Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() without acl.CheckAccessHTTP(r, acl.DEBUGGING), unlike comparable debug endpoints. A remote caller who can reach the vttablet debug HTTP port can bypass the configured security policy and stream VrLogStats data produced from NewVrLogStats().Send(), including literal SQL statements and bound application values from MoveTables, Reshard, Materialize, and vitess-strategy Online DDL workflows
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() without acl.CheckAccessHTTP(r, acl.DEBUGGING), unlike comparable debug endpoints. A remote caller who can reach the vttablet debug HTTP port can bypass the configured security policy and stream VrLogStats data produced from NewVrLogStats().Send(), including literal SQL statements and bound application values from MoveTables, Reshard, Materialize, and vitess-strategy Online DDL workflows
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() without acl.CheckAccessHTTP(r, acl.DEBUGGING), unlike comparable debug endpoints. A remote caller who can reach the vttablet debug HTTP port can bypass the configured security policy and stream VrLogStats data produced from NewVrLogStats().Send(), including literal SQL statements and bound application values from MoveTables, Reshard, Materialize, and vitess-strategy Online DDL workflows
## Vulnerability Details
**File**: `go/vt/vttablet/tabletmanager/vreplication/vrlog.go`
### Summary
`vttablet`'s `/debug/vrlog` HTTP endpoint streams live VReplication event data — including the literal SQL DML statements being replicated by MoveTables, Reshard, Materialize, and "vitess"-strategy Online DDL workflows — with no authorization check at all. Every comparable "debugging" HTTP endpoint in vttablet/vtgate (querylogz, queryz, txlogz, livequeryz, schemaz, debugenv, hotrows, tablet_plans, query_stats, query_rules) calls `acl.CheckAccessHTTP(r, acl.DEBUGGING)` before serving data, so that the cluster operator's configured `--security-policy` (e.g. `deny-all`, `read-only`, or a custom plugin) is actually honored. `vrlog.go` is the one exception: it has zero references to the `acl` package.
### Root Cause
`addHttpEndpoint()` registers `/debug/vrlog` via `servenv.HTTPHandleFunc`, and `vrlogStatsHandler()` immediately starts streaming subscribed `VrLogStats` events to the response writer without first calling `acl.CheckAccessHTTP(r, acl.DEBUGGING)`, unlike every sibling handler in the same family (see e.g. `go/vt/vttablet/tabletserver/querylogz.go`'s `querylogzHandler`, which calls the check first).
The data streamed is sensitive: `go/vt/vttablet/tabletmanager/vreplication/vplayer.go` calls `NewVrLogStats(...).Send(sql)` / `.Send(event.Statement)` for every row change and statement event flowing through a VReplication stream (vplayer.go:339, 700, 771, 785) — i.e. the literal SQL (including bound data values) being copied/replicated by MoveTables, Reshard, Materialize, and Online DDL.
### Attack Scenario
1. A cluster operator configures `--security-policy=deny-all` (or `read-only`, or a custom policy) specifically to lock down debugging/admin HTTP endpoints on vttablet, relying on this being uniformly enforced.
2. An attacker who can reach the vttablet debug HTTP port (common in Kubernetes/Prometheus-scraping deployments where this port is exposed beyond localhost) — but who does NOT have the `DEBUGGING`/admin role the policy requires — sends `GET /debug/vrlog`.
3. Every other debug endpoint correctly returns `403 Forbidden`. `/debug/vrlog` returns `200 OK` and streams live VReplication event data, including raw SQL DML statements containing application data values, for as long as the attacker keeps the connection open (bounded by `timeout`/`limit` query params, repeatable).
### Impact
Confidentiality impact: disclosure of live replicated application data (potentially including PII or other sensitive column values) to an unauthorized actor, bypassing an access control the operator explicitly configured. No write/modify capability; this is a read-only information-disclosure / access-control-bypass issue, scoped to the vttablet debug HTTP listener.
### Vulnerable Code
```go
// go/vt/vttablet/tabletmanager/vreplication/vrlog.go
func addHttpEndpoint() {
servenv.HTTPHandleFunc("/debug/vrlog", func(w http.ResponseWriter, r *http.Request) {
ch := vrLogStatsLogger.Subscribe("vrlogstats")
defer vrLogStatsLogger.Unsubscribe(ch)
vrlogStatsHandler(ch, w, r)
})
}
func vrlogStatsHandler(ch chan *VrLogStats, w http.ResponseWriter, r *http.Request) {
timeout, limit := parseTimeoutLimitParams(r)
// no acl.CheckAccessHTTP(r, acl.DEBUGGING) call anywhere in this file
...
```
### Recommended Fix
```go
import "vitess.io/vitess/go/acl"
func vrlogStatsHandler(ch chan *VrLogStats, w http.ResponseWriter, r *http.Request) {
if err := acl.CheckAccessHTTP(r, acl.DEBUGGING); err != nil {
acl.SendError(w, err)
return
}
timeout, limit := parseTimeoutLimitParams(r)
...
```
This mirrors the exact pattern already used by `querylogz.go`, `queryz.go`, `txlogz.go`, `livequeryz.go`, `schemaz.go`, `debugenv.go`, and `tx_serializer.go` (hotrows) in the same codebase.
### Verification
Built v24.0.1 from source and wrote a standalone Go test that: (1) activates the real `deny-all` security policy via `acl.RegisterFlags`, (2) registers `/debug/vrlog` via the real, unmodified `vreplication.NewVrLogStats(...)` call path (the same one `vplayer.go` uses for live replication traffic), (3) serves the real `servenv` HTTP mux on a loopback listener, and (4) issues a real `GET /debug/vrlog` while emitting a simulated replicated statement.
Result:
```
acl.CheckAccessHTTP(DEBUGGING) under deny-all => err=not allowed: deny-all security-policy enforced
GET /debug/vrlog under deny-all => status=200 body="ROWCHANGE Event\tINSERT INTO secret_table (ssn) VALUES ('leaked-via-vrlog')\t2026-06-22T11:01:55\t147101\n"
```
The identical ACL check that protects every sibling endpoint correctly rejected the request, while `/debug/vrlog` returned 200 and streamed the simulated sensitive content — confirming the bypass against real, unmodified v24.0.1 code.
Also confirmed via the GitHub Contents API that `vrlog.go` has zero references to the `acl` package on `main`, `release-23.0`, and `release-22.0` as well, so all currently supported release lines appear affected.
The vttablet /debug/vrlog HTTP endpoint streams live VReplication event data, including SQL statements and table row changes, without verifying authorization via acl.CheckAccessHTTP. An unauthenticated actor with network access to the debug endpoint can stream live replicated SQL data.
CISA KEV lists vulnerabilities known to be exploited. FIRST EPSS estimates the likelihood of exploitation in the next 30 days. An unavailable score does not mean zero risk.
Which CVEs have scores here?
This view includes predictions for CVE identifiers explicitly listed in public NuGet advisories. Microsoft release pages have their own prediction coverage. Check source status for the scope and latest checks.
No exploitation or prediction data is available here for these CVEs. This does not mean they are safe or unlikely to be exploited.
Risk history
Compare earlier KEV listings and EPSS scores. Advisory wording and affected versions have a separate history.
When does an EPSS score appear as a change?
A prediction change is reported when EPSS rises from below 10% to 10% or higher. Every daily score change from the last 30 days stays in this history. Before that, it keeps KEV changes and moves between EPSS bands (0.1%, 1%, 5%, 10%, 20%, 50% and 90%).
Advisory history
17 saved versions
See how the publisher's advisory changed over time. Each link opens the version saved at that time. The publisher's update date can differ from the date shown here.
How advisory history works
Each link identifies one complete publisher record. History can include advisories that previously shared an identifier; an identifier removed by a publisher no longer establishes a current relationship. A historical link shows only that record and earlier saved versions. Failed checks and unchanged reads do not create new versions.
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…
Vitess is a database clustering system for horizontal scaling of MySQL. In 24.0.2 and earlier, the /debug/vrlog endpoint registered by addHttpEndpoint() in go/vt/vttablet/tabletmanager/vreplication/vrlog.go invokes vrlogStatsHandler() witho…