RFD 189: Console Access - vmadmd console proxy #1159

nwilkens · 2025-10-08T17:27:52Z

Summary

Implements console proxy infrastructure in vmadmd to support RFD 189 Console Access Through CloudAPI.

Add TCP-based console proxy for all VM brands (KVM, Bhyve, Joyent, Joyent-minimal, LX)
Support both KVM unix socket consoles and zone-based zoneadmd consoles
Add console to runtime_info for all brands to enable vmadm info queries
Implement proper handshake protocol for zoneadmd console connections
Add automatic lifecycle management (spawn on start, cleanup on stop)
Include resource leak prevention with timeouts and proper socket cleanup

Changes

8 commits implementing console proxy support
Modified: src/vm/sbin/vmadmd.js - Console proxy infrastructure
Modified: src/vm/node_modules/proptable.js - Runtime_info support for all brands

Related PRs

RFD: https://github.com/TritonDataCenter/rfd/pull/[RFD_0189_PR_NUMBER]
CNAPI: [To be created]
CloudAPI: [To be created]
node-triton: [To be created]

Commits

Add console proxy support to vmadmd
Add console to runtime_info for all VM brands
Spawn console proxy for all running VMs at startup
Fix console proxy to use fs streams for device files
Use socat to proxy zone console devices
Use zoneadmd console socket instead of /dev/zcons device
Add zoneadmd console handshake for non-KVM brands
Fix resource leaks and add handshake timeout in console proxy

Implement TCP proxy for console access across all VM brands: - KVM: Proxy to unix socket at /tmp/vm.console (serial console) - Bhyve/Joyent/LX: Proxy to zone console device at /dev/zcons/<zonename>/zoneconsole Changes: - Add CONSOLE object to track console proxy state - Add spawnConsoleProxy() to create TCP listeners for console access - Add clearConsoleProxy() and reloadConsoleProxy() for lifecycle management - Add infoConsole() to return console connection details via vm.info() - Add console proxy spawning alongside VNC/SPICE in VM lifecycle - Add vm.info() support for joyent, joyent-minimal, and lx brands (console only) Console proxies are automatically created when VMs start and cleaned up when VMs stop. Connection details (host, port, type) are available via vm.info(['console']). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

Add 'console' to runtime_info feature list for all brands to enable vmadm info queries for console connection details. Changes: - KVM: Add 'console' to existing runtime_info array - Bhyve: Add 'console' to runtime_info array - Joyent: Add runtime_info: ['console'] (new) - Joyent-minimal: Add runtime_info: ['console'] (new) - LX: Add runtime_info: ['console'] (new) This allows vm.info(['console']) to work for all brands, which is required by CNAPI console endpoint and vmadmd console proxy. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

Fix issue where non-KVM/Bhyve VMs (Joyent, LX) were not getting console proxies spawned at vmadmd startup. Previously only KVM/Bhyve called loadVM() which spawned the console proxy. Now all running VMs get console proxies regardless of brand. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

Zone console devices (/dev/zcons/*/zoneconsole) are character devices, not unix sockets. net.connect() fails with ENOTSOCK when trying to connect to them. Fix by using different connection methods based on brand: - KVM: net.Stream().connect() for unix socket - Others: fs.createReadStream/WriteStream for device file This allows proper bidirectional data flow for zone console devices. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

fs.createReadStream doesn't work properly with character devices like /dev/zcons/*/zoneconsole. Use socat instead to handle the device I/O properly. For non-KVM brands, spawn socat process per connection to proxy between TCP socket and zone console device with proper raw/echo settings. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

The correct console socket for zlogin -C is managed by zoneadmd at /var/run/zones/<zonename>.console_sock, not the /dev/zcons device. All brands now use unix sockets: - KVM: /zones/<uuid>/root/tmp/vm.console (serial console) - Others: /var/run/zones/<zonename>.console_sock (zoneadmd console) This allows simple net.Stream() connection for all brands. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

The zoneadmd console_sock requires a protocol handshake before data flows. Send 'IDENT C 0\n' and wait for 'OK' response before proxying data. This matches the zlogin -C handshake protocol. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

Critical fixes for console connection handler: 1. Replace deprecated net.Stream() with new net.Socket() 2. Add cleanup() function with idempotent guard to prevent double-cleanup 3. Add 5-second timeout for zone console handshake 4. Ensure socket.destroy() called on all error paths: - Client close - Console socket error - Client error - Handshake timeout - Handshake failure 5. Rename 'console' variable to 'consoleSocket' to avoid shadowing global 6. Truncate handshake error messages to prevent log spam These fixes prevent file descriptor leaks and hanging connections that could exhaust system resources under error conditions. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

Copilot

Pull Request Overview

This PR implements console proxy infrastructure in vmadmd to support RFD 189 Console Access Through CloudAPI. The implementation adds TCP-based console proxy functionality for all VM brands (KVM, Bhyve, Joyent, Joyent-minimal, LX) with proper lifecycle management and resource leak prevention.

Adds console proxy infrastructure with TCP server that proxies to console sockets
Implements proper handshake protocol for zoneadmd console connections for non-KVM brands
Integrates console proxy into VM lifecycle management (spawn on start, cleanup on stop)

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

Copilot · 2025-10-08T17:42:12Z

src/vm/sbin/vmadmd.js

+    server.listen(port, SDC.sysinfo.admin_ip, function () {
+        addr = server.address();
+
+        CONSOLE[vmobj.uuid] = {
+            'host': SDC.sysinfo.admin_ip,
+            'port': addr.port,
+            'server': server,
+            'type': 'socket',
+            'path': consolePath
+        };


The CONSOLE object is being populated inside the async callback, but the function returns immediately without waiting for the server to start listening. This creates a race condition where other code trying to access CONSOLE[vmobj.uuid] might find it undefined.

A JS person should reality-check this.

When this moves to non-draft we'll need >= two reviewers for JS reality checking.

Copilot · 2025-10-08T17:42:13Z

src/vm/sbin/vmadmd.js

+                    clearTimeout(handshakeTimer);
+                    handshakeTimer = null;
+
+                    if (data.toString().indexOf('OK') === 0) {


Using indexOf for protocol validation is fragile. The response should be validated more strictly to ensure it matches the expected protocol format exactly.

Suggested change

if (data.toString().indexOf('OK') === 0) {

if (data.toString().trim() === 'OK') {

A JS person should reality-check this.

Copilot · 2025-10-08T17:42:13Z

src/vm/sbin/vmadmd.js

+                        consoleSocket.pipe(c);
+                    } else {
+                        log.error('console handshake failed for VM ' + vmobj.uuid +
+                            ': ' + data.toString().substring(0, 100));


Magic number 100 for substring length should be defined as a named constant to make the truncation limit explicit and configurable.

Suggested change

': ' + data.toString().substring(0, 100));

': ' + data.toString().substring(0, CONSOLE_HANDSHAKE_LOG_TRUNCATE_LEN));

Copilot · 2025-10-08T17:42:13Z

src/vm/sbin/vmadmd.js

+                    cleanup();
+                    c.end();
+                }
+            }, 5000);


Magic number 5000 for handshake timeout should be defined as a named constant to make the timeout configurable and self-documenting.

Hey, copilot is making its case well so far.

danmcd · 2025-10-08T18:11:25Z

1.) Thank you for the draft PR.

2.) Copilot does not count as an official reviewer for our purposes. It'll be interesting to see if its feedback is helpful or distracting.

danmcd · 2025-10-08T18:11:31Z

1.) Thank you for the draft PR.

2.) Copilot does not count as an official reviewer for our purposes. It'll be interesting to see if its feedback is helpful or distracting.

- Add CONSOLE_HANDSHAKE_TIMEOUT (5000ms) and CONSOLE_LOG_TRUNCATE_LEN (100) named constants to replace magic numbers - Use stricter protocol validation: data.toString().trim() === 'OK' instead of indexOf('OK') === 0 for more robust handshake checking The stricter validation prevents false positives if the response contains "OK" as a substring of a different message. Addresses PR review feedback from #1159 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

danmcd

If I'm reading this correctly, this will always open a TCP console port in the global zone for any BHYVE or KVM VM IF AND ONLY IF the "console" property in the VM object exists.

We should have a documentation update in the PR if I'm understanding correctly, and it should be off by default. OMG vnc is ENABLED BY DEFAULT. AND HAS BEEN SINCE THE DAYS OF JOYENT?!?

This is not your fault, but I just noticed this (eeesh).

There is an easy way to disable it (vmadm update $UUID vnc_port=-1), and if we go down this path we need to behave like VNC does, or even MORE strict, because there's no password-authentication save the console itself, unlike VNC.

danmcd · 2026-01-12T15:15:38Z

src/vm/sbin/vmadmd.js

+    server.listen(port, SDC.sysinfo.admin_ip, function () {
+        addr = server.address();
+
+        CONSOLE[vmobj.uuid] = {
+            'host': SDC.sysinfo.admin_ip,
+            'port': addr.port,
+            'server': server,
+            'type': 'socket',
+            'path': consolePath
+        };


When this moves to non-draft we'll need >= two reviewers for JS reality checking.

danmcd · 2026-01-12T15:16:48Z

src/vm/sbin/vmadmd.js

    // Start Remote Display
    spawnRemoteDisplay(vmobj);
+
+    // Start Console Proxy


This and other always-on calls to spawnRemoteDisplay() have me curious about an overarching question. Will ask it in the overarching comment.

nwilkens and others added 8 commits September 30, 2025 16:45

nwilkens requested a review from Copilot October 8, 2025 17:41

Copilot AI reviewed Oct 8, 2025

View reviewed changes

danmcd requested a review from a team October 8, 2025 18:09

danmcd requested changes Jan 12, 2026

View reviewed changes

	if (data.toString().indexOf('OK') === 0) {
	if (data.toString().trim() === 'OK') {

	': ' + data.toString().substring(0, 100));
	': ' + data.toString().substring(0, CONSOLE_HANDSHAKE_LOG_TRUNCATE_LEN));

RFD 189: Console Access - vmadmd console proxy #1159

Are you sure you want to change the base?

RFD 189: Console Access - vmadmd console proxy #1159

Uh oh!

Conversation

nwilkens commented Oct 8, 2025

Summary

Changes

Related PRs

Commits

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Uh oh!

Copilot AI Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

danmcd Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

danmcd Jan 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

danmcd Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

danmcd Oct 8, 2025

Choose a reason for hiding this comment

Uh oh!

danmcd commented Oct 8, 2025

Uh oh!

danmcd commented Oct 8, 2025

Uh oh!

danmcd left a comment

Choose a reason for hiding this comment

Uh oh!

danmcd Jan 12, 2026

Choose a reason for hiding this comment

Uh oh!

danmcd Jan 12, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants