The physical unplug/replug remount was never actually bench-run — the claim was
carried over from S3. QEMU cannot re-present a usb-storage device_add, so this
path is only reachable on real hardware. Correct both docs to say bench-pending,
keeping the honest distinction: the re-adopt+remount CODE PATH is QEMU-proven by
the driver-crash rebuild; only the physical-replug end-to-end awaits a bench pass.
Storage removal is now robust to all three ways a volume can leave, and the docs
say so. storage-architecture.md and storage-design-rationale.md move medium_changed
from "planned to be consumed" to consumed, and record the third trigger:
- the DEVICE leaving the tree (a pulled stick) — presence polling
- the MEDIUM leaving while its device stays (a reader) — the volume manager
now consumes the pushed medium_changed event
- the storage DRIVER crashing while its device stays — a channel-liveness
geometry() probe reaps the volume and rebuilds it on the restarted driver's
fresh channel; presence polling alone cannot see this (the V4 open edge)
The re-adopt-and-remount path is QEMU-proven by the driver-crash rebuild; a
physical unplug/replug is bench-verified (QEMU cannot re-present a usb-storage
device_add). The transport-native eject signal (SCSI UNIT ATTENTION, AHCI
PxSSTS, NVMe namespace-change AER) in place of the TEST UNIT READY poll stays
the documented future refinement.
The adversarial S5 review found a real, unrecoverable defect: onMediumEvent
deduped medium_changed events on a module-global last_medium_change compared by
equality against the event's change_count. But change_count is a PER-DRIVER
counter that restarts at 0 in every usb-storage instance — it is bumped +%=1 and
published only on a real medium transition, so within one instance every count
is unique and monotonic and an equality dedup can never legitimately fire.
The global was carried across a driver restart — S5's OWN crash-rebuild path —
so a fresh instance's first eject (count=1) collided with a stale last==1 and was
dropped. removeDevice never ran; the filesystem kept serving I/O against absent
media forever, and nothing else recovered it: onGeometry answers from a cached
block_count so channelAlive stays true, and isDevicePresent stays true (the
device never left the tree). medium_changed is the sole eject oracle there.
The dedup guarded a re-delivery the driver already makes impossible, and its
only observable effect was this bug. Remove it: react to each present-edge
directly. Both branches are idempotent (a freed device stops matching dev.used)
and the poll reconciles, so acting on every genuine edge is safe — and a fresh
driver instance's counter can no longer be mistaken for the previous one's.
Verified: build + bounds green; volume-removal, volume-medium-change and
volume-driver-restart all pass, so both removal paths survive the change. The
driver-restart-then-eject intersection that triggered the collision cannot be
staged in QEMU — the internal driver-kill cannot be ordered against a QMP eject,
and a usb-storage device_add is not re-presented — so the guarantee rests on the
driver's one-publish-per-transition-with-unique-count contract.
The V4 review's open edge: a storage driver that crashes while its device
stays in the tree left fat wedged on a dead channel — device-presence
polling (a device-manager enumerate) still reported the device present,
so nothing reaped it. pollTick now also probes channelAlive(dev), a
geometry() on the block channel that fails fast on the dead endpoint; a
present device with a dead channel is reaped like a pull, and the adopt
loop re-adopts it on the restarted driver's fresh channel — the rebuild.
The manager's own liveness probe makes fat self-detection unnecessary:
it rebuilds regardless of the wedged filesystem's state.
The drill: the device manager gains a test-storage-restart mode that
kills usb-storage once, ~2s after its hello (post-mount); a new
volume-driver-restart kernel case boots a manual tree with it, and the
QEMU case asserts a SECOND mount of the same id-path after the reap —
the rebuild. A pre-S5 manager, checking only device presence, never
reaps, so the second mount never appears. The manually-spawned volume
manager needed kernel-supervisor protocol grants (bind its name, open
the device manager), as the other manual-tree services already have.
Full suite 133/133.
The block client gains subscribeMedium / unsubscribeMedium /
decodeMediumChanged (the reserved subscribe/unsubscribe verbs plus the
MediumChanged decode), so a consumer never hand-rolls the wire format.
The volume manager subscribes to each device it adopts and consumes the
event through the new on_buffered_message seam — never the protocol
dispatch, whose op numbers collide with the manager's own hello.
A medium leaving while its device stays in the tree (a card reader, an
eject) now runs the SAME kill-retire-remount path as a pulled stick:
absent retires the volume, present re-probes it. That closes the "two
triggers, one lifecycle" the architecture specifies — device-presence
polling alone could never see a medium leave under a present device.
dropDevice unsubscribes before closing so the driver's bounded
subscriber table frees the slot; on a dead channel (a real pull) the
call fails fast, proven by volume-removal still passing.
New volume-medium-change case: eject the medium (not the device) ->
"medium absent" -> the manager unmounts. Fails against a pre-S5 manager
that never subscribed. Suite 132/132 (device-authority is the known
child-cleanup flake, green on rerun).
An additive, behavior-neutral callback. A buffered async message
(Received.isMessage — a pushed event from a provider this service
subscribed to) carries a payload in the receive buffer; run() now hands
it to on_buffered_message before falling through to on_notification with
the badge, so a coalesced timer/exit riding the same wake is not lost. A
service that does not set the callback (all of them today) is unchanged:
the isMessage branch is a no-op and on_notification still runs, exactly
as before.
This is the seam the volume manager needs to consume block
medium_changed: the event's reserved op number collides with the volume
manager's own hello, so it must be decoded by hand here, never through
the protocol dispatch. Full suite neutral (the lone device-authority
miss is a known child-cleanup race that passes on rerun).
Flip the storage docs to record exFAT as a built second engine reusing
the shared filesystem harness wholesale (the reuse the architecture
promised): full read + write, directories, rename, on-disk up-case
folding, routed by VBR to fat or exfat at an exfat-<serial> id-path.
Note the two surface limits shared by BOTH engines as vfs-layer concerns,
not exFAT shortcuts: u32 file offsets (a 4 GiB cap) and ASCII-only names.
A 7-dimension adversarial review of the engine, tool, and routing found
nine real defects (host tests + the in-VM drill missed them). Fixed:
- geometryOf now rejects a crafted VBR whose cluster shift exceeds the
exFAT ceiling (bytes+sectors shift > 25) or whose cluster_count exceeds
the spec max (0xFFFFFFF5) — either would overflow the engine's u32
cluster-byte / cluster-bounds arithmetic and panic under ReleaseSafe on
untrusted removable media. validCluster/allocateCluster widened to u64,
and writeFile's clusters_needed widened, for a >4 GiB file near the u32
offset boundary.
- writeFile no longer claims valid_data_length = size unconditionally: a
sparse write past a foreign file's old valid boundary now zero-fills the
skipped gap on disk, so a read there returns zero, not stale bytes.
- ensureDirCapacity rewrites a grown subdirectory's own DataLength, so a
spec-compliant reader that bounds a directory by DataLength sees the new
entries (danos itself bounds by the end marker, but chkdsk / other OSes
do not).
- make-exfat-image lays the allocation bitmap across as many clusters as
it needs; a >128 MiB image (whose bitmap exceeds one cluster) was
self-inconsistent. Verified: the engine mounts+reads both the 48 MiB
fixture and a 256 MiB image.
Documented (not fixed here — a shared vfs-layer limit, like the u32
offset cap): non-ASCII names fold to '?', the same as the FAT engine.
New host tests pin each fix (crafted-VBR rejection, sparse-gap zero,
subdir-grows-and-records-size). Full suite 131/131, bounds green.
exfat-volume attaches a bare exFAT data device (serial e0fa0001, from
make-exfat-image.py) beside the FAT boot volume. The volume manager
content-routes it to the exFAT service — not fat — which mounts it at
its id-path /volumes/exfat-e0fa0001; the exfat-test client then reads the
seeded HELLO.TXT and mutates through the mount (mkdir/write/rename/read/
remove). The second engine reusing the shared harness is now proven end
to end, on a real device, in QEMU.
The drill caught what host tests could not: the exfat service was denied
openEndpoint("volume-manager") — it had no protocol.csv grant — so its
hello never reached the manager and its volume never mounted (a silent
spin, no fault). Added the grant mirroring fat's. A new exfatVolumeTest
kernel case boots the tree and spawns the fixture; the run harness grows
an exfat data-volume flavor. Fails against pre-S4 (no exfat binary, csv
row, VBR recognizer, or grant).
Each engine now refuses the other's volume at mount(): the exFAT engine
rejects a FAT boot sector (a non-zero byte where exFAT keeps MustBeZero),
and the FAT engine rejects an exFAT one (MustBeZero reads as a zero
bytes-per-sector), each mounting its own as a control. The two engines
can never claim the same medium.
exfat-test is the mount round-trip client, cloned from fat-test: it waits
for /volumes/exfat-e0fa0001, reads the seeded HELLO.TXT, then exercises
mkdir + write + rename + read-back + remove through the VFS and reports
"exfat-test: ok". It ships as a lazy fixture in test builds (the
exfat-volume QEMU case, step 8, drives it). build + zig build test +
bounds green.
partition.zig gains an exFAT VBR recognizer: the "EXFAT " name (where a
FAT BPB keeps its OEM string, so the two never collide) plus the 0x55AA
signature, with VolumeSerialNumber (offset 100) as a new exfat_serial
identity rung. A `recognize` helper tries exFAT, then FAT, then the MBR
disk-signature fallback, and sets each volume's FilesystemKind — so
allVolumes/firstVolume and the GPT path all tag a volume with the engine
its content needs. volume-map renders exfat-<serial> as the id-path, and
filesystems.csv adds the exfat -> /system/services/exfat row: an exFAT
stick now spawns the exFAT service, at its own content id-path.
Host tests: a bare exFAT volume recognized with its serial; a FAT VBR
still recognized as fat (the discrimination); the exfat-<serial> id
render. build + zig build test + bounds green.
Pure stdlib, no mkfs.exfat: writes a Main Boot Sector + its boot-region
checksum + a backup region, the 32-bit FAT, an allocation bitmap, an
up-case table (with its checksum), and a root directory carrying the
bitmap/up-case/label entries plus a seeded HELLO.TXT file set. --serial
sets the volume serial (the content id-path); --verify re-reads the VBR,
recomputes the boot checksum, and validates each file set's checksum.
Cross-checked against the engine: the Zig exFAT engine mounts a
Python-produced image and reads HELLO.TXT through a case-insensitive
resolve ("exfat hello danos"), proving the tool and engine agree on the
layout — VBR, geometry, bitmap, up-case, and the entry-set checksum. No
image is committed; the drill (step 8) generates one per run.
exfat.zig is a near-clone of fat.zig — the reuse the architecture
promised, now real: same IpcBlock DMA-bounce wrapper, same
acquire-volume-by-hello to the volume manager, same content-conditional
boot rewrites (resolve /system/configuration to decide the system
volume), all differences confined to the engine it wraps. That the
harness's Server(engine.FileSystem) compiles is the proof the exFAT
engine meets the same pub-fn contract as FAT — a comptime check, not a
hope.
The exfat package (build.zig + build.zig.zon) mirrors fat's: it ships in
production_ship, its on-disk + engine host tests run through the root
test aggregate (replacing the temporary standalone entry), and the root
zon declares it. build + zig build test + bounds all green.
create / write / truncate / remove / mkdir / rename, completing the
engine's pub-fn contract with the shared harness. The allocation BITMAP
is the authority: setAllocated IS the allocation (a set bit), and the
32-bit FAT only records a fragmented chain's order — forgetting the bit
would hand a live cluster out twice, so the write path never touches the
FAT without also owning the bit.
This engine writes FAT-linked (no_fat_chain=0) files: createFile makes
an empty set; writeFile allocates+links+zeroes clusters (so a sparse gap
reads zero and valid_data_length can honestly equal data_length) and
rewrites the Stream entry; a contiguous file opened for growth is first
threaded through the FAT. truncate frees the chain; removeFile clears
each set entry's InUse bit and frees the chain (a non-empty directory is
refused); rename re-homes the same clusters under a new name set.
createDirectory allocates one zeroed cluster — exFAT directories carry no
"." / ".." entries. Every entry-set mutation recomputes the set
checksum. Offsets clamp to the vfs u32 surface.
Ten engine host tests now (read + write across clusters, truncate+reuse,
remove, subdir+inner file, rename); 17 total with on-disk. bounds green.
mount + resolve + list + read over a BlockDevice, host-tested against a
RAM-backed image the tests build with formatExfat. mount reads the VBR,
loads the geometry, and scans the root for the Allocation Bitmap (0x81)
and Up-case Table (0x82); the up-case prefix is decompressed (0xFFFF
identity runs) into a bounded table so names fold correctly.
A directory is read as consecutive 32-byte entries via readChain, so a
File/Stream/Name SET that straddles a sector or cluster boundary
assembles cleanly; each set's checksum is validated before it counts as
a file. A stream's no_fat_chain flag picks contiguous-arithmetic vs
FAT-follow cluster walking. reads honor valid_data_length (allocated-
but-unwritten tail reads as zero) and clamp data_length to the vfs u32
offset surface.
Tests cover mount + up-case fold, listing (skipping the metadata
entries), case-insensitive resolve, and reads across a cluster boundary
on both a contiguous and a FAT-fragmented file. Wired via engine.zig
(which imports on-disk.zig) into the host-test aggregate. 12/12, bounds
green. Write path is step 3.
The pure, host-testable byte layer of the second engine: the Main Boot
Sector (VBR) and the six 32-byte directory-entry types — Allocation
Bitmap, Up-case Table, Volume Label, File, Stream Extension, File Name —
as align(1) extern structs, plus the three exFAT checksums (boot region,
up-case table, directory-entry set), the name hash, and the packed
timestamp <-> Unix-epoch conversion.
geometryOf accepts only "EXFAT " + 0xAA55 + an all-zero MustBeZero
region; that last guard is the mutual exclusion with FAT — a FAT prober
reads a zero bytes-per-sector there and rejects the volume, and this one
rejects a FAT boot sector for want of the exFAT name. Wire-format widths
are named consts (spec facts, no bare literals) so the bounds gate stays
green; the layout is pinned by @offsetOf/@sizeOf tests.
Wired into the host-test aggregate directly for now; it moves into the
exfat package's own test step when that lands (step 4). 7/7 host tests,
bounds green.
Flip the storage docs from "one FAT volume today" to the built state:
the manager adopts every device, probes each device's whole partition
table, and spawns one range-confined FAT per volume — several volumes
across several devices, or several partitions on one device's channel.
The boot volume is identified by content; a filesystem installs the
/system rewrites only when it resolves /system/configuration on its own
media. No filesystem binds a shared service name any more — clients route
through the kernel mount table.
Correct two now-stale claims in the rationale, in the honest direction:
"multi-volume providers are reserved, not implemented" becomes built
(multi-namespace-per-provider is the untried NVMe case); and the cloned-
duplicate "second mounts suffixed" was never built — S3 mounts both,
they collide on the shared content id-path (last wins), each boot claim
logged, and distinguishing them is S4 arbitration.
The two-volumes case proves multiple DEVICES; this proves multiple
volumes on ONE device. make-partitioned-image.py lays several FAT32
partitions (each from make-fat-image) behind a classic MBR; the new
partitioned-volume case attaches one such disk (two partitions,
da7a0001 at lba 2048, da7a0002 at lba 83968) as a single usb-storage
device.
partition.allVolumes walks the table and the manager spawns a confined
fat per partition on the SAME block channel, each clamped to its own LBA
range by usb-storage's per-badge range table — so partition B's fat
cannot read partition A's blocks. The case asserts two mount lines at
two distinct non-zero base_lbas on one device; against the pre-uncap
allVolumes (S3 step 2, capped to one partition) only da7a0001 mounts.
No image is committed — the disk is generated per run. Full suite
130/130 (128 + two-volumes + partitioned-volume).
A second usb-storage device (a generated data volume, serial da7a0001,
an empty FAT with no /system) plugged in beside the boot volume: the
volume manager adopts both devices and spawns a confined fat per volume,
each mounted at its own content id-path.
The test surfaced a real coexistence bug. Every filesystem bound the
single "vfs" contract name under /protocol; the second volume's fat lost
the race, service.run refused-and-exited on the held name, and that
volume never mounted. Clients don't reach filesystems by that name —
fs_resolve routes a path to its backing endpoint through the kernel
mount table by prefix — and nothing consumes "vfs", so the fix is to
bind no shared name: the harness's service_name now defaults to null.
This is the "this fades" the harness comment anticipated for the
volume-manager era; a filesystem's endpoint still serves as its mount
backend without a name.
fat logs "is a data volume" for the non-system branch so the test can
positively assert content-based detection. make-fat-image gains
--serial/--label (default unchanged) so a second image gets a distinct
id-path; the data image is generated per run, never committed. The case
fails against the pre-fix harness (the data volume's fat exits on the
refused bind) — toggle-demonstrated.
Full suite 129/129 (128 + two-volumes); the single-volume path is
unaffected by dropping the vestigial name bind.
Multi-volume makes the mount table the bottleneck: each volume installs
one id-path mount and the system volume two FHS rewrites, so at the
volume manager's maximum_volumes (16) the old cap of 8 is far too low.
Raise maximum_mounts to 32 (headroom over the ~20-mount worst case) and
declare its bounds block; drop it from the bounds allowlist.
Fix a latent bug the higher pressure would expose: installMount silently
dropped a mount when the table was full, and mountBackend returned true
anyway — a full table was reported as a successful mount. installMount
now returns whether it placed the mount, and mountBackend propagates a
false so the mounting filesystem's harness logs "could not mount
<prefix>". At-limit is now a refusal that is observed, not a silent
success. (The full-table path has no host unit test: vfs.zig's tests are
not wired into the host aggregate — its import graph reaches the
freestanding kernel — so the correction rests on the propagated return
and the truthful bounds block.)
A volume backs /system/configuration and /system/logs only when it
actually carries the /system tree — decided by content (resolve
/system/configuration on its own media at mount), not by spawn order.
The boot volume takes the branch and installs the two rewrites; a data
volume resolves null, mounts only at its id-path, and never shadows the
running system's config or logs with a dead mount.
This retires the "resolve-at-bring-up quirk" the single-volume step
deferred around: that diagnosis was wrong. A boot probe confirmed
resolve() works the instant mount() returns — /system, /system/
configuration, /system/kernel, /system/services all resolve at bring-up
(mount reads LBA 0 through the same block path, so a directory read
cannot fail where the boot-sector read succeeded). No deferral needed.
Behavior-preserving on the single boot volume (it carries the system
tree, so it still installs all three mounts): suite stays 128/128.
Lift the one-device/one-volume cap. bringUpVolume now probes the whole
partition table (allVolumes uncapped) and spawns a confined filesystem
per volume; pollTick loops it to adopt every present, not-yet-adopted
device each tick.
The subtlety is adopt-once-and-keep: a device is recorded in the table
the first time it is seen and kept until it leaves the tree, even when it
carries no servable volume or its geometry cannot be read. Dropping an
unservable device would make openAnyStorage hand back the same one every
tick and starve the devices behind it; keeping it lets the scan advance
past it. A genuine removal frees the slot; a re-insert (fresh device id)
is probed anew.
The boot image is a single bare-FAT volume, so the full suite is
unchanged at 128/128.
Replace the single `var volume: ?Volume` and file-global supervision
state with two fixed tables: devices[maximum_devices] owning each adopted
block channel once, and volumes[maximum_volumes] each carrying its own
identity, id, mount prefix, and supervision fields (restarts, spawn_ns,
failed, restart_pending, restart_due_ns). A monotonic next_volume_id
never reuses ids, so a stale hello can't address the wrong child.
Lookups (deviceById, volumeById, volumeByPid, firstUsedVolume) and
claims (claimDevice, claimVolumeIndex) replace the ad-hoc singletons.
pollTick reconciles devices first (removeDevice drops their volumes),
then per-volume restarts, then idle bring-up.
This step stays one-device/one-volume on purpose: bringUpVolume adopts
the first device and caps allVolumes to a single partition, so behavior
is identical and the full suite stays 128/128. Uncapping and adopt-all
land next.
The multi-volume enabler. allVolumes(reader, device_blocks, out) appends every
volume on the device to the caller's buffer and returns the count: GPT
enumerates all valid entries (gptFirstVolume becomes gptAllVolumes), the MBR walk
collects all fitting partitions, and a bare FAT is the single whole-device volume
— each with the same per-entry overflow-safe range validation (the confinement
invariant the driver's clamp rests on) and fatIdentity-over-disk-signature
preference. firstVolume is now the one-element case of allVolumes, so the S1
behavior and its ten tests are unchanged. New host test: a two-partition MBR
yields two volumes with distinct identities (index 0 vs 1); it FAILS when
allVolumes is capped to one (the old firstVolume semantics), passes at 11/11.
Flip the storage docs to match S2: filesystems.csv (signature -> binary) and
volumes.csv (identity -> optional override) are built; a volume's mount path IS
its content id (/volumes/<id>), never a port name; the label is display metadata
a `volumes` query returns. fat receives its mount path via argv[2] rather than
hardcoding it. Still pending: rung 2 (filesystem UUID, needs a non-FAT engine),
multi-volume (fat's boot rewrites stay unconditional until S3), medium_changed
consumption, remount bench-verification.
The flip that makes the mount path the volume's content id. fat retires its
hardcoded fat_mounts: it reads its mount path from argv[2] (the volume manager
hands it the id-path, e.g. /volumes/fat-12345678, from the FAT serial), mounts
its volume root there, and installs the /system/configuration + /system/logs FHS
rewrites so /system/logs persistence stays decoupled from which volume backs it.
The rewrites are unconditional this increment (the single volume IS the boot
volume); S3 makes them content-conditional across N volumes. Every /volumes/usb
reference migrates to /volumes/fat-12345678 in one commit — the fat-test,
badge-scope-test, and vfs-test fixtures and the four QEMU regexes — plus a new
volume-identity-name case asserting the id-path mount and the /system/logs
rewrite. Discrimination: the regexes now require /volumes/fat-12345678, which the
old hardcoded fat never emitted (it mounted /volumes/usb). Full suite 128/128.
The mechanism the id/label split needs: a `volumes` verb whose reply packs the
mounted volume's {id, mount_path, label} into the tail (VolumeInfo.encode/decode
— three length-prefixed strings). Software keys on the id (the mount path is
/volumes/<id>); a shell or file manager shows the label — the database id/name
split made a query. The VM's onVolumes answers from the mounted volume, empty
reply if none. Two host round-trip tests (encode/decode; too-small buffer and
short-tail rejection). No runtime consumer yet — the first is a userspace shell;
the hello handshake is unaffected (fat-mount/volume-probe green).
The volume manager reads its policy from configuration at boot (loadTables,
mirroring the device-manager registry load): filesystems.csv (content signature
-> service binary) and volumes.csv (optional id -> mount-prefix override), each
held in a static source buffer with declared bounds. On probe it picks the
binary from the volume's signature (unserved + logged if no row matches, like an
unbound device) and composes the mount path — a volumes.csv override, else the
default /volumes/<id> from volume-map.idString — then spawns that binary with
argv {volume-id, mount-prefix}. Behavior-preserving: fat still ignores argv[2..]
and uses its hardcoded mounts, the binary resolves to /system/services/fat, so
the FULL suite stays green (127/127); the flip to argv-driven mounts and the
/volumes/usb -> id-path migration land in step 5.
NEW volume-map.zig: idString(identity) renders a volume's content identity into
its stable mount id-string — gpt-<32hex>, fat-<8hex>, mbr-<sig>-<index> — the
token whose default mount path is /volumes/<id>, so the path IS the id and never
a port or a label; two volumes that share a label get distinct ids
automatically. parse() reads volumes.csv (id, mount_prefix) into OPTIONAL
overrides; overrideFor returns a pinned prefix or null (the volume takes its
default /volumes/<id>). id_maximum is a declared bound; the fixture sizes are
named. Three host tests (each rung's id token; override hit/miss; malformed rows
counted), wired into the VM package test step with csv. Not yet consumed by the
binary — that lands when the VM loads the tables and composes paths (step 4).
partition.Volume gains a FilesystemKind signature (today .fat for every probed
volume; S4 adds a real VBR recognizer for exFAT) — the seam filesystems.csv keys
on to choose a service binary. New filesystem-map.zig parses
`/system/configuration/filesystems.csv` (signature, binary) into rules and
match()es a signature to its binary, mirroring the device registry: a signature
no row matches goes unserved, never guessed; slices point into the source
buffer. Three host tests (fat->binary, the binary is data-driven not hardcoded,
malformed rows counted); the test rule buffer is a named fixture size so the
bounds gate stays quiet. The VM's build gains the csv dependency and wires the
filesystem-map test into its package test step. Not yet consumed by the binary —
that lands when the VM loads the tables (step 4).
The S1 adversarial boundary review found the off = (i*entry_size) % 512
arithmetic tested only for 128-byte entries. Add a test with 256-byte entries
and the sole valid entry at index 1 (offset 256), exercising the non-zero-offset
path. No code change — the parser was already correct (off is always a multiple
of entry_size >= 128, so off + 128 <= 512); this closes the coverage gap.
Add volume-manager to build.zig's root package-test loop, so `zig build test`
runs the partition parser's nine host tests and a fixture added there can never
be silently skipped. Flip the identity-ladder build status in
storage-design-rationale.md: rungs 1 (GPT partition GUID) and 3 (FAT serial +
label) are built, joining rung 4; rung 2 (filesystem UUID) waits on a non-FAT
engine; the volumes.csv map and the id-derived mount path land with S2.
A pre-existing stale assertion, surfaced by wiring volume-manager into the root
test aggregate (its partition fixtures now run under `zig build test`).
entries_per_reply is computed as (packet_maximum 256 - prefix 16) / sizeof
ChildEntry 32 = 7; the test asserted 10, the value the old count-header layout
carried. No behavior change — the enumerate producer and consumers already page
by the real capacity; only the test documented an obsolete number.
Rung 3, stronger than the MBR disk signature. fatIdentity reads the VBR at the
partition start — 0x55AA plus a 0x28/0x29 extended boot signature; FAT32 iff
fat_size_16 == 0; BS_VolID and BS_VolLab at the FAT12/16 vs FAT32 EBR offsets,
cross-checked against fat/on-disk.zig. firstVolume now prefers it over
mbrIdentity in both the MBR-entry path and the bare-FAT fallback, keeping the
rung-4 id when the VBR is not an extended FAT. The serial becomes the identity
key (the id); the label becomes the display name. Two host tests — a bare FAT32
reports its serial + label; an MBR FAT partition prefers the serial while a
non-FAT partition keeps rung 4 — both FAIL with the preference neutralized (2/9)
and pass with it (9/9). On-image witness: the volume-probe QEMU regex tightens to
the boot image's real serial 0x12345678, which before rung 3 was the ~0x0
pseudo-signature read from VBR offset 440.
Rung 1 of the identity ladder. A protective MBR (a type-0xEE entry) routes
probing to the GPT, authoritatively: gptFirstVolume verifies the LBA-1 header's
'EFI PART' signature and a header CRC-32 (inline reflected poly 0xEDB88320,
shared with the fixtures so parser and tests never drift onto a magic constant),
then walks the entry array — bounded by the declared gpt_entry_scan_maximum —
for the first entry with a non-zero type GUID and an overflow-safe in-device
range. That range check is the confinement-safety guard the driver's clamp
rests on, the invariant firstVolume already enforces for MBR, extended to
untrusted GPT metadata. The unique partition GUID becomes the identity key (the
id / mount-path handle); the 36-char partition name becomes the display label.
Three host tests (GUID-as-id; entry-past-device skipped and an all-out-of-range
table is null; a broken header/CRC is not a volume) — all three FAIL with the
GPT branch neutralized (3/7) and pass with it (7/7). Entry-array CRC deferred
(correctness-only; the range check carries the safety property).
The identity ladder's flag-day — no behavior change. partition.firstVolume stops
taking one preloaded block-0 slice and takes a SectorReader (a read-one-sector
fn), so it can reach GPT metadata at LBA 1 and each partition's VBR on demand
(the next commits). The u64 identity becomes Identity{rung,key,label}: key is the
id (the mount path derives from it), label is display metadata (empty at rung 4).
Identity equality is id-only (rung+key) — the label never enters it. Only rung-4
(MBR sig+index / bare-FAT index 0) is produced, byte-identical to before; the
four host tests port to a RAM-disk reader, and fat-mount/volume-probe/
volume-removal stay green.
Refine the naming decision: a volume's mount path IS its identity id (GPT GUID,
else fat-<serial>/mbr-<sig>-<index>) — a stable, unique, content-derived handle
software uses. The label (FAT volume label / GPT partition name) is mutable
display metadata, NOT in the path, exposed by a volume-manager `volumes` verb
returning {id, mount_path, label} — the database id/name split. This dissolves
the label-collision problem: same-label-different-id volumes get distinct paths
automatically; only identical ids (dd-clones) hit first-wins-and-log. S1 carries
both key (id) and label (display); S2 derives the id-path and adds the query.
Two decisions settled: (1) a volume's mount name is its own content identity
(FAT label / GPT name, else identity-hex), never a port name (/volumes/usb) or
role name (/volumes/boot); volumes.csv stays as an explicit override. This makes
S1 precede S2 and folds the /volumes/usb fixture+regex migration into S2. (2)
exFAT is a complete implementation — full read+write, directories, rename, and
the on-disk up-case table — not a read-first/ASCII-only subset; the only limit
is the vfs u32 offset surface (a 4 GiB cap on all filesystems), flagged as a
separate vfs change.
The phased plan taking the volume-manager track from one FAT volume to any
filesystem, N volumes, content identity, remount, and driver-crash survival:
S1 identity ladder (GPT GUID + FAT serial), S2 volumes.csv/filesystems.csv mount
map, S3 multi-volume, S4 exFAT (the second engine proving the harness reuse),
S5 removal robustness (medium_changed consumption + driver-crash rebuild +
QEMU-verified remount). Code-grounded: real functions, commit-granular steps,
discrimination tests shown to fail against today's behavior, per-phase risks.
Design decisions taken with recommended defaults; two forks flagged for review
(the /volumes/usb transitional naming and the exFAT write scope). Not started.
An all-tracks docs-vs-code audit (the same method that caught the storage
drift) found 23 confirmed inaccuracies where a doc's build-status claim no
longer matches the source — status markers that were never flipped after a
track landed, and a few paths left over from completed flag-days. All verified
against the code before editing; docs only, no behavior change.
The systemic ones:
- IOMMU enforcement (driver-model.md, drivers.md): docs said enforcement was
not built and "device_claim = ring 0" / "memory-safe is not true yet". It is
built (per-device VT-d/AMD-Vi domains programmed at device_claim, -ECONFINE
rollback, dma_alloc buffers bound and torn down at death; fail-open only with
no IOMMU). Restated; M16 marker flipped to done.
- The FHS flag-day paths: /etc/devices.csv -> /system/configuration/devices.csv
(devices-csv.md, new-driver-checklist.md, device-manager.md), /var/log ->
/system/logs (logging.md, new-driver-checklist.md), /mnt/usb -> /volumes/usb
(process-management.md). Following the old paths silently breaks driver match.
- protocol-namespace P4 "remaining" -> landed (only P5 remains); shared-fate
fan-out "not yet enforced" -> enforced; wall_clock "not built" -> built;
SMP affinity + fault-recovery "left" -> built; process_enumerate raw-pointer
trust model -> checked copyToUser/EFAULT; bounds.md maximum_devices static
hole -> dynamic per-registrar quota; init spawns fat -> volume-manager;
config "hardcoded, move to /etc" -> already CSV data files; vdso.md three-
value call; zig-self-hosting library/ layout; python argv "new" -> built.
Found and fixed by a multi-agent audit across 12 doc clusters, each finding
adversarially verified against the source.
A V5 close-out audit (docs against the actual code) found the storage docs
overclaiming in both directions: the status header was flipped to "built" but
several body markers were not, and two passages describe mechanisms the code
never implemented. Nine confirmed, adversarially verified against the source:
Underclaims (marked Planned, actually built):
- storage-architecture: driver named sub-ranges / range confinement (V2, tested
by the block-range case); the pushed medium_changed event (published today
from a TEST UNIT READY poll); ownership-gated fs_unmount (V0, process.zig
gates it with EPERM); the filesystem-harness extraction (V1). Scoped the
remaining *planned* to the genuinely-pending parts (native-signal translation,
the volume manager consuming medium_changed).
Overclaims (described, never built):
- storage-architecture: the "FAT dirty flag on disk" guarantee — no on-disk
dirty/clean-shutdown bit exists; only an in-memory device-dirty bool gating a
device write-cache flush on close. Fixed in all three places.
- storage-architecture: fat "acquires its own volume (first mass-storage child
by enumeration order)" — the V3b flip removed self-acquisition; fat is handed
its volume id and channel by the volume manager.
- rationale + plan: the FAT engine's base_lba "deleted rather than moved" — it
and the engine's MBR walk still exist as now-inert legacy; the authoritative
walk lives in partition.zig.
- rationale: NVMe namespaces "decision 4 settles as endpoint-per-volume" —
decision 4 settles the opposite (per-sender confinement, one endpoint);
endpoint-per-volume is named only as an unbuilt future refactor.
- rationale: the five-rung identity ladder and volumes.csv map stated in flat
present tense — only rung 4 (MBR signature + index) is built; added the
build-status hedge and marked each rung.
Docs only; no code or behavior change. Suite unaffected (127/127).
The V4 adversarial review found removeVolume's comment overclaiming: it said
"the kernel sweeps a dead backend's mounts", which reads as an eager death-time
sweep. There is no such sweep. Killing the filesystem marks its backend endpoint
dead (killOwnedEndpointsLocked), and the VFS router retires each mount that
endpoint backed lazily, on the next path resolution under it (resolvePath sees
the dead backend, frees the slot, returns not_found). The functional guarantee
the comment promised — killing the filesystem retires its mounts — holds; only
the described mechanism was wrong. Comment-only; no behavior change.
Inline V4 review (the boundary-review workflow stalled): the poll ran a due
fat-restart before the presence check and returned, so a fat death followed
by a device removal would respawn fat against the now-dead channel and churn
until the crash cap before the removal was noticed. Reorder: check the
specific device's presence first (unmount if gone), and only fire a due
restart once the device is confirmed present. Neutral: fat-mount,
volume-removal, amd-iommu-usb-storage green.
Noted V4 limitations (not fixed here, edge cases outside the user unplug
case): a usb-storage DRIVER crash (device stays, driver restarts with a new
endpoint) leaves fat holding a dead channel — the device is still present so
removal is not detected; fat would need to observe its channel death and
exit. Deferred with the medium_changed subscription and multi-volume.
The volume manager, per-sender range confinement + gate, medium_changed
event, per-volume spawning + supervision, ownership-gated fs_unmount, and
the removal half of the lifecycle are built. Left honestly pending: the
volumes.csv/filesystems.csv maps, the fuller identity ladder, multi-volume,
the VM consuming medium_changed (removal uses device-presence polling), and
remount-on-replug end-to-end (bench-pending — QEMU can't re-present the
boot-controller device). Full suite 127/127.
The full suite caught a V4 regression: under AMD-Vi the device-manager tree
carries more than one mass-storage-identity entry (a phantom no driver is
bound to, which answers a consumer hello with NO channel). V4 split presence
from acquisition and picked the FIRST identity match blindly, so it kept
helloing the phantom (device 27) and never reached the real storage (device
31). V3's inline loop had skipped no-channel entries with `orelse continue`;
the split lost that.
Restore it: openAnyStorage tries each matching entry and takes the first whose
channel opens, recording its device id. Removal detection then watches THAT
specific device id leave the tree (isDevicePresent), not "any mass-storage" —
so a phantom that never leaves cannot mask a real removal. Both are bare
enumerates; the hello only happens while bringing a volume up.
Green: amd-iommu-usb-storage, fat-mount, volume-removal.
The volume manager stops probing-once and polls storage presence for the life
of the boot: findStorageDevice enumerates the device-manager tree (presence
only, no consumer-hello, so it is cheap and leaks nothing). The volume is now
a field that goes null and back — the whole lifecycle:
- storage present + no volume -> open the channel, probe, confine + spawn the
filesystem (openStorage is the one consumer-hello, on the insertion edge);
- storage gone + have volume -> kill the filesystem (its mounts retire via
the kernel dead-backend sweep), close the dead channel, clear the volume;
- fat crash -> the same supervised backoff/cap as before,
folded into the poll (one timer).
This also subsumes the V3-review leak fix (no per-poll consumer-hello) and the
no-volume retry (a present-but-unreadable device keeps polling).
The user's case — pull the boot stick, plug it back — is a DEVICE unplug (the
stick IS the device), so the mass-storage child leaves the device-manager tree
and the poll catches it. volume-removal asserts the unmount and discriminates:
against the V3 probe-once volume manager the removal is never noticed (0/1).
The re-mount on replug is the VM's bringUpVolume firing when the device
returns — correct and in place, but not QEMU-testable here: device_add of
usb-storage to the boot xHCI controller is not re-presented to the guest (no
port-connect on any port), a harness quirk, not a VM issue. On real hardware
the bus's per-tick port poll catches a reconnect (H1 proves reconnect on a
second controller); bench-verify the full round trip.
The mount-failure diagnostic wrote through a fixed [96]u8 bufPrint buffer,
which the bounds gate flags as an undeclared ceiling. Three ring appends
instead — no buffer, no fixed length to justify. This fix was made during
the V2a bounds work but never git-added, so the committed harness still
carried the flagged buffer; committing it now cleans the tip.
Five confirmed defects from the boundary review:
1. (security) The VM never checked a partition fit inside the device, so a
crafted MBR could hand the driver a range whose base+lba wraps past a u32
— panicking usb-storage in a loop, and at multi-volume overlapping a
neighbour. This is the exact invariant the clamp's overflow-safety rests
on. partition.firstVolume now skips any entry that runs past the device
(host-tested), establishing the invariant where the untrusted bytes are
first read.
2. (leak) The probe re-acquired a fresh block channel on every 500 ms retry,
leaking a handle each time on a medium-absent device. The channel is now
acquired once and kept.
3. (wedge) A failed spawn or defineRange stranded the volume with no retry;
both now arm a backoff restart.
4. (loop) fat respawn had no exit-reason gate, no backoff, no crash-loop cap
— a faulting filesystem respawned in a zero-delay loop, and a clean exit
was resurrected. Supervision now mirrors the device manager: a clean exit
is not restarted, a fault backs off, three fast deaths give up.
5. (removable) A device that parsed to no volume was terminal; it now keeps
polling so an inserted medium is picked up — the removal-lifecycle trigger.
Known limitation (noted, not fixed here): if the VM itself crashes and init
restarts it, the orphaned fat keeps serving vfs while the new VM spawns a
second fat whose bind is refused — the same "manager restart re-learns the
world" gap the device manager also defers. The old fat keeps storage working.
Neutral: partition unit tests + fat-mount, volume-probe, block-range, logger
all green.
The load-bearing step. The FAT service stops acquiring its own volume: the
volume manager spawns it (per volume), defines its partition range on the
storage driver BEFORE it runs, and answers its startup hello with the
range-confined block channel over a new volume-manager protocol. fat never
finds its storage by name and never sees the whole device — establishment
by lineage, one layer up from the driver tree.
- New library/protocol/volume-manager: one verb, hello(volume-id) -> the
block channel as the reply capability (the P0 reply-cap path).
- The volume manager becomes the confinement CONTROLLER: it defines the first
range on usb-storage, so no other party can confine a filesystem. It
supervises the filesystems it spawns and respawns one on death (the reap-
and-rebuild the device manager proved, one layer up).
- fat: drops acquireVolume(device-manager); hellos the volume manager for its
channel; reads its volume id from argv[1]. main takes process.Init now.
- init.csv no longer spawns fat (the volume manager does); protocol.csv
rewires fat to be supervised by the volume manager (bind vfs, open
volume-manager) and drops fat open device-manager.
- The block-range fixture boots registry + device-manager only (not the full
tree), so the volume manager is absent and the fixture stays the sole
confinement definer — otherwise the volume manager would take the
controller first and refuse it.
Verified end to end (VM probes -> spawns fat -> confines it -> hands over the
channel -> fat mounts) and neutral: 18/18 across the fat family, logging,
shutdown, both IOMMU variants, usb restart, vfs, conformance, confinement.
The storage layer gains its policy home (storage-architecture.md): a new
system/services/volume-manager, spawned by init, that acquires the mass-
storage block channel through the device manager (the same lineage a
filesystem uses), reads block 0, and parses the first volume out of it. The
partition-table walk that lived in the FAT engine moves here, above the
driver where it belongs (partition.zig, host-tested: MBR entry, bare-FAT,
no-signature). Identity is the MBR disk signature + partition index — the
weak rung of the ladder; GPT GUID and FAT serial refine identityOf without
changing shape.
This increment is discovery + probe + log only, additive: the FAT service
still acquires its own volume, so nothing changes for it. Confining each
filesystem to its partition and spawning one per volume (the flip) lands
next, keeping fat working throughout.
Grants + wiring: init.csv spawns it after the device manager; protocol.csv
grants bind volume-manager + open device-manager. Verified: volume-probe
asserts the parse (bare-FAT volume at lba 0), neutral 10/10 across storage,
restart, display, logging, confinement — the volume manager now runs in
every boot and disturbs nothing.
Two defects the V2 boundary review confirmed:
1. The per-badge range table was never reclaimed. usb-storage receives exit
notifications (via the subscriber watch), but onNotification handled only
the timer and dropped child-exits, so a dead filesystem left its range
slot used forever. The medium-removal lifecycle churns filesystems, so
after maximum_ranges confine/die cycles define_range would return ENOSPC
and no volume could be confined again until reboot. onNotification now
frees the dead badge's slot, mirroring fat's open-node sweep.
2. define_range checked only that the CALLER was unconfined, never that it
owned the target badge — so any unconfined opener could install a range
for another live client and silently redirect its I/O. Confinement now has
a single controller: the first unconfined party to define a range (the
volume manager, which confines every filesystem before handing it a
channel). Only the controller may thereafter; the slot releases on its
death so a restarted manager re-takes it. This is the mechanism half; V3
adds the grant half (only the volume manager gets an unconfined channel).
Neutral: block-range (the fixture is the sole definer -> controller),
fat-mount, usb-report all green. End-to-end exercise of both lands in V3/V4
(the VM+filesystem relationship and the remount churn).
The naive bound `lba + count > r.count` wraps for an lba near u64 max: the
sum overflows to a small value, sails under the check, and `base + lba`
wraps to an absolute block OUTSIDE the range. Calibrated, it is a real
confinement escape — a process confined to [1,3) reads absolute block 0
(the boot sector) with lba = maxInt(u64), since base + lba wraps to 0.
The bound is rewritten as two subtractions that cannot overflow: lba within
the range, and count within what remains. block-range gains a wrap-refused
assertion calibrated to be exploitable against the naive form — it FAILS
against the old bound (reads block 0) and passes against the fix (verified
by reverting the clamp). Caught pre-emptively before the V2 boundary review.
The block protocol gains a pushed medium_changed event (present + a monotonic
change counter; presence only, never content). usb-storage becomes a
Subscribers provider and runs a slow TEST UNIT READY poll (1 s): success is
present, failure absent, and a transition bumps the counter and publishes.
This is the second removal trigger — the DEVICE stays while the MEDIUM leaves
(card readers, ATAPI trays) — which channel death cannot see
(storage-architecture.md, two triggers one lifecycle).
The subscriber is the volume manager (V3); until it exists the publish is a
no-op fan-out, so this commit is behaviour-neutral, and its end-to-end test
(eject -> medium_changed -> unmount/remount) lands in V4 with the real
consumer rather than a throwaway subscriber fixture (recorded sequencing).
Sense-key inspection to tell medium-absent from other transport errors is a
noted refinement; a clean eject reads correctly as not-ready.
Neutral: 12/12 across the block-serving surface, restart, confinement,
conformance, and logging.
A process acquires a block channel the way a filesystem does (consumer-hello
the device manager), confines ITSELF to blocks [1,3), then proves the clamp
and the gate: volume-relative LBA 0 maps inside the range and reads; a read
reaching past the range is refused; geometry reports the confined size; and
a confined caller can no longer call define_range (no widening, no escape).
It gates on argv so the ramdisk sweep leaves it silent in other boots, and
coexists with fat (ranges are per-badge).
Discrimination (verified by reverting usb-storage to pre-clamp f1bdce2~1):
the unconfined read still succeeds but define_range returns ENOSYS, so the
fixture cannot arm confinement and the case fails — exactly the property
the clamp adds. With the clamp: block-range 1/1.
The block protocol gains define_range (appended, numbers hold): confine the
process named by `badge` to blocks [base, base+count). usb-storage keeps a
per-badge range table and, in read/write, translates volume-relative LBAs
(base added) and refuses any transfer past the volume end. geometry returns
the confined size, so a filesystem mounts against what it may actually touch.
The security seam (decision 4, settled): the clamp lives at the PROVIDER, so
a channel carries exactly the authority it grants — handing a filesystem the
whole disk plus a base offset would let it reach the neighbouring partition.
The gate: a confined caller may NOT call define_range, so a filesystem cannot
widen its own range or confine anyone; only an unconfined party (the volume
manager, whole-device) may. The volume manager defines a filesystem's range
before handing it the channel, so the ordering holds by construction.
Default (no range for a badge) is the whole device — behaviour-neutral for a
single-volume boot and what the volume manager itself uses to probe
partitions. The range table is declared through bounds.md as a runaway
detector (ours, refuse at limit), not a real-partition cap. Neutral:
fat-mount, usb-storage, iommu-usb-storage green. The discrimination fixture
(a confined process reads past its range and is refused) follows next.
V1 boundary review (4 reviewers converged): the extraction routed mount-
FAILURE logs through std.log where old fat used logging.write. That matters
precisely when the failed mount IS /system/logs — a std.log record would
then have nowhere to land. Restore the direct kernel-ring write for the
failure branch (generic prefix, since the harness is filesystem-agnostic
now); success stays std.log.info as before. Unexercised error path; fat-mount
still green. Everything substantive in the extraction verified neutral.
fat was one binary doing four jobs; the three that are not FAT-specific move
to library/kernel/file-system-harness, a Server(comptime Engine) generic over
the engine type: the badge-scoped open-node table, the nine vfs handlers, the
not-mounted politeness, the exit sweep, mount registration, and durable-on-
close. A filesystem is now an engine plus a main that hands the harness a
mounted volume; a second engine reuses the harness wholesale.
Placement note: the plan said library/file-system, but the harness is a
specialization of `service` (its sibling) and needs nothing from the device
domain, so it lives beside service in library/kernel and stays block-free —
durability rides a caller closure (Volume.flush), no backwards kernel->device
dependency, no new-domain scaffolding. The engine type is inferred from
resolve()'s return, so engine.zig is untouched (its Node stays module-scope).
fat keeps only its FAT-specific bring-up (acquireVolume, DMA, engine.mount,
the attach/detach round trip) and the three mount prefixes as data. Behavior-
neutral: 13/13 across the fat/vfs/logger/IOMMU surface, nothing observable
changed. This lands first so every later phase touches the harness once.
Per-sender range confinement at the provider, one serving endpoint: the
badge-scoped provider pattern the xHCI bus already uses (the per-client
device-token table), applied to blocks. Endpoint-per-volume would buy the
same enforced property only by inventing a multi-endpoint harness; it stays
available as a future refactor, same wire contract. The plan's flag-for-veto
is gone: nothing in the track waits on a choice.
fs_unmount was gated by nothing but the /protocol carve-out: any process
could unmount any prefix — latent with one mount owner, an obvious
cross-tenant hole once volumes multiply. Each backend mount now records the
mounting task, and the syscall layer enforces two rules that keep the
restart story intact: only the owner unmounts (a dead owner's mount is
swept lazily by resolution — strangers gain nothing by racing that), and a
mount may be REPLACED only by its live owner or after its owner died (the
respawned-filesystem path; displacement of a live mount would be worse than
unmounting it). Kernel-installed mounts are never displaceable.
The vfs-test park role is the discrimination: with the volume provably
mounted it attempts the foreign unmount, requires the refusal AND the
subtree still resolving, and withholds its "parked" marker otherwise —
against the ungated kernel the unmount was ALLOWED and vfs-client-death
fails; with the gate, green. (Its verification handle closes immediately:
the kernel test string-matches "released 1 handle(s)".)
The sharpest instance of the return story, folded into the architecture: the
boot volume is a recorded content identity (the volume carrying
/system/configuration and /system/logs), the system runs on without it (the
ramdisk is the OS; only log persistence pauses), the kernel ring is the
buffer during absence — bounded, so a wrapped ring is a data loss window
that gets MARKED in the file on resume, never spliced silently — and on
return at any port the same identity remounts the same prefixes and the
logger appends into the same boot-stamp tree. The logger requirement is
named: failed flushes retry on the patient cadence, never abandoned after
the first not_found. The dirty-honesty rule stands; the boot volume gets no
exemption.
Decision 8 in the rationale, mirrored into the architecture's volume-manager
section: the mount map keys on CONTENT identity, never port or arrival order
— Linux's /dev/sda1-era fstab broke on every port move until UUID= replaced
it, and danos skips that era. The identity ladder the prober reads off the
medium: GPT partition GUID, filesystem UUID, FAT serial+label, MBR
signature+index, anonymous. Consequences are mechanical: port moves change
nothing (USB, hub, SATA bay, or transport swaps), replug remounts at the
same path, the boot volume is a recorded identity findable anywhere, and
cloned duplicates are a loud policy case instead of silent shadowing.
/volumes/<name> stands as the hierarchy's home for attached media; minting
identifiers (formatting, entropy) stays deliberately out of scope.
storage-stack-discussion.md was misplaced at the docs root — that level is
for track plans; this is the file-system domain's design record. Moved to
file-system-development/storage-design-rationale.md, renamed to say what it
is, cross-references updated.
The NVMe/SATA assessment folded into the storage discussion: what transfers
untouched (the block contract and everything above it — one driver binary
plus one devices.csv row per transport), NVMe as the shorter stack whose
namespaces are the reserved multi-volume case, AHCI as an open shape choice
(leaning per-port processes, the matrix-proven granularity), and the three
pressure points named honestly: multi-volume is reserved-not-implemented,
synchronous call/reply bottlenecks NVMe until the shm-ring data plane, and
media lifecycle is not device lifecycle.
That last one becomes decision 7 and enters the architecture doc: the
removal path has TWO TRIGGERS, ONE LIFECYCLE — channel death (device leaves)
and a planned pushed medium_changed event (medium leaves, device stays: card
readers and trays, USB ones today), translated by the storage driver from
its transport's native signal, presence never content, consumed by the
volume manager into the same kill-retire-remount path. Without it a swapped
card would be served with the previous card's filesystem state.
The kernel was always symmetric (dma_bind 51 / dma_unbind 52); the two
protocols that forward an attachment up the stack were one-way, so a live
client could grant a device reach into its buffer but never revoke it while
alive — exactly the one-way lifecycle the storage architecture's enforcement
section forbids. Death stays the mechanical backstop; detach is the living
process's path.
Both verbs are appended, so every existing number holds. The shape mirrors
attach precisely: the same region capability rides the cap slot again — the
kernel matches the region, so no layer retains anything between the calls
(the bus never kept the handle; now it never needs to).
fat's bring-up does attach -> detach -> attach, exercising both verbs
through the whole chain (fat -> storage -> bus -> kernel) on every boot: a
broken detach fails every fat case instead of lying dormant until the first
buffer replacement. Honest scope: the round trip proves the plumbing; unbind
semantics are the kernel iommu tests' (map/unmap/translationOf); the full
composition (detach then DMA faults) is a future iommu-fault extension.
The enforcement section of the storage architecture: the three levers the
device lifecycle already proved, mapped onto volumes — a filesystem can only
be GIVEN its volume (no establishment grants, channel at spawn), the volume
manager supervises with teeth (deadline, kill-on-removal, crash-loop cap),
and the shared harness makes every engine inherit the state machine by
construction (engines never see channels). Kernel backstops: ownership-gated
fs_unmount + the lazy dead-endpoint sweep. Checkable via a lifecycle
conformance drill parameterized over filesystems — supporting a filesystem
MEANS passing it.
The settled shape from the storage-stack discussion, written as the
reference: the data path (vfs -> filesystem service -> block -> driver) as
the application/service/protocol/driver model applied twice; the three kinds
of boundary (protocol between processes, library inside them, control-plane
beside them); the volume manager as the policy home (planned — the FAT
service squats on its duties today, marked as such); adding a filesystem as
engine + shared harness + one configuration row; and the per-layer
responsibility table for removable media — one removal path, kill/retire/
respawn, dirty data lost and SAID to be lost. Indexed from docs/README.md.
The hot-plug matrix's last two cells. H4: unplug from hub port 1.1, replug
on 1.2 — per-port identity means the old child is removed and reaped while
the new port binds a fresh driver; nothing ties a driver to the old port.
H5: three unplug/replug cycles on one port — three reaps, a bind after the
last, proving slots, the bus open table, and the manager's driver entries
are all reusable across generations.
hub -> hub -> keyboard, one device_del of the outer hub: the H2 recursion
runs at depth two (the inner hub is torn down as a child, its keyboard
first), the keyboard's driver is reaped, and re-adding all three rebinds.
The hot-plug matrix's predicted bug, found on first contact: tearDownPort
never recursed into a departing hub's children — only tearDownHubDevice
(a hub leaving one level down) did. Yank a populated hub from a root port
and the downstream slots stayed live against vanished hardware, their class
drivers were never reaped, and the replugged hub found its port still
occupied, so nothing ever re-enumerated: the subtree was gone for the boot.
tearDownPort now recurses children-first, exactly like tearDownHubDevice.
The usb-hub-yank case is the discrimination: one device_del removes a hub
carrying a keyboard AND a mouse, both drivers must be reaped, and the
re-added hub must rebind both — it failed against the unfixed bus and
passes with the recursion.
The tearDownPort path, exercised for the first time with a real removal and
return: QEMU raises no root-port change events, so the bus's 250 ms
reconcile tick notices the PORTSC change alone — and it does. Reap, re-add,
rebind, second generation binds on the same port. Discrimination: against
the pre-reap manager (4a2df58~1) the case fails at the dedupe wall.
The hot-unplug path (onChildRemoved, a report from a live bus) cleared the
child but left the bound class driver: a process blocked on reports that
will never come, whose stale entry made the matcher's dedupe refuse the
respawn when the device was plugged back in — the same wall the restart
zombie hit, one path over. Unplug now reaps exactly like reporter death.
The usb-hub-unplug case grows the replug: device_del the hub keyboard, then
device_add it back (qmp_sequence); the ordered tail — child removed, reaping,
delegated, ok — can only be satisfied by the second generation, since every
boot keyboard's ok precedes the unplug. Discrimination: without the reap the
replug never rebinds and the case times out (verified by stash run). Hub
family, restart drill, and the two-controller proof all green (8/8).
P4 of docs/establishment-planes-plan.md. The new usb-two-controllers case is
the Ryzen mouse bug pinned in the suite: a second xHCI controller with its
own keyboard while the boot controller keeps the default one — both must
come up, on different device ids, which requires each class driver to reach
ITS OWN controller. Discrimination: at 72807c2 (name-based establishment)
the case fails — one keyboard is unreachable, exactly the bench failure —
verified against a checkout of that merge; with lineage routing it passes.
The existing second-controller cases could not prove this: the boot bus
always carries a keyboard, so their expects were satisfiable by it.
Docs updated with the code: device-manager.md (hello moves the channels,
paged enumerate, delegation as built, reap-and-rebuild in the restart
sequence), device-authority.md (the fourth as-built decision: driver-layer
channels ride the hello; hello is no longer only the liveness handshake).
Full suite: 118/119, the one failure being iommu-fault's fixed 200 ms
fault window under end-of-run host load — 3/3 green standalone, deflake
flagged separately. usb-two-controllers passed inside the full run.
P3 of docs/establishment-planes-plan.md — the restart-zombie fix. A class
driver cannot observe its provider's death: an HID driver blocks on
interrupt reports that will simply never come, and storage answers its
callers with refusals forever. Worse, the dead generation's still-used
entries made the matcher's dedupe refuse the respawn when the restarted bus
re-reported — the subtree was a permanent zombie, which is the exact
opposite of the restart-a-driver-live goal the driver model exists for.
pruneChildrenOf now reaps: each pruned child's bound driver is killed and
its entry cleared (the exit notification finds no entry, so the death is
never double-counted; its device returns by the loan rule; its stored
endpoint handle is closed). The re-report then spawns a fresh generation
whose hellos fetch the successor's channel.
The usb-report drill now asserts the subtree WORKS after the restart: the
respawned storage opens its device on the NEW bus instance and reads block
0. Discrimination: against the pre-reap manager the drill fails — no reap
line, no post-restart respawn (the survivors were zombies), verified by a
stash run. Note the scenario boots no input service, so the HID drivers of
BOTH generations exit after their input lookup times out — storage is the
functional proof.
P2 of docs/establishment-planes-plan.md. usb-storage serves nameless — one
process per stick cannot share an exclusive bind, and a second stick used to
die silently on -EBUSY before ever helloing. Its one hello now moves both
directions at once: the block-serving endpoint up, its controller channel
down. fat finds its volume through the manager — a new `.consumer` role asks
for the channel of the driver BOUND TO a device (distinct from the device's
reporter), found by enumerating the tree for the mass-storage identity.
fat stays single-volume; the boot-volume-by-content choice is M21.
The conversion immediately caught a live truncation of exactly the audit's
shape: ChildEntry grew to 32 bytes, one enumerate reply holds ~7, and a real
tree carries a dozen ACPI nodes before the first USB child — the storage
entry silently never fit (the protocol comment already said "paging joins
the protocol if a tree ever outgrows one packet"). enumerate is now paged:
Header.target is the start cursor, a short page is the end; device-list's
page-0 read is unchanged.
Grant rows move with the code: the block bind and fat's block open die, fat
gains open device-manager. Gate: 18 cases green including the registry
trio, device-list, and both IOMMU storage variants.
P1 of docs/establishment-planes-plan.md. The bus no longer binds
/protocol/usb-transfer — the bind race made whichever instance came second
unreachable, which on a real three-controller Ryzen meant a mouse no class
driver could reach ("could not open device 50"). Each instance hands its
serving endpoint up in the hello that already delegates its controller, and
class drivers receive their OWN controller's channel from helloForChannel —
routed by the manager's lineage, retried while a provider is mid-restart,
on one manager handle so retries never spend handle-table slots.
usb.open(bus, id) now takes the channel it used to look up; the name rows
leave protocol.csv with the code (bind 67, opens 119/120/122); the
conformance fixture's prose stops claiming the bus binds; and the stale
input-client import leaves the bus with the channel one.
Gate: 19 QEMU cases green (usb family, hubs, both IOMMU variants, fat chain,
boot-from-USB, orderly shutdown, conformance).
P0 of docs/establishment-planes-plan.md; no behavior changes yet, nothing
sets the new flag or sends a hello capability.
- service.run gains a reply-capability out-slot (replyWithCapability), the
registry idiom init already uses, lifted into the harness; null stays the
untouched common path. Subscribers gains claimArrival() so a provider
handler can keep a turn capability through the same flag the reserved
subscribe uses.
- Hello wire struct: the padding byte becomes wants_channel — old callers
wire-compatibly say 0, and the manager nominates a reply capability ONLY
when asked, because a capability sent to a caller that never reads one is
a leaked slot in that caller's table.
- driver.helloExchange: one handshake can hand a serving endpoint up and
receive the device's provider channel down (usb-storage will need both at
once). No channel in the reply is retryable, never a verdict.
- The manager stores each instance's serving endpoint on its Driver entry,
routes consumer hellos by lineage (child -> reporter -> endpoint), replaces
on re-hello, and closes the stale handle on death - the 32-slot table is
the bound that makes forgetting this boot-fatal.
The namespace holds protocols, never instances; what multiplies is provider
processes, and their establishment routes through the device manager, which
owns the topology. communication.md gets the model; the plan converts the
usb-transfer and block seams in one flag-day, closes the restart-zombie hole
the scoping found, and pins the three-controller mouse bug as a QEMU case.
Two command encodings checked against the specification:
- COMPLETION_WAIT set bit 1 (I, interrupt) instead of bit 0 (S, store), so the
IOMMU was never asked to write the sentinel, the poll always exhausted its
spins, and completeAndWait returned without any guarantee the preceding
invalidation had executed — no invalidation barrier has ever existed, on QEMU
or on silicon. The in-code claim that "QEMU's amd-iommu does not implement
the store form" was a misdiagnosis of this bug: with S set, QEMU stores the
sentinel fine, and the amd-iommu cases now run without the warn line. On real
hardware, which fetches commands asynchronously, the missing barrier was an
IOTLB use-after-free window: unmap returned before the invalidation was
confirmed and the caller freed the frames.
- INVALIDATE_IOMMU_PAGES "invalidate everything" used address bits 51:12
all-ones; the architected encoding is bits 62:12 all-ones (the spec's literal
0x7FFF_FFFF_FFFF_F000). Real silicon is free to misread the non-architected
form as a bounded range.
Three holes from the real-AMD audit, one shared root: the delegation flag-day
added paths that touch the broker table and the IOMMU records without the big
kernel lock, and a loan-return rule whose re-delegation skipped confinement.
- device_enumerate walked the table with no lock. The table stopped being a
static array in the bounds track — reserve() regrows it through realloc on
every boot — so an unlocked reader can be mid-copy out of a slice that
device_register on another core has already freed and reused, or pair a fresh
count with a stale slice. Each chunk is now snapshotted under the lock; the
copy to the user stays outside it.
- system_spawn's give ran entirely unlocked — the comment claiming "the lock
has not been dropped" was false (spawnProcessSupervised takes and releases it
internally). The ownership pre-check now only spares creating a doomed child;
the give itself re-checks, confines and moves in one lock hold, and a give
that fails after the spawn kills the child rather than leaving it running
without the hardware it was spawned for.
- A re-delegated device after a driver death was never re-confined. Death tears
the domain down before the loan returns to the lender, so the next give found
no active record, reassign no-op'd, and the respawned driver ran the device
with a V=0 device-table entry and no domain — silently unconfined, the exact
fail-open the fail-closed claim was built to remove. Both give paths now share
one body (giveDeviceLocked): check first, confine afresh when no record is
active — refusing with ECONFINE like the claim — and move last, when nothing
can fail.
The iommu test drives the death-and-respawn sequence directly; with the old
reassign-only behaviour its two confinement checks fail, with this change the
suite is 118/118.
An AMD Ryzen booted to a working compositor with no USB and no storage. The
cause was maximum_children_per_parent = 16 — a number nobody had justified,
in a kernel where 235 compile-time ceilings turned out to exist, 139 of them
on quantities the machine or a file decides rather than us, and 171 silent
when reached.
Both invented ceilings are deleted rather than raised. The device table
grows, because no specification bounds how many devices a machine has, and a
runaway is bounded per-registrar so it costs only the driver that caused it.
Device authority became real rather than advisory. Matching was always
policy, but claiming was first-come, so any process could take any device
nobody happened to be holding. Now every driver receives its hardware in the
spawn that creates it — it never runs without it — a grant is a loan that
returns to its lender when the borrower dies, and nothing firmware-discovered
is left unheld except the framebuffer, which the compositor owns.
Along the way: the AMD boot fix itself (STAR's SYSRET base carries the RPL,
which Intel forgives and AMD does not); one errno space with every refusal
naming the rule that refused it; the IOMMU confinement moving with its
device, and refusing rather than silently succeeding at the edge of its
table; and PCI apertures derived without copying the firmware memory map
into a fixed array, which had let a large map turn occupied RAM into a
window a driver could map.
A gate now stands behind the audit: zig build bounds refuses a new ceiling
that will not say what it counts, who decides its size, what happens when it
is reached, and how anyone finds out. The 269 that predate the rule are
allowlisted and that list can only shrink.
Suite 114 -> 118, green at every step.
Docs: fixed-bounds-audit.md, os-development/bounds.md, bounds-track-plan.md,
os-development/device-authority.md.
The name list and its predicate existed so drivers could move to delegation
one at a time with the suite green throughout. Every driver is delegated
now, so the manager simply hands over whatever device a driver was assigned.
Deleting it caught a real consequence: crash-test finally got delegated too,
and it was still claiming its device — so it got AlreadyClaimed because it
already held it, exited, and the restart drill had nothing to restart. Its
own comment named what the case was really checking: "the respawn only
reaches this line because the kernel released the previous instance's claim
at death". That property still holds, by a different mechanism — the device
reverts to the manager on death and is handed to the replacement, which is
the same guarantee without the race it used to rely on.
All four delegation paths verified: the xHCI controller, the PCI bridge, the
PS/2 two-node singleton, and virtio-gpu's restart re-attach.
Run 3 complete. Suite 118/118.
A device nobody holds can be claimed by anyone, so the manager now takes
every firmware-discovered device that carries mappable resources, whether or
not a driver wants it. The real gap was the HPET: an MMIO window, an IRQ, no
user-space driver, and there for the taking. Held by the manager it is
inert; unheld it was a way into physical memory.
Two deliberate exclusions. The loader's framebuffer, which the compositor
claims and which the manager must not take because it starts first. And
anything with no resources, which grants nothing worth holding.
Scope is the boot snapshot. A device reported later and matched to no driver
stays claimable — pci-cap-test and iommu-fault-test both reach an unmatched
NIC that way, so narrowing it is a separate change with those fixtures in
scope. Recorded in the plan rather than left implied.
The attacker fixture gains the assertion deferred since D2: after the system
settles, nothing with resources may be taken.
That assertion defeated itself twice before it worked, and both failures are
worth remembering. First it swept at 0.029 while the manager did not bind
its protocol until 0.047, so it reported a hole that closed a millisecond
later. The retry loop that "fixed" that was worse: the first pass TAKES the
device, so the second finds it unavailable because this process now holds
it, and concludes all is well — it passed with the manager's claiming
removed entirely. It now settles once and sweeps once, and fails when the
claiming is removed.
Suite 118/118.
The rule as planned: a device that was given to someone may be handed on,
never taken. Implemented, and honest about what it is worth.
Writing the test showed the plan had the wrong step doing the work. A
delegated device is HELD, so an attempt to take it is refused as
AlreadyClaimed before the giver is ever consulted; and once a borrower's
death returns the device to its lender — or clears both when the lender is
gone — there is no state where a device is unheld and still on loan. The
window a stranger could have used stops existing at E2. This check is
unreachable.
It stays anyway: one comparison, failing closed, guarding any future path
that frees a device without clearing its giver, which is exactly the hole
this run closed. The comment says it is unreachable rather than implying a
protection it does not provide.
The attacker fixture does not gain the assertion that was deferred to this
step, and its header records why: there is no refusal for it to observe, and
on a bare boot with no device manager nothing is delegated at all, so the
assertion had nothing to bite on. It failed loudly on its first run rather
than passing quietly, which is the only reason this was noticed.
It also leaves the loader's framebuffer alone without naming it: nobody
delegates the framebuffer, so it has no giver, so the display service claims
it exactly as before.
Suite 118/118.
When a driver dies, a device it was *given* now goes back to whoever lent
it, rather than to nobody. The device manager gets its hardware back the
instant a driver dies and hands it to the replacement, with no window in
between.
That window was real: the kernel released the claim to no one and the
manager re-claimed first-come, so every driver restart reopened the hole
this run is closing. It also becomes load-bearing at the next step — once
claim refuses a device that has a giver, releasing to nobody would strand a
dead driver's hardware permanently, because nobody could ever take it again.
A dead lender is no lender: the claim and the giver clear together, so a
device is never owed to a ghost. A device nobody lent is released outright,
exactly as before.
The broker cannot see the task table, so liveness arrives through the same
hook idiom the scheduler already uses. Null means assume dead, so a kernel
built without the hook frees claims rather than handing them to a ghost.
A stale binary nearly passed as proof for the third time this session: the
first discrimination patch left `alive` unused, the build failed with three
errors, and the old binary reported every assertion passing. Checking the
build before reading results is what caught it.
Suite 118/118.
One field, and the rest of the run follows from it. A device that was given
to someone is delegated hardware: it may be handed on, never taken, and when
its holder dies it goes back to whoever lent it instead of becoming free for
anyone to grab.
It also settles the framebuffer without mentioning it. Nobody delegates the
loader's framebuffer, so it has no giver, so the display service claims it
exactly as it always has — no exemption and no reference to display anywhere
in the rule.
No behaviour changes here; the field is recorded and read by nothing yet.
The test found a real bug on its first run, before the discrimination check.
The sentinel for "nobody gave this" was 0 — and task 0 is a real task, the
kernel's own, so a device given away by task 0 read back as belonging to
nobody. Both giver and registrar are optionals now. The second was a latent
bug from D9: the per-registrar allowance would have miscounted every device
task 0 registered.
Suite 118/118.
One field settles both outstanding questions. The kernel records who gave
each device. device_claim then refuses anything that has a giver, and on
death a device reverts to its giver rather than to nobody.
The framebuffer needs no exemption: nobody delegates it, so it has no giver,
so the display service claims it exactly as today. The rule never mentions
display.
And restart stops racing. Today a dying driver releases its claim to nobody
and the manager re-claims first-come, so every restart reopens the hole this
run closes; a loan that reverts to its lender removes the window entirely.
The zero-resource inventory question is dropped from the plan rather than
carried as a blocker. It is real but nothing depends on it.
The last claimant. The kernel seeds the acpi-tables node, so it sits in the
same boot snapshot the manager already scans to find the PCI host bridge —
there was never a bootstrap problem, only a lookup nobody had written. The
manager claims it and names it in the spawn; the service stops claiming.
Every driver in the system now receives its hardware rather than taking it.
Two failures on the way, both mine. addDriver puts the device id in argv[1],
and the acpi service read argv[1] as a self-verify device-count floor — so
handed device 7 it decided it was in test mode, printed "acpi-parse: ok",
and never reported a device. The test argument is now floor:N, which a bare
id cannot be mistaken for.
And acpi-parse spawns the service directly rather than through the manager,
so nothing handed it the node. That test now claims and transfers it exactly
as the manager does, which is the right shape: the test plays the manager's
role instead of the service reaching for hardware.
device_claim now has two callers left: the manager, which is the acquirer
and should have it, and the display service's GOP path. That is recorded as
question 10 — the framebuffer is not a device, so the answer is likely that
it leaves the device table rather than being exempted from its rules.
Suite 118/118.
The 8042 is a single controller described by two ACPI nodes — PNP0303
carries the 0x60/0x64 ports, PNP0F13 is the mouse — so it cannot be split
across two processes without them fighting over the same registers. That is
why ps2-bus is a singleton, and why it used to find and claim both nodes
itself.
The manager now gives it every matching node instead. The keyboard node
rides the spawn, because it holds the ports and is needed immediately; the
mouse node is transferred to the already-running instance. Late arrival is
safe here and the ordering is natural rather than lucky: the mouse is not
touched until after the controller handshakes and identify. Measured, the
handover lands at 0.336 and the driver reaches the mouse at 0.456.
Because the count is however many matched, a machine with no PS/2 ports or
only one works without a special case — which matters, since the bus is
mostly emulated now and machines vary.
ps2-bus claims nothing. irq_bind on the mouse node is the proof it holds it:
that call is ownership-gated, so a failure means the handover did not land
rather than a hardware fault, and the log says so.
acpi-ps2 asserts both delegations with the spawned device backreferenced, so
the node that rides the spawn must be the one the driver was spawned for.
Disabling the second delegation fails it.
Two things worth recording. The first discrimination patch was not valid Zig,
so nothing ran and a stale binary reported a pass — checked the build before
believing it. And with the second delegation disabled, acpi-ps2 fails while
input still passes: the mouse works without its IRQ binding, so exactly one
case covers that path.
Suite 118/118.
The second invented ceiling. It was written to stop a driver looping
device_register and exhausting a shared table — but there is no shared table
to exhaust any more, and each registrar already has its own allowance, so a
runaway costs only itself.
It never bounded a determined caller in the first place: 16 children per
parent, and nothing stopped it claiming more parents. What it reliably did
was refuse a real PCI bus with more than 16 functions, which is how an AMD
Ryzen booted with a working display, no USB and no storage.
The constant, its check, and the now-unused childCount all go. TooManyChildren
survives with one meaning instead of two: the caller is at its per-registrar
allowance.
This was unblocked from the moment D9 landed. The plan said so — "once the
quota exists the per-parent cap is redundant whether or not D6 has landed" —
in the same edit that left the step tagged "blocked on D6". Three iterations
were then spent re-reading that tag instead of the sentence beside it.
The containment test now asserts 64 children under one parent, four times
the old ceiling; restoring the cap fails it.
Suite 118/118.
reassign was added at D4 to fix a regression and has been proven only
indirectly since — three IOMMU+USB cases going green. That covered the
visible symptom (a driver's DMA rings unbound) and neither of the latent
ones: the confinement still naming the giver, so the giver's death would
tear down a domain a live driver was using, and the receiver's death would
leave one behind. Those are now asserted.
confinementOwner exposes the record's owner so the suite can see it. The
sequence is the delegation in miniature: unconfined, confine as this task,
reassign to another, confirm the new holder owns it and the old one does
not, then kill the new holder and confirm the domain goes with it.
Two attempts at this test could not have failed. The first found no PCI
function to confine — pciAddressOf needs a pci_device entry and this case
runs no pci-bus — so every assertion skipped silently while the case stayed
green. It now synthesizes a function the way pci-bus does, a 4 KiB config
window inside the bridge's ECAM, and asserts that precondition explicitly so
a skip is a failure.
Verified to discriminate: making reassign a no-op flips three assertions,
including the domain surviving its holder's death.
Suite 118/118.
Daniel Samson