microsoft/nvx のほかの issue

microsoft/nvx の issue をすべて見る

似ている issue

Rust の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。