Files
relicario/docs/FORMATS.md
adlee-was-taken db0ab1d82e docs(formats): org collection-scoped attachment layout + default cap
Document the attachments/<slug>/<item-id>/<att-id>.enc layout (exactly 3
segments, slug-authorized by the pre-receive hook, never decrypted
server-side) and DEFAULT_ORG_ATTACHMENT_MAX_BYTES = 10 MiB, citing
org_session.rs:24 and the mirrored personal default settings.rs:116.
2026-06-20 21:08:23 -04:00

170 lines
11 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Relicario Wire Formats
> **Audience:** anyone implementing a compatible client or reading raw vault bytes. This doc owns the `.enc` blob layout, `params.json` / `salt` / `devices.json` / `revoked.json` shapes, the manifest JSON schema, the `.relbak` envelope, item-ID formats, and the settings JSON schema. **Does NOT own:** why these formats look this way (see [CRYPTO.md](CRYPTO.md)), threat model around them (see [SECURITY.md](SECURITY.md)), or Rust struct internals (see [../crates/relicario-core/ARCHITECTURE.md](../crates/relicario-core/ARCHITECTURE.md)).
> Quick-reference for the load-bearing binary and JSON formats. Check this file before touching serialization, versioning, or storage layout code. Full diagrams and invariants live in the per-crate `ARCHITECTURE.md` files.
## Encrypted blob (`.enc` files)
Every encrypted file — `manifest.enc`, `settings.enc`, `items/<id>.enc`, `attachments/<item-id>/<aid>.enc` — uses the layout produced by `relicario_core::crypto::encrypt` (`crypto.rs`):
```
┌─────────┬────────────────────────┬──────────────────┬──────────────────┐
│ version │ nonce │ ciphertext │ auth tag │
│ 1 byte │ 24 bytes │ N bytes │ 16 bytes │
│ 0x02 │ random per write │ XChaCha20 stream │ Poly1305 MAC │
└─────────┴────────────────────────┴──────────────────┴──────────────────┘
```
- `VERSION_BYTE = 0x02` (`crypto.rs:59`). Any blob starting with `0x01` is rejected with `UnsupportedFormatVersion { found: 0x01, expected: 0x02 }`.
- Minimum valid blob length: 41 bytes (1 + 24 + 0 + 16).
- Nonces are always fresh from `OsRng` — no caller-supplied nonces.
- Full diagram: `docs/CRYPTO.md` § "Encrypted File Format".
## `.relicario/params.json`
```json
{
"format_version": 2,
"aead": "xchacha20-poly1305",
"salt_path": ".relicario/salt",
"kdf": {
"argon2_m": 65536,
"argon2_t": 3,
"argon2_p": 4
}
}
```
Parsed via `ParamsFile { kdf: KdfParams }` in `session.rs`. The `kdf` nesting is intentional — `format_version`, `aead`, and `salt_path` co-exist for forward-compat probing. Do not flatten. Production defaults: `m=65536` (64 MiB), `t=3`, `p=4`. Tests use `m=256, t=1, p=1`.
## `.relicario/salt`
32 raw bytes. Not secret. Generated once at vault init via `OsRng`. Feeds Argon2id as the KDF salt.
## Manifest (`manifest.enc`)
Decrypts to JSON matching the `Manifest` struct (`manifest.rs`).
- **Schema version:** `MANIFEST_SCHEMA_VERSION = 2` (`manifest.rs:12`). v1 manifests (pre-typed-items) fail to parse and are not supported.
- **`ManifestEntry` fields** (declared order in `manifest.rs:21-38`): `id`, `type`, `title`, `tags`, `favorite`, `group`, `icon_hint`, `modified`, `trashed_at`, `attachment_summaries`. The `type` field is `r#type: ItemType` in Rust but serializes as the bare JSON key `"type"` (no serde rename — `r#` is just the raw-identifier escape). `group`, `icon_hint`, and `trashed_at` are `#[serde(skip_serializing_if = "Option::is_none")]`; `tags`, `favorite`, and `attachment_summaries` use `#[serde(default)]`.
- The manifest is rebuilt from scratch on every `upsert` — it can never drift from the source-of-truth item files.
- Supports case-insensitive title/tag search without decrypting any item.
## `.relicario/devices.json`
```json
[
{ "name": "laptop", "public_key": "<hex-encoded ed25519 public key>" }
]
```
An empty array (`[]`) puts the pre-receive hook in bootstrap mode (all pushes accepted). Both `devices.json` and `revoked.json` must be empty for bootstrap mode to activate — a non-empty `revoked.json` alone forces strict verification.
## `.relicario/revoked.json`
```json
[
{ "name": "old-laptop", "public_key": "<hex>", "revoked_at": 1746000000 }
]
```
Commits by `public_key` at or after `revoked_at` (Unix seconds) are rejected by the pre-receive hook. Commits before `revoked_at` remain valid (they were authorized at the time).
## Org vault repo formats
The org vault is a **separate git repository** alongside the personal vault. It is not nested inside `.relicario/`. Its layout:
```
org.json # OrgMeta (schema_version, org_id, display_name, created_at)
members.json # PUBLIC/unencrypted member directory
collections.json # collection definitions
keys/<member-id>.enc # org master key wrapped to that member's device key
manifest.enc # OrgManifest (schema_version 1, per-member-filtered)
items/<collection-slug>/<item-id>.enc # collection-scoped item blobs
attachments/<collection-slug>/<item-id>/<att-id>.enc # Document attachment blobs (collection-scoped)
```
### `org.json` — OrgMeta
Unencrypted JSON (`OrgMeta`, `org.rs:164`). `schema_version: 1` (`org.rs:174`). Fields: `schema_version`, `org_id`, `display_name`, `created_at` (Unix seconds).
### `members.json` — OrgMembers
Unencrypted JSON array of `OrgMember` records (`org.rs:72`); container type `OrgMembers` carries `schema_version: 1` (`org.rs:93`). Per-member fields: `member_id` (16 lowercase hex chars), `display_name`, `role` (one of `owner | admin | member`), `ed25519_pubkey` (OpenSSH wire string), `collections` (array of granted slug strings), `added_at`, `added_by`. Roles are not secrets — authorization to read this file is not required to verify signatures.
### `collections.json` — OrgCollections
Unencrypted JSON; `schema_version: 1` (`org.rs:138`). Contains a list of `CollectionDef` records (`org.rs:123`). Validation (`org.rs:145`) rejects slugs that are empty, contain `/`, or equal `.`.
### `keys/<member-id>.enc` — wrapped org master key
Binary blob; NOT a standard `.enc` blob. Layout (`org.rs:264`):
```
┌──────────────────────────┬─────────┬────────┬──────────────────────┐
│ ephemeral_x25519_pubkey │ version │ nonce │ ciphertext + tag │
│ 32 bytes │ 1 byte │24 bytes│ N + 16 bytes │
└──────────────────────────┴─────────┴────────┴──────────────────────┘
```
- The wrapping key is `SHA-256(dh_shared || ephemeral_pubkey || recipient_pubkey)` (`org.rs:278281`), held in `Zeroizing<Vec<u8>>`.
- The inner AEAD (`version || nonce || ciphertext+tag`) is produced by `crate::crypto::encrypt` — the same XChaCha20-Poly1305 framing used for personal `.enc` blobs (see **Encrypted blob** above). `VERSION_BYTE = 0x02` applies here too.
- The X25519 private scalar is derived from the device ed25519 seed via `SHA-512(seed)[..32]` with RFC 7748 clamping (`org.rs:242`). Argon2id is **not** involved — the wrapping key is derived entirely from the X25519 DH exchange.
### `manifest.enc` — OrgManifest
Encrypted with the org master key using `crypto::encrypt` (standard `.enc` framing). Decrypts to `OrgManifest` JSON (`org.rs:199`); `schema_version: 1` (`org.rs:206`). Each `OrgManifestEntry` (`org.rs:185`) carries: `id`, `type`, `title`, `tags`, `modified`, `trashed_at`, and a `collection` slug field. The `collection` field distinguishes this type from `ManifestEntry` in the personal vault.
Contrast with the personal vault manifest: `Manifest` uses `MANIFEST_SCHEMA_VERSION = 2` (`manifest.rs:12`) and `ManifestEntry` has no `collection` field. The two types are distinct and do not share a schema.
### `items/<collection-slug>/<item-id>.enc`
Standard `.enc` blob (see **Encrypted blob** above), encrypted under the org master key. The blob itself does **not** name its collection — the directory path segment carries the slug. This allows the pre-receive hook (`relicario-server`) to authorize a write by path segment without decrypting the blob.
These blobs are written and read by the `relicario org` item commands (`org add` / `get` / `list` / `edit` / `rm` / `restore` / `purge`), all collection-scoped and grant-enforced. `org add` currently creates Login / SecureNote / Identity items; `get` / `list` display any item type present.
### `attachments/<collection-slug>/<item-id>/<att-id>.enc`
Standard `.enc` blob (see **Encrypted blob** above), encrypted under the org master key — the encrypted file payload of a Document item. As with item blobs, the blob does **not** name its collection; the leading `<collection-slug>` path segment carries it, so the pre-receive hook (`relicario-server`, `classify_path`) authorizes the write by slug without decrypting — reusing the same grant + slug-existence check as the `items/` branch. The path is **exactly three segments** after `attachments/` (`<collection-slug>/<item-id>/<att-id>.enc`); the hook rejects any other shape (segment-count and `.`-free slug guards). `<att-id>` is the content-addressed `AttachmentId` (see **Item IDs and Field IDs** below).
Per-attachment size is capped at `DEFAULT_ORG_ATTACHMENT_MAX_BYTES = 10 * 1024 * 1024` (10 MiB) (`org_session.rs:24`), mirroring the personal-vault default `AttachmentCaps::per_attachment_max_bytes` (`settings.rs:116`). Org vaults have no `settings.enc`, so this cap is a fixed default rather than per-org configurable. Blobs are persisted / read / removed by `UnlockedOrgVault::save_attachment` / `load_attachment` / `remove_item_attachments` (`org_session.rs:137`, `:147`, `:156`). The storage primitives back the org **Document** item type; the `org add document` / Document-edit commands that produce these blobs land in v0.8.1 (see the item-type-parity note below).
**TODO (extension follow-up):** extension UI for browsing and editing org vault items. **Deferred:** `org add` / `edit` parity for Card / Key / Document / Totp item types (landing in v0.8.1; Document file payloads use the attachment layout above).
## Item IDs and Field IDs
| Kind | Length | Entropy | Source |
|---|---|---|---|
| `ItemId` | 16 hex chars | 64 bits | `OsRng` |
| `FieldId` | 16 hex chars | 64 bits | `OsRng` |
| `AttachmentId` | 32 hex chars | 128 bits | first 16 bytes (32 hex chars) of `SHA-256` over the plaintext |
`AttachmentId` is content-addressed — identical plaintexts deduplicate in git automatically. The 128-bit truncation (`ids.rs:59-69`) was widened from 64 bits per audit I2/B4 to put birthday-collision risk out of reach.
## `.relbak` backup format
A zstd-compressed tar archive containing a bare git clone of the vault. Designed for `relicario backup export/restore`.
Full spec: `docs/superpowers/specs/2026-04-27-relicario-import-export-design.md`.
## `ItemCore` JSON (internal)
`ItemCore` uses `#[serde(tag = "type")]` — the outer JSON object gets a `"type"` discriminator key. No `*Core` struct may have a field named `"type"` (use `"kind"` instead — see `CardKind`, `TotpKind`).
Full item type inventory: `crates/relicario-core/ARCHITECTURE.md` § "Module map".
## KDF input construction
The password fed to Argon2id is length-prefixed to prevent extension attacks:
```
u64_be(len(passphrase)) || passphrase_bytes || u64_be(32) || image_secret
```
NFC-normalized before hashing. Covered in `crypto.rs:229-236` and tested in `tests/format_v2.rs:44-54`.
---
**Next:** [SECURITY.md](SECURITY.md) — the threat model.