MFC r311804: Rewrite CTL statistics in more simple and scalable way.

Instead of collecting statistics for each combination of ports and logical
units, that consumed ~45KB per LU with present number of ports, collect
separate statistics for every port and every logical unit separately, that
consume only 176 bytes per each single LU/port.  This reduces struct
ctl_lun size down to just 6KB.

Also new IOCTL API/ABI does not hardcode number of LUs/ports, and should
allow handling of very large quantities.

Old API is still enabled in stable branches for compatibility reasons.

6 files changed, 242 insertions, 130 deletions

diff --git a/sys/cam/ctl/ctl.c b/sys/cam/ctl/ctl.c
index 8962f03..30082cb 100644
--- a/sys/cam/ctl/ctl.c
+++ b/sys/cam/ctl/ctl.c
@@ -1,7 +1,7 @@
 /*-
  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
  * Copyright (c) 2012 The FreeBSD Foundation
- * Copyright (c) 2015 Alexander Motin <mav@FreeBSD.org>
+ * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
  * All rights reserved.
  *
  * Portions of this software were developed by Edward Tomasz Napierala
@@ -2558,6 +2558,7 @@ ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 	  struct thread *td)
 {
 	struct ctl_softc *softc = dev->si_drv1;
+	struct ctl_port *port;
 	struct ctl_lun *lun;
 	int retval;
 
@@ -2769,6 +2770,7 @@ ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 #endif /* CTL_IO_DELAY */
 		break;
 	}
+#ifdef CTL_LEGACY_STATS
 	case CTL_GETSTATS: {
 		struct ctl_stats *stats = (struct ctl_stats *)addr;
 		int i;
@@ -2781,26 +2783,26 @@ ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 		stats->status = CTL_SS_OK;
 		stats->fill_len = 0;
 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
-			if (stats->fill_len + sizeof(lun->stats) >
+			if (stats->fill_len + sizeof(lun->legacy_stats) >
 			    stats->alloc_len) {
 				stats->status = CTL_SS_NEED_MORE_SPACE;
 				break;
 			}
-			retval = copyout(&lun->stats, &stats->lun_stats[i++],
-					 sizeof(lun->stats));
+			retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
+					 sizeof(lun->legacy_stats));
 			if (retval != 0)
 				break;
-			stats->fill_len += sizeof(lun->stats);
+			stats->fill_len += sizeof(lun->legacy_stats);
 		}
 		stats->num_luns = softc->num_luns;
-#ifdef CTL_TIME_IO
-		stats->flags = CTL_STATS_FLAG_TIME_VALID;
-#else
 		stats->flags = CTL_STATS_FLAG_NONE;
+#ifdef CTL_TIME_IO
+		stats->flags |= CTL_STATS_FLAG_TIME_VALID;
 #endif
 		getnanouptime(&stats->timestamp);
 		break;
 	}
+#endif /* CTL_LEGACY_STATS */
 	case CTL_ERROR_INJECT: {
 		struct ctl_error_desc *err_desc, *new_err_desc;
 
@@ -3388,6 +3390,72 @@ ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 			ctl_isc_announce_port(port);
 		break;
 	}
+	case CTL_GET_LUN_STATS: {
+		struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
+		int i;
+
+		/*
+		 * XXX KDM no locking here.  If the LUN list changes,
+		 * things can blow up.
+		 */
+		i = 0;
+		stats->status = CTL_SS_OK;
+		stats->fill_len = 0;
+		STAILQ_FOREACH(lun, &softc->lun_list, links) {
+			if (lun->lun < stats->first_item)
+				continue;
+			if (stats->fill_len + sizeof(lun->stats) >
+			    stats->alloc_len) {
+				stats->status = CTL_SS_NEED_MORE_SPACE;
+				break;
+			}
+			retval = copyout(&lun->stats, &stats->stats[i++],
+					 sizeof(lun->stats));
+			if (retval != 0)
+				break;
+			stats->fill_len += sizeof(lun->stats);
+		}
+		stats->num_items = softc->num_luns;
+		stats->flags = CTL_STATS_FLAG_NONE;
+#ifdef CTL_TIME_IO
+		stats->flags |= CTL_STATS_FLAG_TIME_VALID;
+#endif
+		getnanouptime(&stats->timestamp);
+		break;
+	}
+	case CTL_GET_PORT_STATS: {
+		struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
+		int i;
+
+		/*
+		 * XXX KDM no locking here.  If the LUN list changes,
+		 * things can blow up.
+		 */
+		i = 0;
+		stats->status = CTL_SS_OK;
+		stats->fill_len = 0;
+		STAILQ_FOREACH(port, &softc->port_list, links) {
+			if (port->targ_port < stats->first_item)
+				continue;
+			if (stats->fill_len + sizeof(port->stats) >
+			    stats->alloc_len) {
+				stats->status = CTL_SS_NEED_MORE_SPACE;
+				break;
+			}
+			retval = copyout(&port->stats, &stats->stats[i++],
+					 sizeof(port->stats));
+			if (retval != 0)
+				break;
+			stats->fill_len += sizeof(port->stats);
+		}
+		stats->num_items = softc->num_ports;
+		stats->flags = CTL_STATS_FLAG_NONE;
+#ifdef CTL_TIME_IO
+		stats->flags |= CTL_STATS_FLAG_TIME_VALID;
+#endif
+		getnanouptime(&stats->timestamp);
+		break;
+	}
 	default: {
 		/* XXX KDM should we fix this? */
 #if 0
@@ -4382,7 +4450,7 @@ ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
 	struct scsi_vpd_id_descriptor *desc;
 	struct scsi_vpd_id_t10 *t10id;
 	const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
-	int lun_number, i, lun_malloced;
+	int lun_number, lun_malloced;
 	int devidlen, idlen1, idlen2 = 0, len;
 
 	if (be_lun == NULL)
@@ -4604,13 +4672,16 @@ ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
 	ctl_softc->num_luns++;
 
 	/* Setup statistics gathering */
-	lun->stats.device_type = be_lun->lun_type;
-	lun->stats.lun_number = lun_number;
-	lun->stats.blocksize = be_lun->blocksize;
+#ifdef CTL_LEGACY_STATS
+	lun->legacy_stats.device_type = be_lun->lun_type;
+	lun->legacy_stats.lun_number = lun_number;
+	lun->legacy_stats.blocksize = be_lun->blocksize;
 	if (be_lun->blocksize == 0)
-		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
-	for (i = 0;i < CTL_MAX_PORTS;i++)
-		lun->stats.ports[i].targ_port = i;
+		lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
+	for (len = 0; len < CTL_MAX_PORTS; len++)
+		lun->legacy_stats.ports[len].targ_port = len;
+#endif /* CTL_LEGACY_STATS */
+	lun->stats.item = lun_number;
 
 	mtx_unlock(&ctl_softc->ctl_lock);
 
@@ -6676,9 +6747,7 @@ ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
 {
 	struct ctl_lun *lun = CTL_LUN(ctsio);
 	struct stat_page *data;
-	uint64_t rn, wn, rb, wb;
-	struct bintime rt, wt;
-	int i;
+	struct bintime *t;
 
 	data = (struct stat_page *)page_index->page_data;
 
@@ -6686,28 +6755,21 @@ ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
 	data->sap.hdr.param_control = SLP_LBIN;
 	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
 	    sizeof(struct scsi_log_param_header);
-	rn = wn = rb = wb = 0;
-	bintime_clear(&rt);
-	bintime_clear(&wt);
-	for (i = 0; i < CTL_MAX_PORTS; i++) {
-		rn += lun->stats.ports[i].operations[CTL_STATS_READ];
-		wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
-		rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
-		wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
-		bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
-		bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
-	}
-	scsi_u64to8b(rn, data->sap.read_num);
-	scsi_u64to8b(wn, data->sap.write_num);
-	if (lun->stats.blocksize > 0) {
-		scsi_u64to8b(wb / lun->stats.blocksize,
-		    data->sap.recvieved_lba);
-		scsi_u64to8b(rb / lun->stats.blocksize,
-		    data->sap.transmitted_lba);
-	}
-	scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
+	scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
+	    data->sap.read_num);
+	scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
+	    data->sap.write_num);
+	if (lun->be_lun->blocksize > 0) {
+		scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
+		    lun->be_lun->blocksize, data->sap.recvieved_lba);
+		scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
+		    lun->be_lun->blocksize, data->sap.transmitted_lba);
+	}
+	t = &lun->stats.time[CTL_STATS_READ];
+	scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
 	    data->sap.read_int);
-	scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
+	t = &lun->stats.time[CTL_STATS_WRITE];
+	scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
 	    data->sap.write_int);
 	scsi_u64to8b(0, data->sap.weighted_num);
 	scsi_u64to8b(0, data->sap.weighted_int);
@@ -13044,13 +13106,13 @@ static void
 ctl_process_done(union ctl_io *io)
 {
 	struct ctl_softc *softc = CTL_SOFTC(io);
+	struct ctl_port *port = CTL_PORT(io);
 	struct ctl_lun *lun = CTL_LUN(io);
 	void (*fe_done)(union ctl_io *io);
 	union ctl_ha_msg msg;
-	uint32_t targ_port = io->io_hdr.nexus.targ_port;
 
 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
-	fe_done = softc->ctl_ports[targ_port]->fe_done;
+	fe_done = port->fe_done;
 
 #ifdef CTL_TIME_IO
 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
@@ -13153,11 +13215,13 @@ ctl_process_done(union ctl_io *io)
 	 */
 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
 	    io->io_hdr.io_type == CTL_IO_SCSI) {
-#ifdef CTL_TIME_IO
-		struct bintime cur_bt;
-#endif
 		int type;
+#ifdef CTL_TIME_IO
+		struct bintime bt;
 
+		getbinuptime(&bt);
+		bintime_sub(&bt, &io->io_hdr.start_bt);
+#endif
 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
 		    CTL_FLAG_DATA_IN)
 			type = CTL_STATS_READ;
@@ -13167,18 +13231,38 @@ ctl_process_done(union ctl_io *io)
 		else
 			type = CTL_STATS_NO_IO;
 
-		lun->stats.ports[targ_port].bytes[type] +=
+#ifdef CTL_LEGACY_STATS
+		uint32_t targ_port = port->targ_port;
+		lun->legacy_stats.ports[targ_port].bytes[type] +=
 		    io->scsiio.kern_total_len;
-		lun->stats.ports[targ_port].operations[type]++;
+		lun->legacy_stats.ports[targ_port].operations[type] ++;
+		lun->legacy_stats.ports[targ_port].num_dmas[type] +=
+		    io->io_hdr.num_dmas;
 #ifdef CTL_TIME_IO
-		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
+		bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
 		   &io->io_hdr.dma_bt);
-		getbinuptime(&cur_bt);
-		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
-		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
+		bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
+		    &bt);
 #endif
-		lun->stats.ports[targ_port].num_dmas[type] +=
-		    io->io_hdr.num_dmas;
+#endif /* CTL_LEGACY_STATS */
+
+		lun->stats.bytes[type] += io->scsiio.kern_total_len;
+		lun->stats.operations[type] ++;
+		lun->stats.dmas[type] += io->io_hdr.num_dmas;
+#ifdef CTL_TIME_IO
+		bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
+		bintime_add(&lun->stats.time[type], &bt);
+#endif
+
+		mtx_lock(&port->port_lock);
+		port->stats.bytes[type] += io->scsiio.kern_total_len;
+		port->stats.operations[type] ++;
+		port->stats.dmas[type] += io->io_hdr.num_dmas;
+#ifdef CTL_TIME_IO
+		bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
+		bintime_add(&port->stats.time[type], &bt);
+#endif
+		mtx_unlock(&port->port_lock);
 	}
 
 	/*
diff --git a/sys/cam/ctl/ctl_backend.h b/sys/cam/ctl/ctl_backend.h
index 4177e2d..d8ada83 100644
--- a/sys/cam/ctl/ctl_backend.h
+++ b/sys/cam/ctl/ctl_backend.h
@@ -1,6 +1,6 @@
 /*-
  * Copyright (c) 2003 Silicon Graphics International Corp.
- * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org>
+ * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -40,54 +40,7 @@
 #ifndef	_CTL_BACKEND_H_
 #define	_CTL_BACKEND_H_
 
-/*
- * XXX KDM move this to another header file?
- */
-#define	CTL_BE_NAME_LEN		32
-
-/*
- * The ID_REQ flag is used to say that the caller has requested a
- * particular LUN ID in the req_lun_id field.  If we cannot allocate that
- * LUN ID, the ctl_add_lun() call will fail.
- *
- * The STOPPED flag tells us that the LUN should default to the powered
- * off state.  It will return 0x04,0x02 until it is powered up.  ("Logical
- * unit not ready, initializing command required.")
- *
- * The NO_MEDIA flag tells us that the LUN has no media inserted.
- *
- * The PRIMARY flag tells us that this LUN is registered as a Primary LUN
- * which is accessible via the Master shelf controller in an HA. This flag
- * being set indicates a Primary LUN. This flag being reset represents a
- * Secondary LUN controlled by the Secondary controller in an HA
- * configuration. Flag is applicable at this time to T_DIRECT types. 
- *
- * The SERIAL_NUM flag tells us that the serial_num field is filled in and
- * valid for use in SCSI INQUIRY VPD page 0x80.
- *
- * The DEVID flag tells us that the device_id field is filled in and
- * valid for use in SCSI INQUIRY VPD page 0x83.
- *
- * The DEV_TYPE flag tells us that the device_type field is filled in.
- *
- * The EJECTED flag tells us that the removable LUN has tray open.
- *
- * The UNMAP flag tells us that this LUN supports UNMAP.
- *
- * The OFFLINE flag tells us that this LUN can not access backing store.
- */
-typedef enum {
-	CTL_LUN_FLAG_ID_REQ		= 0x01,
-	CTL_LUN_FLAG_STOPPED		= 0x02,
-	CTL_LUN_FLAG_NO_MEDIA		= 0x04,
-	CTL_LUN_FLAG_PRIMARY		= 0x08,
-	CTL_LUN_FLAG_SERIAL_NUM		= 0x10,
-	CTL_LUN_FLAG_DEVID		= 0x20,
-	CTL_LUN_FLAG_DEV_TYPE		= 0x40,
-	CTL_LUN_FLAG_UNMAP		= 0x80,
-	CTL_LUN_FLAG_EJECTED		= 0x100,
-	CTL_LUN_FLAG_READONLY		= 0x200
-} ctl_backend_lun_flags;
+#include <cam/ctl/ctl_ioctl.h>
 
 typedef enum {
 	CTL_LUN_SERSEQ_OFF,
diff --git a/sys/cam/ctl/ctl_frontend.c b/sys/cam/ctl/ctl_frontend.c
index 97f4e0b..03ee9da 100644
--- a/sys/cam/ctl/ctl_frontend.c
+++ b/sys/cam/ctl/ctl_frontend.c
@@ -1,5 +1,6 @@
 /*-
  * Copyright (c) 2003 Silicon Graphics International Corp.
+ * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -192,13 +193,14 @@ error:
 		mtx_unlock(&softc->ctl_lock);
 		return (retval);
 	}
+	port->targ_port = port_num;
 	port->ctl_pool_ref = pool;
-
 	if (port->options.stqh_first == NULL)
 		STAILQ_INIT(&port->options);
+	port->stats.item = port_num;
+	mtx_init(&port->port_lock, "CTL port", NULL, MTX_DEF);
 
 	mtx_lock(&softc->ctl_lock);
-	port->targ_port = port_num;
 	STAILQ_INSERT_TAIL(&port->frontend->port_list, port, fe_links);
 	for (tport = NULL, nport = STAILQ_FIRST(&softc->port_list);
 	    nport != NULL && nport->targ_port < port_num;
@@ -218,17 +220,11 @@ int
 ctl_port_deregister(struct ctl_port *port)
 {
 	struct ctl_softc *softc = port->ctl_softc;
-	struct ctl_io_pool *pool;
-	int retval, i;
-
-	retval = 0;
-
-	pool = (struct ctl_io_pool *)port->ctl_pool_ref;
+	struct ctl_io_pool *pool = (struct ctl_io_pool *)port->ctl_pool_ref;
+	int i;
 
-	if (port->targ_port == -1) {
-		retval = 1;
-		goto bailout;
-	}
+	if (port->targ_port == -1)
+		return (1);
 
 	mtx_lock(&softc->ctl_lock);
 	STAILQ_REMOVE(&softc->port_list, port, ctl_port, links);
@@ -251,9 +247,9 @@ ctl_port_deregister(struct ctl_port *port)
 	for (i = 0; i < port->max_initiators; i++)
 		free(port->wwpn_iid[i].name, M_CTL);
 	free(port->wwpn_iid, M_CTL);
+	mtx_destroy(&port->port_lock);
 
-bailout:
-	return (retval);
+	return (0);
 }
 
 void
diff --git a/sys/cam/ctl/ctl_frontend.h b/sys/cam/ctl/ctl_frontend.h
index 3e42ac3..37e6824 100644
--- a/sys/cam/ctl/ctl_frontend.h
+++ b/sys/cam/ctl/ctl_frontend.h
@@ -1,5 +1,6 @@
 /*-
  * Copyright (c) 2003 Silicon Graphics International Corp.
+ * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -39,6 +40,8 @@
 #ifndef	_CTL_FRONTEND_H_
 #define	_CTL_FRONTEND_H_
 
+#include <cam/ctl/ctl_ioctl.h>
+
 typedef enum {
 	CTL_PORT_STATUS_NONE		= 0x00,
 	CTL_PORT_STATUS_ONLINE		= 0x01,
@@ -243,6 +246,8 @@ struct ctl_port {
 	struct ctl_devid *port_devid;		/* passed to CTL */
 	struct ctl_devid *target_devid;		/* passed to CTL */
 	struct ctl_devid *init_devid;		/* passed to CTL */
+	struct ctl_io_stats stats;		/* used by CTL */
+	struct mtx	port_lock;		/* used by CTL */
 	STAILQ_ENTRY(ctl_port) fe_links;	/* used by CTL */
 	STAILQ_ENTRY(ctl_port) links;		/* used by CTL */
 };
diff --git a/sys/cam/ctl/ctl_ioctl.h b/sys/cam/ctl/ctl_ioctl.h
index e8f6b87..ee4ce18 100644
--- a/sys/cam/ctl/ctl_ioctl.h
+++ b/sys/cam/ctl/ctl_ioctl.h
@@ -1,6 +1,7 @@
 /*-
  * Copyright (c) 2003 Silicon Graphics International Corp.
  * Copyright (c) 2011 Spectra Logic Corporation
+ * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -80,6 +81,9 @@
 /* Hopefully this won't conflict with new misc devices that pop up */
 #define	CTL_MINOR	225
 
+/* Legacy statistics accumulated for every port for every LU. */
+#define CTL_LEGACY_STATS	1
+
 typedef enum {
 	CTL_DELAY_TYPE_NONE,
 	CTL_DELAY_TYPE_CONT,
@@ -117,6 +121,18 @@ typedef enum {
 #define	CTL_STATS_NUM_TYPES	3
 
 typedef enum {
+	CTL_SS_OK,
+	CTL_SS_NEED_MORE_SPACE,
+	CTL_SS_ERROR
+} ctl_stats_status;
+
+typedef enum {
+	CTL_STATS_FLAG_NONE		= 0x00,
+	CTL_STATS_FLAG_TIME_VALID	= 0x01
+} ctl_stats_flags;
+
+#ifdef CTL_LEGACY_STATS
+typedef enum {
 	CTL_LUN_STATS_NO_BLOCKSIZE	= 0x01
 } ctl_lun_stats_flags;
 
@@ -137,17 +153,6 @@ struct ctl_lun_io_stats {
 	struct ctl_lun_io_port_stats	ports[CTL_MAX_PORTS];
 };
 
-typedef enum {
-	CTL_SS_OK,
-	CTL_SS_NEED_MORE_SPACE,
-	CTL_SS_ERROR
-} ctl_stats_status;
-
-typedef enum {
-	CTL_STATS_FLAG_NONE		= 0x00,
-	CTL_STATS_FLAG_TIME_VALID	= 0x01
-} ctl_stats_flags;
-
 struct ctl_stats {
 	int			alloc_len;	/* passed to kernel */
 	struct ctl_lun_io_stats	*lun_stats;	/* passed to/from kernel */
@@ -157,6 +162,27 @@ struct ctl_stats {
 	ctl_stats_flags		flags;		/* passed to userland */
 	struct timespec		timestamp;	/* passed to userland */
 };
+#endif /* CTL_LEGACY_STATS */
+
+struct ctl_io_stats {
+	uint32_t			item;
+	uint64_t			bytes[CTL_STATS_NUM_TYPES];
+	uint64_t			operations[CTL_STATS_NUM_TYPES];
+	uint64_t			dmas[CTL_STATS_NUM_TYPES];
+	struct bintime			time[CTL_STATS_NUM_TYPES];
+	struct bintime			dma_time[CTL_STATS_NUM_TYPES];
+};
+
+struct ctl_get_io_stats {
+	struct ctl_io_stats	*stats;		/* passed to/from kernel */
+	size_t			alloc_len;	/* passed to kernel */
+	size_t			fill_len;	/* passed to userland */
+	int			first_item;	/* passed to kernel */
+	int			num_items;	/* passed to userland */
+	ctl_stats_status	status;		/* passed to userland */
+	ctl_stats_flags		flags;		/* passed to userland */
+	struct timespec		timestamp;	/* passed to userland */
+};
 
 /*
  * The types of errors that can be injected:
@@ -342,12 +368,54 @@ typedef enum {
 	CTL_LUNREQ_MODIFY,
 } ctl_lunreq_type;
 
+/*
+ * The ID_REQ flag is used to say that the caller has requested a
+ * particular LUN ID in the req_lun_id field.  If we cannot allocate that
+ * LUN ID, the ctl_add_lun() call will fail.
+ *
+ * The STOPPED flag tells us that the LUN should default to the powered
+ * off state.  It will return 0x04,0x02 until it is powered up.  ("Logical
+ * unit not ready, initializing command required.")
+ *
+ * The NO_MEDIA flag tells us that the LUN has no media inserted.
+ *
+ * The PRIMARY flag tells us that this LUN is registered as a Primary LUN
+ * which is accessible via the Master shelf controller in an HA. This flag
+ * being set indicates a Primary LUN. This flag being reset represents a
+ * Secondary LUN controlled by the Secondary controller in an HA
+ * configuration. Flag is applicable at this time to T_DIRECT types. 
+ *
+ * The SERIAL_NUM flag tells us that the serial_num field is filled in and
+ * valid for use in SCSI INQUIRY VPD page 0x80.
+ *
+ * The DEVID flag tells us that the device_id field is filled in and
+ * valid for use in SCSI INQUIRY VPD page 0x83.
+ *
+ * The DEV_TYPE flag tells us that the device_type field is filled in.
+ *
+ * The EJECTED flag tells us that the removable LUN has tray open.
+ *
+ * The UNMAP flag tells us that this LUN supports UNMAP.
+ *
+ * The OFFLINE flag tells us that this LUN can not access backing store.
+ */
+typedef enum {
+	CTL_LUN_FLAG_ID_REQ		= 0x01,
+	CTL_LUN_FLAG_STOPPED		= 0x02,
+	CTL_LUN_FLAG_NO_MEDIA		= 0x04,
+	CTL_LUN_FLAG_PRIMARY		= 0x08,
+	CTL_LUN_FLAG_SERIAL_NUM		= 0x10,
+	CTL_LUN_FLAG_DEVID		= 0x20,
+	CTL_LUN_FLAG_DEV_TYPE		= 0x40,
+	CTL_LUN_FLAG_UNMAP		= 0x80,
+	CTL_LUN_FLAG_EJECTED		= 0x100,
+	CTL_LUN_FLAG_READONLY		= 0x200
+} ctl_backend_lun_flags;
 
 /*
  * LUN creation parameters:
  *
- * flags:		Various LUN flags, see ctl_backend.h for a
- *			description of the flag values and meanings.
+ * flags:		Various LUN flags, see above.
  *
  * device_type:		The SCSI device type.  e.g. 0 for Direct Access,
  *			3 for Processor, etc.  Only certain backends may
@@ -465,6 +533,7 @@ union ctl_lunreq_data {
  * kern_be_args:	For kernel use only.
  */
 struct ctl_lun_req {
+#define	CTL_BE_NAME_LEN		32
 	char			backend[CTL_BE_NAME_LEN];
 	ctl_lunreq_type		reqtype;
 	union ctl_lunreq_data	reqdata;
@@ -761,6 +830,8 @@ struct ctl_lun_map {
 #define	CTL_PORT_REQ		_IOWR(CTL_MINOR, 0x26, struct ctl_req)
 #define	CTL_PORT_LIST		_IOWR(CTL_MINOR, 0x27, struct ctl_lun_list)
 #define	CTL_LUN_MAP		_IOW(CTL_MINOR, 0x28, struct ctl_lun_map)
+#define	CTL_GET_LUN_STATS	_IOWR(CTL_MINOR, 0x29, struct ctl_get_io_stats)
+#define	CTL_GET_PORT_STATS	_IOWR(CTL_MINOR, 0x2a, struct ctl_get_io_stats)
 
 #endif /* _CTL_IOCTL_H_ */
 
diff --git a/sys/cam/ctl/ctl_private.h b/sys/cam/ctl/ctl_private.h
index 5b71f9a..3f46769 100644
--- a/sys/cam/ctl/ctl_private.h
+++ b/sys/cam/ctl/ctl_private.h
@@ -1,6 +1,6 @@
 /*-
  * Copyright (c) 2003, 2004, 2005, 2008 Silicon Graphics International Corp.
- * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org>
+ * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -404,7 +404,10 @@ struct ctl_lun {
 	struct callout			ie_callout;	/* INTERVAL TIMER */
 	struct ctl_mode_pages		mode_pages;
 	struct ctl_log_pages		log_pages;
-	struct ctl_lun_io_stats		stats;
+#ifdef CTL_LEGACY_STATS
+	struct ctl_lun_io_stats		legacy_stats;
+#endif /* CTL_LEGACY_STATS */
+	struct ctl_io_stats		stats;
 	uint32_t			res_idx;
 	uint32_t			pr_generation;
 	uint64_t			*pr_keys[CTL_MAX_PORTS];