Phase 1.0S: hbldr BigApp launch and safety contract

Status: EXISTING_ROUTE_UNSAFE_FOR_CURRENT_DEVICE_GATE

v0.19 callgraph

host telnet client
  -> shsrv TCP listener on port 2323
     -> accepted connection
        -> elfldr_spawn(embedded sh.elf)
           -> shell parses command
              -> builtin hbldr
                 -> elfldr_spawn(embedded hbldr.elf)
                    -> read requested ELF from device filesystem
                    -> ensure /system_ex/app/FAKE00000 exists
                       -> optionally remount /system_ex
                       -> optionally mkdir/write param.json and eboot.bin
                    -> initialize foreground-user context
                    -> kill currently running BigApp, if present
                    -> attach to SceSysCore and follow fork
                    -> sceSystemServiceLaunchApp("FAKE00000", ...)
                    -> follow child exec
                    -> set child rootdir/jaildir through kernel helpers
                    -> replace BigApp process image with requested ELF
                    -> detach; target continues

The v0.7 path has the same essential distinction from raw elfldr but directly launches VideoPlayer title PPSA01659. It does not contain the later FAKE00000 creation/remount path. It still kills a running BigApp, reads a device-resident target ELF, changes target-process root/jail state, and injects the ELF into a SystemService-launched BigApp.

Raw elfldr versus hbldr

Property	Hardened raw elfldr	Official hbldr
input	host-streamed raw ELF	path to ELF already on device
process basis	`SceSpZeroConf` child	SystemService-launched BigApp
foreground user	not established by inspected route	queried and passed to launch context
app/title launch	none	v0.7 `PPSA01659`; v0.19 `FAKE00000`
running BigApp	left untouched by constructor	may be killed before launch
persistent writes	none in raw launch route	possible v0.19 fake-app creation
kernel/ptrace process mutation	yes	yes, plus BigApp substitution
display permission	unproven	unproven; architecture is materially different

The difference is source-proven. Its effect on VideoOut is not.

Side-effect matrix

Effect	v0.7	v0.19	Classification
read target ELF from device	yes	yes	requires staging; current hard blocker
terminate active BigApp	yes	yes	`HIGH_SESSION_DISRUPTION`
launch SystemService BigApp	VideoPlayer	fake game	`HIGH_FUNCTIONAL`
attach/follow SysCore child	event tracking	ptrace follow-fork/exec	`HIGH_RUNTIME_MUTATION`
kernel root/jail changes	yes	yes	outside current safety scope
executable mapping replacement	yes	yes	expected launcher mutation, not persistent file proof
remount `/system_ex`	no	possible	`BRICK_RELEVANT_HARD_BLOCKER`
create fake-app directories/files	no	possible	`PERSISTENT_SYSTEM_MUTATION_HARD_BLOCKER`
autoload change	not found	not found	source-negative only
bounded deadline	no	no	`SESSION_HANG_RISK`
automatic retry	not found	not found	source-negative only
rollback/power-loss protocol	absent	absent	v0.19 `POWER_LOSS_UNSAFE_OR_UNPROVEN`

Persistent-write audit for v0.19

fakeapp_create_if_missing() checks fixed paths, creates directories, writes a fixed param.json, and copies the existing PSNow eboot into the fake app. If the initial attempt fails, remount_system_ex() performs an MNT_UPDATE and the creation is attempted again.

The writes have no temporary sibling, atomic rename, fsync, byte-exact post-write verification, backup, rollback, or power-loss recovery. open() uses O_CREAT|O_WRONLY, without O_EXCL or an atomic replacement protocol. This is incompatible with Chimera GFX's current anti-brick gate regardless of whether a particular target already has the fake app.

Robustness findings

readfile() has no maximum input size and allocates the whole ELF;
process waits, child-await, and pthread_join have no hard deadline;
the port-2323 listener binds all IPv4 interfaces and the inspected source shows no authentication or transport protection;
splitstring() initially allocates bufsize * sizeof(char) but stores pointers, a source-visible memory-corruption risk;
which() uses unbounded strcpy/sprintf into PATH_MAX storage;
failure branches do not establish a complete restoration contract for every traced process and modified breakpoint;
target authenticity, size, and SHA-256 are not verified before launch.

These are SESSION_CRASH_OR_INTEGRITY_RISK unless a persistent write is involved. A session crash is not automatically a brick, but it does not satisfy the bounded one-shot Chimera test contract.

Decision

The official hbldr route cannot be reused, imitated, or treated as an approved launcher in the current phase. It requires device staging and contains operations explicitly outside the active authorization and safety boundary. No hbldr command, port-2323 connection, shsrv deployment, target build, transfer, installation, app termination, system remount, file creation, or execution is authorized.

5.1 KiB Raw Permalink Blame History