Description

Klever-Go: /log controls global node logging An unauthenticated client can connect to GET /log, send an arbitrary logger profile as the first WebSocket message, and mutate the node's global logging configuration before receiving live logs from the process. I confirmed this against a local validator built from this repository: an unauthenticated client set the global log level to *:NONE, the node accepted the profile, and the node stopped emitting normal slot logs while the WebSocket connection remained open.

This is not a duplicate of the published KVM or P2P advisories. It is a management-plane flaw in the public WebSocket logging endpoint.

Vulnerability details

Affected code

Route exposed by default in config/node/api.yaml

Route registration in network/api/api.go

Unauthenticated upgrade in network/api/api.go

First client message is parsed as a logger profile and applied globally in network/api/logs/logSender.go

Global logger mutation happens in dependency github.com/klever-io/klever-go-logger, profile.go, Apply()

Root cause

/log is enabled by default and does not require authentication. After the WebSocket upgrade, the server reads the first client message and treats it as a logger Profile. That profile is then applied process-wide through profile.Apply(), which changes global log level patterns and output formatting options for the whole node.

After that handshake, the same unauthenticated connection is registered as a log observer and receives live logs from the running process.

Reproduction steps

Environment

Validator built from the repository at commit 9640d63265e910e166dfa694c8e5ddeb53018ffd

REST API bound locally for validation

Tested on 2026-05-30

Step 1: Build and run a local validator from source

cd <repo-root>

go build -o ./bin/validator ./cmd/node

./bin/validator \
  --rest-api-interface=127.0.0.1:18080 \
  --port=18083 \
  --config=./config/node/config.yaml \...
Show more

The node exposes GET /log and a plain HTTP request already shows it is a live WebSocket endpoint:

curl -i http://127.0.0.1:18080/log

Observed response:

HTTP/1.1 400 Bad Request
Sec-Websocket-Version: 13

Step 2: Confirm normal node logging before the attack

Before the attack, the validator emits periodic slot logs such as:

#################################### SLOT 14 BEGINS ####################################
#################################### SLOT 15 BEGINS ####################################

Step 3: Connect to /log without authentication and apply a global mute profile

Run the PoC file:

cd <repo-root>
go run ./poc-log-profile-control.go \
  -url ws://127.0.0.1:18080/log \
  -profile none \
  -hold 12s

Full PoC source:

package main

import (
	"flag"
	"fmt"
	"log"
	"time"
...
Show more

Save the PoC as poc-log-profile-control.go in the repository root, or run it from any directory with access to the Go module cache.

Step 4: Observe the validator accepts and applies the unauthenticated profile

While the PoC is connected, the validator prints:

websocket log profile received  profile = [pattern=*:NONE, with correlation=false, with logger name=false]

Step 5: Observe logging is suppressed while the attacker connection remains open

In my local reproduction, the validator emitted:

SLOT 14 BEGINS
websocket log profile received profile = [pattern=*:NONE, ...]
reverted log profile profile = [pattern=*:INFO, ...]
SLOT 18 BEGINS

The expected slot logs for the interval while *:NONE was active did not appear. This proves that an unauthenticated client can suppress process logs globally while the WebSocket remains connected.

Step 6: Observe the profile is restored only after the attacker disconnects

After the PoC closes the WebSocket, the validator prints:

reverted log profile  profile = [pattern=*:INFO, with correlation=false, with logger name=false]

The revert happens because the server stores the previous profile and restores it only on disconnect. During the lifetime of the attacker connection, the attacker-controlled profile remains active.

Impact

An unauthenticated attacker can:

read live process logs over /log

mute node logging completely with *:NONE

increase verbosity to *:TRACE and force noisy logging

toggle correlation and logger-name settings process-wide

This affects both confidentiality and operational integrity.

In the reproduced case, the attacker hid normal validator slot logs for multiple slot intervals. In real deployments, logs commonly contain operational details, peer information, error traces, and occasionally secrets or credentials emitted by adjacent components. Even when no secrets are present, the ability to suppress or distort logs from the public network is a meaningful security impact because it degrades detection, incident response, and operator visibility while an attacker is active.

This issue is distinct from:

GHSA-jc6w-wmfc-fh33 (KVM read-only execution side effects)

GHSA-87m7-qffr-542v (MultiDataInterceptor remote OOM)

GHSA-74m6-4hjp-7226 (MultiDataInterceptor throttler slot leak)

Those are VM/P2P-path flaws. This finding is an unauthenticated management-plane flaw in the WebSocket logging endpoint.

Recommended fix

Immediate

Remove /log from the default open: true route set.

Require authentication before upgrading the WebSocket.

Reject unauthenticated clients before any profile message is processed.

Short term

Do not apply client-provided logger profiles to the process-global logger.

If remote log viewing is required, allow only a fixed server-side profile or a strict allowlist of safe settings.

Enforce origin checks and, if possible, bind /log to localhost-only or a dedicated admin interface.

Long term

Separate log streaming from global logger configuration.

Move any profile mutation capability behind an authenticated admin-only channel with explicit authorization and audit logging.

Mitigation

Aliases

References