Attachment 292553 Details for Bug 195685 – Updated rhel4.7 patch - currently under test and looking promising

Updated rhel4.7 patch - currently under test and looking promising

rhel4u7-second.patch (text/x-patch), 12.43 KB, created by Dave Wysochanski on 2008-01-22 19:46:29 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Dave Wysochanski

Created: 2008-01-22 19:46:29 UTC

Size: 12.43 KB

patch

obsolete

>Index: linux-2.6.9/drivers/md/Kconfig
>===================================================================
>--- linux-2.6.9.orig/drivers/md/Kconfig
>+++ linux-2.6.9/drivers/md/Kconfig
>@@ -230,5 +230,11 @@ config DM_MULTIPATH_EMC
> 	---help---
> 	  Multipath support for EMC CX/AX series hardware.
> 
>+config DM_MULTIPATH_HP
>+        tristate "HP MSA multipath support (EXPERIMENTAL)"
>+        depends on DM_MULTIPATH && BLK_DEV_DM && EXPERIMENTAL
>+        ---help---
>+          Multipath support for HP MSA (Active/Passive) series hardware.
>+
> endmenu
> 
>Index: linux-2.6.9/drivers/md/Makefile
>===================================================================
>--- linux-2.6.9.orig/drivers/md/Makefile
>+++ linux-2.6.9/drivers/md/Makefile
>@@ -7,6 +7,7 @@ dm-mod-objs	:= dm.o dm-table.o dm-target
> dm-multipath-objs := dm-hw-handler.o dm-path-selector.o dm-mpath.o
> dm-snapshot-objs := dm-snap.o dm-exception-store.o
> dm-mirror-objs	:= dm-log.o dm-raid1.o
>+dm-hp-sw-objs	:= dm-mpath-hp-sw.o
> raid6-objs	:= raid6main.o raid6algos.o raid6recov.o raid6tables.o \
> 		   raid6int1.o raid6int2.o raid6int4.o \
> 		   raid6int8.o raid6int16.o raid6int32.o \
>@@ -32,6 +33,7 @@ obj-$(CONFIG_BLK_DEV_DM)	+= dm-mod.o
> obj-$(CONFIG_DM_CRYPT)		+= dm-crypt.o
> obj-$(CONFIG_DM_MULTIPATH)	+= dm-multipath.o dm-round-robin.o
> obj-$(CONFIG_DM_MULTIPATH_EMC)	+= dm-emc.o
>+obj-$(CONFIG_DM_MULTIPATH_HP)	+= dm-hp-sw.o
> obj-$(CONFIG_DM_SNAPSHOT)	+= dm-snapshot.o
> obj-$(CONFIG_DM_MIRROR)		+= dm-mirror.o
> obj-$(CONFIG_DM_ZERO)		+= dm-zero.o
>Index: linux-2.6.9/drivers/md/dm-mpath-hp-sw.c
>===================================================================
>--- /dev/null
>+++ linux-2.6.9/drivers/md/dm-mpath-hp-sw.c
>@@ -0,0 +1,227 @@
>+/*
>+ * Copyright (C) 2005 Mike Christie, All rights reserved.
>+ * Copyright (C) 2007 Red Hat, Inc. All rights reserved.
>+ * Authors: Mike Christie
>+ *          Dave Wysochanski
>+ *
>+ * This file is released under the GPL.
>+ *
>+ * This module implements the specific path activation code for
>+ * HP StorageWorks and FSC FibreCat Asymmetric (Active/Passive)
>+ * storage arrays.
>+ * These storage arrays have controller-based failover, not
>+ * LUN-based failover.  However, LUN-based failover is the design
>+ * of dm-multipath. Thus, this module is written for LUN-based failover.
>+ */
>+#include <linux/blkdev.h>
>+#include <linux/list.h>
>+#include <linux/types.h>
>+#include <scsi/scsi.h>
>+#include <scsi/scsi_cmnd.h>
>+#include <scsi/scsi_request.h>
>+#include <scsi/scsi_device.h>
>+#include <scsi/scsi_dbg.h>
>+
>+#include "dm.h"
>+#include "dm-hw-handler.h"
>+
>+#define DM_MSG_PREFIX "multipath hp-sw: "
>+#define DM_HP_HWH_NAME "hp-sw"
>+#define DM_HP_HWH_VER "1.0.0"
>+
>+/*
>+ * hp_sw_error_is_retryable - Is an HP-specific check condition retryable?
>+ * @sc: scsi_cmnd completing with error status
>+ *
>+ * Examine error codes of request and determine whether the error is retryable.
>+ * Some error codes are already retried by scsi-ml (see
>+ * scsi_decide_disposition), but some HP specific codes are not.
>+ * The intent of this routine is to supply the logic for the HP specific
>+ * check conditions.
>+ *
>+ * Returns:
>+ *  1 - command completed with retryable error
>+ *  0 - command completed with non-retryable error
>+ *
>+ * Possible optimizations
>+ * 1. More hardware-specific error codes
>+ */
>+static int hp_sw_error_is_retryable(struct scsi_cmnd *sc)
>+{
>+	struct scsi_request *srq = sc->sc_request;
>+
>+	/*
>+	 * NOT_READY is known to be retryable
>+	 * For now we just dump out the sense data and call it retryable
>+	 */
>+	if ((status_byte(sc->result) == CHECK_CONDITION) &&
>+	    (driver_byte(srq->sr_result) & DRIVER_SENSE))
>+			scsi_print_sense("", sc);
>+
>+	/*
>+	 * At this point we don't have complete information about all the error
>+	 * codes from this hardware, so we are just conservative and retry
>+	 * when in doubt.
>+	 */
>+	return 1;
>+}
>+
>+/*
>+ * hp_sw_cmd_done - Completion handler for HP path activation.
>+ * @sc: scsi_cmnd that is completed
>+ *
>+ *  Check sense data, free scsi_request structure, and notify dm that
>+ *  pg initialization has completed.
>+ *
>+ * Context: scsi-ml softirq
>+ *
>+ */
>+static void hp_sw_cmd_done(struct scsi_cmnd * sc)
>+{
>+	struct scsi_request *srq;
>+	struct path *path;
>+	unsigned err_flags = 0;
>+
>+	srq = sc->sc_request;
>+	path = srq->upper_private_data;
>+	if (!sc->result)
>+		goto out;
>+
>+	if (hp_sw_error_is_retryable(sc)) {
>+		err_flags = MP_RETRY;
>+		goto out;
>+	}
>+
>+	DMWARN(DM_MSG_PREFIX "%s path activation fail - error=0x%x",
>+	       path->dev->name, srq->sr_result);
>+	err_flags = MP_FAIL_PATH;
>+
>+out:
>+	scsi_release_request(srq);
>+	dm_pg_init_complete(path, err_flags);
>+}
>+
>+/*
>+ * hp_sw_get_request - Allocate an HP specific path activation request
>+ * @path: path on which request will be sent (needed for request queue)
>+ * @cmd: scsi START_STOP cdb
>+ * @cmd_size: size of 'cmd'
>+ *
>+ * The START command is used for path activation request.
>+ * These arrays are controller-based failover, not LUN based.
>+ * One START command issued to a single path will fail over all
>+ * LUNs for the same controller.
>+ *
>+ * Possible optimizations
>+ * 1. Make timeout configurable
>+ * 2. Preallocate request
>+ */
>+static struct scsi_request *hp_sw_get_request(struct path *path,
>+					      unsigned char *cdb,
>+					      unsigned cdb_size)
>+{
>+	struct scsi_device *sdp;
>+	struct scsi_request *srq;
>+
>+	/*
>+	 * Get scsi_device from sysfs struct device, then allocate the
>+	 * request and fill in the cdb.
>+	 */
>+	sdp = to_scsi_device(path->dev->bdev->bd_disk->driverfs_dev);
>+
>+	srq = scsi_allocate_request(sdp, GFP_ATOMIC);
>+	if (!srq) {
>+		DMERR(DM_MSG_PREFIX "scsi_allocate_request() failed.");
>+		return NULL;
>+	}
>+	srq->upper_private_data = path;
>+
>+	memset(cdb, 0, cdb_size);
>+	cdb[0] = START_STOP;
>+	cdb[4] = 1;
>+
>+	return srq;
>+}
>+
>+/*
>+ * hp_sw_pg_init - HP path activation implementation.
>+ * @hwh: hardware handler specific data
>+ * @bypassed: unused; is the path group bypassed? (see dm-mpath.c)
>+ * @path: path to send initialization command
>+ *
>+ * Send an HP-specific path activation command on 'path'.
>+ * Do not try to optimize in any way, just send the activation command.
>+ * More than one path activation command may be sent to the same controller.
>+ * This seems to work fine for basic failover support.
>+ *
>+ * Possible optimizations
>+ * 1. Detect an in-progress activation request and avoid submitting another one
>+ * 2. Model the controller and only send a single activation request at a time
>+ * 3. Determine the state of a path before sending an activation request
>+ *
>+ * Context: kmpathd (see process_queued_ios() in dm-mpath.c)
>+ */
>+static void hp_sw_pg_init(struct hw_handler *hwh, unsigned bypassed,
>+			  struct path *path)
>+{
>+	struct scsi_request *srq;
>+	unsigned char cdb[MAX_COMMAND_SIZE];
>+
>+	srq = hp_sw_get_request(path, cdb, sizeof(cdb));
>+	if (!srq)
>+		goto retry;
>+
>+	scsi_do_req(srq, (void *) cdb, (void *) NULL, 0,
>+		    hp_sw_cmd_done, 60*HZ, 0);
>+	return;
>+
>+retry:
>+	dm_pg_init_complete(path, MP_RETRY);
>+}
>+
>+static int hp_sw_create(struct hw_handler *hwh, unsigned argc, char **argv)
>+{
>+	return 0;
>+}
>+
>+static void hp_sw_destroy(struct hw_handler *hwh)
>+{
>+}
>+
>+static struct hw_handler_type hp_sw_hwh = {
>+	.name = DM_HP_HWH_NAME,
>+	.module = THIS_MODULE,
>+	.create = hp_sw_create,
>+	.destroy = hp_sw_destroy,
>+	.pg_init = hp_sw_pg_init,
>+};
>+
>+static int __init hp_sw_init(void)
>+{
>+	int r;
>+
>+	r = dm_register_hw_handler(&hp_sw_hwh);
>+	if (r < 0)
>+		DMERR(DM_MSG_PREFIX "register failed %d", r);
>+	else
>+		DMINFO(DM_MSG_PREFIX "version " DM_HP_HWH_VER " loaded");
>+
>+	return r;
>+}
>+
>+static void __exit hp_sw_exit(void)
>+{
>+	int r;
>+
>+	r = dm_unregister_hw_handler(&hp_sw_hwh);
>+	if (r < 0)
>+		DMERR(DM_MSG_PREFIX "unregister failed %d", r);
>+}
>+
>+module_init(hp_sw_init);
>+module_exit(hp_sw_exit);
>+
>+MODULE_DESCRIPTION("DM Multipath HP StorageWorks / FSC FibreCat (A/P) support");
>+MODULE_AUTHOR("Mike Christie, Dave Wysochanski <dm-devel@redhat.com>");
>+MODULE_LICENSE("GPL");
>+MODULE_VERSION(DM_HP_HWH_VER);
>Index: linux-2.6.9/drivers/md/dm-hw-handler.h
>===================================================================
>--- linux-2.6.9.orig/drivers/md/dm-hw-handler.h
>+++ linux-2.6.9/drivers/md/dm-hw-handler.h
>@@ -57,5 +57,6 @@ unsigned dm_scsi_err_handler(struct hw_h
> #define MP_FAIL_PATH 1
> #define MP_BYPASS_PG 2
> #define MP_ERROR_IO  4	/* Don't retry this I/O */
>+#define MP_RETRY 8
> 
> #endif
>Index: linux-2.6.9/drivers/md/dm-mpath.c
>===================================================================
>--- linux-2.6.9.orig/drivers/md/dm-mpath.c
>+++ linux-2.6.9/drivers/md/dm-mpath.c
>@@ -74,6 +74,8 @@ struct multipath {
> 	unsigned queue_io;		/* Must we queue all I/O? */
> 	unsigned queue_if_no_path;	/* Queue I/O if last path fails? */
> 	unsigned saved_queue_if_no_path;/* Saved state during suspension */
>+	unsigned pg_init_retries;	/* Number of times to retry pg_init */
>+	unsigned pg_init_count;		/* Number of times pg_init called */
> 
> 	struct work_struct process_queued_ios;
> 	struct bio_list queued_ios;
>@@ -228,6 +230,8 @@ static void __switch_pg(struct multipath
> 		m->pg_init_required = 0;
> 		m->queue_io = 0;
> 	}
>+
>+	m->pg_init_count = 0;
> }
> 
> static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg)
>@@ -432,6 +436,7 @@ static void process_queued_ios(void *dat
> 		must_queue = 0;
> 
> 	if (m->pg_init_required && !m->pg_init_in_progress) {
>+		m->pg_init_count++;
> 		m->pg_init_required = 0;
> 		m->pg_init_in_progress = 1;
> 		init_required = 1;
>@@ -695,9 +700,11 @@ static int parse_features(struct arg_set
> {
> 	int r;
> 	unsigned argc;
>+	const char *param_name;
> 
> 	static struct param _params[] = {
>-		{0, 1, ESTR("invalid number of feature args")},
>+		{0, 3, "invalid number of feature args"},
>+		{1, 50, "pg_init_retries must be between 1 and 50"},
> 	};
> 
> 	r = read_param(_params, shift(as), &argc, &ti->error);
>@@ -707,12 +714,28 @@ static int parse_features(struct arg_set
> 	if (!argc)
> 		return 0;
> 
>-	if (!strnicmp(shift(as), MESG_STR("queue_if_no_path")))
>-		return queue_if_no_path(m, 1, 0);
>-	else {
>+	do {
>+		param_name = shift(as);
>+		argc--;
>+
>+		if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) {
>+			r = queue_if_no_path(m, 1, 0);
>+			continue;
>+		}
>+
>+		if (!strnicmp(param_name, MESG_STR("pg_init_retries")) &&
>+		    (argc >= 1)) {
>+			r = read_param(_params + 1, shift(as),
>+				       &m->pg_init_retries, &ti->error);
>+			argc--;
>+			continue;
>+		}
>+
> 		ti->error = "Unrecognised multipath feature request";
>-		return -EINVAL;
>-	}
>+		r = -EINVAL;
>+	} while (argc && !r);
>+
>+	return r;
> }
> 
> static int multipath_ctr(struct dm_target *ti, unsigned int argc,
>@@ -989,6 +1012,26 @@ 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 path *path, unsigned err_flags)
>@@ -998,8 +1041,14 @@ void dm_pg_init_complete(struct path *pa
> 	struct multipath *m = pg->m;
> 	unsigned long flags;
> 
>-	/* We insist on failing the path if the PG is already bypassed. */
>-	if (err_flags && pg->bypassed)
>+	/*
>+	 * 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)
>@@ -1009,7 +1058,7 @@ void dm_pg_init_complete(struct path *pa
> 		bypass_pg(m, pg, 1);
> 
> 	spin_lock_irqsave(&m->lock, flags);
>-	if (err_flags) {
>+	if (err_flags & ~MP_RETRY) {
> 		m->current_pgpath = NULL;
> 		m->current_pg = NULL;
> 	} else if (!m->pg_init_required)
>@@ -1163,11 +1212,15 @@ static int multipath_status(struct dm_ta
> 
> 	/* Features */
> 	if (type == STATUSTYPE_INFO)
>-		DMEMIT("1 %u ", m->queue_size);
>-	else if (m->queue_if_no_path)
>-		DMEMIT("1 queue_if_no_path ");
>-	else
>-		DMEMIT("0 ");
>+		DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count);
>+	else {
>+		DMEMIT("%u ", m->queue_if_no_path +
>+			      (m->pg_init_retries > 0) * 2);
>+		if (m->queue_if_no_path)
>+			DMEMIT("queue_if_no_path ");
>+		if (m->pg_init_retries)
>+			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);

Actions: View

Attachments on bug 195685: 131053 | 131073 | 131074 | 132388 | 140056 | 140721 | 140741 | 142204 | 142207 | 148162 | 148204 | 155210 | 292333 | 292553