CVE-2026-64192

*** Pendiente de traducción *** In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> bpf: Reject BPF_MAP_TYPE_INODE_STORAGE creation if BPF LSM is uninitialized<br /> <br /> When CONFIG_BPF_LSM=y is set, BPF inode storage maps<br /> (BPF_MAP_TYPE_INODE_STORAGE) are compiled into the kernel. However,<br /> if the BPF LSM is not explicitly enabled at boot time (e.g. omitted<br /> from the "lsm=" boot parameter), lsm_prepare() is never executed for<br /> the BPF LSM.<br /> <br /> Consequently, the BPF inode security blob offset<br /> (bpf_lsm_blob_sizes.lbs_inode) is never initialized and remains at<br /> its default compiled size of 8 bytes instead of being updated to a<br /> valid offset past the reserved struct rcu_head (typically 16 bytes<br /> or more).<br /> <br /> When a privileged user creates and updates a BPF_MAP_TYPE_INODE_STORAGE<br /> map, bpf_inode() evaluates inode-&gt;i_security + 8. This erroneously<br /> aliases the struct rcu_head.func callback pointer at the beginning<br /> of the inode-&gt;i_security blob. During subsequent map element cleanup<br /> or inode destruction, writing NULL to owner_storage clears the queued<br /> RCU callback pointer. When rcu_do_batch() later executes the queued<br /> callback, it attempts an instruction fetch at address 0x0, triggering<br /> an immediate kernel panic.<br /> <br /> Fix this by introducing a global bpf_lsm_initialized boolean flag<br /> marked with __ro_after_init. Set this flag to true inside bpf_lsm_init()<br /> when the LSM framework successfully registers the BPF LSM. Gate map<br /> allocation in inode_storage_map_alloc() on this flag, returning<br /> -EOPNOTSUPP if the BPF LSM is in turn uninitialized.<br /> <br /> This fail-fast approach prevents userspace from allocating inode<br /> storage maps when the supporting BPF LSM infrastructure is absent,<br /> avoiding zombie map states.