docs: rewrite security/operator docs for the multi-user panel (P2)

The CSRF ADR, guide.md, architecture.md route table, the internal/backup
package comment, roadmap/dmarc-reports admin.dmarc_report_email references,
the e2e coverage summary, and the setupManager/store package comments still
described the pre-1.2.0 single-administrator panel or stale implementation
details. Closes P2 of docs/plans/code-review.md; no behaviour change.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

docs/architecture.md

@@ -161,8 +161,10 @@ unless noted. The table below is a summary — HTMX fragment endpoints
 | Route | Purpose |
 |---|---|
 | `/healthz` | Liveness (no auth) |
+| `/license` | Embedded `LICENSE` text (no auth) |
 | `/setup/*` | One-time admin bootstrap |
 | `/login`, `/logout` | Session auth |
+| `/account` | 308 redirect to `/settings` (pre-1.2.3 route, kept as a compat shim) |
 | `/status` | Process, cert, socket, PTR checks; machine CPU/memory/network |
 | `/domains`, `/domains/*` | Domain and application CRUD, DKIM, L2 limits |
 | `/domains/import` | Domain import (`POST`; form on the Backup page) |

docs/development.md

@@ -265,7 +265,11 @@ operator would actually use.
 **Coverage (summary):** bootstrap → SMTP AUTH → delivery → DKIM verify →
 send-log `queued → sent`; negatives (no AUTH, relay, sender/login mismatch,
 L1/L2 limits, milter fail-open, bad `SELFPOST_HOSTNAME`, session survives
-`docker restart`). Polling with timeouts only — no fixed `sleep`.
+`docker restart`); startup checks that supervisord actually brought up
+OpenDKIM, the panel, and Postfix (`checkSupervisorProcesses`), plus logrotate
+config-mode and forced-rotation checks (`checkLogrotateConfigMode`,
+`checkLogrotateRotation` — [test/e2e/logrotate_check.go](../test/e2e/logrotate_check.go)).
+Polling with timeouts only — no fixed `sleep`.
 
 Requires **Docker + Compose v2** on the machine running the suite.
 

docs/guide.md

@@ -230,8 +230,22 @@ service healthy and will mail be accepted?"
 - **Backup** (`/backup`) — download a full-server backup; the same page hosts
   the domain-import form (`POST /domains/import`). See
   [Backup, restore, and moving a single domain](#backup-restore-and-moving-a-single-domain).
-- **Settings** (`/settings`) — change the administrator username and/or password.
-  Application SASL logins are separate and are not changed here.
+- **Settings** (`/settings`) — change the administrator username and/or
+  password, and the panel-wide default DMARC report address (`rua=`) offered
+  when a domain doesn't set its own — see [DNS setup](#dns-setup). Application
+  SASL logins are separate and are not changed here.
+- **Users** (`/users`, global administrator only) — create, edit, and delete
+  panel users. There are two roles:
+  - **Global administrator** — full access to every page and every domain,
+    including Users, Backup, Status, Mail queue, and System log.
+  - **Domain-admin** — scoped to one or more domains assigned by a global
+    administrator. Sees only those domains' pages, applications, and
+    Deliveries rows; `/users`, `/backup`, `/status`, `/mail-queue`, and
+    `/system-log` are not reachable (404). A domain-admin can *export* the
+    domains assigned to them — see the note on working credentials below.
+
+  The panel refuses to remove or demote the **last** global administrator, so
+  it can never end up with none.
 
 **Sessions.** A login survives a container restart: sessions live in SQLite, not
 in memory. Expiry is a sliding idle window (`PANEL_SESSION_IDLE_DAYS`, default
@@ -296,6 +310,12 @@ An application override without trusted IPs is inactive.
 When a level-2 ceiling is exceeded, Postfix returns a 4xx and the refusal is
 recorded in Deliveries as `rejected`.
 
+**Level 2 is best-effort, not a guarantee.** It runs inside the journal-milter
+and is deliberately fail-open: if the rate-limit lookup hits a store error, or
+the connecting client's IP is not available to the milter, level 2 is skipped
+and the message is accepted rather than held up. Level 1 (the Postfix anvil
+limit above) is the backstop that keeps working even when level 2 cannot run.
+
 ## Backup, restore, and moving a single domain
 
 Two related but distinct operations
@@ -317,6 +337,12 @@ Two related but distinct operations
   `:latest`: without a known version, there'd be no way to tell which image
   restoring a given backup actually requires.
 
+  Restoring an archive taken **before** you invalidated a session (password
+  change, logout everywhere) can bring that session back: session rows travel
+  with the backup, and a browser that still holds the matching cookie is
+  logged in again once the idle timeout allows it. If a restore might do this,
+  changing every user's password afterwards clears it out.
+
   **Alternative: archive `./data` while stopped.** If the service can be taken
   offline, `docker compose down` then `tar czf selfpost-data.tar.gz ./data` on
   the host is safe — nothing is writing to SQLite. Unlike the panel/CLI backup
@@ -332,7 +358,12 @@ Two related but distinct operations
   file, *Backup* → *Import a domain* to read it back in): moves one domain — its DKIM key and its applications' **working**
   SASL passwords — to a different SelfPost instance without regenerating
   anything, so DNS (the DKIM TXT record) doesn't need to change. Unlike a full
-  restore, this works across different hostnames/instances.
+  restore, this works across different hostnames/instances. *Import* is
+  global-administrator only; *export* is available to any user who can access
+  the domain, **including a domain-admin** for a domain assigned to them — so
+  a domain-admin can walk away with that domain's working SASL passwords in
+  the clear. Weigh that when deciding which domains to assign to a
+  domain-admin account.
 
 Both files are **secrets** — they contain the admin password hash (full
 backup) or working application credentials (domain export) in the clear or in

docs/plans/code-review.md

@@ -522,21 +522,21 @@ the list, the fragment, or query parameters. `go test ./...` green.
 
 **Model: Sonnet** (English docs). No code behaviour change except copy.
 
-- [ ] Rewrite the CSRF ADR in [security.md](../security.md) for a panel that
+- [x] Rewrite the CSRF ADR in [security.md](../security.md) for a panel that
       already has global + domain-admin. Keep the origin-check mechanism
       unless a new decision says otherwise. New revisit trigger (e.g. untrusted
       domain-admins, or a requirement that does not depend on browser
       headers).
-- [ ] Drop “or argon2” unless argon2 is implemented.
-- [ ] [guide.md](../guide.md): Users / roles; Settings DMARC field; L2
+- [x] Drop “or argon2” unless argon2 is implemented.
+- [x] [guide.md](../guide.md): Users / roles; Settings DMARC field; L2
       fail-open; restore can resurrect sessions; domain-admin can export
       working SASL passwords for assigned domains.
-- [ ] Architecture route table: `/license`, `/account` → `/settings`.
-- [ ] Fix `internal/backup` package comment (restore = extract tarball +
+- [x] Architecture route table: `/license`, `/account` → `/settings`.
+- [x] Fix `internal/backup` package comment (restore = extract tarball +
       `CheckRestore`; maps come from the archive; Reload heals drift).
-- [ ] `admin.dmarc_report_email` → `settings` in roadmap + dmarc-reports plan.
-- [ ] development.md e2e summary: logrotate + process checks.
-- [ ] `setupManager` / `store` package comments: `users`, not `admin` row.
+- [x] `admin.dmarc_report_email` → `settings` in roadmap + dmarc-reports plan.
+- [x] development.md e2e summary: logrotate + process checks.
+- [x] `setupManager` / `store` package comments: `users`, not `admin` row.
 
 **Done when:** an operator who reads only the guide knows the panel has two
 roles, and security.md no longer calls the panel single-user.

docs/plans/dmarc-reports.md

@@ -18,7 +18,8 @@ operator.
 - gzip + XML aggregate parsing → SQLite summaries per sending domain.
 - Panel page and/or per-domain section: recent reports, third-party senders,
   delivery health of report ingestion.
-- Reuse `admin.dmarc_report_email` and `domains.dmarc_rua` for DNS templates;
+- Reuse the `dmarc_report_email` setting (moved off the old `admin` table into
+  `settings` by migration `0005`) and `domains.dmarc_rua` for DNS templates;
   when enabled, suggest a SelfPost-hosted report address.
 
 **Out:**

docs/roadmap.md

@@ -118,9 +118,10 @@ reports (`ruf=`) out of scope for v1.
 
 **Dependencies / risks:** inbound SMTP in the image (may share infrastructure
 with [inbound-relay](plans/inbound-relay.md) but must not require backup-MX);
-storage and retention of parsed summaries; the `admin.dmarc_report_email` and
-`domains.dmarc_rua` settings added in the DMARC template work must stay the
-source of truth for `rua=` in DNS guidance.
+storage and retention of parsed summaries; the `dmarc_report_email` setting
+(migration `0005` moved it off the old `admin` table into `settings`) and
+`domains.dmarc_rua` added in the DMARC template work must stay the source of
+truth for `rua=` in DNS guidance.
 
 **Order:** after the DMARC `rua=` settings ship; may follow or overlap with
 inbound-relay depending on how port 25 acceptance is structured.

docs/security.md

@@ -34,8 +34,8 @@ The panel is exposed to the internet — the items below are **not optional**.
 - Failed attempts do **not** invalidate the token early (protects setup from
   being DoS-ed).
 - Once the administrator exists the token is void forever, `/setup/*` → 404.
-- The administrator password is bcrypt (or argon2) in SQLite only; no plaintext
-  and no MD5.
+- The administrator password is bcrypt in SQLite only; no plaintext and no
+  MD5.
 - `PANEL_USERNAME` / `PANEL_PASSWORD_HASH` in env are **not used**.
 
 ### Application SASL passwords
@@ -112,19 +112,21 @@ deferred item from the roadmap.
 - **A `POST` with neither `Sec-Fetch-Site` nor `Origin` is allowed through.**
   A client that sends neither — a genuinely old browser, or a webview with a
   frozen engine — stays vulnerable to CSRF from any site. Accepted
-  deliberately: the panel is single-user, the administrator picks the browser,
-  and a strict mode would not "protect" such a client, it would simply break the
-  panel in it. Tightening is one line in `originAllowed`
+  deliberately: every panel user (global or domain-admin) is an operator who
+  picks their own browser, not an untrusted party the panel needs to defend
+  against, and a strict mode would not "protect" such a client, it would
+  simply break the panel in it. Tightening is one line in `originAllowed`
   ([internal/web/security.go](../internal/web/security.go)): return `false`
   instead of `true` in the "neither header present" branch.
 - **Session-bound CSRF tokens are not implemented.** The origin check closes the
   neighbouring-subdomain case but depends on browser behaviour; a token does
   not. The price is a hidden field in roughly two dozen forms. The trigger to
-  revisit is a requirement for protection that holds regardless of the browser.
-  A token would not save the panel from XSS inside it either: code executing in
-  the panel's origin sends the request itself — against that, `html/template`
-  auto-escaping and CSP do the work, which is why templates must contain no
-  inline scripts and no inline styles.
+  revisit is a requirement for protection that holds regardless of the browser,
+  or a domain-admin population the global administrator does not fully trust
+  (see the ADR below). A token would not save the panel from XSS inside it
+  either: code executing in the panel's origin sends the request itself —
+  against that, `html/template` auto-escaping and CSP do the work, which is why
+  templates must contain no inline scripts and no inline styles.
 - **Encrypting backups and exports is an option, not the default.** With the
   checkbox cleared the file downloads in the clear, as in 1.0. Otherwise an
   operator with nowhere to keep a password would lose the ability to take a
@@ -158,24 +160,35 @@ deferred item from the roadmap.
 
 **Context.** The panel is forms (`POST`) with a cookie session — the classic
 CSRF surface. What is needed is a way to tell a request from the panel's own
-page apart from one initiated by a third-party site in the logged-in
-administrator's browser.
+page apart from one initiated by a third-party site in a logged-in user's
+browser. The panel is multi-user since 1.2.0 (a global administrator plus
+zero or more domain-admin users, each scoped to their assigned domains), but
+that is an authorization boundary (who can see or change what), not a change
+to the CSRF threat: the attacker in scope here is still an external site
+riding a legitimate user's cookie, not one panel user attacking another
+through the browser.
 
 **Decision.** `originAllowed` in
 [internal/web/security.go](../internal/web/security.go) checks `Sec-Fetch-Site`
 (when the browser sends it) or `Origin` (fallback) against the panel's host; a
 request carrying neither header is **allowed through** rather than rejected.
-There are no session-bound tokens embedded in forms.
+There are no session-bound tokens embedded in forms. The check applies the same
+way regardless of the requesting user's role.
 
-**Why not tokens.** The panel is single-user (one administrator per instance) —
-the threat model does not include cross-user CSRF inside the panel itself, only
-an external site making the administrator's browser send a request. The origin
-check covers that without touching a single template: a token would need a
-hidden field in roughly two dozen forms and synchronisation with every new form,
-and it would still not protect against XSS inside the panel — code executing in
-the panel's origin reads the token and sends the request itself. XSS is handled
-by `html/template` auto-escaping and CSP, so that is a separate line of defence,
-not a CSRF token.
+**Why not tokens.** Cross-user CSRF is not the threat model here: a
+domain-admin's browser sending a request still needs that domain-admin's own
+cookie, so a token would not add a boundary between roles that the
+authorization checks (`Principal.CanAccessDomain`,
+[internal/web/auth/principal.go](../internal/web/auth/principal.go); route
+gating in [internal/web/auth/middleware.go](../internal/web/auth/middleware.go))
+don't already enforce. The remaining case is an external site making a
+logged-in user's browser send a request, which the origin check covers without
+touching a single template.
+A token would need a hidden field in roughly two dozen forms and
+synchronisation with every new form, and it would still not protect against
+XSS inside the panel — code executing in the panel's origin reads the token
+and sends the request itself. XSS is handled by `html/template` auto-escaping
+and CSP, so that is a separate line of defence, not a CSRF token.
 
 **Trade-off.** A client that sends neither `Sec-Fetch-Site` nor `Origin` (a
 genuinely old browser, or a webview with a frozen engine) stays vulnerable — see
@@ -183,7 +196,11 @@ genuinely old browser, or a webview with a frozen engine) stays vulnerable — s
 such a client, at the price of a narrow residual surface.
 
 **Revisit if:** a requirement appears for protection that does not depend on
-browser behaviour, or the panel becomes multi-user.
+browser behaviour, or domain-admin accounts stop being trusted operators (for
+example, if a future release lets a global administrator invite domain-admins
+whose browsers/devices are not vetted) — at that point cross-role request
+forgery inside the panel would need its own analysis, separate from the
+external-site case this ADR covers.
 
 ## How this list grows