danos/system/drivers/usb-storage/usb-storage.zig

//! USB mass-storage class driver (Bulk-Only Transport + transparent SCSI).
//!
//! Spawned by the device manager when the xHCI bus driver reports a mass-storage
//! / SCSI / bulk-only interface (class 8, subclass 6, protocol 0x50); its device
//! id arrives as argv[1]. It owns no hardware: it opens its device through the
//! USB transfer protocol (`usb`), then drives it with the BOT command
//! cycle — CBW out, an optional data stage, CSW in — carrying SCSI commands
//! (READ CAPACITY, READ(10), WRITE(10)). Upward it is a block device: it serves
//! the block protocol under `.block`, the storage a FAT filesystem sits on.
//!
//! Block data never crosses IPC: read/write name a caller-owned DMA buffer by
//! physical address, which the data stage DMAs straight to/from.

const std = @import("std");
const ipc = @import("ipc");
const process = @import("process");
const service = @import("service");
const time = @import("time");
const device_manager = @import("driver");
const memory = @import("memory");
const logging = @import("logging");
const usb = @import("usb");
const scsi = @import("scsi.zig");
const bot = @import("bulk-only-transport.zig");
const envelope = @import("envelope");
const block_protocol = @import("block-protocol");

/// The generated block dispatch plus the subscriber machinery the harness owns
/// (subscribe/unsubscribe, the exit sweep, the fan-out) — usb-storage publishes
/// `medium_changed`, so it is a Subscribers provider, not a bare Provider. One
/// device per process, so the handler context is empty.
const Serve = service.Subscribers(block_protocol.Protocol, void);

const Invocation = envelope.Invocation;
const Answer = envelope.Answer;

var device_id: u64 = 0;
var device: usb.Device = undefined;
var bulk_in: usb.Endpoint = undefined;
var bulk_out: usb.Endpoint = undefined;

// DMA buffers for the transport: the 31-byte CBW, the 13-byte CSW, and a page
// for the small command data (INQUIRY / READ CAPACITY / the self-check sector).
var command_wrapper: memory.DmaRegion = undefined;
var status_wrapper: memory.DmaRegion = undefined;
var command_data: memory.DmaRegion = undefined;

var next_tag: u32 = 1;
var block_size: u32 = 512;
var block_count: u64 = 0;

// --- medium presence --------------------------------------------------------
//
// A slow TEST UNIT READY poll tracks whether the medium is present; on a
// transition the driver publishes `medium_changed` to its subscribers (the
// volume manager). This is the second removal trigger — the DEVICE stays while
// the MEDIUM leaves (a card reader, an ATAPI tray) — which channel death cannot
// see (docs/file-system-development/storage-architecture.md). Presence only,
// never content. A device that is genuinely unplugged is reaped by the device
// manager instead; a poll failure just before that death publishes absent
// harmlessly.
var service_endpoint: ipc.Handle = 0;
var medium_present: bool = true; // a successful bring-up means the medium is here
var medium_change_count: u32 = 0;
const presence_poll_ms = 1000;

/// One Bulk-Only-Transport command: send the CBW, run the data stage (to/from
/// `data_physical`), read and validate the CSW. Returns true on a passed status.
fn transact(cdb: []const u8, direction_in: bool, data_physical: u64, data_length: u32) bool {
    const tag = next_tag;
    next_tag +%= 1;

    const wrapper: *bot.CommandBlockWrapper = @ptrFromInt(command_wrapper.virtual);
    wrapper.* = .{
        .tag = tag,
        .data_transfer_length = data_length,
        .flags = if (direction_in) bot.flag_data_in else 0,
        .lun = 0,
        .cdb_length = @intCast(cdb.len),
    };
    @memcpy(wrapper.cdb[0..cdb.len], cdb);

    if (device.bulk(bulk_out.address, command_wrapper.physical, @sizeOf(bot.CommandBlockWrapper)) == null) return false;
    if (data_length > 0) {
        const endpoint = if (direction_in) bulk_in.address else bulk_out.address;
        if (device.bulk(endpoint, data_physical, data_length) == null) return false;
    }
    if (device.bulk(bulk_in.address, status_wrapper.physical, @sizeOf(bot.CommandStatusWrapper)) == null) return false;

    const status: *const bot.CommandStatusWrapper = @ptrFromInt(status_wrapper.virtual);
    if (status.signature != bot.csw_signature or status.tag != tag) return false;
    return status.status == @intFromEnum(bot.CommandStatus.passed);
}

/// Set when bring-up failed with the device PRESENT (an opened device that then
/// failed a step): main exits nonzero, and the device manager restarts us with
/// backoff — a transient failure heals instead of leaving storage down forever.
/// Device-absent paths stay clean exits: nothing to serve, nothing to retry.
var bring_up_failed = false;

fn initialise(endpoint: ipc.Handle) bool {
    service_endpoint = endpoint;
    // One hello, both directions: the block-serving endpoint goes UP (the
    // manager routes fat's consumer hello here — this driver serves one
    // volume, one process per stick, so `block` is never a registry name),
    // and the channel to THIS device's controller comes DOWN, routed by
    // lineage. A second stick used to die silently on the exclusive bind;
    // now every instance is reachable through its lineage.
    const bus = device_manager.helloForChannel(.device, device_id, endpoint) orelse return false;
    device = usb.open(bus, device_id) orelse {
        std.log.info("could not open device {d}", .{device_id});
        return false;
    };
    bulk_in = device.findEndpoint(usb.transfer_type_bulk, true) orelse {
        _ = logging.write("/system/drivers/usb-storage: no bulk-IN endpoint\n");
        bring_up_failed = true;
        return false;
    };
    bulk_out = device.findEndpoint(usb.transfer_type_bulk, false) orelse {
        _ = logging.write("/system/drivers/usb-storage: no bulk-OUT endpoint\n");
        bring_up_failed = true;
        return false;
    };
    // Shareable, so each buffer's capability can be handed to the controller: usb-storage
    // owns no device, so its buffers are not auto-bound anywhere — the controller reaches
    // them only once attached. (No-op binding when no IOMMU is enforcing.)
    command_wrapper = memory.dmaAlloc(4096, memory.dma_coherent | memory.dma_shareable) orelse return false;
    status_wrapper = memory.dmaAlloc(4096, memory.dma_coherent | memory.dma_shareable) orelse return false;
    command_data = memory.dmaAlloc(4096, memory.dma_coherent | memory.dma_shareable) orelse return false;
    for ([_]memory.DmaRegion{ command_wrapper, status_wrapper, command_data }) |region| {
        if (region.handle) |handle| {
            if (!device.attachDma(handle)) {
                _ = logging.write("/system/drivers/usb-storage: could not attach a DMA buffer to the controller\n");
                bring_up_failed = true;
                return false;
            }
            _ = ipc.close(handle); // the binding holds its own reference now
        }
    }

    // Bring the LUN up: wait for it to be ready (clearing the initial unit-attention
    // with REQUEST SENSE), identify it, and read its capacity.
    var tries: u32 = 0;
    while (tries < 10) : (tries += 1) {
        const ready = scsi.testUnitReady();
        if (transact(&ready, false, 0, 0)) break;
        const sense = scsi.requestSense(18);
        _ = transact(&sense, true, command_data.physical, 18);
        time.sleepMillis(50);
    }
    const inquiry = scsi.inquiry(36);
    _ = transact(&inquiry, true, command_data.physical, 36);

    const capacity_command = scsi.readCapacity10();
    if (!transact(&capacity_command, true, command_data.physical, 8)) {
        _ = logging.write("/system/drivers/usb-storage: READ CAPACITY failed\n");
        bring_up_failed = true;
        return false;
    }
    var capacity_bytes: [8]u8 = undefined;
    const capacity_source: [*]const u8 = @ptrFromInt(command_data.virtual);
    @memcpy(&capacity_bytes, capacity_source[0..8]);
    const capacity = scsi.parseCapacity(capacity_bytes);
    block_size = capacity.block_size;
    block_count = @as(u64, capacity.last_lba) + 1;
    std.log.info("ready ({d} blocks x {d} bytes)", .{ block_count, block_size });

    // Self-check: read block 0 and log its trailing signature (0x55AA for a boot
    // sector) — proof READ(10) works end to end over the bulk path.
    const read0 = scsi.read10(0, 1);
    if (block_size <= 4096 and transact(&read0, true, command_data.physical, block_size)) {
        const sector: [*]const u8 = @ptrFromInt(command_data.virtual);
        std.log.info("block 0 signature 0x{x:0>2}{x:0>2}", .{ sector[510], sector[511] });
    }
    // Bring-up succeeded, so the medium is present; start the presence poll.
    _ = time.timerOnce(endpoint, presence_poll_ms);
    return true;
}

// --- serving the block protocol ---------------------------------------------
//
// Geometry, and whole-block read/write to and from the caller's DMA buffer
// (named by physical address). One device per process, so `Header.target` is
// always 0 and no handler reads it.

/// A transfer the device refused. Every failure here is the same one — the SCSI
/// command did not complete — so there is one errno for all of them.
const refused: isize = -envelope.ENOENT;

// --- per-sender range confinement -------------------------------------------
//
// The volume manager confines each filesystem to the partition it mounts
// (define_range); a confined sender addresses volume-relative LBAs from 0 and
// the driver translates and bounds-checks against its range. A sender with no
// range is unconfined — the whole device — which is the default until a range
// is defined (behaviour-neutral for a single-volume boot), and is what the
// volume manager itself uses to probe partitions before it confines anyone.

/// bound: filesystem processes confined to sub-ranges of this device at once
/// decided-by: ours
/// protects: the per-badge range table below
/// at-limit: refuse - define_range past it returns -ENOSPC; a runaway detector for
///   a compromised volume manager, not a real-partition limit (real disks carry a
///   handful of volumes, far under this)
/// observed-by: the -ENOSPC a define_range caller gets when the table is full
const maximum_ranges = 64;

const Range = struct { used: bool = false, badge: u32 = 0, base: u64 = 0, count: u64 = 0 };
var ranges = [_]Range{.{}} ** maximum_ranges;

fn rangeFor(badge: u32) ?*Range {
    for (&ranges) |*r| {
        if (r.used and r.badge == badge) return r;
    }
    return null;
}

/// Resolve a caller's transfer to an absolute LBA, or null if it falls outside
/// the caller's confinement. Unconfined callers (no range) pass through against
/// the whole device.
fn resolveTransfer(sender: u32, lba: u64, count: u32) ?u64 {
    const r = rangeFor(sender) orelse return lba; // unconfined: whole device
    if (lba + count > r.count) return null; // past the volume's end
    return r.base + lba;
}

fn onGeometry(_: void, invocation: Invocation(void), answer: Answer(block_protocol.Geometry)) isize {
    // A confined caller sees ITS volume's size, not the device's — so a
    // filesystem mounts against the geometry it is actually allowed to touch.
    if (rangeFor(invocation.sender)) |r| {
        answer.set(.{ .block_size = block_size, .block_count = r.count });
        return 0;
    }
    answer.set(.{ .block_size = block_size, .block_count = block_count });
    return 0;
}

fn onRead(_: void, invocation: Invocation(block_protocol.Transfer), answer: Answer(block_protocol.Transferred)) isize {
    const request = invocation.request;
    const abs = resolveTransfer(invocation.sender, request.lba, request.count) orelse return refused;
    const cdb = scsi.read10(@intCast(abs), @intCast(request.count));
    if (!transact(&cdb, true, request.physical, request.count * block_size)) return refused;
    answer.set(.{ .count = request.count });
    return 0;
}

fn onWrite(_: void, invocation: Invocation(block_protocol.Transfer), answer: Answer(block_protocol.Transferred)) isize {
    const request = invocation.request;
    const abs = resolveTransfer(invocation.sender, request.lba, request.count) orelse return refused;
    const cdb = scsi.write10(@intCast(abs), @intCast(request.count));
    if (!transact(&cdb, false, request.physical, request.count * block_size)) return refused;
    answer.set(.{ .count = request.count });
    return 0;
}

/// Confine a sender to a block sub-range (the volume manager's per-volume grant).
/// Refused if the CALLER is itself confined — a filesystem cannot widen its own
/// range or confine anyone; only an unconfined party (the volume manager) may.
fn onDefineRange(_: void, invocation: Invocation(block_protocol.DefineRange), _: Answer(void)) isize {
    if (rangeFor(invocation.sender) != null) return -envelope.EPERM;
    const request = invocation.request;
    const slot = rangeFor(request.badge) orelse free: {
        for (&ranges) |*r| {
            if (!r.used) break :free r;
        }
        break :free null;
    } orelse return -envelope.ENOSPC;
    slot.* = .{ .used = true, .badge = request.badge, .base = request.base_lba, .count = request.block_count };
    return 0;
}

/// SYNCHRONIZE CACHE: commit the device's write cache to flash. No data stage.
/// Makes prior writes durable before a caller (init at shutdown) cuts power. A
/// device without a volatile cache reports success anyway.
fn onFlush(_: void, _: Invocation(void), _: Answer(void)) isize {
    const cdb = scsi.synchronizeCache10();
    return if (transact(&cdb, false, 0, 0)) 0 else refused;
}

/// The filesystem's DMA buffer: forward its capability to the controller so the
/// device can reach it. Never claimed — the binding holds its own reference, so
/// our copy is the turn's to close, on this path and on the refusal alike.
fn onAttach(_: void, invocation: Invocation(void), _: Answer(void)) isize {
    const handle = invocation.capability orelse return -envelope.EPROTO;
    return if (device.attachDma(handle)) 0 else refused;
}

/// The reverse: forward the same region capability so the controller unbinds
/// the buffer. As with attach, our copy stays the turn's to close.
fn onDetach(_: void, invocation: Invocation(void), _: Answer(void)) isize {
    const handle = invocation.capability orelse return -envelope.EPROTO;
    return if (device.detachDma(handle)) 0 else refused;
}

const handlers = Serve.Handlers{
    .geometry = onGeometry,
    .read = onRead,
    .write = onWrite,
    .flush = onFlush,
    .attach = onAttach,
    .detach = onDetach,
    .define_range = onDefineRange,
};

fn onMessage(message: []const u8, reply: []u8, sender: u32, arrived: *ipc.Arrival) usize {
    // The harness peeks the arrival and takes it only if a handler (subscribe)
    // claimed it; attach/detach forward the capability without claiming, so the
    // turn still closes their copy after the controller took its own reference.
    return Serve.dispatch({}, handlers, message, sender, arrived, reply);
}

/// A slow TEST UNIT READY poll: success means the medium is present, failure
/// means it is not. On a transition, bump the counter and publish. (Sense-key
/// inspection to tell "medium absent" from other transport errors is a
/// refinement; a clean eject — what QEMU and a card reader produce — makes
/// TEST UNIT READY report not-ready, which this reads correctly.)
fn pollPresence() void {
    const ready = scsi.testUnitReady();
    const now = transact(&ready, false, 0, 0);
    if (now == medium_present) return;
    medium_present = now;
    medium_change_count +%= 1;
    std.log.info("medium {s}", .{if (now) "present" else "absent"});
    Serve.publish(.medium_changed, 0, .{ .present = @intFromBool(now), .change_count = medium_change_count });
}

fn onNotification(badge: u64) void {
    const got = ipc.Received{ .len = 0, .badge = badge, .cap = null };
    if (got.isTimer()) {
        pollPresence();
        _ = time.timerOnce(service_endpoint, presence_poll_ms);
    }
    // Subscriber deaths are swept by the harness (Serve.hooks); nothing else here.
}

pub fn main(init: process.Init) void {
    const argument = init.arguments.get(1) orelse {
        _ = logging.write("/system/drivers/usb-storage: missing device id (argv[1])\n");
        return;
    };
    device_id = std.fmt.parseInt(u64, argument, 10) catch {
        std.log.info("malformed device id '{s}'", .{argument});
        return;
    };
    // No `.service` name: several instances provide the block contract (one
    // per stick), so consumers are routed here by the device manager's
    // lineage, never by a registry bind — the endpoint goes up in the hello.
    service.run(block_protocol.message_maximum, .{
        .init = initialise,
        .on_message = onMessage,
        .on_notification = onNotification,
        .subscribers = Serve.hooks,
    });
    // A failure exit (nonzero -> .aborted) tells the device manager to restart
    // us with backoff; a clean return means there was nothing to serve.
    if (bring_up_failed) process.exit(1);
}