GHSA-R39G-X868-W7R7
Quadratic-time (O(n^2)) ReDoS in CSS length unit parsing (RE_UNIT) via a style attribute
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L
Fuente: GitHub
Ciclo de vida de la vulnerabilidad
| Fecha | Fase | Evento | Detalle |
|---|---|---|---|
| Divulgación | Publicada | GHSA-R39G-X868-W7R7 divulgada · CVSS 5.3 (GitHub) | |
| Monitoreo | Detectada | Ingresó al índice | |
| Monitoreo | Verificada | Última verificación contra las fuentes |
Summary
CssParse._get_em parses a CSS length with RE_UNIT = re.compile(r"(-?[0-9.]+)(\w+)")
via .search(). On a style value that is a long run of '.' characters, this search
is quadratic: [0-9.]+ matches the whole run at every start position and then the
required \w+ fails, backtracking each time. get_text() runs on untrusted HTML by
default, so a single small document hangs the process. This is a CPU/ReDoS issue
in the regex itself, distinct from the memory issue already reported for CSS
lengths.
Details
`src/inscriptis/model/css.py:33` and its use in `_get_em`:
RE_UNIT = re_compile(r"(-?[0-9.]+)(\w+)")
...
_m = CssParse.RE_UNIT.search(length) # line 70`length` is the value of a length-based CSS property in an untrusted `style` attribute. `attr_style()` dispatches `margin-top` / `margin-bottom` / `padding-left` (and aliases) to `_get_em(value)`. `RE_UNIT.search` over a value of only `.` characters never matches (`[0-9.]+` consumes the dots, `\w+` then fails because `.` is not a word char), and `re.search` retries from every start position and backtracks the greedy `[0-9.]+`, giving O(n^2) in the value length. Style attribute values have no length limit, and this is reachable in 2.7.6.
This is separate from the previously reported "CSS length drives unbounded string allocation" (a memory issue from the numeric value used as a repeat count). Here the cost is CPU inside the regex, before any number is used.
Suggested fix: cap the CSS value length before `RE_UNIT.search`, or anchor the match.
Proof of concept
import time
from inscriptis import get_text
for n in (10_000, 20_000, 40_000):
html = '<p style="margin-top:' + '.' * n + '">x</p>'
t0 = time.perf_counter()
get_text(html)
print(n, round(time.perf_counter() - t0, 1), "s")Scales as O(n^2) (doubling n ~= 4x time): 5,000 -> 0.22 s | 10,000 -> 0.89 s | 20,000 -> 3.52 s | 40,000 -> 14.1 s.
Impact
Anything that runs get_text()/get_annotated_text() on untrusted HTML (the
documented use) can be made to consume quadratic CPU and hang on a single small
document, a denial of service. No configuration or auth is required.
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 |
|---|---|---|---|---|
PyPIinscriptis |
Sin parche | — | — | — |
Arsenal público
Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.
https://github.com/weblyzard/inscriptis/security/advisories/GHSA-r39g-x868-w7r7Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.
