mirror of
https://github.com/mountain-loop/yaak.git
synced 2026-08-24 12:24:01 +02:00
Serve the web client from yaak-web (#582)
This commit is contained in:
@@ -1,139 +0,0 @@
|
||||
# yaak-send-proxy
|
||||
|
||||
The network half of Yaak in a browser.
|
||||
|
||||
A tab can't see an HTTP response the way a desktop app can: CORS hides most
|
||||
headers (2 of 8 in a typical response), redirects are followed silently, and
|
||||
there is no timeline. So the tab renders the request and posts it here, and this
|
||||
process puts it on the network with the desktop's own engine (`yaak-http`) and
|
||||
streams back everything that happened — every header, every redirect hop, DNS
|
||||
timing, the body — for the tab to store.
|
||||
|
||||
It is a **stateless executor**. It keeps nothing: no database, no files, no
|
||||
sessions, no cookies between calls. Every byte it sees comes from the tab in the
|
||||
request, and every byte it returns is stored by the tab. Restart it any time.
|
||||
|
||||
## Running it
|
||||
|
||||
```shell
|
||||
cargo run -p yaak-send-proxy
|
||||
```
|
||||
|
||||
Listens on `127.0.0.1:9227`. Then run the web build against it:
|
||||
|
||||
```shell
|
||||
YAAK_TARGET=web npm run dev --workspace @yaakapp/yaak-client
|
||||
```
|
||||
|
||||
The tab looks for the proxy at `http://127.0.0.1:9227` unless
|
||||
`VITE_YAAK_SEND_PROXY_URL` says otherwise at build time.
|
||||
|
||||
Every flag has a `YAAK_PROXY_*` environment variable, so a container needs no
|
||||
arguments; `--help` lists them all.
|
||||
|
||||
| Flag | Default | What |
|
||||
| --- | --- | --- |
|
||||
| `--bind` | `127.0.0.1:9227` | Listen address. `0.0.0.0:9227` inside a container. |
|
||||
| `--allowed-origins` | `*` | CORS origins, comma-separated. A hosted instance should name its web origin. |
|
||||
| `--max-request-bytes` | 16 MiB | Largest rendered request accepted from the tab. |
|
||||
| `--max-response-bytes` | 64 MiB | Largest upstream body relayed before the send is cut off. |
|
||||
| `--max-timeout-secs` | 60 | Ceiling on a send's timeout; a request asking for more (or none) gets this. |
|
||||
| `--rate-limit-per-minute` | 120 | Sends per client IP per minute; 0 disables. |
|
||||
| `--max-concurrent` | 256 | Sends in flight at once. |
|
||||
| `--trust-forwarded-for` | off | Take the client IP from `X-Forwarded-For`. Only behind a load balancer that sets it. |
|
||||
|
||||
## What it refuses, and why
|
||||
|
||||
A hosted proxy is, by construction, a machine that makes HTTP requests on
|
||||
behalf of strangers. Left alone that is an open relay into whatever network it
|
||||
sits on. So it refuses, always, to connect to:
|
||||
|
||||
- loopback (`127/8`, `::1`), private (`10/8`, `172.16/12`, `192.168/16`,
|
||||
`fc00::/7`), link-local (`169.254/16` — where cloud metadata lives — and
|
||||
`fe80::/10`), carrier-grade NAT, multicast, reserved and unspecified ranges,
|
||||
IPv4 addresses carried inside IPv6 forms (`::ffff:a.b.c.d`, the well-known
|
||||
NAT64 prefix, 6to4), and the whole NAT64 local-use range;
|
||||
- anything not `http://` or `https://`.
|
||||
|
||||
The check runs **on the resolved addresses, after DNS**, for every hop of a
|
||||
redirect chain, so a public hostname that points at an internal address is
|
||||
caught, and so is a `Location:` header that points at one. It also refuses body
|
||||
types that would read files on the proxy's disk (`binary`, multipart file
|
||||
fields), since no browser tab could legitimately mean those.
|
||||
|
||||
Refusals are logged with the reason. There is no switch to turn this off: the
|
||||
proxy's private network is the cloud's, not the user's, so a `localhost` or LAN
|
||||
API can never be reached through it — that is what the desktop app is for.
|
||||
|
||||
## Deploying
|
||||
|
||||
One binary, no dependencies:
|
||||
|
||||
```shell
|
||||
cargo build --release -p yaak-send-proxy
|
||||
YAAK_PROXY_BIND=0.0.0.0:9227 \
|
||||
YAAK_PROXY_ALLOWED_ORIGINS=https://yaak.example.com \
|
||||
./target/release/yaak-send-proxy
|
||||
```
|
||||
|
||||
There is no authentication: an instance is anonymous and protected by the
|
||||
per-client rate limit and the destination policy, which is what the hosted
|
||||
funnel wants. Anything more (a shared token, per-user quotas) is a later slice
|
||||
and would sit in front of `send_http` in `main.rs`. Put TLS in front of it (a
|
||||
reverse proxy). If the reverse proxy buffers responses, tell it not to: the reply
|
||||
is a stream and the `X-Accel-Buffering: no` header it sets is honoured by
|
||||
nginx-shaped ones.
|
||||
|
||||
## The wire
|
||||
|
||||
`POST /v1/http/send` with a JSON body:
|
||||
|
||||
```json
|
||||
{
|
||||
"request": { "url": "https://…", "method": "GET", "headers": […], "body": {…}, "bodyType": null, "urlParameters": […] },
|
||||
"settings": { "validateCertificates": true, "followRedirects": true, "timeoutMs": 0, "sendCookies": true, "storeCookies": true },
|
||||
"cookies": [ … ]
|
||||
}
|
||||
```
|
||||
|
||||
`request` is a Yaak `HttpRequest` in the desktop's own model shape with every
|
||||
template already rendered by the tab; the proxy builds the URL, headers and
|
||||
body from it exactly the way the desktop does after rendering. `cookies` is the
|
||||
jar's contents (or `null` for no jar).
|
||||
|
||||
The reply is `application/x-ndjson`, one JSON frame per line, in the order things
|
||||
happened:
|
||||
|
||||
| `type` | When | Carries |
|
||||
| --- | --- | --- |
|
||||
| `event` | as the engine produces them | one timeline event, in the desktop's `http_response_event.event` shape |
|
||||
| `response` | once, when the final hop's headers arrive | status, all headers, request headers as sent, remote address, HTTP version, timing |
|
||||
| `body` | as the body is read | a decompressed chunk, base64 |
|
||||
| `done` | last, on success | elapsed, byte counts, and the cookie jar as the send left it |
|
||||
| `error` | last, on failure | the reason, and any cookies collected before the failure |
|
||||
|
||||
Refusals that happen before anything is sent (a blocked destination, a bad body,
|
||||
rate limit, capacity) are plain HTTP errors (`403`, `400`, `429`, `503`) with
|
||||
`{"error": "…"}`, not streams.
|
||||
|
||||
Why a streamed HTTP response and not a WebSocket: one `POST` is stateless by
|
||||
construction, cancellable by closing the connection, readable with `curl`, and
|
||||
needs no upgrade handling on either side. A WebSocket only earns its keep when
|
||||
traffic is bidirectional, which a single send is not.
|
||||
|
||||
The TypeScript side of this contract is generated from `src/wire.rs` by ts-rs
|
||||
into `bindings/` (run `cargo test -p yaak-send-proxy` after changing a frame)
|
||||
and published to the tab as `@yaakapp-internal/send-proxy`, so a change to the
|
||||
wire on one side is a type error on the other.
|
||||
|
||||
`GET /v1/health` reports the version and the effective limits.
|
||||
|
||||
## What comes later
|
||||
|
||||
Not built, by design, but the router is shaped for it: a WebSocket relay
|
||||
(`/v1/ws/relay`) and a gRPC relay (`/v1/grpc/relay`) would be long-lived,
|
||||
bidirectional endpoints on the same binary, behind the same destination policy
|
||||
and limits. They differ from this endpoint in holding per-connection
|
||||
in-memory state while a connection is open (never persisted), which brings
|
||||
connection limits and a larger abuse surface — the reason they are separate
|
||||
work.
|
||||
@@ -1,4 +0,0 @@
|
||||
// The send proxy's wire contract, generated by ts-rs from src/wire.rs
|
||||
// (`cargo test -p yaak-send-proxy`). The tab imports these so a change to a
|
||||
// frame on the Rust side is a type error in packages/platform/src/web.
|
||||
export type { Frame, SendRequest } from "./bindings/gen_send_proxy";
|
||||
@@ -1,52 +0,0 @@
|
||||
use clap::Parser;
|
||||
use std::net::SocketAddr;
|
||||
|
||||
/// A stateless HTTP send executor for Yaak running in a browser.
|
||||
///
|
||||
/// The tab renders the request and owns the data; this binary only puts bytes on the
|
||||
/// network and streams back what came back. Nothing is written to disk or a database.
|
||||
#[derive(Parser, Debug, Clone)]
|
||||
#[command(name = "yaak-send-proxy", version, about, long_about = None)]
|
||||
pub struct Config {
|
||||
/// Address to listen on. 127.0.0.1 for a local instance; 0.0.0.0 inside a container.
|
||||
#[arg(long, env = "YAAK_PROXY_BIND", default_value = "127.0.0.1:9227")]
|
||||
pub bind: SocketAddr,
|
||||
|
||||
/// Browser origins allowed to call this proxy (CORS), comma-separated. `*` allows any.
|
||||
/// A local dev instance wants the Vite origin; a hosted instance wants its own web origin.
|
||||
#[arg(
|
||||
long,
|
||||
env = "YAAK_PROXY_ALLOWED_ORIGINS",
|
||||
default_value = "*",
|
||||
value_delimiter = ','
|
||||
)]
|
||||
pub allowed_origins: Vec<String>,
|
||||
|
||||
/// Largest request the proxy accepts from the tab (the rendered request JSON, body included).
|
||||
#[arg(long, env = "YAAK_PROXY_MAX_REQUEST_BYTES", default_value_t = 16 * 1024 * 1024)]
|
||||
pub max_request_bytes: usize,
|
||||
|
||||
/// Largest upstream response body the proxy will relay before cutting the send off.
|
||||
#[arg(long, env = "YAAK_PROXY_MAX_RESPONSE_BYTES", default_value_t = 64 * 1024 * 1024)]
|
||||
pub max_response_bytes: usize,
|
||||
|
||||
/// Ceiling on a send's timeout, in seconds. A request asking for longer (or for no timeout)
|
||||
/// gets this instead.
|
||||
#[arg(long, env = "YAAK_PROXY_MAX_TIMEOUT_SECS", default_value_t = 60)]
|
||||
pub max_timeout_secs: u64,
|
||||
|
||||
/// Sends allowed per client IP per minute. 0 disables the limit. This and the concurrency
|
||||
/// cap are the whole of what protects an instance: there is no authentication.
|
||||
#[arg(long, env = "YAAK_PROXY_RATE_LIMIT_PER_MINUTE", default_value_t = 120)]
|
||||
pub rate_limit_per_minute: u32,
|
||||
|
||||
/// Sends in flight at once across all clients.
|
||||
#[arg(long, env = "YAAK_PROXY_MAX_CONCURRENT", default_value_t = 256)]
|
||||
pub max_concurrent: usize,
|
||||
|
||||
/// Take the client IP from `X-Forwarded-For` (first hop) instead of the socket. Only turn
|
||||
/// this on behind a load balancer that sets the header; otherwise anyone can spoof their way
|
||||
/// past the rate limit.
|
||||
#[arg(long, env = "YAAK_PROXY_TRUST_FORWARDED_FOR", default_value_t = false)]
|
||||
pub trust_forwarded_for: bool,
|
||||
}
|
||||
@@ -1,9 +1,9 @@
|
||||
[package]
|
||||
name = "yaak-send-proxy"
|
||||
name = "yaak-web"
|
||||
version = "0.1.0"
|
||||
edition = "2024"
|
||||
publish = false
|
||||
description = "Stateless HTTP send executor for Yaak in the browser"
|
||||
description = "The server behind Yaak in the browser: executes sends, and can serve the app"
|
||||
|
||||
# The send engine (yaak-http) and the model types it speaks (yaak-models, for
|
||||
# HttpRequest / Cookie / HttpResponseEventData). Deliberately NOT yaak (the
|
||||
@@ -13,7 +13,7 @@ description = "Stateless HTTP send executor for Yaak in the browser"
|
||||
# its query layer.
|
||||
|
||||
[[bin]]
|
||||
name = "yaak-send-proxy"
|
||||
name = "yaak-web"
|
||||
path = "src/main.rs"
|
||||
|
||||
[dependencies]
|
||||
@@ -28,7 +28,7 @@ log = { workspace = true }
|
||||
serde = { workspace = true, features = ["derive"] }
|
||||
serde_json = { workspace = true }
|
||||
tokio = { workspace = true, features = ["rt-multi-thread", "macros", "signal", "sync", "io-util", "time", "net"] }
|
||||
tower-http = { version = "0.6", features = ["cors"] }
|
||||
tower-http = { version = "0.6", features = ["compression-gzip", "compression-zstd", "cors", "fs"] }
|
||||
ts-rs = { workspace = true }
|
||||
url = "2"
|
||||
uuid = { version = "1", features = ["v4"] }
|
||||
@@ -0,0 +1,219 @@
|
||||
# yaak-web
|
||||
|
||||
The network half of Yaak in a browser — and, with `--serve`, the half that
|
||||
hands the browser the app in the first place.
|
||||
|
||||
A tab can't see an HTTP response the way a desktop app can: CORS hides most
|
||||
headers (2 of 8 in a typical response), redirects are followed silently, and
|
||||
there is no timeline. So the tab renders the request and posts it here, and this
|
||||
process puts it on the network with the desktop's own engine (`yaak-http`) and
|
||||
streams back everything that happened — every header, every redirect hop, DNS
|
||||
timing, the body — for the tab to store.
|
||||
|
||||
It is a **stateless executor**. It keeps nothing: no database, no files, no
|
||||
sessions, no cookies between calls. Every byte it sees comes from the tab in the
|
||||
request, and every byte it returns is stored by the tab. Restart it any time.
|
||||
|
||||
## Self-hosting it
|
||||
|
||||
One container, no configuration, nothing behind it:
|
||||
|
||||
```shell
|
||||
docker run -p 8080:8080 ghcr.io/mountain-loop/yaak-web
|
||||
```
|
||||
|
||||
Open <http://localhost:8080>. The image carries the built web client and this
|
||||
binary, which serves it — so the app and its sends are on one origin, and the
|
||||
tab's send URL is a path (`/v1/http/send`) rather than an address anyone has to
|
||||
configure. The image is `linux/amd64` and `linux/arm64`, built from
|
||||
`Dockerfile.web` at the repo root.
|
||||
|
||||
Your data lives in your browser (SQLite compiled to wasm, in IndexedDB), not in
|
||||
the container. The container is stateless: nothing is written to disk, so
|
||||
upgrading is `docker pull` and nothing else.
|
||||
|
||||
Two settings are worth knowing about:
|
||||
|
||||
```shell
|
||||
docker run -p 8080:8080 \
|
||||
-e YAAK_WEB_ALLOW_PRIVATE_NETWORKS=true \
|
||||
-e YAAK_WEB_RATE_LIMIT_PER_MINUTE=0 \
|
||||
ghcr.io/mountain-loop/yaak-web
|
||||
```
|
||||
|
||||
- **`YAAK_WEB_ALLOW_PRIVATE_NETWORKS=true`** lets sends reach loopback,
|
||||
private and link-local addresses. Off by default, and it should stay off on
|
||||
anything strangers can reach — see [What it refuses](#what-it-refuses-and-why).
|
||||
Turn it on for an instance on your own network, where calling the API on the
|
||||
next machine is the whole point. Note that "private" is relative to the
|
||||
*container*: `127.0.0.1` is the container itself, and reaching the Docker
|
||||
host means `host.docker.internal` (or `--network host`).
|
||||
- **`YAAK_WEB_RATE_LIMIT_PER_MINUTE`** defaults to 120 sends per client IP,
|
||||
which suits a public instance and not a team of your own; `0` disables it.
|
||||
|
||||
Behind a reverse proxy, add `YAAK_WEB_TRUST_FORWARDED_FOR=true` so the rate
|
||||
limit sees real client addresses instead of its own — and only then, since
|
||||
otherwise anyone can spoof the header. If the reverse proxy buffers responses,
|
||||
tell it not to: sends are streamed, and the `X-Accel-Buffering: no` header this
|
||||
binary sets is honoured by nginx-shaped ones.
|
||||
|
||||
## Running it from source
|
||||
|
||||
```shell
|
||||
cargo run -p yaak-web -- --serve dist/apps/yaak-client
|
||||
```
|
||||
|
||||
after a `YAAK_TARGET=web SKIP_WASM_BUILD=1 npx vp -C apps/yaak-client build`.
|
||||
Without `--serve` it is the send executor alone, which is what the frontend
|
||||
dev server wants:
|
||||
|
||||
```shell
|
||||
cargo run -p yaak-web
|
||||
YAAK_TARGET=web npm run dev --workspace @yaakapp/yaak-client
|
||||
```
|
||||
|
||||
A dev build looks for the server at `http://127.0.0.1:9227` (the Vite server is a
|
||||
different origin and serves no `/v1`); a production build sends to its own
|
||||
origin unless `VITE_YAAK_WEB_URL` was set when it was built.
|
||||
|
||||
## Configuration
|
||||
|
||||
Every flag has a `YAAK_WEB_*` environment variable, so a container needs no
|
||||
arguments; `--help` lists them all.
|
||||
|
||||
| Flag | Default | What |
|
||||
| --- | --- | --- |
|
||||
| `--serve` | off | Also serve a built web client from this directory, on the same origin. |
|
||||
| `--bind` | `127.0.0.1:9227` | Listen address. The image sets `0.0.0.0:8080`. |
|
||||
| `--allow-private-networks` | off | Allow sends to loopback, private and link-local addresses. |
|
||||
| `--allowed-origins` | `*` | CORS origins, comma-separated. Unused when the app is served from here: same origin, no CORS. |
|
||||
| `--max-request-bytes` | 16 MiB | Largest rendered request accepted from the tab. |
|
||||
| `--max-response-bytes` | 64 MiB | Largest upstream body relayed before the send is cut off. |
|
||||
| `--max-timeout-secs` | 60 | Ceiling on a send's timeout; a request asking for more (or none) gets this. |
|
||||
| `--rate-limit-per-minute` | 120 | Sends per client IP per minute; 0 disables. |
|
||||
| `--max-concurrent` | 256 | Sends in flight at once. |
|
||||
| `--trust-forwarded-for` | off | Take the client IP from `X-Forwarded-For`. Only behind a load balancer that sets it. |
|
||||
|
||||
## Serving the app
|
||||
|
||||
`--serve DIR` puts a file server behind the API routes: `/v1/*` is matched
|
||||
first, everything else comes from `DIR`, and a path with no file behind it gets
|
||||
`index.html` so the app's own routes survive a refresh. Responses are compressed
|
||||
(gzip or zstd) on the fly. `/assets/*` is cached forever — Vite content-hashes
|
||||
those names — and everything else is `no-cache`, so a new deploy arrives on the
|
||||
next reload.
|
||||
|
||||
Serving files changes nothing about sending: the same rendered request, the same
|
||||
destination policy, the same stateless executor. It exists so that a
|
||||
self-hosted Yaak is one thing to run rather than two.
|
||||
|
||||
## Split deployments
|
||||
|
||||
The app and the sender can still be separate services — one CDN-hosted bundle and
|
||||
one server elsewhere, or one server shared by several fronts. Then the bundle has
|
||||
to be told where to send, at build time:
|
||||
|
||||
```shell
|
||||
docker build -f Dockerfile.web \
|
||||
--build-arg VITE_YAAK_WEB_URL=https://send.example.com .
|
||||
```
|
||||
|
||||
and the server needs the CORS origins its callers use, since the requests are no
|
||||
longer same-origin:
|
||||
|
||||
```shell
|
||||
docker run -p 8080:8080 \
|
||||
-e YAAK_WEB_ALLOWED_ORIGINS=https://yaak.example.com \
|
||||
ghcr.io/mountain-loop/yaak-web \
|
||||
yaak-web
|
||||
```
|
||||
|
||||
The trailing `yaak-web` is a command override: the same image run without
|
||||
`--serve`, so it executes sends and serves no app.
|
||||
|
||||
## What it refuses, and why
|
||||
|
||||
A hosted sender is, by construction, a machine that makes HTTP requests on
|
||||
behalf of strangers. Left alone that is an open relay into whatever network it
|
||||
sits on. So by default it refuses to connect to:
|
||||
|
||||
- loopback (`127/8`, `::1`), private (`10/8`, `172.16/12`, `192.168/16`,
|
||||
`fc00::/7`), link-local (`169.254/16` — where cloud metadata lives — and
|
||||
`fe80::/10`), carrier-grade NAT, multicast, reserved and unspecified ranges,
|
||||
IPv4 addresses carried inside IPv6 forms (`::ffff:a.b.c.d`, the well-known
|
||||
NAT64 prefix, 6to4), and the whole NAT64 local-use range;
|
||||
- anything not `http://` or `https://`.
|
||||
|
||||
The check runs **on the resolved addresses, after DNS**, for every hop of a
|
||||
redirect chain, so a public hostname that points at an internal address is
|
||||
caught, and so is a `Location:` header that points at one. It also refuses body
|
||||
types that would read files on its own disk (`binary`, multipart file
|
||||
fields), since no browser tab could legitimately mean those.
|
||||
|
||||
Refusals are logged with the reason. On a public instance (`web.yaak.app`, or
|
||||
anything else strangers can reach) this must stay on: the machine's private
|
||||
network is the host's, not the user's, so a `localhost` or LAN API is not the
|
||||
user's to reach through it — the desktop app is what reaches those. On an
|
||||
instance you run for yourself, that reasoning is inverted, and
|
||||
`--allow-private-networks` inverts the policy with it. It allows every range
|
||||
above, including `169.254.169.254`, so use it only where the network on the
|
||||
other side is one the users are entitled to.
|
||||
|
||||
There is no authentication either way: an instance is anonymous, protected by
|
||||
the per-client rate limit and the destination policy. Anything more (a shared
|
||||
token, per-user quotas) is a later slice and would sit in front of `send_http`
|
||||
in `main.rs`. Put TLS in front of a public instance.
|
||||
|
||||
## The wire
|
||||
|
||||
`POST /v1/http/send` with a JSON body:
|
||||
|
||||
```json
|
||||
{
|
||||
"request": { "url": "https://…", "method": "GET", "headers": […], "body": {…}, "bodyType": null, "urlParameters": […] },
|
||||
"settings": { "validateCertificates": true, "followRedirects": true, "timeoutMs": 0, "sendCookies": true, "storeCookies": true },
|
||||
"cookies": [ … ]
|
||||
}
|
||||
```
|
||||
|
||||
`request` is a Yaak `HttpRequest` in the desktop's own model shape with every
|
||||
template already rendered by the tab; the server builds the URL, headers and
|
||||
body from it exactly the way the desktop does after rendering. `cookies` is the
|
||||
jar's contents (or `null` for no jar).
|
||||
|
||||
The reply is `application/x-ndjson`, one JSON frame per line, in the order things
|
||||
happened:
|
||||
|
||||
| `type` | When | Carries |
|
||||
| --- | --- | --- |
|
||||
| `event` | as the engine produces them | one timeline event, in the desktop's `http_response_event.event` shape |
|
||||
| `response` | once, when the final hop's headers arrive | status, all headers, request headers as sent, remote address, HTTP version, timing |
|
||||
| `body` | as the body is read | a decompressed chunk, base64 |
|
||||
| `done` | last, on success | elapsed, byte counts, and the cookie jar as the send left it |
|
||||
| `error` | last, on failure | the reason, and any cookies collected before the failure |
|
||||
|
||||
Refusals that happen before anything is sent (a blocked destination, a bad body,
|
||||
rate limit, capacity) are plain HTTP errors (`403`, `400`, `429`, `503`) with
|
||||
`{"error": "…"}`, not streams.
|
||||
|
||||
Why a streamed HTTP response and not a WebSocket: one `POST` is stateless by
|
||||
construction, cancellable by closing the connection, readable with `curl`, and
|
||||
needs no upgrade handling on either side. A WebSocket only earns its keep when
|
||||
traffic is bidirectional, which a single send is not.
|
||||
|
||||
The TypeScript side of this contract is generated from `src/wire.rs` by ts-rs
|
||||
into `bindings/` (run `cargo test -p yaak-web` after changing a frame)
|
||||
and published to the tab as `@yaakapp-internal/web`, so a change to the
|
||||
wire on one side is a type error on the other.
|
||||
|
||||
`GET /v1/health` reports the version and the effective limits.
|
||||
|
||||
## What comes later
|
||||
|
||||
Not built, by design, but the router is shaped for it: a WebSocket relay
|
||||
(`/v1/ws/relay`) and a gRPC relay (`/v1/grpc/relay`) would be long-lived,
|
||||
bidirectional endpoints on the same binary, behind the same destination policy
|
||||
and limits. They differ from this endpoint in holding per-connection
|
||||
in-memory state while a connection is open (never persisted), which brings
|
||||
connection limits and a larger abuse surface — the reason they are separate
|
||||
work.
|
||||
crates-server/yaak-send-proxy/bindings/gen_models.ts → crates-server/yaak-web/bindings/gen_models.ts
Generated
+1
-1
@@ -28,7 +28,7 @@ export type HttpResponseHeader = { name: string, value: string, };
|
||||
/**
|
||||
* The resolved send settings, values only: what an executor has to obey, with the sources
|
||||
* (which model each came from) left behind in [`ResolvedHttpRequestSettings`]. This is what
|
||||
* crosses from a tab to the send proxy, and what the proxy reads.
|
||||
* crosses from a tab to the Yaak server, and what the server reads.
|
||||
*/
|
||||
export type HttpSendSettings = { validateCertificates: boolean, followRedirects: boolean,
|
||||
/**
|
||||
+2
-2
@@ -49,13 +49,13 @@ cookies: Array<Cookie> | null, } | { "type": "error", message: string, cookies:
|
||||
export type SendRequest = {
|
||||
/**
|
||||
* The request to send, in the desktop's own model shape but with every template already
|
||||
* rendered by the tab. The proxy builds the URL, headers and body from it exactly the way
|
||||
* rendered by the tab. The server builds the URL, headers and body from it exactly the way
|
||||
* the desktop does after rendering.
|
||||
*/
|
||||
request: HttpRequest,
|
||||
/**
|
||||
* The resolved settings, values only. Where they came from is the tab's to record in
|
||||
* its timeline; the proxy only needs to obey them.
|
||||
* its timeline; the server only needs to obey them.
|
||||
*/
|
||||
settings: HttpSendSettings,
|
||||
/**
|
||||
@@ -0,0 +1,4 @@
|
||||
// The server's wire contract, generated by ts-rs from src/wire.rs
|
||||
// (`cargo test -p yaak-web`). The tab imports these so a change to a
|
||||
// frame on the Rust side is a type error in packages/platform/src/web.
|
||||
export type { Frame, SendRequest } from "./bindings/gen_web";
|
||||
@@ -1,5 +1,5 @@
|
||||
{
|
||||
"name": "@yaakapp-internal/send-proxy",
|
||||
"name": "@yaakapp-internal/web",
|
||||
"version": "1.0.0",
|
||||
"private": true,
|
||||
"main": "index.ts"
|
||||
@@ -0,0 +1,67 @@
|
||||
use clap::Parser;
|
||||
use std::net::SocketAddr;
|
||||
use std::path::PathBuf;
|
||||
|
||||
/// The server behind Yaak running in a browser.
|
||||
///
|
||||
/// The tab renders the request and owns the data; this binary puts the bytes on the network
|
||||
/// and streams back what came back, and with `--serve` hands the browser the app as well.
|
||||
/// Nothing is written to disk or a database.
|
||||
#[derive(Parser, Debug, Clone)]
|
||||
#[command(name = "yaak-web", version, about, long_about = None)]
|
||||
pub struct Config {
|
||||
/// Address to listen on. 127.0.0.1 for a local instance; 0.0.0.0 inside a container.
|
||||
#[arg(long, env = "YAAK_WEB_BIND", default_value = "127.0.0.1:9227")]
|
||||
pub bind: SocketAddr,
|
||||
|
||||
/// Also serve a built web client from this directory, on the same origin as the API.
|
||||
/// Unknown paths fall back to `index.html` so the app's own routes work on a refresh.
|
||||
/// Without this the binary is only the send executor.
|
||||
#[arg(long, env = "YAAK_WEB_SERVE", value_name = "DIR")]
|
||||
pub serve: Option<PathBuf>,
|
||||
|
||||
/// Allow sends to loopback, private and link-local addresses. Off by default, because a
|
||||
/// server reachable by strangers is an open relay into the network it sits on. Turn it on
|
||||
/// only for an instance whose users are meant to reach that network — a self-hosted one
|
||||
/// on a LAN, where the point is to call the API on the next machine.
|
||||
#[arg(long, env = "YAAK_WEB_ALLOW_PRIVATE_NETWORKS", default_value_t = false)]
|
||||
pub allow_private_networks: bool,
|
||||
|
||||
/// Browser origins allowed to call this server (CORS), comma-separated. `*` allows any.
|
||||
/// A local dev instance wants the Vite origin; a hosted instance wants its own web origin.
|
||||
#[arg(
|
||||
long,
|
||||
env = "YAAK_WEB_ALLOWED_ORIGINS",
|
||||
default_value = "*",
|
||||
value_delimiter = ','
|
||||
)]
|
||||
pub allowed_origins: Vec<String>,
|
||||
|
||||
/// Largest request the server accepts from the tab (the rendered request JSON, body included).
|
||||
#[arg(long, env = "YAAK_WEB_MAX_REQUEST_BYTES", default_value_t = 16 * 1024 * 1024)]
|
||||
pub max_request_bytes: usize,
|
||||
|
||||
/// Largest upstream response body the server will relay before cutting the send off.
|
||||
#[arg(long, env = "YAAK_WEB_MAX_RESPONSE_BYTES", default_value_t = 64 * 1024 * 1024)]
|
||||
pub max_response_bytes: usize,
|
||||
|
||||
/// Ceiling on a send's timeout, in seconds. A request asking for longer (or for no timeout)
|
||||
/// gets this instead.
|
||||
#[arg(long, env = "YAAK_WEB_MAX_TIMEOUT_SECS", default_value_t = 60)]
|
||||
pub max_timeout_secs: u64,
|
||||
|
||||
/// Sends allowed per client IP per minute. 0 disables the limit. This and the concurrency
|
||||
/// cap are the whole of what protects an instance: there is no authentication.
|
||||
#[arg(long, env = "YAAK_WEB_RATE_LIMIT_PER_MINUTE", default_value_t = 120)]
|
||||
pub rate_limit_per_minute: u32,
|
||||
|
||||
/// Sends in flight at once across all clients.
|
||||
#[arg(long, env = "YAAK_WEB_MAX_CONCURRENT", default_value_t = 256)]
|
||||
pub max_concurrent: usize,
|
||||
|
||||
/// Take the client IP from `X-Forwarded-For` (first hop) instead of the socket. Only turn
|
||||
/// this on behind a load balancer that sets the header; otherwise anyone can spoof their way
|
||||
/// past the rate limit.
|
||||
#[arg(long, env = "YAAK_WEB_TRUST_FORWARDED_FOR", default_value_t = false)]
|
||||
pub trust_forwarded_for: bool,
|
||||
}
|
||||
@@ -20,13 +20,20 @@ use yaak_http::dns::AddressFilter;
|
||||
use yaak_http::sender::{HttpResponse, HttpResponseEvent, HttpSender};
|
||||
use yaak_http::types::SendableHttpRequest;
|
||||
|
||||
/// The destination policy, shared by every send: public addresses only, always. A hosted
|
||||
/// proxy's "private network" is the cloud's, not the user's, so there is no configuration
|
||||
/// that makes reaching it right.
|
||||
/// The destination policy, shared by every send: public addresses only, unless the operator
|
||||
/// has said otherwise. A hosted server's "private network" is the cloud's, not the user's, so
|
||||
/// the default is public-only; a self-hosted instance on a LAN can be told that its private
|
||||
/// network *is* the user's, which is what `--allow-private-networks` means.
|
||||
#[derive(Clone, Default)]
|
||||
pub struct DestinationPolicy;
|
||||
pub struct DestinationPolicy {
|
||||
allow_private: bool,
|
||||
}
|
||||
|
||||
impl DestinationPolicy {
|
||||
pub fn new(allow_private: bool) -> Self {
|
||||
Self { allow_private }
|
||||
}
|
||||
|
||||
/// Check a URL before a hop is attempted: scheme and literal IPs. A hostname that passes
|
||||
/// here still has its resolved addresses checked by [`Self::address_filter`].
|
||||
pub fn check_url(&self, raw: &str) -> Result<(), String> {
|
||||
@@ -53,9 +60,12 @@ impl DestinationPolicy {
|
||||
}
|
||||
|
||||
pub fn check_ip(&self, ip: IpAddr) -> Result<(), String> {
|
||||
if self.allow_private {
|
||||
return Ok(());
|
||||
}
|
||||
match non_public_reason(ip) {
|
||||
Some(reason) => Err(format!(
|
||||
"Refusing to connect to {ip}: {reason}. This proxy only sends to public addresses"
|
||||
"Refusing to connect to {ip}: {reason}. This server only sends to public addresses"
|
||||
)),
|
||||
None => Ok(()),
|
||||
}
|
||||
@@ -236,15 +246,24 @@ mod tests {
|
||||
|
||||
#[test]
|
||||
fn literal_private_addresses_in_urls_are_refused() {
|
||||
let policy = DestinationPolicy;
|
||||
let policy = DestinationPolicy::new(false);
|
||||
assert!(policy.check_url("http://127.0.0.1/").is_err());
|
||||
assert!(policy.check_url("http://[::1]/").is_err());
|
||||
assert!(policy.check_url("http://169.254.169.254/latest/meta-data").is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn allow_private_networks_opens_the_local_ranges_but_not_other_schemes() {
|
||||
let policy = DestinationPolicy::new(true);
|
||||
assert!(policy.check_url("http://127.0.0.1/").is_ok());
|
||||
assert!(policy.check_ip(ip("10.0.0.1")).is_ok());
|
||||
assert!(policy.check_ip(ip("169.254.169.254")).is_ok());
|
||||
assert!(policy.check_url("file:///etc/passwd").is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn only_http_schemes() {
|
||||
let policy = DestinationPolicy;
|
||||
let policy = DestinationPolicy::new(false);
|
||||
assert!(policy.check_url("ftp://example.com/").is_err());
|
||||
assert!(policy.check_url("file:///etc/passwd").is_err());
|
||||
assert!(policy.check_url("https://example.com/").is_ok());
|
||||
@@ -1,4 +1,4 @@
|
||||
//! yaak-send-proxy: the network half of Yaak in a browser.
|
||||
//! yaak-web: the network half of Yaak in a browser.
|
||||
//!
|
||||
//! A tab can't see a response the way a desktop app can — CORS hides most
|
||||
//! headers, redirects are followed silently, there is no timeline. So the tab
|
||||
@@ -6,7 +6,7 @@
|
||||
//! with the desktop's own engine and streams back everything that happened,
|
||||
//! for the tab to store. It keeps nothing: no database, no files, no session.
|
||||
//!
|
||||
//! One binary, configured by flags or `YAAK_PROXY_*` environment variables.
|
||||
//! One binary, configured by flags or `YAAK_WEB_*` environment variables.
|
||||
//! See README.md for running and deploying it, and `guard.rs` for what it
|
||||
//! refuses to talk to.
|
||||
|
||||
@@ -18,8 +18,9 @@ mod wire;
|
||||
|
||||
use axum::Router;
|
||||
use axum::body::Body;
|
||||
use axum::extract::{ConnectInfo, DefaultBodyLimit, State};
|
||||
use axum::extract::{ConnectInfo, DefaultBodyLimit, Request, State};
|
||||
use axum::http::{HeaderMap, HeaderValue, Method, StatusCode, header};
|
||||
use axum::middleware::{self, Next};
|
||||
use axum::response::{IntoResponse, Json, Response};
|
||||
use axum::routing::{get, post};
|
||||
use clap::Parser;
|
||||
@@ -30,10 +31,13 @@ use log::{info, warn};
|
||||
use send::{Refusal, SendLimits};
|
||||
use serde_json::json;
|
||||
use std::net::{IpAddr, SocketAddr};
|
||||
use std::path::Path;
|
||||
use std::sync::Arc;
|
||||
use std::time::{Duration, Instant};
|
||||
use tokio::sync::Semaphore;
|
||||
use tower_http::compression::CompressionLayer;
|
||||
use tower_http::cors::{AllowOrigin, CorsLayer};
|
||||
use tower_http::services::{ServeDir, ServeFile};
|
||||
use wire::SendRequest;
|
||||
|
||||
#[derive(Clone)]
|
||||
@@ -49,7 +53,13 @@ async fn main() {
|
||||
env_logger::Builder::from_env(env_logger::Env::default().default_filter_or("info")).init();
|
||||
let config = Config::parse();
|
||||
|
||||
let policy = DestinationPolicy;
|
||||
let policy = DestinationPolicy::new(config.allow_private_networks);
|
||||
if config.allow_private_networks {
|
||||
warn!(
|
||||
"Sends to loopback, private and link-local addresses are ALLOWED. Only run this way \
|
||||
on an instance strangers cannot reach"
|
||||
);
|
||||
}
|
||||
let state = AppState {
|
||||
limits: Arc::new(SendLimits {
|
||||
policy,
|
||||
@@ -66,7 +76,7 @@ async fn main() {
|
||||
.allow_headers([header::CONTENT_TYPE])
|
||||
.allow_origin(allowed_origins(&state.config.allowed_origins));
|
||||
|
||||
let app = Router::new()
|
||||
let api = Router::new()
|
||||
.route("/v1/health", get(health))
|
||||
// A WebSocket or gRPC relay would sit beside this as `/v1/ws/relay` and `/v1/grpc/relay`
|
||||
// on the same router, behind the same policy, limits and auth. Not built; see README.
|
||||
@@ -75,13 +85,21 @@ async fn main() {
|
||||
.layer(cors)
|
||||
.with_state(state.clone());
|
||||
|
||||
let app = match &state.config.serve {
|
||||
Some(dir) => {
|
||||
info!("Serving the web client from {}", dir.display());
|
||||
api.merge(web_router(dir))
|
||||
}
|
||||
None => api,
|
||||
};
|
||||
|
||||
let bind = state.config.bind;
|
||||
let listener = tokio::net::TcpListener::bind(bind).await.unwrap_or_else(|e| {
|
||||
eprintln!("Failed to bind {bind}: {e}");
|
||||
std::process::exit(1);
|
||||
});
|
||||
info!(
|
||||
"yaak-send-proxy listening on http://{bind} (rate limit: {}/min)",
|
||||
"yaak-web listening on http://{bind} (rate limit: {}/min)",
|
||||
state.config.rate_limit_per_minute,
|
||||
);
|
||||
|
||||
@@ -94,6 +112,48 @@ async fn main() {
|
||||
.expect("server error");
|
||||
}
|
||||
|
||||
/// The built web client, served on the same origin as the API.
|
||||
///
|
||||
/// This is what makes a single container zero-configuration: the tab's send URL is a path on
|
||||
/// the page's own origin, so there is no CORS, no second service and no URL to bake in. It is
|
||||
/// only a file server — a send behaves exactly as it does without this flag.
|
||||
///
|
||||
/// Merged as a fallback, so the `/v1` routes are matched first and a request that matches no
|
||||
/// file at all gets `index.html` (the app routes client-side; a deep link must survive a
|
||||
/// refresh).
|
||||
fn web_router(dir: &Path) -> Router {
|
||||
let index = ServeFile::new(dir.join("index.html"));
|
||||
Router::new()
|
||||
// `fallback`, not `not_found_service`: the app's own routes are real pages, so
|
||||
// index.html is served with the 200 the browser expects, not a 404 carrying HTML.
|
||||
.fallback_service(ServeDir::new(dir).fallback(index))
|
||||
.layer(middleware::from_fn(cache_control))
|
||||
.layer(CompressionLayer::new())
|
||||
}
|
||||
|
||||
/// Vite gives everything in `/assets` a content-hashed name, so those can be cached forever.
|
||||
/// Everything else — `index.html` above all, including the copy served for an unknown path —
|
||||
/// must be revalidated, or a browser keeps serving the deploy before last.
|
||||
async fn cache_control(req: Request, next: Next) -> Response {
|
||||
let hashed_name = req.uri().path().starts_with("/assets/");
|
||||
let mut res = next.run(req).await;
|
||||
if !res.status().is_success() {
|
||||
return res;
|
||||
}
|
||||
let is_html = res
|
||||
.headers()
|
||||
.get(header::CONTENT_TYPE)
|
||||
.and_then(|v| v.to_str().ok())
|
||||
.is_some_and(|v| v.starts_with("text/html"));
|
||||
let value = if hashed_name && !is_html {
|
||||
"public, max-age=31536000, immutable"
|
||||
} else {
|
||||
"no-cache"
|
||||
};
|
||||
res.headers_mut().insert(header::CACHE_CONTROL, HeaderValue::from_static(value));
|
||||
res
|
||||
}
|
||||
|
||||
fn allowed_origins(origins: &[String]) -> AllowOrigin {
|
||||
if origins.iter().any(|o| o.trim() == "*") {
|
||||
return AllowOrigin::any();
|
||||
@@ -149,7 +209,7 @@ async fn send_http(
|
||||
|
||||
let Ok(permit) = state.in_flight.clone().try_acquire_owned() else {
|
||||
warn!("At capacity; refusing {ip}");
|
||||
return error_response(StatusCode::SERVICE_UNAVAILABLE, "This proxy is at capacity");
|
||||
return error_response(StatusCode::SERVICE_UNAVAILABLE, "This server is at capacity");
|
||||
};
|
||||
|
||||
let prepared = match send::prepare(state.limits.clone(), body).await {
|
||||
@@ -42,7 +42,7 @@ pub struct SendLimits {
|
||||
pub enum Refusal {
|
||||
/// The request asks for something a browser-originated send cannot mean.
|
||||
Unsupported(String),
|
||||
/// The destination is not one this proxy will talk to.
|
||||
/// The destination is not one this server will talk to.
|
||||
Destination(String),
|
||||
/// The request could not be turned into something sendable.
|
||||
Invalid(String),
|
||||
@@ -57,10 +57,10 @@ pub async fn prepare(limits: Arc<SendLimits>, send: SendRequest) -> Result<Prepa
|
||||
|
||||
// The engine reads files for these body types. There are no files here that a browser tab
|
||||
// could legitimately mean, and letting a request name a path on this machine would be a
|
||||
// local file read for anyone who can reach the proxy.
|
||||
// local file read for anyone who can reach the server.
|
||||
if request.body_type.as_deref() == Some("binary") {
|
||||
return Err(Refusal::Unsupported(
|
||||
"Binary file bodies can't be sent from the browser: the proxy has no access to your files"
|
||||
"Binary file bodies can't be sent from the browser: the server has no access to your files"
|
||||
.to_string(),
|
||||
));
|
||||
}
|
||||
@@ -74,7 +74,7 @@ pub async fn prepare(limits: Arc<SendLimits>, send: SendRequest) -> Result<Prepa
|
||||
});
|
||||
if names_a_file {
|
||||
return Err(Refusal::Unsupported(
|
||||
"Multipart file fields can't be sent from the browser: the proxy has no access to your files"
|
||||
"Multipart file fields can't be sent from the browser: the server has no access to your files"
|
||||
.to_string(),
|
||||
));
|
||||
}
|
||||
@@ -204,7 +204,7 @@ impl PreparedSend {
|
||||
|
||||
if self.timeout_capped {
|
||||
let _ = event_tx.try_send(HttpResponseEvent::Info(format!(
|
||||
"Timeout set to {:?} (this proxy's ceiling)",
|
||||
"Timeout set to {:?} (this server's ceiling)",
|
||||
self.timeout
|
||||
)));
|
||||
}
|
||||
@@ -276,7 +276,7 @@ impl PreparedSend {
|
||||
total += n;
|
||||
if total > limits.max_response_bytes {
|
||||
break Err(format!(
|
||||
"Response body exceeds this proxy's limit of {} bytes",
|
||||
"Response body exceeds this server's limit of {} bytes",
|
||||
limits.max_response_bytes
|
||||
));
|
||||
}
|
||||
@@ -309,7 +309,7 @@ impl PreparedSend {
|
||||
///
|
||||
/// A connection error from reqwest arrives wrapped several layers deep, and the layer that
|
||||
/// says something useful — "Refusing to connect to ::1: loopback" — is the innermost. The
|
||||
/// desktop shows the outer `Debug`; a stranger reading a proxy's reply deserves the reason.
|
||||
/// desktop shows the outer `Debug`; a stranger reading its reply deserves the reason.
|
||||
fn describe_error(err: &yaak_http::error::Error) -> String {
|
||||
match err {
|
||||
yaak_http::error::Error::Client(e) => {
|
||||
@@ -1,4 +1,4 @@
|
||||
//! What crosses the wire between a tab and this proxy.
|
||||
//! What crosses the wire between a tab and this server.
|
||||
//!
|
||||
//! One `POST /v1/http/send` carries a request the tab has already rendered —
|
||||
//! templates resolved, inheritance applied — plus the send settings and the
|
||||
@@ -6,13 +6,13 @@
|
||||
//! JSON frames: timeline events as they happen, the response head as soon as
|
||||
//! headers arrive, body chunks as they are read, and one terminal frame.
|
||||
//!
|
||||
//! Nothing here names a workspace, a request id, or a response id. The proxy
|
||||
//! Nothing here names a workspace, a request id, or a response id. The server
|
||||
//! does not know what the tab will call this response; it only knows what came
|
||||
//! back.
|
||||
//!
|
||||
//! The TypeScript side of this contract is generated from these types into
|
||||
//! `bindings/` (`cargo test -p yaak-send-proxy`) and published to the tab as
|
||||
//! `@yaakapp-internal/send-proxy`, so a change here is a type error there.
|
||||
//! `bindings/` (`cargo test -p yaak-web`) and published to the tab as
|
||||
//! `@yaakapp-internal/web`, so a change here is a type error there.
|
||||
|
||||
use serde::{Deserialize, Serialize};
|
||||
use ts_rs::TS;
|
||||
@@ -23,14 +23,14 @@ use yaak_models::models::{
|
||||
/// The body of `POST /v1/http/send`.
|
||||
#[derive(Deserialize, Debug, TS)]
|
||||
#[serde(rename_all = "camelCase")]
|
||||
#[ts(export, export_to = "gen_send_proxy.ts")]
|
||||
#[ts(export, export_to = "gen_web.ts")]
|
||||
pub struct SendRequest {
|
||||
/// The request to send, in the desktop's own model shape but with every template already
|
||||
/// rendered by the tab. The proxy builds the URL, headers and body from it exactly the way
|
||||
/// rendered by the tab. The server builds the URL, headers and body from it exactly the way
|
||||
/// the desktop does after rendering.
|
||||
pub request: HttpRequest,
|
||||
/// The resolved settings, values only. Where they came from is the tab's to record in
|
||||
/// its timeline; the proxy only needs to obey them.
|
||||
/// its timeline; the server only needs to obey them.
|
||||
pub settings: HttpSendSettings,
|
||||
/// The cookies to start with. `None` means no jar at all: nothing sent, nothing kept.
|
||||
#[serde(default)]
|
||||
@@ -45,7 +45,7 @@ pub struct SendRequest {
|
||||
rename_all = "snake_case",
|
||||
rename_all_fields = "camelCase"
|
||||
)]
|
||||
#[ts(export, export_to = "gen_send_proxy.ts")]
|
||||
#[ts(export, export_to = "gen_web.ts")]
|
||||
pub enum Frame {
|
||||
/// A timeline event, in the same shape the desktop stores. Interleaved with everything
|
||||
/// else in the order the engine produced it.
|
||||
Reference in New Issue
Block a user