| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
Signed-off-by: David Goulet <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
| |
This enables us to make these configurable: any relays that are not
configured to be an exit will exclude BEGIN and RESOLVE from their list
of allowed commands, causing exit and DNS streams to be rejected as soon
as the BEGIN/RESOLVE cell is received in the circuit reactor.
Context: https://gitlab.torproject.org/tpo/core/arti/-/merge_requests/4145#note_3430345
Part of #2606
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This implements what we discussed in
`doc/dev/notes/relay-streams.md` (lines 218-234):
> Currently, to allow incoming stream requests on a circuit,
> you first need to call `RelayCirc::allow_stream_requests()`
> to install a `CmdChecker` and `IncomingStreamRequestFilter`.
> This is not ideal, because `allow_stream_requests()` will need to be
> called unconditionally, on each `RelayCirc`,
> right after it's created in the `CreateHandler` impl
> (which in turn, would mean making `handle_create()` async too,
> because `allow_stream_requests()` is async, which wouldn't be great).
>
> So, the first step here is to rework the `RelayCirc` API to make relay circuits
> be constructable with a list of allowed `RelayCmd`s and `IncomingStreamRequestFilter`
> from the get-go ([#2582]), and to get rid of `allow_stream_requests()`,
> which will enable the `CREATE*` handler to remain non-`async`.
>
> In any case, the `CREATE*` handler will still require some changes,
> because it needs to be initialized with an `IncomingStreamRequestFilter`,
I am not sure using an `IncomingStreamRequestFilter` "factory" is
necessarily the right approach here, but the circuit `Reactor`'s
constructor needs to take an `IncomingStreamRequestFilter`, and
`IncomingStreamRequestFilter` is not `Clone` (and FWIW, I think it's
better if we don't make it `Clone`).
One obvious limitation is that the `IncomingStreamRequestFilter` of the
circuit reactor is fixed for the entire lifetime of the circuit.
In practice, I don't think this is going to be a problem,
because the arti-relay `IncomingStreamRequestFilter` is only going
be used for
* preventing single-hop exit streams
* per-circuit rate-limiting.
Both of these checks will require the filter to have access to a recent
`NetDir`, which is straightforward if the filter has an Arc<dyn
NetDirProvider> (as mentioned in doc/dev/notes/relay-streams.md,
`NetDirProvider` has a handy non-async `timely_netdir()` function we can
use). And since these checks are based on consensus params, we don't
really need to ever update an already-built circuit with a new
`IncomingStreamRequestFilter` (because all `IncomingStreamRequestFilter`
will have the ability to obtain a fresh `NetDir` as needed).
Nevertheless, I left a TODO about this, because I expect this type to
change once we figure out all the other pieces needed for #1448.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
Signed-off-by: David Goulet <[email protected]>
|
| |
|
|
| |
And update the docs
|
| |
|
|
| |
This just removes an unnecessary `async`.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
To pull this off, ChannelProvider::get_or_launch() needed to change from
"&self" to "self: Arc<Self>" so we could pass self to the spawned task.
This is fine as the caller of ChannelProvider (circuit reactor) has a
Arc<ChanMgr>.
This also removes the PhantomData for the runtime as we now actually use
it.
Signed-off-by: David Goulet <[email protected]>
|
| |
|
|
| |
Signed-off-by: David Goulet <[email protected]>
|
| | |
|
| |
|
|
| |
Signed-off-by: David Goulet <[email protected]>
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
Upcoming code for relay channels are put in the src/relay module and
thus we need visibility into some channel generic things.
Turns out also we don't need to re-export publicly UnverifiedChannel and
VerifiedChannel.
Signed-off-by: David Goulet <[email protected]>
|
| | |
|
| |
|
|
|
| |
This enables us to read the CBT estimates from the circuit reactor (we
need these to compute the half-stream timeouts for #264).
|
| | |
|
| |\
| |
| |
| |
| | |
tor-proto,tor-netdir: add flow control consensus parameters
See merge request tpo/core/arti!3249
|
| |/
|
|
|
| |
`Release.md` gives a wildcard path of `crates/*/semver.md`, so it seems
that these files should go in the top level of each crate.
|
| | |
|
| |
|
|
|
|
|
| |
This is just code motion (I suggest reviewing with `--color-moved`).
This also moves the implementation-agnostic parts from
`tor_proto::client::circuit` to a new `tor_proto::circuit` module.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
The `stream` module is client-specific, for the most part, so I am
moving it under `client`. Later on, we will factor out the parts that
can be shared with the relay implementation.
Note: this is a breaking change as the deleted `stream` module was
`pub`. We could've kept the module and reexported from it the public
types from `tor_proto::client::stream`, but I think it's better to have
this `client` namespacing, because it makes the separation between the
client and relay parts clearer.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
Having this be a `futures::Stream` makes it nicer to work with. For
example we are able to remove a boxed future from `DataReaderState`,
which should be better for performance and makes the code simpler.
As mentioned in a previous commit when this type was named
`StreamReader`, this type is public in the API, but is not actually
accessible. As far as I can tell there is no way to construct it or
access it.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
In rust, the typical nomenclature is to use "receiver" for channels, and
"reader" for byte streams. For example `mpsc::Receiver` for something
that returns objects and `AsyncRead` for something that reads bytes.
Since we also have a `DataReader` for reading bytes, I think renaming
this from `StreamReader` to `StreamReceiver` better describes what it is
(it's not a "reader" in the typical `Read`/`AsyncRead` sense).
This type is public in the API, but is not actually accessible. As far
as I can tell there is no way to construct it or access it.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
These functions have been deprecated for a while, and are now
complicating the `MutableState` changes we need to do for #1840, so it
seems like a good time to remove them.
|
| |
|
|
|
|
|
|
| |
```
for crate in $(./maint/list_crates | rg '^(tor|arti-)'); do
cargo set-version -p $crate 0.30.0
done
```
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
The constructor for rand::distr::Uniform is now fallible,
so it makes sense to bubble up its restrictions.
This is a breaking change.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It is unused but most importantly it allows any RELAY cell to be sent
from anywhere in the code which is really not desirable because it is
skipping congestion control.
It also allows us to remove the `control_tx` from the reactor which is
one less channel to track/understand/think about.
This opens up the door to all sorts of problems especially side channel
that can be exploited if we are not careful.
We can always bring this back if we need it but for now, it is unused
and allows us to remove the `CtrlMsg::SendRelayCell` control message.
No code behavior change.
Signed-off-by: David Goulet <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This rewrites the circuit reactor main loop to use `select_biased!` to poll
multiple futures simultaneously.
The new `run_once()`, like the old, first waits for an initial
`CtrlMsg::Create`. Then, it uses a `select_biased!` to poll the
`chan_sender` sink and shutdown channel for readiness.
When the channel sink is ready, we poll the `control` and `input`
channels like before, as well as the new `ready_streams` `Stream`
(`ready_streams` is a `futures::Stream` that replaces the previous
`send_outbound()` function).
Most of the implementation remains unchanged, except the `handle_input`,
`handle_cell` and `handle_control` functions no longer send anything on
the `chan_sender` channel. Instead, they may do some (synchronous)
processing, and send instructions for the remaining work that needs to
be done (for example, for writing the cell to the `chan_sender`
channel). These instructions are handled at the end of `run_once()`,
and are encoded in the `RunOnceCmdInner` enum.
What this change does **not** do:
* the control channel *still* bypasses congestion control. We could
fix this by making the various reactor functions send the
`RunOnceCmdInner` commands to `run_once()` via a channel
(instead of returning them). This would enable the reactor to stop
reading the commands (except for handle `Sendme`, which would be
handled separately) if it's blocked on congestion control.
|
| |
|
|
|
|
| |
This is unfortunately necessary, because after the channel handshake, we
need to give the channel reactor a `StreamOps` handle to the underlying
stream.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Needed for cases where we wrap an object that implements `StreamOps` in
an external type, thereby losing access to the `StreamOps`
functionality. For example, during the channel handshake, we `.split()`
the stream that implements `StreamOps`, which leaves us with a
`SplitSink` and a `SplitStream`, neither of which implement `StreamOps`.
Getting a handle to the underlying object that implements `StreamOps`
(for example, a file handle) *before* the stream is `.split()` enables
us to use `StreamOps` to manipulate the underlying split stream.
This commit also introduces a special `UnsupportedStreamOpsHandle`,
which is a type that implements `StreamOps`, but always returns an
error. This type is meant to simplify error handling and usage, and is
meant to be used in cases where `StreamOps` is not supported. TODO: the
name of this type is pretty confusing (it's very similar to
`UnsupportedStreamOp`, which is an error type), and should probably be
renamed to something else (`NoOpStreamOpsHandle`,
`BrokenStreamOpsHandle`, `DummyStreamOpsHandle` come to mind...).
Note: this changes the `StreamOps` trait to be slightly different from
what I originally envisioned in !2660 and #1769
|
| |
|
|
| |
The 1.2.4 release is out, so we won't be needing these anymore.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, Channel was a type that you could Clone that implicitly
its state. Now, Channel always appears as an Arc<Channel>.
This change has several benefits:
* It makes the relationship between Channel struct and the
underlying channel more clear.
* It enables Channel to participate in the RPC system,
where everything has to be an Arc<.>
* It enables us to have a Weak<Channel>, if we ever want to.
* It will let us move various members out of ChannelDetails.
We did this change a while ago with ClientCirc.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
This serves three purposes:
* It removes the 'send a cell' method from the channel's public
API. Nothing outside of tor-proto should have to use this.
* It paves the way for giving each circuit a separate handle onto
the channel's send functionality. This will eventually let
the channel multiplex among circuits more intelligently.
* It prepares for the next commit, which will make Channel itself
universally Arc<.>ed.
|
| | |
|
| | |
|
