GHSA-VG43-Q6XR-HP92
--hide-public-info masks public_address but not public_info_human, rendered from the same document
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N
Fuente: GitHub
Ciclo de vida de la vulnerabilidad
| Fecha | Fase | Evento | Detalle |
|---|---|---|---|
| Divulgación | Publicada | GHSA-VG43-Q6XR-HP92 divulgada · CVSS 5.3 (GitHub) | |
| Remediación | Parche 4.5.7 | Glances · 0 d desde divulgación | |
| Monitoreo | Detectada | Ingresó al índice | |
| Monitoreo | Verificada | Última verificación contra las fuentes |
Summary
`--hide-public-info` masks `public_address` and does not mask `public_info_human`, which is rendered from the same fetched document one line below it. The two statements sit inside one `if` block and only the first consults the flag.
`grep -rn hide_public_info glances/` over the whole package returns exactly one line. That is the only call site, and the sibling field does not get it.
Details
`glances/plugins/ip/__init__.py:121-125`:
if self.public_address:
stats['public_address'] = (
self.public_address if not self.args.hide_public_info else self.__hide_ip(self.public_address)
)
stats['public_info_human'] = self.public_info_for_human(self.public_info)`public_info_for_human` renders the unmodified document:
def public_info_for_human(self, public_info):
if not public_info:
return ''
return self.public_template.format(**public_info)What leaks depends on `public_template`, and there are two tiers. We are reporting both rather than leading with the stronger one.
- With the template you ship active — `conf/glances.conf:348`, `public_template={continent_name}/{country_name}/{city_name}` — the flag masks the address and leaves continent, country and city in `public_info_human`. That is still geolocation of the monitored host, surviving a switch named "hide public info".
- With the documented `{ip}` variant — `conf/glances.conf:340`, `# public_template={ip} {continent_name}/{country_name}/{city_name}`, shipped commented as the example — the public IP itself survives the flag in full, in the same response where `public_address` reads `203.0.*.*`.
The field reaches the curses UI, `--stdout`, `/api/4/ip`, XML-RPC and every exporter.
PoC
Released 4.5.6 from PyPI. The installed wheel was diffed against tag `v4.5.6` first and is byte-identical for every file touched.
The public-IP service is a local stub on `127.0.0.1` returning RFC 5737 TEST-NET-3 data — the plugin does `urlopen` + `json_loads`, so a stub is faithful and nothing is sent to a third party:
{"ip": "203.0.113.77", "continent_name": "Europe", "country_name": "France", "city_name": "Paris"}Config uses your documented example template, pointed at the stub:
[ip]
public_api=http://127.0.0.1:61290/json/
public_field=ip
public_template={ip} {continent_name}/{country_name}/{city_name}Run A — flag absent:
"public_address": "203.0.113.77",
"public_info_human": "203.0.113.77 Europe/France/Paris"Run B — same config, `--hide-public-info`:
"public_address": "203.0.*.*",
"public_info_human": "203.0.113.77 Europe/France/Paris"The masking of `public_address` is simultaneously the negative control: it proves the flag was active in run B. Without that line the run would be measuring nothing.
Impact
An operator who turns on the privacy switch keeps broadcasting what it names. In the shipped default template that is the host's geolocation; with the documented `{ip}` template it is the public IP itself, verbatim, beside a masked copy of the same value.
This is not a credential. We are not claiming more than it is: it is a redaction control that covers one field and stops one line short of its sibling.
Affected range
`>= 4.0.0, <= 4.5.6` — all supported 4.x — and still present on `develop`.
Determined by reading `glances/plugins/ip/__init__.py` at each tag, two greps per tag:
v3.2.7 human=0 hide=0 # field does not exist yet
v3.3.0 … v3.4.0.5 human=3 hide=0 # field exists; the flag does not
v4.0.0 human=4 hide=1 # flag introduced -> the gap begins here
v4.1.0 … v4.5.6 human=4 hide=1
develop f05c60d2 lines 121-125 unchanged`v4.0.0` shows the identical two-line shape at `:134-137`. The 3.3–3.4 line carries the field but not the flag, so there is no broken promise there, and `< 4.0` is `:x:` in your `SECURITY.md` anyway.
Suggested fix
Mask the source document rather than one derived field, so the two statements stop disagreeing:
stats['public_info_human'] = self.public_info_for_human(
self.public_info
if not self.args.hide_public_info
else {**self.public_info, self.public_field[0]: self.__hide_ip(self.public_address)}
)Masking the document also covers any future field rendered from it, and any template an operator writes that includes `{ip}`, without needing a second guard.
Severity
`CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N` — 5.3 Medium, for the web-server deployment (`glances -w`, unauthenticated by default).
Scored against your own nearest advisory rather than argued upward: `GHSA-4h34-v6r8-mmjc` / `CVE-2026-68520` — `as_dict_secure()` covering key names and not the value — is the same class, a redaction control that covers one thing and not its sibling, and you scored it at exactly this vector. This finding is that class on less sensitive data, so 5.3 is the ceiling rather than the floor.
We will not argue it upward and we would not object to Low. If you prefer `AV:L` on the grounds that the flag's real audience is the terminal and screenshot user, `CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N` = 3.3 is defensible and we would accept it. Nothing in the finding depends on the score.
Scope
The operator is not configuring a dangerous thing here — the operator is switching a privacy control on. Showing the public IP is the feature; `--hide-public-info` is your own opt-out from it, and what is broken is the extent of the opt-out.
That is the same complaint you accepted in `GHSA-59fj-m2j6-hcxh` (a documented switch covering some but not all of the paths it names) and in `GHSA-73wf-9vmv-5pv9` (a sanitizer covering some but not all of the values it claims).
Recorded, not claimed
Three hunts on adjacent surfaces came back negative and we are reporting that rather than leaving you to wonder whether we looked.
`--disable-config-exec` is complete: all seven exec paths enumerated, `system_exec` has one definition and one call site, inside the gated branch. The action-template sanitizer hunt was not run, because your newest fix there — `secure_popen` taking `render=` and expanding Mustache after tokenisation — is a chokepoint rewrite rather than a per-call-site patch, and a sibling hunt over a chokepoint fix is predicted to come back empty. And REST, XML-RPC and MCP have converged: the membership-versus-equality divergence from `GHSA-fp27-3fvq-9xr7` is fixed at `:303` in 4.5.6, even though that advisory's metadata still shows no patched version — you may want to correct the metadata.
Incluye detalles técnicos publicados por el investigador. Úselos solo para validar y corregir sus propios sistemas.
Desglose de métricas
Rangos afectados
Estado del parche por paquete
| Paquete / producto | Estado | Corregida en | Desde divulgación | Última versión |
|---|---|---|---|---|
pypiGlances |
Corregida | 4.5.7 | 0 d | 4.5.9 no afectada |
4.5.7Arsenal público
Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.
https://github.com/nicolargo/glances/security/advisories/GHSA-vg43-q6xr-hp92Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.
