If there is an error when opening the session for agent forwarding in
the process ssh connention, there is a deadlock when recursively
calling Connect on an internal reattempt. Avoid that, and let the
connection be reattempted externally.
If a connection fails and attempts to reconnect after the keep-alive
loop started, the client will be pulled out from under the keep-alive
requests. Close over a local copy of the client, so that reconnecting
doesn't race with the keepalive loop terminating.
1) Mention the host and port in the "Connecting..." message.
2) Mention the username in the post-connection handshaking message.
3) If handshaking fails, mention the user, host, and port in the error
message that will eventually be returned to the user.
An ssh server should always send a reply packet to the keepalive
request. If we miss those replies for over 2min, consider the connection
dead and abort, rather than block the provisioner indefinitely.
Match the tested behavior, and that of the ssh implementation, where the
communicator automatically connects when starting a command.
Remove unused import from legacy dependency handling.
Most of the time an ssh authentication failure would be non-recoverable,
but some host images can start the ssh service before it is properly
configured, or before user authentication data is available.
Log ssh authentication errors and allow the provisioner to retry until
the connection timeout.
Combine the ExitStatus and Err values from remote.Cmd into an error
returned by Wait, better matching the behavior of the os/exec package.
Non-zero exit codes are returned from Wait as a remote.ExitError.
Communicator related errors are returned directly.
Clean up all the error handling in the provisioners using a
communicator. Also remove the extra copyOutput synchronization that was
copied from package to package.
The remote.Cmd struct could not convey any transport related error to
the caller, meaning that interrupted commands would show that they
succeeded.
Change Cmd.SetExited to accept an exit status, as well as an error to
store for the caller. Make the status and error fields internal,
require serialized access through the getter methods.
Users of remote.Cmd should not check both Cmd.Err() and Cmd.ExitStatus()
until after Wait returns.
Require communicators to call Cmd.Init before executing the command.
This will indicate incorrect usage of the remote.Cmd by causing a panic
in SetExitStatus.
Add `host_key` and `bastion_host_key` fields to the ssh communicator
config for strict host key checking.
Both fields expect the contents of an openssh formated public key. This
key can either be the remote host's public key, or the public key of the
CA which signed the remote host certificate.
Support for signed certificates is limited, because the provisioner
usually connects to a remote host by ip address rather than hostname, so
the certificate would need to be signed appropriately. Connecting via
a hostname needs to currently be done through a secondary provisioner,
like one attached to a null_resource.
It's becoming more common for users to have many ssh keys loaded into an
agent, and with the default max auth attempts of an openssh server at 6,
one often needs to specify which id to use in order to avoid a `too many
authentication failures` error.
Add a connection field called `agent_identity` which will function
similarly to the ssh_config IdentityFile when used in conjunction with
an ssh agent. This uses `agent_identity` rather than `identity_file` to
specify that the file is not used directly for authentication, rather
it's used to choose which identity returned from the agent to
authenticate with first.
This feature tries a number of different methods to match the agent
identity. First the provisioner attempts to read the id file and extract
the public key. If that isn't available, we look for a .pub authorized
key file. Either of these will result in a public key that can be
matched directly against the agent keys. Finally we fall back to
matching the comment string exactly, and the id as a suffix. The only
result of using the agent_identity is the reordering of the public keys
used for authentication, and if there is no exact match the client
will still attempt remaining keys until there is an error.
These are skipped without the SSH_UPLOAD_TEST flag, since they require
connecting to a running ssh server on localhost as the current user, and
one creates a very large file.
Fixes#10463
I'm really surprised this flew under the radar for years...
By having unique PRNGs, the SSH communicator could and would
generate identical ScriptPaths and two provisioners running in parallel
could overwrite each other and execute the same script. This would
happen because they're both seeded by the current time which could
potentially be identical if done in parallel...
Instead, we share the rand now so that the sequence is guaranteed
unique. As an extra measure of robustness, we also multiple by the PID
so that we're also protected against two processes at the same time.
Fixes#3605 and adds the functionality suggested in PR #7440.
This PR is using a different appraoch that (IMHO) feels cleaner and (even more important) adds support for Windows at the same time.
Without a seed, the "random" script path locations for the remote-exec
provisioner were actually deterministic!
Every rand.Int31() would return the same pseudorandom chain starting w/
the numbers: 1298498081, 2019727887, 1427131847, 939984059, ...
So here we properly seed the communicators so the script paths are
actually random, and multiple runs on a single remote host have much
less chance of clobbering each other.
Fixes#4186
Kudos to @DustinChaloupka for the correct hunch leading to this fix!
We've been moving away from config fields expecting file paths that
Terraform will load, instead prefering fields that expect file contents,
leaning on `file()` to do loading from a path.
This helps with consistency and also flexibility - since this makes it
easier to shift sensitive files into environment variables.
Here we add a little helper package to manage the transitional period
for these fields where we support both behaviors.
Also included is the first of several fields being shifted over - SSH
private keys in provisioner connection config.
We're moving to new field names so the behavior is more intuitive, so
instead of `key_file` it's `private_key` now.
Additional field shifts will be included in follow up PRs so they can be
reviewed and discussed individually.
On connections where no second hop is made, there's no problem if the
agent forwarding connection is denied, so we shouldn't treat that as a
fatal error.
In the SSH client configuration, we had SSH Agent authentication listed
before the static PrivateKey loaded from the `key_file` setting.
Switching the default of the `agent` setting exposed the fact that the
SSH agent overrides the `key_file` during the handshake. By listing the
`key_file` first, we catch the provided key before any query goes out to
the agent.
Adds a key-based authentication SSH test to cover this new behavior. It
fails without the reordering on any machine with an SSH agent running.
Fixes#2614
* adds `bastion_*` fields to `connection` which add configuration for a
bastion host
* if `bastion_host` is set, connect to that host first, then jump
through it to make the SSH connection to `host`
* enables SSH Agent forwarding by default
This changes SSH Agent utilization from opt-in to opt-out, bringing
Terraform in line with the behavior of Packer and the `ssh` command
itself.
But skip SSH agent connections if the SSH_AUTH_SOCK env var doesn't exist,
whcih means there's no agent to connect to.
I added a debug log line in the last commit, only to find out it’s now
logging the same info twice. So removed the double entry and tweaked
the existing once.
Jaime Caamaño Ruiz