2.5.54 fix ide-cd/ide-scsi oopses after module unload

Mikael Pettersson posted the following to the linux-kernel mailing list. His patch includes a fix that Willem Riede spotted as well (no driver_unregister() in ide.c). Willem also had other changes in ide-scsi.c which are still pending. See attached file.

Doug Gilbert

Mikael Pettersson wrote on lkml:
vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv

In 2.5 (the bug's been there since 2.5.42), rmmod:ing a modular IDE subdriver like ide-cd or ide-scsi and then rebooting causes an oops in device_shutdown(). This is because the IDE layer doesn't reset the drive->gendev.driver pointer that it previously set up to point to data structures in the subdriver module. device_shutdown() sees a non-NULL ->driver, dereferences it, and oopses.

The patch below for 2.5.54 fixes two generic bugs related to unloading of modular IDE subdrivers, and one specific to ide-scsi:

1. ata_attach() needs to set drive->gendev.driver = NULL when no specific driver claims the drive. This prevents a drive previously owned by a subdriver module from keeping its ->gendev.driver pointing into that module.

2. ide_unregister_driver() needs to unregister &driver->gen_driver; this is to balance the corresponding register call in ide_register_driver(). [This part of the patch is originally by Patrick Mochel.]

3. ide-scsi.c abuses ide_driver_t's busy field as a counter while the field in fact is a single-bit boolean. This causes the busyness of the driver to be incorrect while the driver is active. (From my recent patch for 2.4.20-ac2/2.4.21-pre2.)

With these three fixes modular ide-cd and ide-scsi work quite reliably for me in 2.5.54.

/Mikael

diff -ruN linux-2.5.54/drivers/ide/ide.c linux-2.5.54.ide-fixes/drivers/ide/ide.c
@@ -1346,6 +1346,7 @@
 	spin_unlock(&drivers_lock);
 	spin_lock(&drives_lock);
 	list_add_tail(&drive->list, &ata_unused);
+	drive->gendev.driver = NULL;
 	spin_unlock(&drives_lock);
 	return 1;
 }

@@ -2308,6 +2309,8 @@
 	list_del(&driver->drivers);
 	spin_unlock(&drivers_lock);
 
+	driver_unregister(&driver->gen_driver);
+
 	while(!list_empty(&driver->drives)) {
 		drive = list_entry(driver->drives.next, ide_drive_t, list);
 		if (driver->cleanup(drive)) {

diff -ruN linux-2.5.54/drivers/scsi/ide-scsi.c linux-2.5.54.ide-fixes/drivers/scsi/ide-scsi.c
@@ -590,6 +590,7 @@
 	set_bit(IDESCSI_LOG_CMD, &scsi->log);
 #endif /* IDESCSI_DEBUG_LOG */
 	idescsi_add_settings(drive);
+	DRIVER(drive)->busy--;
 }
 
 static int idescsi_cleanup (ide_drive_t *drive)

@@ -995,7 +996,7 @@
 	for (id = 0; id < MAX_HWIFS * MAX_DRIVES; id++) {
 		drive = idescsi_drives[id];
 		if (drive)
-			DRIVER(drive)->busy--;
+			DRIVER(drive)->busy = 0;
 	}
 	scsi_unregister   (idescsi_host);
 	device_unregister(&idescsi_primary);

- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

@@ -289,6 +289,7 @@
 	pc->timeout = jiffies + WAIT_READY;
 	/* NOTE! Save the failed packet command in "rq->buffer" */
 	rq->buffer = (void *) failed_command->special;
+	pc->scsi_cmd = ((idescsi_pc_t *) failed_command->special)->scsi_cmd;
 	if (test_bit(IDESCSI_LOG_CMD, &scsi->log)) {
 		printk ("ide-scsi: %s: queue cmd = ", drive->name);
 		hexdump(pc->c, 6);

@@ -876,7 +877,8 @@
 			/* is cmd active?
 			 *  need to lock so this stuff doesn't change under us */
 			spin_lock_irqsave(&ide_lock, flags);
-			if (scsi->pc && scsi->pc->scsi_cmd->serial_number == cmd->serial_number) {
+			if (scsi->pc && scsi->pc->scsi_cmd && 
+					scsi->pc->scsi_cmd->serial_number == cmd->serial_number) {
 				/* yep - let's give it some more time - 
 				 * we can do that, we're in _our_ error kernel thread */
 				spin_unlock_irqrestore(&ide_lock, flags);