CVE-2026-112077

A flaw was found in radare2. When the embedded HTTP server is running, the sandbox mechanism fails to restrict process execution due to overly permissive default policy settings: http.sandbox defaults to true, but cfg.sandbox.grain defaults to 'all', leaving exec (and sockets, network, filesystem, and environment access) allowed. A remote attacker with network access to an exposed HTTP command service can send commands via /cmd that bypass the intended sandbox restrictions and execute arbitrary system processes.

Red Hat rates this flaw IMPORTANT (CVSS 8.1, CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H). The AC:H reflects the exploitation precondition: the embedded HTTP server must be started and exposed to the attacker; the maintainer states it is not remotely exploitable by default CLI usage. The C/I/A:H reflects the outcome of successful exploitation: the sandbox bypass permits execution of arbitrary system processes, as confirmed by the fix commit 18fd12efa98a7d1977de7bdd379d8a848777b90f, which explicitly blocks process execution, sockets, network, filesystem, and environment access across HTTP commands.

The upstream VulDB CNA rating of MODERATE (V4 6.3 / V3 5.6) uses C/I/A:L, which is inconsistent with the advisory's own description ('execute arbitrary system processes'). Red Hat's higher rating is therefore justified. The affected community streams (epel-all, fedora-all) ship radare2 5.9.8, which contains the vulnerable code path (r_core_cmd_str_pipe in libr/core/rtr_http.inc.c under the permissive default sandbox grain) fixed in radare2 6.2.4.