diff --git a/AUDIT-2026-09-22.md b/AUDIT-2026-09-22.md index 1a7c7ef..16fcfee 100644 --- a/AUDIT-2026-09-22.md +++ b/AUDIT-2026-09-22.md @@ -24,6 +24,7 @@ IPv6-only 接入(`ssh -6 -i ~/.ssh/id_ed25519 root@fdb2:2c26:f4e4:0:21c:42ff:f | 20 | **验证邮箱唯一性只存在于注释里**:errors.go/repo.go 都声称迁移 0010 建了 `users_verified_email_unique` 部分唯一索引、`VerifyEmailOTP` 会返回 `ErrEmailTaken`;实际上**索引从未创建**、`ErrEmailTaken` 全仓库从未被返回 → 两个账号可同时验证一个邮箱,而登录门正是按 verified email 解析账号 → 同一邮箱的登录码归谁由数据库任意决定 | pgint 首跑即红:第二账号验证同邮箱返回 `nil`;`grep users_verified_email_unique internal/store/migrations/*.sql` 零命中。真机复验(auditfix17 邮箱 409 演练):同码二次 verify 仍 409(码未被消费) | 新增迁移 `0020_verified_email_unique.sql`(`lower(email) WHERE email_verified`);`VerifyEmailOTP` 在消费码**前**查「他人已验证」→ `ErrEmailTaken`(不消费码、不扣次数),并把并发唯一冲突映射为同一答案;handler 新增 409 `email_taken`。提交 `b6ef27c`;PG 契约测试基建 `2a55a0d`(`-tags pgint`,见 CONTRIBUTING) | | 21 | **admin 改邮箱不清 verified**:`UpdateUser` 写入新地址但保留 `email_verified=true` → 面板改错一个字符就能让登录码寄到别人邮箱(该 flag 正是邮箱登录解析/寄码的依据);fake 同错 | pgint 契约测试 | 改地址时同写 `email_verified = email_verified AND email IS NOT DISTINCT FROM 新值`(同值 no-op 保留证明;换值即清除);fake 同步。提交 `d1ec40f`。真机验证:PATCH→`f`、PATCH 回原值仍 `f`、玩家重验证→`t` | | 22 | **`role='owner'` 是死信**:迁移 0011 加了 owner 角色并把全部用户管理路由压在 `IsOwner()` 上,但**没有任何代码写过 `owner`**——breakGlass(`UpsertOwner`)、setup MC 绑定(`CompleteOwnerSetup`)、重置路径一律写 `admin` → 全新安装的整个 owner 层(用户列表/创建/编辑/禁用/删除/配额/会话)不可达;且新角色没进各处 staff 判定(op-login 只认 admin、玩家邮箱门只拒 admin、reclaim 保护与 AdminExists 只认 admin);面板一旦有 owner 行还可被降级/删除/禁用 | 真机:owner 会话加载 `/api/v1/users` 403;promote 后 200。op-login:修复前 owner start 中立不发码,修复后铸请求+finish 得 `role=owner` 会话 | `UpsertOwner`/`CompleteOwnerSetup` 改写 `role='owner'`(冲突臂重断言,即 0011 文档的 promote 路径);op-login 双端改 `staffRole`;玩家邮箱门改 `role != 'user'`;`IsProtectedAdminLink`/`AdminExists` 计入 owner;面板新增守卫:owner 行不可降级/删除/禁用(用户名/邮箱编辑仍可)。提交 `e0d2378` | +| 23 | **配额门非原子 + storage 缓存被清零**:(a) audit #4:`QuotaCheck` 与 `ClaimServer` 两条语句,同一用户并发认领两台无主服可双双通过 `max_servers`(deferred-seams 曾把这条挂为“只能在真 PG 上关闭”);(b) 更隐蔽:PATCH 资源时 `UpdateServerResources(..., 0)` 把本不能改的 storage 缓存写 0,而缓存列是配额聚合的**唯一**输入 → 此后该服的 storage 维度在配额里凭空消失 | (a) 新增 pgint 并发测试:修复前两台全赢;修复后恰 1 赢 + 1 `ErrQuotaExceeded`,DB 只 1 行 owned;(b) hermetic 测试 `TestPatchServerPreservesStorageCache`(修复前 `resourceUpdates` 里 storage=0) | (a) 门槛进 `ClaimServer`:同一事务内 `pg_advisory_xact_lock(hashtext(user_id))` + 四维重查(与 `QuotaCheck` 共用 `quotaAllows` 防漂移)+ 行 `FOR UPDATE`,两个 claim handler 把 `ErrQuotaExceeded` 映射为与串行一致的 403;(b) resize 前读取现值并透传 storage。提交 `bb68fef`;deferred-seams 对应条目核销 | ## 待决策台账(未修) @@ -75,6 +76,7 @@ IPv6-only 接入(`ssh -6 -i ~/.ssh/id_ed25519 root@fdb2:2c26:f4e4:0:21c:42ff:f - **邮箱唯一性真机 409 演练(auditfix17)**:owner 已验地址被玩家 onboarding 流程二次验证 → 第一次 409 `email_taken`;**同码重放仍是 409**(证明码未被消费,符合契约;若被消费会是 400)。 - **admin 改邮箱清 verified 真机演练(auditfix18)**:PATCH player.test → `player.test+edited@example.net` 后 `email_verified=f`;PATCH 回原地址仍 `f`;玩家走 onboarding 重验证 → `t`。 - **owner 角色真机闭环(auditfix18)**:owner 行按 0011 文档升级路径置 `role='owner'` → `GET /api/v1/users` 200(修复前 403);owner op-login start 铸请求+发码、finish 得 `role=owner` 会话;PATCH owner role / DELETE owner / DISABLE owner 全部 403;玩家邮箱登录回归 200。 +- **配额原子门(第五批,`bb68fef`,auditfix19 已上线)**:pgint 并发实证(真实 PG 上两路并发认领:恰 1 赢 + 1 `ErrQuotaExceeded`;修复前两路全赢);`docs/deferred-seams.md` 的 audit #4 条目核销。 ## 结论:离"生产可用"还差什么(按优先级) diff --git a/docs/deferred-seams.md b/docs/deferred-seams.md index b8e1e8c..96c63ac 100644 --- a/docs/deferred-seams.md +++ b/docs/deferred-seams.md @@ -83,8 +83,11 @@ or a real upstream account to run it against — not an implementation. These are decisions, not backlog. Each names the condition under which it would be worth revisiting. -- `internal/api/pgrepo.go:281` — the quota check and `ClaimServer` are two statements - (audit #4 TOCTOU). Closeable only against a real Postgres. +- ~~`internal/api/pgrepo.go:281` — the quota check and `ClaimServer` are two statements + (audit #4 TOCTOU). Closeable only against a real Postgres.~~ **Closed** — the gate + moved inside `ClaimServer` (advisory lock + re-check + UPDATE in one transaction), + red-then-green in the pgint suite, which is exactly the real-Postgres harness this + line was waiting for. - `internal/api/api.go:671` — `cooldownLimiter` is process-local, so across N api replicas a caller could draw up to N OTP codes per window. The intra-replica burst is closed; cross-replica bounding needs a shared store, out of scope for a