fix: security & privacy (rate limiter fix, csrf session cookie, logout on server, drafts stale token passw, login timing, password over 72b, JWT_SECRET complexity, SMTP TLS, CSV export neutralised, fonts bundled, upload metadata, hourly cleanup job, privacy.md report)

docs/api.md

@@ -21,6 +21,12 @@ for example `https://cms.example/v1/article/`.
   or revoked key answers 401, even on public routes. The site writes (`POST`, `PATCH` and `DELETE`
   under `/v1/site`) and the author writes (`POST`, `PATCH` and `DELETE` under `/v1/author`, `POST /v1/author/me` included) are session only too.
 - Each key can make 120 requests per minute. Beyond that: 429 with a `Retry-After` header.
+- A session lasts 24 hours. `POST /v1/auth/logout` revokes it on the server, so a copied cookie
+  stops working at once; other sessions of the same user stay valid. A password reset ends every
+  session and revokes every key of that user.
+- A browser `POST`, `PUT`, `PATCH` or `DELETE` sent from another origin without an
+  `Authorization` header answers 403 "Cross-origin requests are not allowed." This blocks
+  cross-site request forgery with the cookie; scripts on other origins must send their key.
 
 ## Routes a key can call (30)
 | Group | Routes |

docs/architecture.md

@@ -55,6 +55,14 @@ The use cases `GetPurgeSettings`, `UpdatePurgeSettings`, `PreviewPurge` and `Run
 | `GET /v1/media/purge-preview` | Groups, copies, distinct articles, the first 5 groups, `more` and the fingerprint. |
 | `POST /v1/media/purge` | Runs the purge. Optional `fingerprint`; 409 when stale or when a run is in progress. |
 
+### Sessions and personal data retention
+
+A session is an HS256 token from `api/pkg/authtoken`. `Encode` adds a random `jti` and an `exp` 24 hours ahead; `Decode` refuses a token without `exp`. `authtoken.WithRevocationStore` plugs in `useCases.SessionRevocationUseCase`, which stores a `session.Revocation` (`jti`, `exp`) through the `internal/domain/session` `Repository` (`persistence.RevokedSessionRepository`, table `revoked_sessions`). `LoginHandler.Logout` revokes the presented token; from then on `Decode` answers `ErrRevoked` for it, and a store that cannot be read makes every token invalid rather than valid. Revoking every session of a user stays the job of `token_version`, bumped by a password reset.
+
+`useCases.EnforceRetentionUseCase.Tick(now)` deletes revocations past their expiry, unverified accounts 7 days after their code expired (`user.UnverifiedRetention`) and invites 30 days after they expired or were accepted (`invite.Retention`), and returns a `retention.Result`. `scheduler.NewSystemRetentionRunner` calls it one minute after startup and then hourly, logs one line only when something was removed, and stops with the other runners on shutdown.
+
+Uploads go through the `image.MetadataStripper` port before they are measured for size, hashed and stored. `imaging.MetadataStripper` walks the JPEG segments, PNG chunks and WebP RIFF chunks without re-encoding pixels, drops metadata, keeps colour profiles and re-adds an orientation-only EXIF block when the photo was rotated. A container it cannot walk fails the upload with `image.ErrImageUnreadable`.
+
 ### Activity log
 
 `internal/domain/audit` holds the append-only `Event` (actor, target and failure-reason snapshots, outcome, security flag, source, ip, user agent, `params` and `changes`), the closed vocabularies (actor type, category, outcome, source, change kind), the `Filter`, `Summary` and `Settings` types, the `Repository` interface (`Append`, `List`, `Summary`, `DeleteOlderThan`: no way to edit or delete one event), the `SettingsRepository` (`Get`, `SaveRetention`, `SaveLastCleanup`; no row means 90 days) and the sentinels `ErrInvalidFilter`, `ErrInvalidRetention`, `ErrInvalidEvent` and `ErrActorNotAllowed`. The `Recorder` port is `Record(ctx, event)`; it returns nothing, so a caller can never fail on it.

docs/configuration.md

@@ -11,7 +11,7 @@ Copy [`.env.example`](../.env.example) to `.env` to get started. When a required
 | Name            | Description                                                                                                         | Example                    |
 |-----------------|---------------------------------------------------------------------------------------------------------------------|----------------------------|
 | `PUBLIC_URL`    | Public base URL of the site, including scheme and no trailing slash. Used for links in emails and in the startup log. It is also the CMS's own address: at the first migration, when the sites table is empty, it seeds the default site (name "Default", key "default") with its scheme and host (path dropped, non-loopback `http` upgraded to `https`); an unparseable value stops the migration. It is never read for sites afterwards. | `https://cms.example.com`  |
-| `JWT_SECRET`    | Secret used to sign the auth JWT. Use a long random value; changing it logs everyone out.                            | `openssl rand -hex 32`     |
+| `JWT_SECRET`    | Secret used to sign the auth JWT. It must be at least 32 bytes, and the example values from `.env.example` and `docker.env.example` are refused: the server stops at startup with `JWT_SECRET is too weak` and how to generate one. Changing it logs everyone out. | `openssl rand -hex 32`     |
 | `SMTP_LOGIN`    | SMTP username used to authenticate with the server. Some providers (e.g. Brevo) issue a login distinct from the sender address. | `noreply@example.com` |
 | `SMTP_SENDER`   | Sender address used in the `From` header. Must be an address/domain the SMTP provider allows that login to send as.  | `noreply@example.com`      |
 | `SMTP_PASSWORD` | SMTP password.                                                                                                      |                            |
@@ -32,6 +32,7 @@ Copy [`.env.example`](../.env.example) to `.env` to get started. When a required
 | `CORS_ALLOWED_ORIGINS` | *(empty)*            | Semicolon-separated list of origins allowed to call the API. Matching is case-insensitive. An entry may contain one `*` (`https://*.preview.example.com`), and a lone `*` allows every origin. Credentials are never allowed cross-origin. When empty, no CORS headers are sent. That is the right setting whenever the SPA is served by RenewCMS itself (production) or through the Vite proxy (development). |
 | `AUTH_RATE_LIMIT_MAX_ATTEMPTS` | `5`           | Maximum number of requests a single client IP may send to `POST /v1/auth/login`, `POST /v1/auth/password-reset` or the invite-accept pair (`GET /v1/auth/invite` and `POST /v1/auth/invite/accept`, sharing one counter) within the window, tracked independently per route.                                                          |
 | `AUTH_RATE_LIMIT_WINDOW_SECONDS` | `60`        | Length, in seconds, of the fixed rate-limit window for the routes above.                                                                                                                                                    |
+| `SMTP_ALLOW_INSECURE`  | `false`              | `true` lets RenewCMS deliver mail without encryption to a server on another host that does not offer STARTTLS (a relay on a private network, a local Mailpit container). By default such delivery fails with an error naming this variable; port `465` always uses implicit TLS and `localhost`, `127.0.0.1` and `::1` are always allowed in clear text. Credentials (`SMTP_LOGIN`/`SMTP_PASSWORD`) are never sent unencrypted to another host, whatever this says. Any value other than `true` or `false` (case-insensitive) stops the server at boot. |
 | `TRUSTED_PROXIES`      | *(empty)*            | Semicolon-separated list of exact IP addresses (no CIDR) of reverse proxies allowed to supply the client IP via `X-Real-IP` for the rate limiting of the routes above. When a request's connection address is not in this list, any `X-Real-IP` or `X-Forwarded-For` header it sends is ignored and the connection address is used instead. See [deployment.md](deployment.md).      |
 | `RENEWCMS_CONTAINER` | `false` | Set to `true` (case-insensitive; any other value means `false`) when the app runs in a container. Settings then tells the owner to update by pulling a new image instead of offering browser updates, and `can_apply` is always `false`. The Dockerfile and `docker-compose.yml` set it to `true`. Update checks still work. |
 | `IMAGE_VARIANT_MAX_SOURCE_BYTES` | `10485760` | Largest stored image, in bytes, that an [image variant](image-variants.md) may be generated from. A request for a bigger source answers 413 with a plain-text body. The original is never affected, and SVG and animated GIF are always served as they are. A value that is not a whole number greater than zero stops the server at boot with an error naming the variable. |

docs/deployment.md

@@ -274,7 +274,7 @@ Either way, set `TRUSTED_PROXIES` (see [configuration.md](configuration.md)) to
 
 RenewCMS sends `Content-Security-Policy` (including `frame-ancestors 'none'`), `X-Content-Type-Options`, `X-Frame-Options`, `Referrer-Policy` and `Cross-Origin-Opener-Policy` itself, so the proxy doesn't need to add them. With `ENVIRONMENT=production` it also sends `Strict-Transport-Security: max-age=31536000` (browsers ignore it over plain HTTP, so serve the site over HTTPS only). Don't add the same headers in the proxy, or they will be duplicated.
 
-The CSP allows Google Fonts (`fonts.googleapis.com`, `fonts.gstatic.com`) and `https:` images, because the SPA loads its fonts from Google and articles may embed remote images. `style-src` keeps `'unsafe-inline'` for the article editor; `script-src` is `'self'` only.
+The CSP serves fonts from `'self'` only (the SPA bundles its fonts, so no visitor IP address is sent to a font CDN) and allows `https:` images, because articles may embed remote images. `style-src` keeps `'unsafe-inline'` for the article editor; `script-src` is `'self'` only.
 
 ## Using MySQL or MariaDB
 

docs/privacy.md

@@ -0,0 +1,77 @@
+# Privacy and personal data
+
+This page is for operators: whoever runs a RenewCMS instance is the data controller for it under the GDPR. It lists what the software stores about people, where, for how long, and what it sends elsewhere, so you can fill in your record of processing (art. 30), write your privacy notice (art. 13) and answer requests from the people concerned. It describes the software, not your obligations; it is not legal advice.
+
+## Who is concerned
+
+- **Back-office users**: the owner and the admins they invite.
+- **Invitees** who have not accepted yet.
+- **People named in content**: authors, and anyone your editors write about or photograph.
+- **Anyone who performs an audited action** (logging in, failing to log in, changing anything) has their IP address and browser recorded in the activity log.
+
+Readers of your public site are not tracked by RenewCMS: the public routes set no cookie and log nothing in production.
+
+## What is stored, and for how long
+
+| Data | Where | Kept |
+|---|---|---|
+| Username, email, display name, role, bcrypt password hash | `users` table | Until the account is deleted (see [Gaps](#gaps)). |
+| Verification and password-reset codes | `users` table, bcrypt-hashed | Verification: until the account is verified. Reset: one hour, then useless; cleared when used. |
+| Unverified sign-ups | `users` table | Deleted automatically 7 days after their verification code expired. |
+| Invites: invitee email, hashed code, inviter | `invites` table | Deleted automatically 30 days after they expired or were accepted. |
+| Sessions | The `jwt` cookie in the user's browser, valid 24 hours | Logging out revokes the session on the server; the revocation (a random id and an expiry, nothing else) is deleted once the session would have expired anyway. |
+| API keys: name, prefix, SHA-256 hash, creator, last use | `api_keys` table | Until revoked. A revoked key stays listed. |
+| Activity log: who (user id and name, API key name, or the username typed at a failed login), what, when, IP address, browser (user agent, truncated to 255 characters) | `audit_events` table | Settings → Activity: 30, 90 (default), 365 days or forever, and an optional size cap. Cleaned up once a night, at 03:00 server time. |
+| Authors: name, slug, bio, avatar, website, social links; the linked user's email is shown to admins | `authors` table | Until deleted. |
+| Articles and custom entries | `articles`, `entries` tables | Deleted articles are soft-deleted: the row and its content stay in the database (see [Gaps](#gaps)). |
+| Uploaded images | `UPLOAD_DIR` or the S3 bucket, plus the `images` table (file name, size, uploader) | Until deleted from the media library. Deleted images are soft-deleted in the table; the file itself is removed. |
+
+### Photo metadata
+
+Since this version, uploads are cleaned before they are stored. JPEG, PNG and WebP files lose their EXIF (GPS position, camera serial number, date), XMP, IPTC, comments and text chunks; only the orientation is kept, so rotated photos still display correctly. The pixels are not re-encoded. SVG files are stored as uploaded. **Images uploaded with an earlier version are not cleaned retroactively**: if they may carry GPS positions, download, clean and re-upload them, or merge them into a cleaned copy from the media library.
+
+### Activity log retention
+
+For security logs, the CNIL generally recommends keeping access logs between six months and a year. The default of 90 days is shorter; choose 365 days in Settings if that suits your purpose. Avoid "Forever": it keeps IP addresses indefinitely, which the storage-limitation principle rarely justifies.
+
+## Cookies and browser storage
+
+- `jwt`: the session cookie. HttpOnly, `SameSite=Lax`, `Secure` in production, 24 hours. Strictly necessary for the back office, so it needs no consent (ePrivacy art. 5(3); in France, Loi Informatique et Libertés art. 82).
+- `localStorage` on the user's device holds interface choices only: language (`renewcms-language`), page sizes, open or collapsed panels, dismissed hints and columns. `sessionStorage` holds unsaved article drafts until the tab closes. None of it is sent to the server or to anyone else, and it needs no consent either.
+
+RenewCMS sets no analytics, advertising or third-party cookie, so no consent banner is needed for the back office. Your public site may need one if *it* adds trackers.
+
+## What leaves the server
+
+| Destination | When | What |
+|---|---|---|
+| Your SMTP provider (`SMTP_HOST`) | Sign-up, invite, password reset | Recipient address, the message (with a one-time link). STARTTLS or implicit TLS is required for any non-loopback host unless you set `SMTP_ALLOW_INSECURE=true`. |
+| Your S3 provider, if `STORAGE_DRIVER=s3` | Every upload and image read | The image files. |
+| `UPDATE_FEED_URL` (default `https://git.floriansylvain.fr`) | Only when the owner clicks "Check for updates" or downloads one | The server's IP address and a standard HTTP request. No account or content data. |
+
+The back office loads nothing from third parties: fonts, icons and scripts are served by RenewCMS itself, and the Content-Security-Policy forbids any other source. Pick EU-based SMTP and storage providers, or make sure transfer safeguards (art. 44 and following) cover them, and list them as processors.
+
+## Security measures you can cite (art. 32)
+
+- Passwords are hashed with bcrypt (cost 12). One-time codes are random UUIDs, stored hashed, single-use and short-lived. API keys are 256-bit random values stored as SHA-256 hashes.
+- Login, password-reset and invite routes are rate-limited per client IP; failed logins are recorded as security events.
+- Sessions are revocable: logout ends one session, a password reset ends all of them and revokes the user's API keys.
+- Cross-site writes are refused, cookies are `SameSite=Lax`, and a strict Content-Security-Policy, `X-Frame-Options: DENY` and HSTS (in production) are sent.
+- The server refuses to start with a short or example `JWT_SECRET`.
+- Release binaries are signed (ed25519) and checked before a browser update is installed.
+
+## Answering requests from the people concerned
+
+- **Access and portability (art. 15 and 20)**: there is no self-service export yet. The data of a user is their row in `users`, their author profile, their API keys and the activity events where they are the actor; an owner can export the activity log as CSV from the Activity page and filter it by person.
+- **Rectification (art. 16)**: users can change their display name and author profile themselves. Changing an email address currently needs a database edit.
+- **Erasure (art. 17)**: an unverified account can cancel its own registration. Deleting a verified user has no button yet; see [Gaps](#gaps).
+
+## Gaps
+
+These are known and tracked; until they are fixed, handle them by hand.
+
+- No way to delete a verified user or remove an admin from the interface.
+- No self-service export or email change.
+- Soft-deleted articles and images keep their rows indefinitely.
+- The username typed at a failed login is stored as typed, so a password typed in the wrong field can end up in the activity log; clean such events by hand if it happens.
+- Images uploaded before metadata stripping keep their metadata.