PSB Fields

§5.2.1 Process identity (fixed at fork) #

The Process GUID is distinct from the PID. PIDs are recycled; Process GUIDs are unique within a boot (globally unique preferred). The GUID is the stable correlation key that KMES uses to attribute events to a process across its lifetime.

§5.2.2 Protection (set at exec, fixed) #

PIP fields are signing-based. At exec, the kernel MUST verify the binary's cryptographic signature and determine pip_type and pip_trust from the signer's identity. The verification algorithm and key model are specified in §6.1. The parent process MUST NOT be able to influence PIP determination — even a compromised peinit running as SYSTEM MUST NOT be able to forge PIP protection for an unsigned binary.

When pip_type is set to a value other than None at exec, the kernel MUST also set lsv (Library Signature Verification) on the PSB automatically. PIP without LSV is never correct — a Protected process that loads unsigned libraries has a code injection path that defeats the purpose of PIP. This is the only mitigation that is coupled to PIP; all other mitigations remain independently controlled by peinit.

The public verification key is compiled into the kernel image. The kernel only verifies signatures; it MUST NOT sign.

§5.2.3 Process mitigations (one-way) #

Mitigations are inherited from the parent at fork and MAY be set via syscall (typically by peinit between fork and exec). They are one-way: once set, they MUST NOT be cleared. Exec MUST NOT reset mitigations — a mitigation set by the process launcher persists regardless of what binary is loaded.

Setting a mitigation bit is activation-backed. Before changing a mitigation bit from clear to set, KACS MUST either activate the underlying protection for the target process or verify that the target process already satisfies the mitigation's invariant. If any requested mitigation cannot be activated or verified, the operation MUST fail closed and MUST NOT mutate any PSB mitigation bits from that request.

After a mitigation bit is committed, KACS MUST reject later operations that would disable the underlying protection or make the process violate the mitigation. Re-requesting an already-set mitigation MUST NOT clear or weaken the committed invariant.

For runtime memory mitigations, set-time activation MUST cover existing executable state as well as future transitions. Enabling wxp MUST fail if the target process already has any mapping that is simultaneously writable and executable or otherwise already violates the WXP invariant KACS can observe. Enabling tlp MUST fail if the target process already has a file-backed executable mapping whose current kernel-resolved path is missing, unresolvable, outside the approved prefix cache, or would otherwise be denied by TLP. Enabling lsv MUST fail if the target process already has a file-backed executable mapping whose signing material is missing, invalid, or below the target process's required PIP trust. Anonymous executable mappings are governed by wxp; tlp and lsv apply only to file-backed executable mappings.

For architecture-backed mitigations, set-time activation MUST use the architecture's kernel interface to place the process in the protected state and lock or otherwise prevent later process-controlled disablement. Enabling cfif, cfib, or sml MUST fail closed if the platform cannot make the requested protection true for the target process. If a platform proves that a protection is unconditionally active and not process-disableable, that condition satisfies activation for the corresponding bit.

The pie and no_child_process mitigations are event-gated rather than retroactive over existing state: pie is enforced at subsequent exec, and no_child_process is enforced at subsequent process creation. They still follow the same one-way commit rule and MUST be set before the event they are intended to constrain.

This is distinct from PIP, which is determined at exec from the binary's signature. Mitigations are set by the process launcher as policy; PIP is determined by the kernel as a property of the binary.

§5.2.4 UI access (one-way) #

§5.2.5 Process restrictions (one-way) #

Unlike the exec-time fields, no_child_process MAY be set at two points:

1. Between fork and exec. The parent's code, running in the freshly forked child, calls a KACS syscall to set the flag on itself before exec. The new binary loads with the restriction already in place.
2. At runtime by the process itself. A process MAY restrict itself at any time (e.g., after spawning worker processes).

§5.2.6 TLP cache (machine-wide) #

The approved directory prefixes for TLP enforcement are stored in a global kernel cache, not on individual PSBs. The PSB tlp flag controls whether a process is subject to TLP enforcement; the paths themselves are shared.

Cache structure: an array of absolute directory prefix byte strings evaluated against the kernel-resolved Linux path bytes. Maximum 64 entries, maximum 4096 bytes per path. Each prefix MUST begin with /, MUST end with / so /usr/lib/ does not match /usr/libevil, and MUST NOT contain an embedded NUL byte. Empty, relative, NUL-containing, or non-slash-terminated prefixes are invalid and MUST be rejected without mutating the existing cache. The mechanism for populating and updating the cache is defined by the registry subsystem (loregd) and is outside the scope of this section.

At mmap(PROT_EXEC) time, when the process has TLP enabled, the kernel checks whether the mapped file's current kernel-resolved backing file->f_path path starts with any approved prefix. If the backing path cannot be resolved, no prefix matches, or the cache is empty, the mmap is rejected. TLP applies only to file-backed executable mappings; anonymous executable mappings have no path and are governed by WXP only.

§5.2.7 Identity virtualization (reserved) #

§5.2.8 Process security descriptor #

Every process carries a security descriptor that controls who can perform operations on it. The process SD is stored on the PSB alongside the PIP and mitigation fields.