Implement USB mass storage and batch OHCI bulk transfers

doc/src/driver/storage/ctl/driver.storage.ctl.usb.md

@@ -1,49 +1,68 @@
-# driver.storage.ctl.usb
+# USB mass storage
 
-USB Mass Storage Class driver stub.
+`driver.storage.ctl.usb` implements SCSI transparent commands over USB Mass
+Storage Bulk-Only Transport (interface class/subclass/protocol `08/06/50`).
 
-## Overview
+It probes INQUIRY, TEST UNIT READY and READ CAPACITY(10), registers LUN 0 with
+the storage manager, and exposes READ(10), WRITE(10), REQUEST SENSE and
+SYNCHRONIZE CACHE(10). Existing partition discovery and filesystem mounting
+apply to the resulting USB disk. Writes complete only after cache synchronization.
 
-This unit registers a USB class driver with the driver manager (`driver.mgr`) to claim USB Mass Storage Class interfaces (class `$08`, subclass `$06` SCSI, protocol `$50` Bulk-Only). When the USB stack enumerates a matching interface, the `load` callback is invoked. Currently the callback only logs a detection message; no I/O infrastructure is established.
+## Execution and ownership
 
-Full Bulk-Only Transport (BOT) support — including Max LUN query, endpoint setup, CBW/CSW command exchange, and `driver.storage.mgr` device registration — is planned but not yet implemented.
+An attachment task probes and registers each disk after scheduling starts; boot
+initialization never waits on a worker that cannot run yet.
 
-## Boot Registration
+Storage dispatch only publishes a request. One persistent `usb.storage` worker
+performs I/O in task context, with one command active per device and a reusable,
+page-aligned 4 KiB bounce buffer. Caller buffers are never submitted for DMA.
+Commands are split at the bounce-buffer boundary; all LBA/count and buffer-end
+arithmetic is checked before submission. CBW/CSW wire sizes, tags, status,
+residue, and actual transfer lengths are validated.
 
-Registered with `boot.mgr` as `driver.storage.ctl.usb` at the `device` barrier. The barrier order places `device` after `bus`, ensuring `driver.bus.usb.core.init` has run before the storage class driver registers with the driver manager.
+OHCI batches bulk data into spans of at most 4 KiB per transfer descriptor,
+using the endpoint's hardware toggle carry across packets. A 4 KiB full-speed
+read uses one data descriptor instead of 64. Short packets still halt the chain
+and report the actual byte count; interrupt endpoint packet sizing is unchanged.
 
-## Dependencies
+The worker performs reset recovery (class reset, clear bulk-IN halt, clear
+bulk-OUT halt) after malformed status or phase errors. Failed SCSI commands
+request sense and return storage errors, including write protection.
 
-- `boot.mgr`
+On disconnect the disk is marked removed immediately. Accepted requests are
+completed with errors, and idle volumes are invalidated/unmounted by the worker;
+busy volumes are retried. Storage-manager records remain stable tombstones.
+Class-driver buffers and context are freed after pending hardware work drains.
 
-- `driver.mgr`
+Host controllers currently have no cancellation API. A timed-out transfer is
-- `memory.heap`
+therefore quarantined with its private device snapshot and DMA memory. No further
-- `driver.storage.types`
+commands are issued to that device. If hardware never completes, those allocations
-- `io.syslog`
+remain retained until reboot; freeing them would permit DMA into reused memory.
-- `debug.tracer`
+The timeout uses the PIT's actual frequency and a bounded polling fallback.
-- `driver.bus.usb.core`
-- `driver.bus.usb.types`
-- `core.util`, `arch.x86.util`
 
-## Constants
+## Supported scope
 
-| Constant | Value | Description |
+- LUN 0 on a single-interface device with one bulk-IN and one bulk-OUT endpoint.
-|---|---|---|
+- Direct-access disks with power-of-two sectors from 512 through 4096 bytes.
-| `USB_CLASS_MASS_STORAGE` | `$08` | USB interface class for mass storage |
+- Capacity representable by the existing 32-bit storage-sector API; the
-| `USB_SC_SCSI` | `$06` | SCSI transparent command set subclass |
+  READ CAPACITY(16) sentinel is rejected.
-| `USB_PROTO_BBB` | `$50` | Bulk-Only (BOT) transport protocol |
+- OHCI, EHCI and xHCI bulk transfers. OHCI/EHCI report actual data lengths and
+  terminate bulk reads on short packets.
 
-## Functions and Procedures
+UAS, additional LUNs, composite/alternate-setting interfaces, UHCI storage,
+optical devices, and media replacement without USB reconnect are not supported.
+The current xHCI driver lacks endpoint-ring reset support: a halted bulk endpoint
+fails the device closed rather than attempting an unsafe ring reuse.
 
-### init
+## Validation
 
-```pascal
+`UnitTest` is called by the boot test suite and is available as `USBMSCTEST`.
-procedure init;
+Tests use a memory-only mock host controller: wire layout, byte order, capacity
-```
+limits, reads, writes, sense errors, short transfers, status stalls, reset ordering,
+and request bounds. They perform no disk I/O.
 
-Constructs a `TDeviceIdentifier` matching any USB device with `bInterfaceClass = $08`, `bInterfaceSubClass = $06`, `bInterfaceProtocol = $50`, and registers the `load` callback with `driver.mgr`.
+For hardware/VM checks, boot with a disposable USB disk, confirm `USBMSC` discovery
+and filesystem mounting, test file read/write/read-back, then unplug while idle
+and during I/O. Do not run destructive disk tests against a disk containing data.
 
-## Notes
+Protocol: [USB-IF Bulk-Only Transport 1.0](https://www.usb.org/sites/default/files/usbmassbulk_10.pdf).
-
-- The `load` callback currently returns true unconditionally after logging a detection message.
-- Once full BOT support is implemented, `load` should query the Max LUN, allocate endpoint structures, and call `driver.storage.mgr.register_device` for each logical unit.

src/driver/bus/usb/driver.bus.usb.core.pas

@@ -658,6 +658,7 @@ begin
 
     { Step 2: Create device at address 0 }
     dev := PUSBDevice(kalloc(sizeof(TUSBDevice)));
+    if dev = nil then begin pop_trace; exit; end;
     dev^.Address     := 0;
     dev^.Speed       := speed;
     dev^.MaxPacket0  := 8;  { Default, will be updated from descriptor }
@@ -667,6 +668,8 @@ begin
     dev^.ParentPort  := parentPort;
     dev^.NumEndpoints := 0;
     dev^.Configured  := false;
+    dev^.InterfaceNumber := 0;
+    dev^.InterfaceCount := 0;
     dev^.fnDisconnect := nil;
 
     { Step 3: Get first 8 bytes of device descriptor to learn MaxPacketSize0 }
@@ -730,7 +733,14 @@ begin
     end;
 
     totalLen := cfgDesc.wTotalLength;
+    if totalLen < sizeof(TUSBConfigDescriptor) then begin
+        kfree(void(dev));
+        pop_trace;
+        exit;
+    end;
+    dev^.InterfaceCount := cfgDesc.bNumInterfaces;
     cfgBuf := Pointer(kalloc(totalLen));
+    if cfgBuf = nil then begin kfree(void(dev)); pop_trace; exit; end;
 
     status := usb_get_descriptor(dev, USB_DESC_CONFIGURATION, 0, cfgBuf, totalLen);
     if status <> tsSuccess then begin
@@ -744,12 +754,14 @@ begin
     { Step 7: Parse interface and endpoint descriptors }
     offset := sizeof(TUSBConfigDescriptor);
     dev^.NumEndpoints := 0;
-    while offset < totalLen do begin
+    while offset + 2 <= totalLen do begin
         { Read descriptor length and type }
-        if PUint8(uint32(cfgBuf) + offset)^ = 0 then break; { Safety: zero-length = corrupt }
+        if PUint8(uint32(cfgBuf) + offset)^ < 2 then break;
+        if offset + PUint8(uint32(cfgBuf) + offset)^ > totalLen then break;
 
         if PUint8(uint32(cfgBuf) + offset + 1)^ = USB_DESC_ENDPOINT then begin
-            if dev^.NumEndpoints < USB_MAX_ENDPOINTS then begin
+            if (PUint8(uint32(cfgBuf) + offset)^ >= sizeof(TUSBEndpointDescriptor)) and
+               (dev^.NumEndpoints < USB_MAX_ENDPOINTS) then begin
                 epDesc := PUSBEndpointDescriptor(uint32(cfgBuf) + offset);
                 dev^.Endpoints[dev^.NumEndpoints].Address   := usb_ep_number(epDesc^.bEndpointAddress);
                 dev^.Endpoints[dev^.NumEndpoints].Direction  := usb_ep_direction(epDesc^.bEndpointAddress);
@@ -780,16 +792,22 @@ begin
     { Find the first interface descriptor for class/subclass/protocol }
     offset := sizeof(TUSBConfigDescriptor);
     ifDesc := nil;
-    while offset < totalLen do begin
+    while offset + 2 <= totalLen do begin
-        if PUint8(uint32(cfgBuf) + offset)^ = 0 then break;
+        if PUint8(uint32(cfgBuf) + offset)^ < 2 then break;
-        if PUint8(uint32(cfgBuf) + offset + 1)^ = USB_DESC_INTERFACE then begin
+        if offset + PUint8(uint32(cfgBuf) + offset)^ > totalLen then break;
+        if (PUint8(uint32(cfgBuf) + offset + 1)^ = USB_DESC_INTERFACE) and
+           (PUint8(uint32(cfgBuf) + offset)^ >= sizeof(TUSBInterfaceDescriptor)) then begin
             ifDesc := PUSBInterfaceDescriptor(uint32(cfgBuf) + offset);
+            dev^.InterfaceNumber := ifDesc^.bInterfaceNumber;
             break;
         end;
         offset := offset + PUint8(uint32(cfgBuf) + offset)^;
     end;
 
     DevID := PDeviceIdentifier(kalloc(sizeof(TDeviceIdentifier)));
+    if DevID = nil then begin
+        kfree(void(cfgBuf)); kfree(void(dev)); pop_trace; exit;
+    end;
     DevID^.Bus := biUSB;
     DevID^.id0 := (uint32(devDesc.idVendor) SHL 16) OR devDesc.idProduct;
     DevID^.id1 := devDesc.bDeviceClass;

src/driver/bus/usb/driver.bus.usb.ehci.pas

@@ -963,6 +963,8 @@ end;
 
 { ========================= Submit Transfer ========================= }
 
+procedure ehci_free_qtd_chain(first : PEHCI_qTD); forward;
+
 function ehci_submit_control(hc : PUSBHCDriver; transfer : PUSBTransfer) : boolean;
 var
     priv     : PEHCI_PrivData;
@@ -1004,7 +1006,7 @@ begin
 
     { SETUP qTD }
     qtd := ehci_alloc_qtd;
-    if qtd = nil then begin pop_trace; exit; end;
+    if qtd = nil then begin ehci_free_qtd_chain(firstqTD); pop_trace; exit; end;
     firstqTD := qtd;
     qtd^.Token := ehci_make_qtd_token(EHCI_QTD_PID_SETUP, false, 8, false, 3);
     ehci_fill_qtd_buffers(qtd, @transfer^.Setup, 8);
@@ -1021,7 +1023,7 @@ begin
 
     while remaining > 0 do begin
         qtd := ehci_alloc_qtd;
-        if qtd = nil then begin pop_trace; exit; end;
+        if qtd = nil then begin ehci_free_qtd_chain(firstqTD); pop_trace; exit; end;
 
         if remaining > maxPkt then
             pktLen := maxPkt
@@ -1034,6 +1036,7 @@ begin
             pid := EHCI_QTD_PID_OUT;
 
         qtd^.Token := ehci_make_qtd_token(pid, toggle, pktLen, false, 3);
+        qtd^.SWPad0 := pktLen; { Original data length, excluding SETUP/status }
         ehci_fill_qtd_buffers(qtd, Pointer(uint32(transfer^.Buffer) + offset), pktLen);
         qtd^.NextqTD    := EHCI_QTD_T;
         qtd^.AltNextqTD := EHCI_QTD_T;
@@ -1052,7 +1055,7 @@ begin
 
     { STATUS qTD (opposite direction, DATA1) }
     qtd := ehci_alloc_qtd;
-    if qtd = nil then begin pop_trace; exit; end;
+    if qtd = nil then begin ehci_free_qtd_chain(firstqTD); pop_trace; exit; end;
 
     if (transfer^.Setup.bmRequestType AND $80) <> 0 then
         pid := EHCI_QTD_PID_OUT
@@ -1070,7 +1073,7 @@ begin
 
     { Build QH }
     qh := ehci_alloc_qh;
-    if qh = nil then begin pop_trace; exit; end;
+    if qh = nil then begin ehci_free_qtd_chain(firstqTD); pop_trace; exit; end;
 
     qh^.EPChars  := ehci_make_qh_epchars(devAddr, 0, speedBits, maxPkt,
                                            true, false, isCtrlEP, 4);
@@ -1152,7 +1155,7 @@ begin
 
     repeat
         qtd := ehci_alloc_qtd;
-        if qtd = nil then begin pop_trace; exit; end;
+        if qtd = nil then begin ehci_free_qtd_chain(firstqTD); pop_trace; exit; end;
 
         if firstqTD = nil then firstqTD := qtd;
 
@@ -1164,6 +1167,7 @@ begin
             pktLen := 0;
 
         qtd^.Token := ehci_make_qtd_token(pid, toggle, pktLen, false, 3);
+        qtd^.SWPad0 := pktLen; { Original data length, excluding SETUP/status }
         if pktLen > 0 then
             ehci_fill_qtd_buffers(qtd, Pointer(uint32(transfer^.Buffer) + offset), pktLen);
         qtd^.NextqTD    := EHCI_QTD_T;
@@ -1200,7 +1204,7 @@ begin
 
     { Build QH }
     qh := ehci_alloc_qh;
-    if qh = nil then begin pop_trace; exit; end;
+    if qh = nil then begin ehci_free_qtd_chain(firstqTD); pop_trace; exit; end;
 
     qh^.EPChars := ehci_make_qh_epchars(devAddr, epNum, speedBits, maxPkt,
                                           true, false, false, 4);
@@ -1297,6 +1301,11 @@ begin
             ehci_transfer_done := true;
             exit;
         end;
+        if (qh^.SWTransfer <> nil) and
+           (PUSBTransfer(qh^.SWTransfer)^.PipeType = ptBulk) and
+           (((qtd^.Token SHR 16) AND $7FFF) <> 0) then begin
+            ehci_transfer_done := true; exit;
+        end;
         qtd := qtd^.SWNext;
     end;
     ehci_transfer_done := true;
@@ -1385,6 +1394,8 @@ var
     transfer : PUSBTransfer;
     head     : PEHCI_QH;
     listIdx  : uint32;
+    qtd      : PEHCI_qTD;
+    residual : uint32;
 begin
     if (hc = nil) or (hc^.PrivData = nil) then exit;
     priv := PEHCI_PrivData(hc^.PrivData);
@@ -1422,7 +1433,15 @@ begin
                 if transfer^.Status = tsInProgress then begin
                     if ehci_transfer_done(qh) then begin
                         transfer^.Status := ehci_transfer_status(qh);
-                        transfer^.ActualLen := transfer^.BufferLen;
+                        transfer^.ActualLen := 0;
+                        qtd := qh^.SWFirstqTD;
+                        while qtd <> nil do begin
+                            if (qtd^.Token AND EHCI_QTD_STS_ACTIVE) <> 0 then break;
+                            residual := (qtd^.Token SHR 16) AND $7FFF;
+                            if (qtd^.SWPad0 > 0) and (residual <= qtd^.SWPad0) then
+                                inc(transfer^.ActualLen, qtd^.SWPad0 - residual);
+                            qtd := qtd^.SWNext;
+                        end;
                         { Remove QH from schedule and free qTD chain }
                         if listIdx = 0 then
                             ehci_remove_async_qh(priv, qh)

src/driver/bus/usb/driver.bus.usb.types.pas

@@ -249,6 +249,8 @@ type
         Endpoints    : array[0..USB_MAX_ENDPOINTS-1] of TUSBEndpoint;
         NumEndpoints : uint8;
         Configured   : boolean;
+        InterfaceNumber : uint8;   { First interface, alternate setting zero }
+        InterfaceCount  : uint8;   { Class drivers may reject composite devices }
         fnDisconnect : TUSBDisconnectCallback; { Called before device is freed on unplug }
     end;
 

src/test/test.pas

@@ -37,6 +37,7 @@ uses
     driver.storage.metatest,
     driver.storage.perftest,
     driver.storage.ctl.ahci,
+    driver.storage.ctl.usb,
     core.enc.fnv1a,
     core.enc.djb2,
     core.ds.bloom,
@@ -76,6 +77,7 @@ begin
     driver.storage.metatest.UnitTest;
     driver.storage.perftest.UnitTest;
     driver.storage.ctl.ahci.UnitTest;
+    driver.storage.ctl.usb.UnitTest;
     core.enc.fnv1a.UnitTest;
     core.enc.djb2.UnitTest;
     core.ds.bloom.UnitTest;