More beginner tips for SFTP server docs

[Andrew-Chen-Wang]

I was pretty confused by the parameter names of the SSH server, so I thought I'd leave some pointers for anyone else having trouble

To set up an AsyncSSH SFTP server, you can follow these steps:

Generate SSH Keys:

Generate a host key for the server:

ssh-keygen -t rsa -b 2048 -f ssh_host_key

Generate a user key:

ssh-keygen -t rsa -b 2048 -f id_rsa

Write the Server Code:

Use the following example to set up a simple SFTP server:

import asyncio
import asyncssh
import sys

async def start_server() -> None:
    await asyncssh.listen('', 8022, server_host_keys=['path/to/ssh_host_key'],
                          authorized_client_keys='path/to/id_rsa',
                          sftp_factory=True)

loop = asyncio.new_event_loop()

try:
    loop.run_until_complete(start_server())
except (OSError, asyncssh.Error) as exc:
    sys.exit('Error starting server: ' + str(exc))

loop.run_forever()

I needed to run this in pytest so:

import pytest
import asyncssh
import tempfile
from pathlib import Path

class TestBasedSFTPServer(asyncssh.SFTPServer):
    # https://asyncssh.readthedocs.io/en/latest/#sftp-server
    def __init__(self, chan: asyncssh.SSHServerChannel):
        # Doing the following would create a new directory per user
        root = f"/{tempfile.gettempprefix()}/sftp/" + chan.get_extra_info('username')
        os.makedirs(root, exist_ok=True)
        super().__init__(chan, chroot=root.encode())

@pytest.fixture(autouse=True)
async def server():
    async with asyncssh.listen(
        host="localhost",
        port=8022,
        reuse_port=True,
        password_auth=True,
        public_key_auth=True,
        server_host_keys=[Path(__file__).parent / "ssh_host_key"],
        authorized_client_keys=[Path(__file__).parent / "id_rsa"],
        sftp_factory=TestBasedSFTPServer
    ) as server:
        yield server

you may come across Host key is not trusted for host localhost

Run the following ssh-keyscan -p 8022 localhost >> ~/.ssh/known_hosts.

A workaround is setting your SFTP client to include a parameter known_hosts=None, but that's a security danger.

---

I'm actually struggling to find a workaround in my unit testing since I don't want to explicitly set ~/.ssh/known_hosts. I have a client that looks like:

    async with asyncssh.connect(
        host=auth.host,
        port=auth.port,
        # https://stackoverflow.com/questions/67222941/can-not-connect-via-asyncssh-error-host-key-is-not-trusted
        # FAQ: in pytest/devs' environments, we can't add the SSH key everywhere. For production,
        # it is expected that the generated key is trusted in ~/.ssh/known_hosts file.
        # TODO Add known_hosts public key
        #  with open('ssh_host_key.pub', 'rb') as f:
        #      host_key = asyncssh.import_public_key(f.read())
        known_hosts=None if "PYTEST_CURRENT_TEST" in os.environ else (),
        options=SSHClientConnectionOptions(
            username=auth.username,
            password=auth.password,
            client_keys=client_keys,
        ),

I tried adding

    with open('ssh_host_key.pub', 'rb') as f:
        host_key = asyncssh.import_public_key(f.read())

connect(
...
known_hosts=host_key
...

parameter to the client, but it failed with TypeError: 'RSAKey' object is not subscriptable or when specifying known_hosts=[host_key], I get TypeError: 'RSAKey' object is not iterable

Any hints as to what I should do?

[Menyadar]

known_hosts needs to be list, see description of it in docs https://asyncssh.readthedocs.io/en/stable/api.html#asyncssh.SSHClientConnectionOptions

In the Specifying known hosts section you can read that:

known_hosts [...] can be the name of a file or list of files containing known hosts, a byte string containing data in known hosts format, or an SSHKnownHosts object which was previously imported from a string by calling import_known_hosts() or read from files by calling read_known_hosts().

Personaly i'd go for using SSHKnownHosts object, so this should work:

known_hosts = asyncssh.SSHKnownHosts(host_key)

...

connect(
    ...
    known_hosts=known_hosts,
    ....
)

[Andrew-Chen-Wang]

Thanks for the response. I assume host_key here is the public key (.pub) file's content (where the server's private key is loaded in the server_host_key). Still with known_hosts=SSHKnownHosts(public_key) I get Host key is not trusted for host localhost (raised from host_key = client_conn.validate_server_host_key(host_key_data))

[ronf]

You need to keep in mind that the known_hosts format is more than just the contents of a public key. It needs to be preceded by a host pattern, and that can optionally be preceded by directive like @cert-authority or @revoked (but that wouldn't be needed here).

At a minimum, if you wanted to try a specific public key for all hosts (which could be ok in something like a unit test), you'd need to prefix the public key data with a '* ' (a star and a space). You can call asyncssh.import_known_hosts(data) to import data once it has this '* ' prefix, and pass that to known_hosts. If you know the hostname you are connecting to, you could also pass that hostname in place of the star.

If you encode the known hosts data in binary mode (as a bytes value), you can also pass that directly into the known_hosts argument without needing to call import_known_hosts on it. You will still need the leading host pattern, though.

[ronf]

On another note, you don't need to use ssh-keygen to generate keys. If you like, you can use asyncssh.generate_private_key to do that. There are methods on the generated key to write out either the private or public keys, in a variety of formats.

[Andrew-Chen-Wang]

At a minimum, if you wanted to try a specific public key for all hosts (which could be ok in something like a unit test), you'd need to prefix the public key data with a '* ' (a star and a space). You can call asyncssh.import_known_hosts(data) to import data once it has this '* ' prefix, and pass that to known_hosts. If you know the hostname you are connecting to, you could also pass that hostname in place of the star.

Adding * preceding the public ssh key worked, thank you!

On another note, you don't need to use ssh-keygen to generate keys. If you like, you can use asyncssh.generate_private_key to do that. There are methods on the generated key to write out either the private or public keys, in a variety of formats.

This works great! Writing here for future readers

import asyncssh

alg = "ssh-rsa"
user_private_key = asyncssh.generate_private_key(alg)
user_public_key = user_private_key.export_public_key().decode()
server_authorized_keys = asyncssh.SSHAuthorizedKeys()
server_authorized_keys.load(user_public_key)

server_private_key = asyncssh.generate_private_key(alg)
server_public_key = server_private_key.export_public_key().decode()

async with asyncssh.listen(
    ...
    public_key_auth=True,
    server_host_keys=[server_private_key],
    authorized_client_keys=server_authorized_keys,
    ...
)

private_key: str | None
passphrase: str | None
client_keys = []
if private_key:
    client_keys.append(
        asyncssh.import_private_key(private_key, passphrase=passphrase)
    )
async with asyncssh.connect(
    ...
    client_keys=client_keys,
    known_hosts=asyncssh.SSHKnownHosts(f"localhost {server_public_key}"),
    ...
)

Thanks so much for the support; this is such a thorough package 😍

[ronf]

Happy to help!

You should be able to do the following regarding known_hosts:

alg = 'ssh-rsa'
server_private_key = asyncssh.generate_private_key(alg)
server_public_key = server_private_key.export_public_key()

async with asyncssh.connect(
    ...
    known_hosts=b'localhost ' + server_public_key,
    ...
)

Your example here actually had decode() called twice, which I think would cause a problem. By not calling it at all and leaving the value as bytes, though, you can actually pass the resulting data directly as a known_hosts argument once you add the leading host pattern. This avoids you needing call into any AsyncSSH internals.

The other option to avoid including internals would be to call asyncssh.import_known_hosts(). In that case, the data can be a str, but the above is simpler.