Move hardware handlers from dm layer to SCSI layer

Index: linux-2.6.25-rc2-mm1/drivers/scsi/scsi_error.c
===================================================================
@@ -298,6 +298,7 @@ static inline void scsi_eh_prt_fail_stat
  */
 static int scsi_check_sense(struct scsi_cmnd *scmd)
 {
+	struct scsi_device *sdev = scmd->device;
 	struct scsi_sense_hdr sshdr;
 
 	if (! scsi_command_normalize_sense(scmd, &sshdr))

@@ -306,6 +307,15 @@ static int scsi_check_sense(struct scsi_
 	if (scsi_sense_is_deferred(&sshdr))
 		return NEEDS_RETRY;
 
+	if (sdev->sdev_dh && sdev->sdev_dh->check_sense) {
+		int rc;
+
+		rc = sdev->sdev_dh->check_sense(sdev, &sshdr);
+		if (rc != SUCCESS)
+			return rc;
+		/* handler does not care. Drop down to default handling */
+	}
+
 	/*
 	 * Previous logic looked for FILEMARK, EOM or ILI which are
 	 * mainly associated with tapes and returned SUCCESS.

Index: linux-2.6.25-rc2-mm1/include/scsi/scsi_device.h
===================================================================
@@ -161,9 +161,25 @@ struct scsi_device {
 
 	struct execute_work	ew; /* used to get process context on put */
 
+	struct scsi_device_handler *sdev_dh;
+	void			*sdev_dh_data;
 	enum scsi_device_state sdev_state;
 	unsigned long		sdev_data[0];
 } __attribute__((aligned(sizeof(unsigned long))));
+
+struct scsi_device_handler {
+	/* Used by the infrastructure */
+	struct list_head list; /* list of scsi_device_handlers */
+	struct notifier_block nb;
+
+	/* Filled by the hardware handler */
+	struct module *module;
+	const char *name;
+	int (*check_sense)(struct scsi_device *, struct scsi_sense_hdr *);
+	int (*activate)(struct scsi_device *);
+	int (*prep_fn)(struct scsi_device *);
+};
+
 #define	to_scsi_device(d)	\
 	container_of(d, struct scsi_device, sdev_gendev)
 #define	class_to_sdev(d)	\

@@ -229,7 +245,9 @@ extern struct scsi_device *__scsi_add_de
 		uint, uint, uint, void *hostdata);
 extern int scsi_add_device(struct Scsi_Host *host, uint channel,
 			   uint target, uint lun);
+extern int scsi_register_device_handler(struct scsi_device_handler *sdev_dh);
 extern void scsi_remove_device(struct scsi_device *);
+extern int scsi_unregister_device_handler(struct scsi_device_handler *sdev_dh);
 
 extern int scsi_device_get(struct scsi_device *);
 extern void scsi_device_put(struct scsi_device *);

146 lines not on this page · raw message

58 lines not on this page · raw message

Index: linux-2.6.25-rc2-mm1/drivers/scsi/scsi_lib.c
===================================================================
@@ -1156,6 +1156,11 @@ int scsi_setup_fs_cmnd(struct scsi_devic
 	struct scsi_cmnd *cmd;
 	int ret = scsi_prep_state_check(sdev, req);
 
+	if ((ret == BLKPREP_OK) && (req->cmd_type == REQ_TYPE_FS)) {
+		if (sdev->sdev_dh && sdev->sdev_dh->prep_fn)
+			ret = sdev->sdev_dh->prep_fn(sdev);
+	}
+
 	if (ret != BLKPREP_OK)
 		return ret;
 	/*

660 lines not on this page · raw message

193 lines not on this page · raw message

488 lines not on this page · raw message

Index: linux-2.6.25-rc2-mm1/drivers/md/dm-mpath.c
===================================================================
@@ -7,7 +7,6 @@
 
 #include "dm.h"
 #include "dm-path-selector.h"
-#include "dm-hw-handler.h"
 #include "dm-bio-list.h"
 #include "dm-bio-record.h"
 #include "dm-uevent.h"

@@ -20,6 +19,7 @@
 #include <linux/slab.h>
 #include <linux/time.h>
 #include <linux/workqueue.h>
+#include <scsi/scsi_dh.h>
 #include <asm/atomic.h>
 
 #define DM_MSG_PREFIX "multipath"

@@ -61,9 +61,14 @@ struct multipath {
 
 	spinlock_t lock;
 
-	struct hw_handler hw_handler;
 	unsigned nr_priority_groups;
 	struct list_head priority_groups;
+
+	/* Fields used by hardware handler usage */
+	char *hw_handler_name;
+	struct work_struct activate_passive_path;
+	struct dm_path *path_to_activate;
+
 	unsigned pg_init_required;	/* pg_init needs calling? */
 	unsigned pg_init_in_progress;	/* Only one pg_init allowed at once */
 

@@ -106,9 +111,13 @@ typedef int (*action_fn) (struct pgpath 
 
 static struct kmem_cache *_mpio_cache;
 
-static struct workqueue_struct *kmultipathd;
+static struct workqueue_struct *kmultipathd, *hw_handlerd;
 static void process_queued_ios(struct work_struct *work);
 static void trigger_event(struct work_struct *work);
+static void activate_passive_path(struct work_struct *work);
+static void bypass_pg(struct multipath *m, struct priority_group *pg,
+		      int bypassed);
+static int fail_path(struct pgpath *pgpath);
 
 
 /*-----------------------------------------------

@@ -178,6 +187,7 @@ static struct multipath *alloc_multipath
 		m->queue_io = 1;
 		INIT_WORK(&m->process_queued_ios, process_queued_ios);
 		INIT_WORK(&m->trigger_event, trigger_event);
+		INIT_WORK(&m->activate_passive_path, activate_passive_path);
 		m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache);
 		if (!m->mpio_pool) {
 			kfree(m);

@@ -193,18 +203,13 @@ static struct multipath *alloc_multipath
 static void free_multipath(struct multipath *m)
 {
 	struct priority_group *pg, *tmp;
-	struct hw_handler *hwh = &m->hw_handler;
 
 	list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
 		list_del(&pg->list);
 		free_priority_group(pg, m->ti);
 	}
 
-	if (hwh->type) {
-		hwh->type->destroy(hwh);
-		dm_put_hw_handler(hwh->type);
-	}
-
+	kfree(m->hw_handler_name);
 	mempool_destroy(m->mpio_pool);
 	kfree(m);
 }

@@ -216,12 +221,10 @@ static void free_multipath(struct multip
 
 static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
 {
-	struct hw_handler *hwh = &m->hw_handler;
-
 	m->current_pg = pgpath->pg;
 
 	/* Must we initialise the PG first, and queue I/O till it's ready? */
-	if (hwh->type && hwh->type->pg_init) {
+	if (m->hw_handler_name) {
 		m->pg_init_required = 1;
 		m->queue_io = 1;
 	} else {

@@ -405,11 +408,86 @@ static void dispatch_queued_ios(struct m
 	}
 }
 
+static void pg_init_done(struct dm_path *path, int errors)
+{
+	struct pgpath *pgpath = path_to_pgpath(path);
+	struct priority_group *pg = pgpath->pg;
+	struct multipath *m = pg->m;
+	unsigned long flags;
+
+	/* device or driver problems */
+	switch (errors) {
+	case SCSI_DH_OK:
+		break;
+	case SCSI_DH_NOSYS:
+		if (!m->hw_handler_name) {
+			errors = 0;
+			break;
+		}
+		DMERR("Cannot failover device because hw-%s may not be "
+		      "loaded.", m->hw_handler_name);
+		/*
+		 * Fail path for now, so we do not ping poing
+		 */
+		fail_path(pgpath);
+		break;
+	case SCSI_DH_DEV_TEMP_BUSY:
+		/*
+		 * Probably doing something like FW upgrade on the
+		 * controller so try the other pg.
+		 */
+		bypass_pg(m, pg, 1);
+		break;
+	/* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
+	case SCSI_DH_RETRY:
+	case SCSI_DH_IMM_RETRY:
+	case SCSI_DH_RES_TEMP_UNAVAIL:
+		break;
+	default:
+		/*
+		 * We probably do not want to fail the path for a device
+		 * error, but this is what the old dm did. In future
+		 * patches we can do more advanced handling.
+		 */
+		fail_path(pgpath);
+	}
+
+	spin_lock_irqsave(&m->lock, flags);
+	if (errors) {
+		DMERR("Could not failover device. Error %d.", errors);
+		m->current_pgpath = NULL;
+		m->current_pg = NULL;
+	} else if (!m->pg_init_required) {
+		m->queue_io = 0;
+		pg->bypassed = 0;
+	}
+
+	m->pg_init_in_progress = 0;
+	queue_work(kmultipathd, &m->process_queued_ios);
+	spin_unlock_irqrestore(&m->lock, flags);
+}
+
+static void activate_passive_path(struct work_struct *work)
+{
+	int ret;
+	struct multipath *m =
+		container_of(work, struct multipath, activate_passive_path);
+	struct dm_path *path = m->path_to_activate;
+
+	ret = scsi_dh_activate(bdev_get_queue(path->dev->bdev));
+	pg_init_done(path, ret);
+}
+
+static void pg_init(struct multipath *m, struct dm_path *path)
+{
+	m->path_to_activate = path;
+	queue_work(hw_handlerd, &m->activate_passive_path);
+}
+
 static void process_queued_ios(struct work_struct *work)
 {
 	struct multipath *m =
 		container_of(work, struct multipath, process_queued_ios);
-	struct hw_handler *hwh = &m->hw_handler;
 	struct pgpath *pgpath = NULL;
 	unsigned init_required = 0, must_queue = 1;
 	unsigned long flags;

@@ -439,7 +517,7 @@ out:
 	spin_unlock_irqrestore(&m->lock, flags);
 
 	if (init_required)
-		hwh->type->pg_init(hwh, pgpath->pg->bypassed, &pgpath->path);
+		pg_init(m, &pgpath->path);
 
 	if (!must_queue)
 		dispatch_queued_ios(m);

@@ -652,10 +730,9 @@ static struct priority_group *parse_prio
 
 static int parse_hw_handler(struct arg_set *as, struct multipath *m)
 {
+	struct dm_target *ti = m->ti;
 	int r;
-	struct hw_handler_type *hwht;
 	unsigned hw_argc;
-	struct dm_target *ti = m->ti;
 
 	static struct param _params[] = {
 		{0, 1024, "invalid number of hardware handler args"},

@@ -668,25 +745,15 @@ static int parse_hw_handler(struct arg_s
 	if (!hw_argc)
 		return 0;
 
-	hwht = dm_get_hw_handler(shift(as));
-	if (!hwht) {
+	m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL);
+	r = request_module("scsi_dh_%s", m->hw_handler_name);
+	if (r) {
+		kfree(m->hw_handler_name);
+		m->hw_handler_name = NULL;
 		ti->error = "unknown hardware handler type";
 		return -EINVAL;
 	}
-
-	m->hw_handler.md = dm_table_get_md(ti->table);
-	dm_put(m->hw_handler.md);
-
-	r = hwht->create(&m->hw_handler, hw_argc - 1, as->argv);
-	if (r) {
-		dm_put_hw_handler(hwht);
-		ti->error = "hardware handler constructor failed";
-		return r;
-	}
-
-	m->hw_handler.type = hwht;
 	consume(as, hw_argc - 1);
-
 	return 0;
 }
 

@@ -808,6 +875,7 @@ static void multipath_dtr(struct dm_targ
 {
 	struct multipath *m = (struct multipath *) ti->private;
 
+	flush_workqueue(hw_handlerd);
 	flush_workqueue(kmultipathd);
 	free_multipath(m);
 }

@@ -1006,71 +1074,11 @@ static int bypass_pg_num(struct multipat
 }
 
 /*
- * Should we retry pg_init immediately?
- */
-static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
-{
-	unsigned long flags;
-	int limit_reached = 0;
-
-	spin_lock_irqsave(&m->lock, flags);
-
-	if (m->pg_init_count <= m->pg_init_retries)
-		m->pg_init_required = 1;
-	else
-		limit_reached = 1;
-
-	spin_unlock_irqrestore(&m->lock, flags);
-
-	return limit_reached;
-}
-
-/*
- * pg_init must call this when it has completed its initialisation
- */
-void dm_pg_init_complete(struct dm_path *path, unsigned err_flags)
-{
-	struct pgpath *pgpath = path_to_pgpath(path);
-	struct priority_group *pg = pgpath->pg;
-	struct multipath *m = pg->m;
-	unsigned long flags;
-
-	/*
-	 * If requested, retry pg_init until maximum number of retries exceeded.
-	 * If retry not requested and PG already bypassed, always fail the path.
-	 */
-	if (err_flags & MP_RETRY) {
-		if (pg_init_limit_reached(m, pgpath))
-			err_flags |= MP_FAIL_PATH;
-	} else if (err_flags && pg->bypassed)
-		err_flags |= MP_FAIL_PATH;
-
-	if (err_flags & MP_FAIL_PATH)
-		fail_path(pgpath);
-
-	if (err_flags & MP_BYPASS_PG)
-		bypass_pg(m, pg, 1);
-
-	spin_lock_irqsave(&m->lock, flags);
-	if (err_flags & ~MP_RETRY) {
-		m->current_pgpath = NULL;
-		m->current_pg = NULL;
-	} else if (!m->pg_init_required)
-		m->queue_io = 0;
-
-	m->pg_init_in_progress = 0;
-	queue_work(kmultipathd, &m->process_queued_ios);
-	spin_unlock_irqrestore(&m->lock, flags);
-}
-
-/*
  * end_io handling
  */
 static int do_end_io(struct multipath *m, struct bio *bio,
 		     int error, struct dm_mpath_io *mpio)
 {
-	struct hw_handler *hwh = &m->hw_handler;
-	unsigned err_flags = MP_FAIL_PATH;	/* Default behavior */
 	unsigned long flags;
 
 	if (!error)

@@ -1097,19 +1105,8 @@ static int do_end_io(struct multipath *m
 	}
 	spin_unlock_irqrestore(&m->lock, flags);
 
-	if (hwh->type && hwh->type->error)
-		err_flags = hwh->type->error(hwh, bio);
-
-	if (mpio->pgpath) {
-		if (err_flags & MP_FAIL_PATH)
-			fail_path(mpio->pgpath);
-
-		if (err_flags & MP_BYPASS_PG)
-			bypass_pg(m, mpio->pgpath->pg, 1);
-	}
-
-	if (err_flags & MP_ERROR_IO)
-		return -EIO;
+	if (mpio->pgpath)
+		fail_path(mpio->pgpath);
 
       requeue:
 	dm_bio_restore(&mpio->details, bio);

@@ -1194,7 +1191,6 @@ static int multipath_status(struct dm_ta
 	int sz = 0;
 	unsigned long flags;
 	struct multipath *m = (struct multipath *) ti->private;
-	struct hw_handler *hwh = &m->hw_handler;
 	struct priority_group *pg;
 	struct pgpath *p;
 	unsigned pg_num;

@@ -1214,12 +1210,10 @@ static int multipath_status(struct dm_ta
 			DMEMIT("pg_init_retries %u ", m->pg_init_retries);
 	}
 
-	if (hwh->type && hwh->type->status)
-		sz += hwh->type->status(hwh, type, result + sz, maxlen - sz);
-	else if (!hwh->type || type == STATUSTYPE_INFO)
-		DMEMIT("0 ");
+	if (m->hw_handler_name)
+		DMEMIT("1 %s ", m->hw_handler_name);
 	else
-		DMEMIT("1 %s ", hwh->type->name);
+		DMEMIT("0 ");
 
 	DMEMIT("%u ", m->nr_priority_groups);
 

@@ -1422,6 +1416,15 @@ static int __init dm_multipath_init(void
 		return -ENOMEM;
 	}
 
+	hw_handlerd = create_workqueue("khwhandlerd");
+	if (!hw_handlerd) {
+		DMERR("failed to create workqueue khwhandlerd");
+		destroy_workqueue(kmultipathd);
+		dm_unregister_target(&multipath_target);
+		kmem_cache_destroy(_mpio_cache);
+		return -ENOMEM;
+	}
+
 	DMINFO("version %u.%u.%u loaded",
 	       multipath_target.version[0], multipath_target.version[1],
 	       multipath_target.version[2]);

@@ -1433,6 +1436,7 @@ static void __exit dm_multipath_exit(voi
 {
 	int r;
 
+	destroy_workqueue(hw_handlerd);
 	destroy_workqueue(kmultipathd);
 
 	r = dm_unregister_target(&multipath_target);

@@ -1441,8 +1445,6 @@ static void __exit dm_multipath_exit(voi
 	kmem_cache_destroy(_mpio_cache);
 }
 
-EXPORT_SYMBOL_GPL(dm_pg_init_complete);
-
 module_init(dm_multipath_init);
 module_exit(dm_multipath_exit);
 

347 lines not on this page · raw message

248 lines not on this page · raw message

701 lines not on this page · raw message

Index: linux-2.6.24/drivers/md/dm-hw-handler.c
===================================================================
@@ -1,213 +0,0 @@
-/*
- * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
- *
- * This file is released under the GPL.
- *
- * Multipath hardware handler registration.
- */
-
-#include "dm.h"
-#include "dm-hw-handler.h"
-
-#include <linux/slab.h>
-
-struct hwh_internal {
-	struct hw_handler_type hwht;
-
-	struct list_head list;
-	long use;
-};
-
-#define hwht_to_hwhi(__hwht) container_of((__hwht), struct hwh_internal, hwht)
-
-static LIST_HEAD(_hw_handlers);
-static DECLARE_RWSEM(_hwh_lock);
-
-static struct hwh_internal *__find_hw_handler_type(const char *name)
-{
-	struct hwh_internal *hwhi;
-
-	list_for_each_entry(hwhi, &_hw_handlers, list) {
-		if (!strcmp(name, hwhi->hwht.name))
-			return hwhi;
-	}
-
-	return NULL;
-}
-
-static struct hwh_internal *get_hw_handler(const char *name)
-{
-	struct hwh_internal *hwhi;
-
-	down_read(&_hwh_lock);
-	hwhi = __find_hw_handler_type(name);
-	if (hwhi) {
-		if ((hwhi->use == 0) && !try_module_get(hwhi->hwht.module))
-			hwhi = NULL;
-		else
-			hwhi->use++;
-	}
-	up_read(&_hwh_lock);
-
-	return hwhi;
-}
-
-struct hw_handler_type *dm_get_hw_handler(const char *name)
-{
-	struct hwh_internal *hwhi;
-
-	if (!name)
-		return NULL;
-
-	hwhi = get_hw_handler(name);
-	if (!hwhi) {
-		request_module("dm-%s", name);
-		hwhi = get_hw_handler(name);
-	}
-
-	return hwhi ? &hwhi->hwht : NULL;
-}
-
-void dm_put_hw_handler(struct hw_handler_type *hwht)
-{
-	struct hwh_internal *hwhi;
-
-	if (!hwht)
-		return;
-
-	down_read(&_hwh_lock);
-	hwhi = __find_hw_handler_type(hwht->name);
-	if (!hwhi)
-		goto out;
-
-	if (--hwhi->use == 0)
-		module_put(hwhi->hwht.module);
-
-	BUG_ON(hwhi->use < 0);
-
-      out:
-	up_read(&_hwh_lock);
-}
-
-static struct hwh_internal *_alloc_hw_handler(struct hw_handler_type *hwht)
-{
-	struct hwh_internal *hwhi = kzalloc(sizeof(*hwhi), GFP_KERNEL);
-
-	if (hwhi)
-		hwhi->hwht = *hwht;
-
-	return hwhi;
-}
-
-int dm_register_hw_handler(struct hw_handler_type *hwht)
-{
-	int r = 0;
-	struct hwh_internal *hwhi = _alloc_hw_handler(hwht);
-
-	if (!hwhi)
-		return -ENOMEM;
-
-	down_write(&_hwh_lock);
-
-	if (__find_hw_handler_type(hwht->name)) {
-		kfree(hwhi);
-		r = -EEXIST;
-	} else
-		list_add(&hwhi->list, &_hw_handlers);
-
-	up_write(&_hwh_lock);
-
-	return r;
-}
-
-int dm_unregister_hw_handler(struct hw_handler_type *hwht)
-{
-	struct hwh_internal *hwhi;
-
-	down_write(&_hwh_lock);
-
-	hwhi = __find_hw_handler_type(hwht->name);
-	if (!hwhi) {
-		up_write(&_hwh_lock);
-		return -EINVAL;
-	}
-
-	if (hwhi->use) {
-		up_write(&_hwh_lock);
-		return -ETXTBSY;
-	}
-
-	list_del(&hwhi->list);
-
-	up_write(&_hwh_lock);
-
-	kfree(hwhi);
-
-	return 0;
-}
-
-unsigned dm_scsi_err_handler(struct hw_handler *hwh, struct bio *bio)
-{
-#if 0
-	int sense_key, asc, ascq;
-
-	if (bio->bi_error & BIO_SENSE) {
-		/* FIXME: This is just an initial guess. */
-		/* key / asc / ascq */
-		sense_key = (bio->bi_error >> 16) & 0xff;
-		asc = (bio->bi_error >> 8) & 0xff;
-		ascq = bio->bi_error & 0xff;
-
-		switch (sense_key) {
-			/* This block as a whole comes from the device.
-			 * So no point retrying on another path. */
-		case 0x03:	/* Medium error */
-		case 0x05:	/* Illegal request */
-		case 0x07:	/* Data protect */
-		case 0x08:	/* Blank check */
-		case 0x0a:	/* copy aborted */
-		case 0x0c:	/* obsolete - no clue ;-) */
-		case 0x0d:	/* volume overflow */
-		case 0x0e:	/* data miscompare */
-		case 0x0f:	/* reserved - no idea either. */
-			return MP_ERROR_IO;
-
-			/* For these errors it's unclear whether they
-			 * come from the device or the controller.
-			 * So just lets try a different path, and if
-			 * it eventually succeeds, user-space will clear
-			 * the paths again... */
-		case 0x02:	/* Not ready */
-		case 0x04:	/* Hardware error */
-		case 0x09:	/* vendor specific */
-		case 0x0b:	/* Aborted command */
-			return MP_FAIL_PATH;
-
-		case 0x06:	/* Unit attention - might want to decode */
-			if (asc == 0x04 && ascq == 0x01)
-				/* "Unit in the process of
-				 * becoming ready" */
-				return 0;
-			return MP_FAIL_PATH;
-
-			/* FIXME: For Unit Not Ready we may want
-			 * to have a generic pg activation
-			 * feature (START_UNIT). */
-
-			/* Should these two ever end up in the
-			 * error path? I don't think so. */
-		case 0x00:	/* No sense */
-		case 0x01:	/* Recovered error */
-			return 0;
-		}
-	}
-#endif
-
-	/* We got no idea how to decode the other kinds of errors ->
-	 * assume generic error condition. */
-	return MP_FAIL_PATH;
-}
-
-EXPORT_SYMBOL_GPL(dm_register_hw_handler);
-EXPORT_SYMBOL_GPL(dm_unregister_hw_handler);
-EXPORT_SYMBOL_GPL(dm_scsi_err_handler);

63 lines not on this page · raw message

Index: linux-2.6.24/drivers/md/dm-mpath.h
===================================================================
@@ -16,7 +16,6 @@ struct dm_path {
 	unsigned is_active;	/* Read-only */
 
 	void *pscontext;	/* For path-selector use */
-	void *hwhcontext;	/* For hw-handler use */
 };
 
 /* Callback for hwh_pg_init_fn to use when complete */