path: root/sys/cam/ctl/ctl.c

/*-
 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
 * Copyright (c) 2012 The FreeBSD Foundation
 * Copyright (c) 2015 Alexander Motin <mav@FreeBSD.org>
 * All rights reserved.
 *
 * Portions of this software were developed by Edward Tomasz Napierala
 * under sponsorship from the FreeBSD Foundation.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions, and the following disclaimer,
 *    without modification.
 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
 *    substantially similar to the "NO WARRANTY" disclaimer below
 *    ("Disclaimer") and any redistribution must be conditioned upon
 *    including a substantially similar Disclaimer requirement for further
 *    binary redistribution.
 *
 * NO WARRANTY
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGES.
 *
 * $Id$
 */
/*
 * CAM Target Layer, a SCSI device emulation subsystem.
 *
 * Author: Ken Merry <ken@FreeBSD.org>
 */

#define _CTL_C

#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/ctype.h>
#include <sys/kernel.h>
#include <sys/types.h>
#include <sys/kthread.h>
#include <sys/bio.h>
#include <sys/fcntl.h>
#include <sys/lock.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/condvar.h>
#include <sys/malloc.h>
#include <sys/conf.h>
#include <sys/ioccom.h>
#include <sys/queue.h>
#include <sys/sbuf.h>
#include <sys/smp.h>
#include <sys/endian.h>
#include <sys/sysctl.h>
#include <vm/uma.h>

#include <cam/cam.h>
#include <cam/scsi/scsi_all.h>
#include <cam/scsi/scsi_da.h>
#include <cam/ctl/ctl_io.h>
#include <cam/ctl/ctl.h>
#include <cam/ctl/ctl_frontend.h>
#include <cam/ctl/ctl_util.h>
#include <cam/ctl/ctl_backend.h>
#include <cam/ctl/ctl_ioctl.h>
#include <cam/ctl/ctl_ha.h>
#include <cam/ctl/ctl_private.h>
#include <cam/ctl/ctl_debug.h>
#include <cam/ctl/ctl_scsi_all.h>
#include <cam/ctl/ctl_error.h>

struct ctl_softc *control_softc = NULL;

/*
 * Template mode pages.
 */

/*
 * Note that these are default values only.  The actual values will be
 * filled in when the user does a mode sense.
 */
const static struct copan_debugconf_subpage debugconf_page_default = {
	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
	DBGCNF_SUBPAGE_CODE,		/* subpage */
	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
	DBGCNF_VERSION,			/* page_version */
	{CTL_TIME_IO_DEFAULT_SECS>>8,
	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
};

const static struct copan_debugconf_subpage debugconf_page_changeable = {
	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
	DBGCNF_SUBPAGE_CODE,		/* subpage */
	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
	0,				/* page_version */
	{0xff,0xff},			/* ctl_time_io_secs */
};

const static struct scsi_da_rw_recovery_page rw_er_page_default = {
	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
	/*read_retry_count*/0,
	/*correction_span*/0,
	/*head_offset_count*/0,
	/*data_strobe_offset_cnt*/0,
	/*byte8*/SMS_RWER_LBPERE,
	/*write_retry_count*/0,
	/*reserved2*/0,
	/*recovery_time_limit*/{0, 0},
};

const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
	/*byte3*/0,
	/*read_retry_count*/0,
	/*correction_span*/0,
	/*head_offset_count*/0,
	/*data_strobe_offset_cnt*/0,
	/*byte8*/0,
	/*write_retry_count*/0,
	/*reserved2*/0,
	/*recovery_time_limit*/{0, 0},
};

const static struct scsi_format_page format_page_default = {
	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
	/*page_length*/sizeof(struct scsi_format_page) - 2,
	/*tracks_per_zone*/ {0, 0},
	/*alt_sectors_per_zone*/ {0, 0},
	/*alt_tracks_per_zone*/ {0, 0},
	/*alt_tracks_per_lun*/ {0, 0},
	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
	/*bytes_per_sector*/ {0, 0},
	/*interleave*/ {0, 0},
	/*track_skew*/ {0, 0},
	/*cylinder_skew*/ {0, 0},
	/*flags*/ SFP_HSEC,
	/*reserved*/ {0, 0, 0}
};

const static struct scsi_format_page format_page_changeable = {
	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
	/*page_length*/sizeof(struct scsi_format_page) - 2,
	/*tracks_per_zone*/ {0, 0},
	/*alt_sectors_per_zone*/ {0, 0},
	/*alt_tracks_per_zone*/ {0, 0},
	/*alt_tracks_per_lun*/ {0, 0},
	/*sectors_per_track*/ {0, 0},
	/*bytes_per_sector*/ {0, 0},
	/*interleave*/ {0, 0},
	/*track_skew*/ {0, 0},
	/*cylinder_skew*/ {0, 0},
	/*flags*/ 0,
	/*reserved*/ {0, 0, 0}
};

const static struct scsi_rigid_disk_page rigid_disk_page_default = {
	/*page_code*/SMS_RIGID_DISK_PAGE,
	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
	/*cylinders*/ {0, 0, 0},
	/*heads*/ CTL_DEFAULT_HEADS,
	/*start_write_precomp*/ {0, 0, 0},
	/*start_reduced_current*/ {0, 0, 0},
	/*step_rate*/ {0, 0},
	/*landing_zone_cylinder*/ {0, 0, 0},
	/*rpl*/ SRDP_RPL_DISABLED,
	/*rotational_offset*/ 0,
	/*reserved1*/ 0,
	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
			   CTL_DEFAULT_ROTATION_RATE & 0xff},
	/*reserved2*/ {0, 0}
};

const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
	/*page_code*/SMS_RIGID_DISK_PAGE,
	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
	/*cylinders*/ {0, 0, 0},
	/*heads*/ 0,
	/*start_write_precomp*/ {0, 0, 0},
	/*start_reduced_current*/ {0, 0, 0},
	/*step_rate*/ {0, 0},
	/*landing_zone_cylinder*/ {0, 0, 0},
	/*rpl*/ 0,
	/*rotational_offset*/ 0,
	/*reserved1*/ 0,
	/*rotation_rate*/ {0, 0},
	/*reserved2*/ {0, 0}
};

const static struct scsi_caching_page caching_page_default = {
	/*page_code*/SMS_CACHING_PAGE,
	/*page_length*/sizeof(struct scsi_caching_page) - 2,
	/*flags1*/ SCP_DISC | SCP_WCE,
	/*ret_priority*/ 0,
	/*disable_pf_transfer_len*/ {0xff, 0xff},
	/*min_prefetch*/ {0, 0},
	/*max_prefetch*/ {0xff, 0xff},
	/*max_pf_ceiling*/ {0xff, 0xff},
	/*flags2*/ 0,
	/*cache_segments*/ 0,
	/*cache_seg_size*/ {0, 0},
	/*reserved*/ 0,
	/*non_cache_seg_size*/ {0, 0, 0}
};

const static struct scsi_caching_page caching_page_changeable = {
	/*page_code*/SMS_CACHING_PAGE,
	/*page_length*/sizeof(struct scsi_caching_page) - 2,
	/*flags1*/ SCP_WCE | SCP_RCD,
	/*ret_priority*/ 0,
	/*disable_pf_transfer_len*/ {0, 0},
	/*min_prefetch*/ {0, 0},
	/*max_prefetch*/ {0, 0},
	/*max_pf_ceiling*/ {0, 0},
	/*flags2*/ 0,
	/*cache_segments*/ 0,
	/*cache_seg_size*/ {0, 0},
	/*reserved*/ 0,
	/*non_cache_seg_size*/ {0, 0, 0}
};

const static struct scsi_control_page control_page_default = {
	/*page_code*/SMS_CONTROL_MODE_PAGE,
	/*page_length*/sizeof(struct scsi_control_page) - 2,
	/*rlec*/0,
	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
	/*eca_and_aen*/0,
	/*flags4*/SCP_TAS,
	/*aen_holdoff_period*/{0, 0},
	/*busy_timeout_period*/{0, 0},
	/*extended_selftest_completion_time*/{0, 0}
};

const static struct scsi_control_page control_page_changeable = {
	/*page_code*/SMS_CONTROL_MODE_PAGE,
	/*page_length*/sizeof(struct scsi_control_page) - 2,
	/*rlec*/SCP_DSENSE,
	/*queue_flags*/SCP_QUEUE_ALG_MASK,
	/*eca_and_aen*/SCP_SWP,
	/*flags4*/0,
	/*aen_holdoff_period*/{0, 0},
	/*busy_timeout_period*/{0, 0},
	/*extended_selftest_completion_time*/{0, 0}
};

#define CTL_CEM_LEN	(sizeof(struct scsi_control_ext_page) - 4)

const static struct scsi_control_ext_page control_ext_page_default = {
	/*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
	/*subpage_code*/0x01,
	/*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
	/*flags*/0,
	/*prio*/0,
	/*max_sense*/0
};

const static struct scsi_control_ext_page control_ext_page_changeable = {
	/*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
	/*subpage_code*/0x01,
	/*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
	/*flags*/0,
	/*prio*/0,
	/*max_sense*/0
};

const static struct scsi_info_exceptions_page ie_page_default = {
	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
	/*info_flags*/SIEP_FLAGS_DEXCPT,
	/*mrie*/0,
	/*interval_timer*/{0, 0, 0, 0},
	/*report_count*/{0, 0, 0, 0}
};

const static struct scsi_info_exceptions_page ie_page_changeable = {
	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
	/*info_flags*/0,
	/*mrie*/0,
	/*interval_timer*/{0, 0, 0, 0},
	/*report_count*/{0, 0, 0, 0}
};

#define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)

const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
	/*subpage_code*/0x02,
	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
	/*flags*/0,
	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
	/*descr*/{}},
	{{/*flags*/0,
	  /*resource*/0x01,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}},
	 {/*flags*/0,
	  /*resource*/0x02,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}},
	 {/*flags*/0,
	  /*resource*/0xf1,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}},
	 {/*flags*/0,
	  /*resource*/0xf2,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}}
	}
};

const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
	/*subpage_code*/0x02,
	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
	/*flags*/0,
	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
	/*descr*/{}},
	{{/*flags*/0,
	  /*resource*/0,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}},
	 {/*flags*/0,
	  /*resource*/0,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}},
	 {/*flags*/0,
	  /*resource*/0,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}},
	 {/*flags*/0,
	  /*resource*/0,
	  /*reserved*/{0, 0},
	  /*count*/{0, 0, 0, 0}}
	}
};

SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
static int worker_threads = -1;
SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
    &worker_threads, 1, "Number of worker threads");
static int ctl_debug = CTL_DEBUG_NONE;
SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
    &ctl_debug, 0, "Enabled debug flags");

/*
 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 */
#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10

static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
				  int param);
static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
static int ctl_init(void);
void ctl_shutdown(void);
static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
			      struct ctl_ooa *ooa_hdr,
			      struct ctl_ooa_entry *kern_entries);
static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
		     struct thread *td);
static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
			 struct ctl_be_lun *be_lun);
static int ctl_free_lun(struct ctl_lun *lun);
static void ctl_create_lun(struct ctl_be_lun *be_lun);
static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);

static int ctl_do_mode_select(union ctl_io *io);
static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
			   uint64_t res_key, uint64_t sa_res_key,
			   uint8_t type, uint32_t residx,
			   struct ctl_scsiio *ctsio,
			   struct scsi_per_res_out *cdb,
			   struct scsi_per_res_out_parms* param);
static void ctl_pro_preempt_other(struct ctl_lun *lun,
				  union ctl_ha_msg *msg);
static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
					 int alloc_len);
static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
					 int alloc_len);
static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
    bool seq);
static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
    union ctl_io *pending_io, union ctl_io *ooa_io);
static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
				union ctl_io *starting_io);
static int ctl_check_blocked(struct ctl_lun *lun);
static int ctl_scsiio_lun_check(struct ctl_lun *lun,
				const struct ctl_cmd_entry *entry,
				struct ctl_scsiio *ctsio);
static void ctl_failover_lun(struct ctl_lun *lun);
static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
			       struct ctl_scsiio *ctsio);
static int ctl_scsiio(struct ctl_scsiio *ctsio);

static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
			    ctl_ua_type ua_type);
static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
			 ctl_ua_type ua_type);
static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
static int ctl_abort_task(union ctl_io *io);
static int ctl_abort_task_set(union ctl_io *io);
static int ctl_query_task(union ctl_io *io, int task_set);
static int ctl_i_t_nexus_reset(union ctl_io *io);
static int ctl_query_async_event(union ctl_io *io);
static void ctl_run_task(union ctl_io *io);
#ifdef CTL_IO_DELAY
static void ctl_datamove_timer_wakeup(void *arg);
static void ctl_done_timer_wakeup(void *arg);
#endif /* CTL_IO_DELAY */

static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
static void ctl_datamove_remote_write(union ctl_io *io);
static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
				    ctl_ha_dt_cb callback);
static void ctl_datamove_remote_read(union ctl_io *io);
static void ctl_datamove_remote(union ctl_io *io);
static void ctl_process_done(union ctl_io *io);
static void ctl_lun_thread(void *arg);
static void ctl_thresh_thread(void *arg);
static void ctl_work_thread(void *arg);
static void ctl_enqueue_incoming(union ctl_io *io);
static void ctl_enqueue_rtr(union ctl_io *io);
static void ctl_enqueue_done(union ctl_io *io);
static void ctl_enqueue_isc(union ctl_io *io);
static const struct ctl_cmd_entry *
    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
static const struct ctl_cmd_entry *
    ctl_validate_command(struct ctl_scsiio *ctsio);
static int ctl_cmd_applicable(uint8_t lun_type,
    const struct ctl_cmd_entry *entry);

static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);

/*
 * Load the serialization table.  This isn't very pretty, but is probably
 * the easiest way to do it.
 */
#include "ctl_ser_table.c"

/*
 * We only need to define open, close and ioctl routines for this driver.
 */
static struct cdevsw ctl_cdevsw = {
	.d_version =	D_VERSION,
	.d_flags =	0,
	.d_open =	ctl_open,
	.d_close =	ctl_close,
	.d_ioctl =	ctl_ioctl,
	.d_name =	"ctl",
};


MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");

static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);

static moduledata_t ctl_moduledata = {
	"ctl",
	ctl_module_event_handler,
	NULL
};

DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
MODULE_VERSION(ctl, 1);

static struct ctl_frontend ha_frontend =
{
	.name = "ha",
};

static void
ctl_ha_datamove(union ctl_io *io)
{
	struct ctl_lun *lun;
	struct ctl_sg_entry *sgl;
	union ctl_ha_msg msg;
	uint32_t sg_entries_sent;
	int do_sg_copy, i, j;

	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
	memset(&msg.dt, 0, sizeof(msg.dt));
	msg.hdr.msg_type = CTL_MSG_DATAMOVE;
	msg.hdr.original_sc = io->io_hdr.original_sc;
	msg.hdr.serializing_sc = io;
	msg.hdr.nexus = io->io_hdr.nexus;
	msg.hdr.status = io->io_hdr.status;
	msg.dt.flags = io->io_hdr.flags;

	/*
	 * We convert everything into a S/G list here.  We can't
	 * pass by reference, only by value between controllers.
	 * So we can't pass a pointer to the S/G list, only as many
	 * S/G entries as we can fit in here.  If it's possible for
	 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
	 * then we need to break this up into multiple transfers.
	 */
	if (io->scsiio.kern_sg_entries == 0) {
		msg.dt.kern_sg_entries = 1;
#if 0
		if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
			msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
		} else {
			/* XXX KDM use busdma here! */
			msg.dt.sg_list[0].addr =
			    (void *)vtophys(io->scsiio.kern_data_ptr);
		}
#else
		KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
		    ("HA does not support BUS_ADDR"));
		msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
#endif
		msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
		do_sg_copy = 0;
	} else {
		msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
		do_sg_copy = 1;
	}

	msg.dt.kern_data_len = io->scsiio.kern_data_len;
	msg.dt.kern_total_len = io->scsiio.kern_total_len;
	msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
	msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
	msg.dt.sg_sequence = 0;

	/*
	 * Loop until we've sent all of the S/G entries.  On the
	 * other end, we'll recompose these S/G entries into one
	 * contiguous list before processing.
	 */
	for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
	    msg.dt.sg_sequence++) {
		msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
		    sizeof(msg.dt.sg_list[0])),
		    msg.dt.kern_sg_entries - sg_entries_sent);
		if (do_sg_copy != 0) {
			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
			for (i = sg_entries_sent, j = 0;
			     i < msg.dt.cur_sg_entries; i++, j++) {
#if 0
				if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
					msg.dt.sg_list[j].addr = sgl[i].addr;
				} else {
					/* XXX KDM use busdma here! */
					msg.dt.sg_list[j].addr =
					    (void *)vtophys(sgl[i].addr);
				}
#else
				KASSERT((io->io_hdr.flags &
				    CTL_FLAG_BUS_ADDR) == 0,
				    ("HA does not support BUS_ADDR"));
				msg.dt.sg_list[j].addr = sgl[i].addr;
#endif
				msg.dt.sg_list[j].len = sgl[i].len;
			}
		}

		sg_entries_sent += msg.dt.cur_sg_entries;
		msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
		    sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
		    sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
		    M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
			io->io_hdr.port_status = 31341;
			io->scsiio.be_move_done(io);
			return;
		}
		msg.dt.sent_sg_entries = sg_entries_sent;
	}

	/*
	 * Officially handover the request from us to peer.
	 * If failover has just happened, then we must return error.
	 * If failover happen just after, then it is not our problem.
	 */
	if (lun)
		mtx_lock(&lun->lun_lock);
	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
		if (lun)
			mtx_unlock(&lun->lun_lock);
		io->io_hdr.port_status = 31342;
		io->scsiio.be_move_done(io);
		return;
	}
	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
	io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
	if (lun)
		mtx_unlock(&lun->lun_lock);
}

static void
ctl_ha_done(union ctl_io *io)
{
	union ctl_ha_msg msg;

	if (io->io_hdr.io_type == CTL_IO_SCSI) {
		memset(&msg, 0, sizeof(msg));
		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
		msg.hdr.original_sc = io->io_hdr.original_sc;
		msg.hdr.nexus = io->io_hdr.nexus;
		msg.hdr.status = io->io_hdr.status;
		msg.scsi.scsi_status = io->scsiio.scsi_status;
		msg.scsi.tag_num = io->scsiio.tag_num;
		msg.scsi.tag_type = io->scsiio.tag_type;
		msg.scsi.sense_len = io->scsiio.sense_len;
		msg.scsi.sense_residual = io->scsiio.sense_residual;
		msg.scsi.residual = io->scsiio.residual;
		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
		    io->scsiio.sense_len);
		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
		    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
		    msg.scsi.sense_len, M_WAITOK);
	}
	ctl_free_io(io);
}

static void
ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
			    union ctl_ha_msg *msg_info)
{
	struct ctl_scsiio *ctsio;

	if (msg_info->hdr.original_sc == NULL) {
		printf("%s: original_sc == NULL!\n", __func__);
		/* XXX KDM now what? */
		return;
	}

	ctsio = &msg_info->hdr.original_sc->scsiio;
	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
	ctsio->io_hdr.status = msg_info->hdr.status;
	ctsio->scsi_status = msg_info->scsi.scsi_status;
	ctsio->sense_len = msg_info->scsi.sense_len;
	ctsio->sense_residual = msg_info->scsi.sense_residual;
	ctsio->residual = msg_info->scsi.residual;
	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
	       msg_info->scsi.sense_len);
	ctl_enqueue_isc((union ctl_io *)ctsio);
}

static void
ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
				union ctl_ha_msg *msg_info)
{
	struct ctl_scsiio *ctsio;

	if (msg_info->hdr.serializing_sc == NULL) {
		printf("%s: serializing_sc == NULL!\n", __func__);
		/* XXX KDM now what? */
		return;
	}

	ctsio = &msg_info->hdr.serializing_sc->scsiio;
	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
	ctl_enqueue_isc((union ctl_io *)ctsio);
}

void
ctl_isc_announce_lun(struct ctl_lun *lun)
{
	struct ctl_softc *softc = lun->ctl_softc;
	union ctl_ha_msg *msg;
	struct ctl_ha_msg_lun_pr_key pr_key;
	int i, k;

	if (softc->ha_link != CTL_HA_LINK_ONLINE)
		return;
	mtx_lock(&lun->lun_lock);
	i = sizeof(msg->lun);
	if (lun->lun_devid)
		i += lun->lun_devid->len;
	i += sizeof(pr_key) * lun->pr_key_count;
alloc:
	mtx_unlock(&lun->lun_lock);
	msg = malloc(i, M_CTL, M_WAITOK);
	mtx_lock(&lun->lun_lock);
	k = sizeof(msg->lun);
	if (lun->lun_devid)
		k += lun->lun_devid->len;
	k += sizeof(pr_key) * lun->pr_key_count;
	if (i < k) {
		free(msg, M_CTL);
		i = k;
		goto alloc;
	}
	bzero(&msg->lun, sizeof(msg->lun));
	msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
	msg->hdr.nexus.targ_lun = lun->lun;
	msg->hdr.nexus.targ_mapped_lun = lun->lun;
	msg->lun.flags = lun->flags;
	msg->lun.pr_generation = lun->PRGeneration;
	msg->lun.pr_res_idx = lun->pr_res_idx;
	msg->lun.pr_res_type = lun->res_type;
	msg->lun.pr_key_count = lun->pr_key_count;
	i = 0;
	if (lun->lun_devid) {
		msg->lun.lun_devid_len = lun->lun_devid->len;
		memcpy(&msg->lun.data[i], lun->lun_devid->data,
		    msg->lun.lun_devid_len);
		i += msg->lun.lun_devid_len;
	}
	for (k = 0; k < CTL_MAX_INITIATORS; k++) {
		if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
			continue;
		pr_key.pr_iid = k;
		memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
		i += sizeof(pr_key);
	}
	mtx_unlock(&lun->lun_lock);
	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
	    M_WAITOK);
	free(msg, M_CTL);

	if (lun->flags & CTL_LUN_PRIMARY_SC) {
		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
			ctl_isc_announce_mode(lun, -1,
			    lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
			    lun->mode_pages.index[i].subpage);
		}
	}
}

void
ctl_isc_announce_port(struct ctl_port *port)
{
	struct ctl_softc *softc = port->ctl_softc;
	union ctl_ha_msg *msg;
	int i;

	if (port->targ_port < softc->port_min ||
	    port->targ_port >= softc->port_max ||
	    softc->ha_link != CTL_HA_LINK_ONLINE)
		return;
	i = sizeof(msg->port) + strlen(port->port_name) + 1;
	if (port->lun_map)
		i += sizeof(uint32_t) * CTL_MAX_LUNS;
	if (port->port_devid)
		i += port->port_devid->len;
	if (port->target_devid)
		i += port->target_devid->len;
	if (port->init_devid)
		i += port->init_devid->len;
	msg = malloc(i, M_CTL, M_WAITOK);
	bzero(&msg->port, sizeof(msg->port));
	msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
	msg->hdr.nexus.targ_port = port->targ_port;
	msg->port.port_type = port->port_type;
	msg->port.physical_port = port->physical_port;
	msg->port.virtual_port = port->virtual_port;
	msg->port.status = port->status;
	i = 0;
	msg->port.name_len = sprintf(&msg->port.data[i],
	    "%d:%s", softc->ha_id, port->port_name) + 1;
	i += msg->port.name_len;
	if (port->lun_map) {
		msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
		memcpy(&msg->port.data[i], port->lun_map,
		    msg->port.lun_map_len);
		i += msg->port.lun_map_len;
	}
	if (port->port_devid) {
		msg->port.port_devid_len = port->port_devid->len;
		memcpy(&msg->port.data[i], port->port_devid->data,
		    msg->port.port_devid_len);
		i += msg->port.port_devid_len;
	}
	if (port->target_devid) {
		msg->port.target_devid_len = port->target_devid->len;
		memcpy(&msg->port.data[i], port->target_devid->data,
		    msg->port.target_devid_len);
		i += msg->port.target_devid_len;
	}
	if (port->init_devid) {
		msg->port.init_devid_len = port->init_devid->len;
		memcpy(&msg->port.data[i], port->init_devid->data,
		    msg->port.init_devid_len);
		i += msg->port.init_devid_len;
	}
	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
	    M_WAITOK);
	free(msg, M_CTL);
}

void
ctl_isc_announce_iid(struct ctl_port *port, int iid)
{
	struct ctl_softc *softc = port->ctl_softc;
	union ctl_ha_msg *msg;
	int i, l;

	if (port->targ_port < softc->port_min ||
	    port->targ_port >= softc->port_max ||
	    softc->ha_link != CTL_HA_LINK_ONLINE)
		return;
	mtx_lock(&softc->ctl_lock);
	i = sizeof(msg->iid);
	l = 0;
	if (port->wwpn_iid[iid].name)
		l = strlen(port->wwpn_iid[iid].name) + 1;
	i += l;
	msg = malloc(i, M_CTL, M_NOWAIT);
	if (msg == NULL) {
		mtx_unlock(&softc->ctl_lock);
		return;
	}
	bzero(&msg->iid, sizeof(msg->iid));
	msg->hdr.msg_type = CTL_MSG_IID_SYNC;
	msg->hdr.nexus.targ_port = port->targ_port;
	msg->hdr.nexus.initid = iid;
	msg->iid.in_use = port->wwpn_iid[iid].in_use;
	msg->iid.name_len = l;
	msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
	if (port->wwpn_iid[iid].name)
		strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
	mtx_unlock(&softc->ctl_lock);
	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
	free(msg, M_CTL);
}

void
ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
    uint8_t page, uint8_t subpage)
{
	struct ctl_softc *softc = lun->ctl_softc;
	union ctl_ha_msg msg;
	int i;

	if (softc->ha_link != CTL_HA_LINK_ONLINE)
		return;
	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
		    page && lun->mode_pages.index[i].subpage == subpage)
			break;
	}
	if (i == CTL_NUM_MODE_PAGES)
		return;
	bzero(&msg.mode, sizeof(msg.mode));
	msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
	msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
	msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
	msg.hdr.nexus.targ_lun = lun->lun;
	msg.hdr.nexus.targ_mapped_lun = lun->lun;
	msg.mode.page_code = page;
	msg.mode.subpage = subpage;
	msg.mode.page_len = lun->mode_pages.index[i].page_len;
	memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
	    msg.mode.page_len);
	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
	    M_WAITOK);
}

static void
ctl_isc_ha_link_up(struct ctl_softc *softc)
{
	struct ctl_port *port;
	struct ctl_lun *lun;
	union ctl_ha_msg msg;
	int i;

	/* Announce this node parameters to peer for validation. */
	msg.login.msg_type = CTL_MSG_LOGIN;
	msg.login.version = CTL_HA_VERSION;
	msg.login.ha_mode = softc->ha_mode;
	msg.login.ha_id = softc->ha_id;
	msg.login.max_luns = CTL_MAX_LUNS;
	msg.login.max_ports = CTL_MAX_PORTS;
	msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
	    M_WAITOK);

	STAILQ_FOREACH(port, &softc->port_list, links) {
		ctl_isc_announce_port(port);
		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
			if (port->wwpn_iid[i].in_use)
				ctl_isc_announce_iid(port, i);
		}
	}
	STAILQ_FOREACH(lun, &softc->lun_list, links)
		ctl_isc_announce_lun(lun);
}

static void
ctl_isc_ha_link_down(struct ctl_softc *softc)
{
	struct ctl_port *port;
	struct ctl_lun *lun;
	union ctl_io *io;
	int i;

	mtx_lock(&softc->ctl_lock);
	STAILQ_FOREACH(lun, &softc->lun_list, links) {
		mtx_lock(&lun->lun_lock);
		if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
			lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
		}
		mtx_unlock(&lun->lun_lock);

		mtx_unlock(&softc->ctl_lock);
		io = ctl_alloc_io(softc->othersc_pool);
		mtx_lock(&softc->ctl_lock);
		ctl_zero_io(io);
		io->io_hdr.msg_type = CTL_MSG_FAILOVER;
		io->io_hdr.nexus.targ_mapped_lun = lun->lun;
		ctl_enqueue_isc(io);
	}

	STAILQ_FOREACH(port, &softc->port_list, links) {
		if (port->targ_port >= softc->port_min &&
		    port->targ_port < softc->port_max)
			continue;
		port->status &= ~CTL_PORT_STATUS_ONLINE;
		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
			port->wwpn_iid[i].in_use = 0;
			free(port->wwpn_iid[i].name, M_CTL);
			port->wwpn_iid[i].name = NULL;
		}
	}
	mtx_unlock(&softc->ctl_lock);
}

static void
ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
{
	struct ctl_lun *lun;
	uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);

	mtx_lock(&softc->ctl_lock);
	if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
	    (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
		mtx_lock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);
		if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
		    msg->ua.ua_set)
			memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
		if (msg->ua.ua_all) {
			if (msg->ua.ua_set)
				ctl_est_ua_all(lun, iid, msg->ua.ua_type);
			else
				ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
		} else {
			if (msg->ua.ua_set)
				ctl_est_ua(lun, iid, msg->ua.ua_type);
			else
				ctl_clr_ua(lun, iid, msg->ua.ua_type);
		}
		mtx_unlock(&lun->lun_lock);
	} else
		mtx_unlock(&softc->ctl_lock);
}

static void
ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
{
	struct ctl_lun *lun;
	struct ctl_ha_msg_lun_pr_key pr_key;
	int i, k;
	ctl_lun_flags oflags;
	uint32_t targ_lun;

	targ_lun = msg->hdr.nexus.targ_mapped_lun;
	mtx_lock(&softc->ctl_lock);
	if ((targ_lun >= CTL_MAX_LUNS) ||
	    ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
		mtx_unlock(&softc->ctl_lock);
		return;
	}
	mtx_lock(&lun->lun_lock);
	mtx_unlock(&softc->ctl_lock);
	if (lun->flags & CTL_LUN_DISABLED) {
		mtx_unlock(&lun->lun_lock);
		return;
	}
	i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
	if (msg->lun.lun_devid_len != i || (i > 0 &&
	    memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
		mtx_unlock(&lun->lun_lock);
		printf("%s: Received conflicting HA LUN %d\n",
		    __func__, msg->hdr.nexus.targ_lun);
		return;
	} else {
		/* Record whether peer is primary. */
		oflags = lun->flags;
		if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
		    (msg->lun.flags & CTL_LUN_DISABLED) == 0)
			lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
		else
			lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
		if (oflags != lun->flags)
			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);

		/* If peer is primary and we are not -- use data */
		if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
		    (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
			lun->PRGeneration = msg->lun.pr_generation;
			lun->pr_res_idx = msg->lun.pr_res_idx;
			lun->res_type = msg->lun.pr_res_type;
			lun->pr_key_count = msg->lun.pr_key_count;
			for (k = 0; k < CTL_MAX_INITIATORS; k++)
				ctl_clr_prkey(lun, k);
			for (k = 0; k < msg->lun.pr_key_count; k++) {
				memcpy(&pr_key, &msg->lun.data[i],
				    sizeof(pr_key));
				ctl_alloc_prkey(lun, pr_key.pr_iid);
				ctl_set_prkey(lun, pr_key.pr_iid,
				    pr_key.pr_key);
				i += sizeof(pr_key);
			}
		}

		mtx_unlock(&lun->lun_lock);
		CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
		    __func__, msg->hdr.nexus.targ_lun,
		    (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
		    "primary" : "secondary"));

		/* If we are primary but peer doesn't know -- notify */
		if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
		    (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
			ctl_isc_announce_lun(lun);
	}
}

static void
ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
{
	struct ctl_port *port;
	struct ctl_lun *lun;
	int i, new;

	port = softc->ctl_ports[msg->hdr.nexus.targ_port];
	if (port == NULL) {
		CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
		    msg->hdr.nexus.targ_port));
		new = 1;
		port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
		port->frontend = &ha_frontend;
		port->targ_port = msg->hdr.nexus.targ_port;
		port->fe_datamove = ctl_ha_datamove;
		port->fe_done = ctl_ha_done;
	} else if (port->frontend == &ha_frontend) {
		CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
		    msg->hdr.nexus.targ_port));
		new = 0;
	} else {
		printf("%s: Received conflicting HA port %d\n",
		    __func__, msg->hdr.nexus.targ_port);
		return;
	}
	port->port_type = msg->port.port_type;
	port->physical_port = msg->port.physical_port;
	port->virtual_port = msg->port.virtual_port;
	port->status = msg->port.status;
	i = 0;
	free(port->port_name, M_CTL);
	port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
	    M_CTL);
	i += msg->port.name_len;
	if (msg->port.lun_map_len != 0) {
		if (port->lun_map == NULL)
			port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
			    M_CTL, M_WAITOK);
		memcpy(port->lun_map, &msg->port.data[i],
		    sizeof(uint32_t) * CTL_MAX_LUNS);
		i += msg->port.lun_map_len;
	} else {
		free(port->lun_map, M_CTL);
		port->lun_map = NULL;
	}
	if (msg->port.port_devid_len != 0) {
		if (port->port_devid == NULL ||
		    port->port_devid->len != msg->port.port_devid_len) {
			free(port->port_devid, M_CTL);
			port->port_devid = malloc(sizeof(struct ctl_devid) +
			    msg->port.port_devid_len, M_CTL, M_WAITOK);
		}
		memcpy(port->port_devid->data, &msg->port.data[i],
		    msg->port.port_devid_len);
		port->port_devid->len = msg->port.port_devid_len;
		i += msg->port.port_devid_len;
	} else {
		free(port->port_devid, M_CTL);
		port->port_devid = NULL;
	}
	if (msg->port.target_devid_len != 0) {
		if (port->target_devid == NULL ||
		    port->target_devid->len != msg->port.target_devid_len) {
			free(port->target_devid, M_CTL);
			port->target_devid = malloc(sizeof(struct ctl_devid) +
			    msg->port.target_devid_len, M_CTL, M_WAITOK);
		}
		memcpy(port->target_devid->data, &msg->port.data[i],
		    msg->port.target_devid_len);
		port->target_devid->len = msg->port.target_devid_len;
		i += msg->port.target_devid_len;
	} else {
		free(port->target_devid, M_CTL);
		port->target_devid = NULL;
	}
	if (msg->port.init_devid_len != 0) {
		if (port->init_devid == NULL ||
		    port->init_devid->len != msg->port.init_devid_len) {
			free(port->init_devid, M_CTL);
			port->init_devid = malloc(sizeof(struct ctl_devid) +
			    msg->port.init_devid_len, M_CTL, M_WAITOK);
		}
		memcpy(port->init_devid->data, &msg->port.data[i],
		    msg->port.init_devid_len);
		port->init_devid->len = msg->port.init_devid_len;
		i += msg->port.init_devid_len;
	} else {
		free(port->init_devid, M_CTL);
		port->init_devid = NULL;
	}
	if (new) {
		if (ctl_port_register(port) != 0) {
			printf("%s: ctl_port_register() failed with error\n",
			    __func__);
		}
	}
	mtx_lock(&softc->ctl_lock);
	STAILQ_FOREACH(lun, &softc->lun_list, links) {
		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
			continue;
		mtx_lock(&lun->lun_lock);
		ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
		mtx_unlock(&lun->lun_lock);
	}
	mtx_unlock(&softc->ctl_lock);
}

static void
ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
{
	struct ctl_port *port;
	int iid;

	port = softc->ctl_ports[msg->hdr.nexus.targ_port];
	if (port == NULL) {
		printf("%s: Received IID for unknown port %d\n",
		    __func__, msg->hdr.nexus.targ_port);
		return;
	}
	iid = msg->hdr.nexus.initid;
	port->wwpn_iid[iid].in_use = msg->iid.in_use;
	port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
	free(port->wwpn_iid[iid].name, M_CTL);
	if (msg->iid.name_len) {
		port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
		    msg->iid.name_len, M_CTL);
	} else
		port->wwpn_iid[iid].name = NULL;
}

static void
ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
{

	if (msg->login.version != CTL_HA_VERSION) {
		printf("CTL HA peers have different versions %d != %d\n",
		    msg->login.version, CTL_HA_VERSION);
		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
		return;
	}
	if (msg->login.ha_mode != softc->ha_mode) {
		printf("CTL HA peers have different ha_mode %d != %d\n",
		    msg->login.ha_mode, softc->ha_mode);
		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
		return;
	}
	if (msg->login.ha_id == softc->ha_id) {
		printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
		return;
	}
	if (msg->login.max_luns != CTL_MAX_LUNS ||
	    msg->login.max_ports != CTL_MAX_PORTS ||
	    msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
		printf("CTL HA peers have different limits\n");
		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
		return;
	}
}

static void
ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
{
	struct ctl_lun *lun;
	int i;
	uint32_t initidx, targ_lun;

	targ_lun = msg->hdr.nexus.targ_mapped_lun;
	mtx_lock(&softc->ctl_lock);
	if ((targ_lun >= CTL_MAX_LUNS) ||
	    ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
		mtx_unlock(&softc->ctl_lock);
		return;
	}
	mtx_lock(&lun->lun_lock);
	mtx_unlock(&softc->ctl_lock);
	if (lun->flags & CTL_LUN_DISABLED) {
		mtx_unlock(&lun->lun_lock);
		return;
	}
	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
		    msg->mode.page_code &&
		    lun->mode_pages.index[i].subpage == msg->mode.subpage)
			break;
	}
	if (i == CTL_NUM_MODE_PAGES) {
		mtx_unlock(&lun->lun_lock);
		return;
	}
	memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
	    lun->mode_pages.index[i].page_len);
	initidx = ctl_get_initindex(&msg->hdr.nexus);
	if (initidx != -1)
		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
	mtx_unlock(&lun->lun_lock);
}

/*
 * ISC (Inter Shelf Communication) event handler.  Events from the HA
 * subsystem come in here.
 */
static void
ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
{
	struct ctl_softc *softc = control_softc;
	union ctl_io *io;
	struct ctl_prio *presio;
	ctl_ha_status isc_status;

	CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
	if (event == CTL_HA_EVT_MSG_RECV) {
		union ctl_ha_msg *msg, msgbuf;

		if (param > sizeof(msgbuf))
			msg = malloc(param, M_CTL, M_WAITOK);
		else
			msg = &msgbuf;
		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
		    M_WAITOK);
		if (isc_status != CTL_HA_STATUS_SUCCESS) {
			printf("%s: Error receiving message: %d\n",
			    __func__, isc_status);
			if (msg != &msgbuf)
				free(msg, M_CTL);
			return;
		}

		CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
		switch (msg->hdr.msg_type) {
		case CTL_MSG_SERIALIZE:
			io = ctl_alloc_io(softc->othersc_pool);
			ctl_zero_io(io);
			// populate ctsio from msg
			io->io_hdr.io_type = CTL_IO_SCSI;
			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
			io->io_hdr.original_sc = msg->hdr.original_sc;
			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
					    CTL_FLAG_IO_ACTIVE;
			/*
			 * If we're in serialization-only mode, we don't
			 * want to go through full done processing.  Thus
			 * the COPY flag.
			 *
			 * XXX KDM add another flag that is more specific.
			 */
			if (softc->ha_mode != CTL_HA_MODE_XFER)
				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
			io->io_hdr.nexus = msg->hdr.nexus;
#if 0
			printf("port %u, iid %u, lun %u\n",
			       io->io_hdr.nexus.targ_port,
			       io->io_hdr.nexus.initid,
			       io->io_hdr.nexus.targ_lun);
#endif
			io->scsiio.tag_num = msg->scsi.tag_num;
			io->scsiio.tag_type = msg->scsi.tag_type;
#ifdef CTL_TIME_IO
			io->io_hdr.start_time = time_uptime;
			getbinuptime(&io->io_hdr.start_bt);
#endif /* CTL_TIME_IO */
			io->scsiio.cdb_len = msg->scsi.cdb_len;
			memcpy(io->scsiio.cdb, msg->scsi.cdb,
			       CTL_MAX_CDBLEN);
			if (softc->ha_mode == CTL_HA_MODE_XFER) {
				const struct ctl_cmd_entry *entry;

				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
				io->io_hdr.flags |=
					entry->flags & CTL_FLAG_DATA_MASK;
			}
			ctl_enqueue_isc(io);
			break;

		/* Performed on the Originating SC, XFER mode only */
		case CTL_MSG_DATAMOVE: {
			struct ctl_sg_entry *sgl;
			int i, j;

			io = msg->hdr.original_sc;
			if (io == NULL) {
				printf("%s: original_sc == NULL!\n", __func__);
				/* XXX KDM do something here */
				break;
			}
			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
			/*
			 * Keep track of this, we need to send it back over
			 * when the datamove is complete.
			 */
			io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
			if (msg->hdr.status == CTL_SUCCESS)
				io->io_hdr.status = msg->hdr.status;

			if (msg->dt.sg_sequence == 0) {
#ifdef CTL_TIME_IO
				getbinuptime(&io->io_hdr.dma_start_bt);
#endif
				i = msg->dt.kern_sg_entries +
				    msg->dt.kern_data_len /
				    CTL_HA_DATAMOVE_SEGMENT + 1;
				sgl = malloc(sizeof(*sgl) * i, M_CTL,
				    M_WAITOK | M_ZERO);
				io->io_hdr.remote_sglist = sgl;
				io->io_hdr.local_sglist =
				    &sgl[msg->dt.kern_sg_entries];

				io->scsiio.kern_data_ptr = (uint8_t *)sgl;

				io->scsiio.kern_sg_entries =
					msg->dt.kern_sg_entries;
				io->scsiio.rem_sg_entries =
					msg->dt.kern_sg_entries;
				io->scsiio.kern_data_len =
					msg->dt.kern_data_len;
				io->scsiio.kern_total_len =
					msg->dt.kern_total_len;
				io->scsiio.kern_data_resid =
					msg->dt.kern_data_resid;
				io->scsiio.kern_rel_offset =
					msg->dt.kern_rel_offset;
				io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
				io->io_hdr.flags |= msg->dt.flags &
				    CTL_FLAG_BUS_ADDR;
			} else
				sgl = (struct ctl_sg_entry *)
					io->scsiio.kern_data_ptr;

			for (i = msg->dt.sent_sg_entries, j = 0;
			     i < (msg->dt.sent_sg_entries +
			     msg->dt.cur_sg_entries); i++, j++) {
				sgl[i].addr = msg->dt.sg_list[j].addr;
				sgl[i].len = msg->dt.sg_list[j].len;

#if 0
				printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
				    __func__, sgl[i].addr, sgl[i].len, j, i);
#endif
			}

			/*
			 * If this is the last piece of the I/O, we've got
			 * the full S/G list.  Queue processing in the thread.
			 * Otherwise wait for the next piece.
			 */
			if (msg->dt.sg_last != 0)
				ctl_enqueue_isc(io);
			break;
		}
		/* Performed on the Serializing (primary) SC, XFER mode only */
		case CTL_MSG_DATAMOVE_DONE: {
			if (msg->hdr.serializing_sc == NULL) {
				printf("%s: serializing_sc == NULL!\n",
				       __func__);
				/* XXX KDM now what? */
				break;
			}
			/*
			 * We grab the sense information here in case
			 * there was a failure, so we can return status
			 * back to the initiator.
			 */
			io = msg->hdr.serializing_sc;
			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
			io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
			io->io_hdr.port_status = msg->scsi.fetd_status;
			io->scsiio.residual = msg->scsi.residual;
			if (msg->hdr.status != CTL_STATUS_NONE) {
				io->io_hdr.status = msg->hdr.status;
				io->scsiio.scsi_status = msg->scsi.scsi_status;
				io->scsiio.sense_len = msg->scsi.sense_len;
				io->scsiio.sense_residual =msg->scsi.sense_residual;
				memcpy(&io->scsiio.sense_data,
				    &msg->scsi.sense_data,
				    msg->scsi.sense_len);
				if (msg->hdr.status == CTL_SUCCESS)
					io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
			}
			ctl_enqueue_isc(io);
			break;
		}

		/* Preformed on Originating SC, SER_ONLY mode */
		case CTL_MSG_R2R:
			io = msg->hdr.original_sc;
			if (io == NULL) {
				printf("%s: original_sc == NULL!\n",
				    __func__);
				break;
			}
			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
			io->io_hdr.msg_type = CTL_MSG_R2R;
			io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
			ctl_enqueue_isc(io);
			break;

		/*
		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
		 * mode.
		 * Performed on the Originating (i.e. secondary) SC in XFER
		 * mode
		 */
		case CTL_MSG_FINISH_IO:
			if (softc->ha_mode == CTL_HA_MODE_XFER)
				ctl_isc_handler_finish_xfer(softc, msg);
			else
				ctl_isc_handler_finish_ser_only(softc, msg);
			break;

		/* Preformed on Originating SC */
		case CTL_MSG_BAD_JUJU:
			io = msg->hdr.original_sc;
			if (io == NULL) {
				printf("%s: Bad JUJU!, original_sc is NULL!\n",
				       __func__);
				break;
			}
			ctl_copy_sense_data(msg, io);
			/*
			 * IO should have already been cleaned up on other
			 * SC so clear this flag so we won't send a message
			 * back to finish the IO there.
			 */
			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;

			/* io = msg->hdr.serializing_sc; */
			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
			ctl_enqueue_isc(io);
			break;

		/* Handle resets sent from the other side */
		case CTL_MSG_MANAGE_TASKS: {
			struct ctl_taskio *taskio;
			taskio = (struct ctl_taskio *)ctl_alloc_io(
			    softc->othersc_pool);
			ctl_zero_io((union ctl_io *)taskio);
			taskio->io_hdr.io_type = CTL_IO_TASK;
			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
			taskio->io_hdr.nexus = msg->hdr.nexus;
			taskio->task_action = msg->task.task_action;
			taskio->tag_num = msg->task.tag_num;
			taskio->tag_type = msg->task.tag_type;
#ifdef CTL_TIME_IO
			taskio->io_hdr.start_time = time_uptime;
			getbinuptime(&taskio->io_hdr.start_bt);
#endif /* CTL_TIME_IO */
			ctl_run_task((union ctl_io *)taskio);
			break;
		}
		/* Persistent Reserve action which needs attention */
		case CTL_MSG_PERS_ACTION:
			presio = (struct ctl_prio *)ctl_alloc_io(
			    softc->othersc_pool);
			ctl_zero_io((union ctl_io *)presio);
			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
			presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
			presio->io_hdr.nexus = msg->hdr.nexus;
			presio->pr_msg = msg->pr;
			ctl_enqueue_isc((union ctl_io *)presio);
			break;
		case CTL_MSG_UA:
			ctl_isc_ua(softc, msg, param);
			break;
		case CTL_MSG_PORT_SYNC:
			ctl_isc_port_sync(softc, msg, param);
			break;
		case CTL_MSG_LUN_SYNC:
			ctl_isc_lun_sync(softc, msg, param);
			break;
		case CTL_MSG_IID_SYNC:
			ctl_isc_iid_sync(softc, msg, param);
			break;
		case CTL_MSG_LOGIN:
			ctl_isc_login(softc, msg, param);
			break;
		case CTL_MSG_MODE_SYNC:
			ctl_isc_mode_sync(softc, msg, param);
			break;
		default:
			printf("Received HA message of unknown type %d\n",
			    msg->hdr.msg_type);
			ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
			break;
		}
		if (msg != &msgbuf)
			free(msg, M_CTL);
	} else if (event == CTL_HA_EVT_LINK_CHANGE) {
		printf("CTL: HA link status changed from %d to %d\n",
		    softc->ha_link, param);
		if (param == softc->ha_link)
			return;
		if (softc->ha_link == CTL_HA_LINK_ONLINE) {
			softc->ha_link = param;
			ctl_isc_ha_link_down(softc);
		} else {
			softc->ha_link = param;
			if (softc->ha_link == CTL_HA_LINK_ONLINE)
				ctl_isc_ha_link_up(softc);
		}
		return;
	} else {
		printf("ctl_isc_event_handler: Unknown event %d\n", event);
		return;
	}
}

static void
ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
{

	memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
	    src->scsi.sense_len);
	dest->scsiio.scsi_status = src->scsi.scsi_status;
	dest->scsiio.sense_len = src->scsi.sense_len;
	dest->io_hdr.status = src->hdr.status;
}

static void
ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
{

	memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
	    src->scsiio.sense_len);
	dest->scsi.scsi_status = src->scsiio.scsi_status;
	dest->scsi.sense_len = src->scsiio.sense_len;
	dest->hdr.status = src->io_hdr.status;
}

void
ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
{
	struct ctl_softc *softc = lun->ctl_softc;
	ctl_ua_type *pu;

	if (initidx < softc->init_min || initidx >= softc->init_max)
		return;
	mtx_assert(&lun->lun_lock, MA_OWNED);
	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
	if (pu == NULL)
		return;
	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
}

void
ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
{
	int i;

	mtx_assert(&lun->lun_lock, MA_OWNED);
	if (lun->pending_ua[port] == NULL)
		return;
	for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
		if (port * CTL_MAX_INIT_PER_PORT + i == except)
			continue;
		lun->pending_ua[port][i] |= ua;
	}
}

void
ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
{
	struct ctl_softc *softc = lun->ctl_softc;
	int i;

	mtx_assert(&lun->lun_lock, MA_OWNED);
	for (i = softc->port_min; i < softc->port_max; i++)
		ctl_est_ua_port(lun, i, except, ua);
}

void
ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
{
	struct ctl_softc *softc = lun->ctl_softc;
	ctl_ua_type *pu;

	if (initidx < softc->init_min || initidx >= softc->init_max)
		return;
	mtx_assert(&lun->lun_lock, MA_OWNED);
	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
	if (pu == NULL)
		return;
	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
}

void
ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
{
	struct ctl_softc *softc = lun->ctl_softc;
	int i, j;

	mtx_assert(&lun->lun_lock, MA_OWNED);
	for (i = softc->port_min; i < softc->port_max; i++) {
		if (lun->pending_ua[i] == NULL)
			continue;
		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
			if (i * CTL_MAX_INIT_PER_PORT + j == except)
				continue;
			lun->pending_ua[i][j] &= ~ua;
		}
	}
}

void
ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
    ctl_ua_type ua_type)
{
	struct ctl_lun *lun;

	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
		mtx_lock(&lun->lun_lock);
		ctl_clr_ua(lun, initidx, ua_type);
		mtx_unlock(&lun->lun_lock);
	}
}

static int
ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
{
	struct ctl_softc *softc = (struct ctl_softc *)arg1;
	struct ctl_lun *lun;
	struct ctl_lun_req ireq;
	int error, value;

	value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
	error = sysctl_handle_int(oidp, &value, 0, req);
	if ((error != 0) || (req->newptr == NULL))
		return (error);

	mtx_lock(&softc->ctl_lock);
	if (value == 0)
		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
	else
		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
	STAILQ_FOREACH(lun, &softc->lun_list, links) {
		mtx_unlock(&softc->ctl_lock);
		bzero(&ireq, sizeof(ireq));
		ireq.reqtype = CTL_LUNREQ_MODIFY;
		ireq.reqdata.modify.lun_id = lun->lun;
		lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
		    curthread);
		if (ireq.status != CTL_LUN_OK) {
			printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
			    __func__, ireq.status, ireq.error_str);
		}
		mtx_lock(&softc->ctl_lock);
	}
	mtx_unlock(&softc->ctl_lock);
	return (0);
}

static int
ctl_init(void)
{
	struct ctl_softc *softc;
	void *other_pool;
	int i, error;

	softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
			       M_WAITOK | M_ZERO);

	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
			      "cam/ctl");
	softc->dev->si_drv1 = softc;

	sysctl_ctx_init(&softc->sysctl_ctx);
	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
		CTLFLAG_RD, 0, "CAM Target Layer");

	if (softc->sysctl_tree == NULL) {
		printf("%s: unable to allocate sysctl tree\n", __func__);
		destroy_dev(softc->dev);
		free(control_softc, M_DEVBUF);
		control_softc = NULL;
		return (ENOMEM);
	}

	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
	softc->open_count = 0;

	/*
	 * Default to actually sending a SYNCHRONIZE CACHE command down to
	 * the drive.
	 */
	softc->flags = CTL_FLAG_REAL_SYNC;

	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
	    OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
	    "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");

	/*
	 * In Copan's HA scheme, the "master" and "slave" roles are
	 * figured out through the slot the controller is in.  Although it
	 * is an active/active system, someone has to be in charge.
	 */
	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
	    "HA head ID (0 - no HA)");
	if (softc->ha_id == 0 || softc->ha_id > NUM_TARGET_PORT_GROUPS) {
		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
		softc->is_single = 1;
		softc->port_cnt = CTL_MAX_PORTS;
		softc->port_min = 0;
	} else {
		softc->port_cnt = CTL_MAX_PORTS / NUM_TARGET_PORT_GROUPS;
		softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
	}
	softc->port_max = softc->port_min + softc->port_cnt;
	softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
	softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;

	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
	    OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
	    "HA link state (0 - offline, 1 - unknown, 2 - online)");

	STAILQ_INIT(&softc->lun_list);
	STAILQ_INIT(&softc->pending_lun_queue);
	STAILQ_INIT(&softc->fe_list);
	STAILQ_INIT(&softc->port_list);
	STAILQ_INIT(&softc->be_list);
	ctl_tpc_init(softc);

	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
	                    &other_pool) != 0)
	{
		printf("ctl: can't allocate %d entry other SC pool, "
		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
		return (ENOMEM);
	}
	softc->othersc_pool = other_pool;

	if (worker_threads <= 0)
		worker_threads = max(1, mp_ncpus / 4);
	if (worker_threads > CTL_MAX_THREADS)
		worker_threads = CTL_MAX_THREADS;

	for (i = 0; i < worker_threads; i++) {
		struct ctl_thread *thr = &softc->threads[i];

		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
		thr->ctl_softc = softc;
		STAILQ_INIT(&thr->incoming_queue);
		STAILQ_INIT(&thr->rtr_queue);
		STAILQ_INIT(&thr->done_queue);
		STAILQ_INIT(&thr->isc_queue);

		error = kproc_kthread_add(ctl_work_thread, thr,
		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
		if (error != 0) {
			printf("error creating CTL work thread!\n");
			ctl_pool_free(other_pool);
			return (error);
		}
	}
	error = kproc_kthread_add(ctl_lun_thread, softc,
	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
	if (error != 0) {
		printf("error creating CTL lun thread!\n");
		ctl_pool_free(other_pool);
		return (error);
	}
	error = kproc_kthread_add(ctl_thresh_thread, softc,
	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
	if (error != 0) {
		printf("error creating CTL threshold thread!\n");
		ctl_pool_free(other_pool);
		return (error);
	}

	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
	    OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
	    softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");

	if (softc->is_single == 0) {
		ctl_frontend_register(&ha_frontend);
		if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
			printf("ctl_init: ctl_ha_msg_init failed.\n");
			softc->is_single = 1;
		} else
		if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
		    != CTL_HA_STATUS_SUCCESS) {
			printf("ctl_init: ctl_ha_msg_register failed.\n");
			softc->is_single = 1;
		}
	}
	return (0);
}

void
ctl_shutdown(void)
{
	struct ctl_softc *softc = control_softc;
	struct ctl_lun *lun, *next_lun;

	if (softc->is_single == 0) {
		ctl_ha_msg_shutdown(softc);
		if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
		    != CTL_HA_STATUS_SUCCESS)
			printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
		if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
			printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
		ctl_frontend_deregister(&ha_frontend);
	}

	mtx_lock(&softc->ctl_lock);

	/*
	 * Free up each LUN.
	 */
	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
		next_lun = STAILQ_NEXT(lun, links);
		ctl_free_lun(lun);
	}

	mtx_unlock(&softc->ctl_lock);

#if 0
	ctl_shutdown_thread(softc->work_thread);
	mtx_destroy(&softc->queue_lock);
#endif

	ctl_tpc_shutdown(softc);
	uma_zdestroy(softc->io_zone);
	mtx_destroy(&softc->ctl_lock);

	destroy_dev(softc->dev);

	sysctl_ctx_free(&softc->sysctl_ctx);

	free(control_softc, M_DEVBUF);
	control_softc = NULL;
}

static int
ctl_module_event_handler(module_t mod, int what, void *arg)
{

	switch (what) {
	case MOD_LOAD:
		return (ctl_init());
	case MOD_UNLOAD:
		return (EBUSY);
	default:
		return (EOPNOTSUPP);
	}
}

/*
 * XXX KDM should we do some access checks here?  Bump a reference count to
 * prevent a CTL module from being unloaded while someone has it open?
 */
static int
ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
{
	return (0);
}

static int
ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
{
	return (0);
}

/*
 * Remove an initiator by port number and initiator ID.
 * Returns 0 for success, -1 for failure.
 */
int
ctl_remove_initiator(struct ctl_port *port, int iid)
{
	struct ctl_softc *softc = port->ctl_softc;

	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);

	if (iid > CTL_MAX_INIT_PER_PORT) {
		printf("%s: initiator ID %u > maximun %u!\n",
		       __func__, iid, CTL_MAX_INIT_PER_PORT);
		return (-1);
	}

	mtx_lock(&softc->ctl_lock);
	port->wwpn_iid[iid].in_use--;
	port->wwpn_iid[iid].last_use = time_uptime;
	mtx_unlock(&softc->ctl_lock);
	ctl_isc_announce_iid(port, iid);

	return (0);
}

/*
 * Add an initiator to the initiator map.
 * Returns iid for success, < 0 for failure.
 */
int
ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
{
	struct ctl_softc *softc = port->ctl_softc;
	time_t best_time;
	int i, best;

	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);

	if (iid >= CTL_MAX_INIT_PER_PORT) {
		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
		free(name, M_CTL);
		return (-1);
	}

	mtx_lock(&softc->ctl_lock);

	if (iid < 0 && (wwpn != 0 || name != NULL)) {
		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
				iid = i;
				break;
			}
			if (name != NULL && port->wwpn_iid[i].name != NULL &&
			    strcmp(name, port->wwpn_iid[i].name) == 0) {
				iid = i;
				break;
			}
		}
	}

	if (iid < 0) {
		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
			if (port->wwpn_iid[i].in_use == 0 &&
			    port->wwpn_iid[i].wwpn == 0 &&
			    port->wwpn_iid[i].name == NULL) {
				iid = i;
				break;
			}
		}
	}

	if (iid < 0) {
		best = -1;
		best_time = INT32_MAX;
		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
			if (port->wwpn_iid[i].in_use == 0) {
				if (port->wwpn_iid[i].last_use < best_time) {
					best = i;
					best_time = port->wwpn_iid[i].last_use;
				}
			}
		}
		iid = best;
	}

	if (iid < 0) {
		mtx_unlock(&softc->ctl_lock);
		free(name, M_CTL);
		return (-2);
	}

	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
		/*
		 * This is not an error yet.
		 */
		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
#if 0
			printf("%s: port %d iid %u WWPN %#jx arrived"
			    " again\n", __func__, port->targ_port,
			    iid, (uintmax_t)wwpn);
#endif
			goto take;
		}
		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
#if 0
			printf("%s: port %d iid %u name '%s' arrived"
			    " again\n", __func__, port->targ_port,
			    iid, name);
#endif
			goto take;
		}

		/*
		 * This is an error, but what do we do about it?  The
		 * driver is telling us we have a new WWPN for this
		 * initiator ID, so we pretty much need to use it.
		 */
		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
		    " but WWPN %#jx '%s' is still at that address\n",
		    __func__, port->targ_port, iid, wwpn, name,
		    (uintmax_t)port->wwpn_iid[iid].wwpn,
		    port->wwpn_iid[iid].name);

		/*
		 * XXX KDM clear have_ca and ua_pending on each LUN for
		 * this initiator.
		 */
	}
take:
	free(port->wwpn_iid[iid].name, M_CTL);
	port->wwpn_iid[iid].name = name;
	port->wwpn_iid[iid].wwpn = wwpn;
	port->wwpn_iid[iid].in_use++;
	mtx_unlock(&softc->ctl_lock);
	ctl_isc_announce_iid(port, iid);

	return (iid);
}

static int
ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
{
	int len;

	switch (port->port_type) {
	case CTL_PORT_FC:
	{
		struct scsi_transportid_fcp *id =
		    (struct scsi_transportid_fcp *)buf;
		if (port->wwpn_iid[iid].wwpn == 0)
			return (0);
		memset(id, 0, sizeof(*id));
		id->format_protocol = SCSI_PROTO_FC;
		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
		return (sizeof(*id));
	}
	case CTL_PORT_ISCSI:
	{
		struct scsi_transportid_iscsi_port *id =
		    (struct scsi_transportid_iscsi_port *)buf;
		if (port->wwpn_iid[iid].name == NULL)
			return (0);
		memset(id, 0, 256);
		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
		    SCSI_PROTO_ISCSI;
		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
		len = roundup2(min(len, 252), 4);
		scsi_ulto2b(len, id->additional_length);
		return (sizeof(*id) + len);
	}
	case CTL_PORT_SAS:
	{
		struct scsi_transportid_sas *id =
		    (struct scsi_transportid_sas *)buf;
		if (port->wwpn_iid[iid].wwpn == 0)
			return (0);
		memset(id, 0, sizeof(*id));
		id->format_protocol = SCSI_PROTO_SAS;
		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
		return (sizeof(*id));
	}
	default:
	{
		struct scsi_transportid_spi *id =
		    (struct scsi_transportid_spi *)buf;
		memset(id, 0, sizeof(*id));
		id->format_protocol = SCSI_PROTO_SPI;
		scsi_ulto2b(iid, id->scsi_addr);
		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
		return (sizeof(*id));
	}
	}
}

/*
 * Serialize a command that went down the "wrong" side, and so was sent to
 * this controller for execution.  The logic is a little different than the
 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
 * sent back to the other side, but in the success case, we execute the
 * command on this side (XFER mode) or tell the other side to execute it
 * (SER_ONLY mode).
 */
static int
ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
{
	struct ctl_softc *softc = control_softc;
	union ctl_ha_msg msg_info;
	struct ctl_port *port;
	struct ctl_lun *lun;
	const struct ctl_cmd_entry *entry;
	int retval = 0;
	uint32_t targ_lun;

	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
	mtx_lock(&softc->ctl_lock);

	/* Make sure that we know about this port. */
	port = ctl_io_port(&ctsio->io_hdr);
	if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
		ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
					 /*retry_count*/ 1);
		goto badjuju;
	}

	/* Make sure that we know about this LUN. */
	if ((targ_lun < CTL_MAX_LUNS) &&
	    ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
		mtx_lock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);
		/*
		 * If the LUN is invalid, pretend that it doesn't exist.
		 * It will go away as soon as all pending I/O has been
		 * completed.
		 */
		if (lun->flags & CTL_LUN_DISABLED) {
			mtx_unlock(&lun->lun_lock);
			lun = NULL;
		}
	} else {
		mtx_unlock(&softc->ctl_lock);
		lun = NULL;
	}
	if (lun == NULL) {
		/*
		 * The other node would not send this request to us unless
		 * received announce that we are primary node for this LUN.
		 * If this LUN does not exist now, it is probably result of
		 * a race, so respond to initiator in the most opaque way.
		 */
		ctl_set_busy(ctsio);
		goto badjuju;
	}

	entry = ctl_get_cmd_entry(ctsio, NULL);
	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
		mtx_unlock(&lun->lun_lock);
		goto badjuju;
	}

	ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
	ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;

	/*
	 * Every I/O goes into the OOA queue for a
	 * particular LUN, and stays there until completion.
	 */
#ifdef CTL_TIME_IO
	if (TAILQ_EMPTY(&lun->ooa_queue))
		lun->idle_time += getsbinuptime() - lun->last_busy;
#endif
	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);

	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
		 ooa_links))) {
	case CTL_ACTION_BLOCK:
		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
				  blocked_links);
		mtx_unlock(&lun->lun_lock);
		break;
	case CTL_ACTION_PASS:
	case CTL_ACTION_SKIP:
		if (softc->ha_mode == CTL_HA_MODE_XFER) {
			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
			ctl_enqueue_rtr((union ctl_io *)ctsio);
			mtx_unlock(&lun->lun_lock);
		} else {
			ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
			mtx_unlock(&lun->lun_lock);

			/* send msg back to other side */
			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
			msg_info.hdr.msg_type = CTL_MSG_R2R;
			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
			    sizeof(msg_info.hdr), M_WAITOK);
		}
		break;
	case CTL_ACTION_OVERLAP:
		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
		mtx_unlock(&lun->lun_lock);
		ctl_set_overlapped_cmd(ctsio);
		goto badjuju;
	case CTL_ACTION_OVERLAP_TAG:
		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
		mtx_unlock(&lun->lun_lock);
		ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
		goto badjuju;
	case CTL_ACTION_ERROR:
	default:
		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
		mtx_unlock(&lun->lun_lock);

		ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
					 /*retry_count*/ 0);
badjuju:
		ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
		msg_info.hdr.serializing_sc = NULL;
		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
		    sizeof(msg_info.scsi), M_WAITOK);
		retval = 1;
		break;
	}
	return (retval);
}

/*
 * Returns 0 for success, errno for failure.
 */
static void
ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
{
	union ctl_io *io;

	mtx_lock(&lun->lun_lock);
	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
	     ooa_links)) {
		struct ctl_ooa_entry *entry;

		/*
		 * If we've got more than we can fit, just count the
		 * remaining entries.
		 */
		if (*cur_fill_num >= ooa_hdr->alloc_num)
			continue;

		entry = &kern_entries[*cur_fill_num];

		entry->tag_num = io->scsiio.tag_num;
		entry->lun_num = lun->lun;
#ifdef CTL_TIME_IO
		entry->start_bt = io->io_hdr.start_bt;
#endif
		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
		entry->cdb_len = io->scsiio.cdb_len;
		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;

		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;

		if (io->io_hdr.flags & CTL_FLAG_ABORT)
			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;

		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;

		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
	}
	mtx_unlock(&lun->lun_lock);
}

static void *
ctl_copyin_alloc(void *user_addr, int len, char *error_str,
		 size_t error_str_len)
{
	void *kptr;

	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);

	if (copyin(user_addr, kptr, len) != 0) {
		snprintf(error_str, error_str_len, "Error copying %d bytes "
			 "from user address %p to kernel address %p", len,
			 user_addr, kptr);
		free(kptr, M_CTL);
		return (NULL);
	}

	return (kptr);
}

static void
ctl_free_args(int num_args, struct ctl_be_arg *args)
{
	int i;

	if (args == NULL)
		return;

	for (i = 0; i < num_args; i++) {
		free(args[i].kname, M_CTL);
		free(args[i].kvalue, M_CTL);
	}

	free(args, M_CTL);
}

static struct ctl_be_arg *
ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
		char *error_str, size_t error_str_len)
{
	struct ctl_be_arg *args;
	int i;

	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
				error_str, error_str_len);

	if (args == NULL)
		goto bailout;

	for (i = 0; i < num_args; i++) {
		args[i].kname = NULL;
		args[i].kvalue = NULL;
	}

	for (i = 0; i < num_args; i++) {
		uint8_t *tmpptr;

		args[i].kname = ctl_copyin_alloc(args[i].name,
			args[i].namelen, error_str, error_str_len);
		if (args[i].kname == NULL)
			goto bailout;

		if (args[i].kname[args[i].namelen - 1] != '\0') {
			snprintf(error_str, error_str_len, "Argument %d "
				 "name is not NUL-terminated", i);
			goto bailout;
		}

		if (args[i].flags & CTL_BEARG_RD) {
			tmpptr = ctl_copyin_alloc(args[i].value,
				args[i].vallen, error_str, error_str_len);
			if (tmpptr == NULL)
				goto bailout;
			if ((args[i].flags & CTL_BEARG_ASCII)
			 && (tmpptr[args[i].vallen - 1] != '\0')) {
				snprintf(error_str, error_str_len, "Argument "
				    "%d value is not NUL-terminated", i);
				goto bailout;
			}
			args[i].kvalue = tmpptr;
		} else {
			args[i].kvalue = malloc(args[i].vallen,
			    M_CTL, M_WAITOK | M_ZERO);
		}
	}

	return (args);
bailout:

	ctl_free_args(num_args, args);

	return (NULL);
}

static void
ctl_copyout_args(int num_args, struct ctl_be_arg *args)
{
	int i;

	for (i = 0; i < num_args; i++) {
		if (args[i].flags & CTL_BEARG_WR)
			copyout(args[i].kvalue, args[i].value, args[i].vallen);
	}
}

/*
 * Escape characters that are illegal or not recommended in XML.
 */
int
ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
{
	char *end = str + size;
	int retval;

	retval = 0;

	for (; *str && str < end; str++) {
		switch (*str) {
		case '&':
			retval = sbuf_printf(sb, "&amp;");
			break;
		case '>':
			retval = sbuf_printf(sb, "&gt;");
			break;
		case '<':
			retval = sbuf_printf(sb, "&lt;");
			break;
		default:
			retval = sbuf_putc(sb, *str);
			break;
		}

		if (retval != 0)
			break;

	}

	return (retval);
}

static void
ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
{
	struct scsi_vpd_id_descriptor *desc;
	int i;

	if (id == NULL || id->len < 4)
		return;
	desc = (struct scsi_vpd_id_descriptor *)id->data;
	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
	case SVPD_ID_TYPE_T10:
		sbuf_printf(sb, "t10.");
		break;
	case SVPD_ID_TYPE_EUI64:
		sbuf_printf(sb, "eui.");
		break;
	case SVPD_ID_TYPE_NAA:
		sbuf_printf(sb, "naa.");
		break;
	case SVPD_ID_TYPE_SCSI_NAME:
		break;
	}
	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
	case SVPD_ID_CODESET_BINARY:
		for (i = 0; i < desc->length; i++)
			sbuf_printf(sb, "%02x", desc->identifier[i]);
		break;
	case SVPD_ID_CODESET_ASCII:
		sbuf_printf(sb, "%.*s", (int)desc->length,
		    (char *)desc->identifier);
		break;
	case SVPD_ID_CODESET_UTF8:
		sbuf_printf(sb, "%s", (char *)desc->identifier);
		break;
	}
}

static int
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_lun *lun;
	int retval;

	retval = 0;

	switch (cmd) {
	case CTL_IO:
		retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
		break;
	case CTL_ENABLE_PORT:
	case CTL_DISABLE_PORT:
	case CTL_SET_PORT_WWNS: {
		struct ctl_port *port;
		struct ctl_port_entry *entry;

		entry = (struct ctl_port_entry *)addr;
		
		mtx_lock(&softc->ctl_lock);
		STAILQ_FOREACH(port, &softc->port_list, links) {
			int action, done;

			if (port->targ_port < softc->port_min ||
			    port->targ_port >= softc->port_max)
				continue;

			action = 0;
			done = 0;
			if ((entry->port_type == CTL_PORT_NONE)
			 && (entry->targ_port == port->targ_port)) {
				/*
				 * If the user only wants to enable or
				 * disable or set WWNs on a specific port,
				 * do the operation and we're done.
				 */
				action = 1;
				done = 1;
			} else if (entry->port_type & port->port_type) {
				/*
				 * Compare the user's type mask with the
				 * particular frontend type to see if we
				 * have a match.
				 */
				action = 1;
				done = 0;

				/*
				 * Make sure the user isn't trying to set
				 * WWNs on multiple ports at the same time.
				 */
				if (cmd == CTL_SET_PORT_WWNS) {
					printf("%s: Can't set WWNs on "
					       "multiple ports\n", __func__);
					retval = EINVAL;
					break;
				}
			}
			if (action == 0)
				continue;

			/*
			 * XXX KDM we have to drop the lock here, because
			 * the online/offline operations can potentially
			 * block.  We need to reference count the frontends
			 * so they can't go away,
			 */
			if (cmd == CTL_ENABLE_PORT) {
				mtx_unlock(&softc->ctl_lock);
				ctl_port_online(port);
				mtx_lock(&softc->ctl_lock);
			} else if (cmd == CTL_DISABLE_PORT) {
				mtx_unlock(&softc->ctl_lock);
				ctl_port_offline(port);
				mtx_lock(&softc->ctl_lock);
			} else if (cmd == CTL_SET_PORT_WWNS) {
				ctl_port_set_wwns(port,
				    (entry->flags & CTL_PORT_WWNN_VALID) ?
				    1 : 0, entry->wwnn,
				    (entry->flags & CTL_PORT_WWPN_VALID) ?
				    1 : 0, entry->wwpn);
			}
			if (done != 0)
				break;
		}
		mtx_unlock(&softc->ctl_lock);
		break;
	}
	case CTL_GET_PORT_LIST: {
		struct ctl_port *port;
		struct ctl_port_list *list;
		int i;

		list = (struct ctl_port_list *)addr;

		if (list->alloc_len != (list->alloc_num *
		    sizeof(struct ctl_port_entry))) {
			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
			       "alloc_num %u * sizeof(struct ctl_port_entry) "
			       "%zu\n", __func__, list->alloc_len,
			       list->alloc_num, sizeof(struct ctl_port_entry));
			retval = EINVAL;
			break;
		}
		list->fill_len = 0;
		list->fill_num = 0;
		list->dropped_num = 0;
		i = 0;
		mtx_lock(&softc->ctl_lock);
		STAILQ_FOREACH(port, &softc->port_list, links) {
			struct ctl_port_entry entry, *list_entry;

			if (list->fill_num >= list->alloc_num) {
				list->dropped_num++;
				continue;
			}

			entry.port_type = port->port_type;
			strlcpy(entry.port_name, port->port_name,
				sizeof(entry.port_name));
			entry.targ_port = port->targ_port;
			entry.physical_port = port->physical_port;
			entry.virtual_port = port->virtual_port;
			entry.wwnn = port->wwnn;
			entry.wwpn = port->wwpn;
			if (port->status & CTL_PORT_STATUS_ONLINE)
				entry.online = 1;
			else
				entry.online = 0;

			list_entry = &list->entries[i];

			retval = copyout(&entry, list_entry, sizeof(entry));
			if (retval != 0) {
				printf("%s: CTL_GET_PORT_LIST: copyout "
				       "returned %d\n", __func__, retval);
				break;
			}
			i++;
			list->fill_num++;
			list->fill_len += sizeof(entry);
		}
		mtx_unlock(&softc->ctl_lock);

		/*
		 * If this is non-zero, we had a copyout fault, so there's
		 * probably no point in attempting to set the status inside
		 * the structure.
		 */
		if (retval != 0)
			break;

		if (list->dropped_num > 0)
			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
		else
			list->status = CTL_PORT_LIST_OK;
		break;
	}
	case CTL_DUMP_OOA: {
		union ctl_io *io;
		char printbuf[128];
		struct sbuf sb;

		mtx_lock(&softc->ctl_lock);
		printf("Dumping OOA queues:\n");
		STAILQ_FOREACH(lun, &softc->lun_list, links) {
			mtx_lock(&lun->lun_lock);
			for (io = (union ctl_io *)TAILQ_FIRST(
			     &lun->ooa_queue); io != NULL;
			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
			     ooa_links)) {
				sbuf_new(&sb, printbuf, sizeof(printbuf),
					 SBUF_FIXEDLEN);
				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
					    (intmax_t)lun->lun,
					    io->scsiio.tag_num,
					    (io->io_hdr.flags &
					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
					    (io->io_hdr.flags &
					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
					    (io->io_hdr.flags &
					    CTL_FLAG_ABORT) ? " ABORT" : "",
			                    (io->io_hdr.flags &
		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
				sbuf_finish(&sb);
				printf("%s\n", sbuf_data(&sb));
			}
			mtx_unlock(&lun->lun_lock);
		}
		printf("OOA queues dump done\n");
		mtx_unlock(&softc->ctl_lock);
		break;
	}
	case CTL_GET_OOA: {
		struct ctl_ooa *ooa_hdr;
		struct ctl_ooa_entry *entries;
		uint32_t cur_fill_num;

		ooa_hdr = (struct ctl_ooa *)addr;

		if ((ooa_hdr->alloc_len == 0)
		 || (ooa_hdr->alloc_num == 0)) {
			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
			       "must be non-zero\n", __func__,
			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
			retval = EINVAL;
			break;
		}

		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
		    sizeof(struct ctl_ooa_entry))) {
			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
			       __func__, ooa_hdr->alloc_len,
			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
			retval = EINVAL;
			break;
		}

		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
		if (entries == NULL) {
			printf("%s: could not allocate %d bytes for OOA "
			       "dump\n", __func__, ooa_hdr->alloc_len);
			retval = ENOMEM;
			break;
		}

		mtx_lock(&softc->ctl_lock);
		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
			mtx_unlock(&softc->ctl_lock);
			free(entries, M_CTL);
			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
			       __func__, (uintmax_t)ooa_hdr->lun_num);
			retval = EINVAL;
			break;
		}

		cur_fill_num = 0;

		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
			STAILQ_FOREACH(lun, &softc->lun_list, links) {
				ctl_ioctl_fill_ooa(lun, &cur_fill_num,
				    ooa_hdr, entries);
			}
		} else {
			lun = softc->ctl_luns[ooa_hdr->lun_num];
			ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
			    entries);
		}
		mtx_unlock(&softc->ctl_lock);

		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
		ooa_hdr->fill_len = ooa_hdr->fill_num *
			sizeof(struct ctl_ooa_entry);
		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
		if (retval != 0) {
			printf("%s: error copying out %d bytes for OOA dump\n", 
			       __func__, ooa_hdr->fill_len);
		}

		getbinuptime(&ooa_hdr->cur_bt);

		if (cur_fill_num > ooa_hdr->alloc_num) {
			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
		} else {
			ooa_hdr->dropped_num = 0;
			ooa_hdr->status = CTL_OOA_OK;
		}

		free(entries, M_CTL);
		break;
	}
	case CTL_CHECK_OOA: {
		union ctl_io *io;
		struct ctl_ooa_info *ooa_info;


		ooa_info = (struct ctl_ooa_info *)addr;

		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
			ooa_info->status = CTL_OOA_INVALID_LUN;
			break;
		}
		mtx_lock(&softc->ctl_lock);
		lun = softc->ctl_luns[ooa_info->lun_id];
		if (lun == NULL) {
			mtx_unlock(&softc->ctl_lock);
			ooa_info->status = CTL_OOA_INVALID_LUN;
			break;
		}
		mtx_lock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);
		ooa_info->num_entries = 0;
		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
		     &io->io_hdr, ooa_links)) {
			ooa_info->num_entries++;
		}
		mtx_unlock(&lun->lun_lock);

		ooa_info->status = CTL_OOA_SUCCESS;

		break;
	}
	case CTL_DELAY_IO: {
		struct ctl_io_delay_info *delay_info;

		delay_info = (struct ctl_io_delay_info *)addr;

#ifdef CTL_IO_DELAY
		mtx_lock(&softc->ctl_lock);

		if ((delay_info->lun_id >= CTL_MAX_LUNS)
		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
		} else {
			lun = softc->ctl_luns[delay_info->lun_id];
			mtx_lock(&lun->lun_lock);

			delay_info->status = CTL_DELAY_STATUS_OK;

			switch (delay_info->delay_type) {
			case CTL_DELAY_TYPE_CONT:
				break;
			case CTL_DELAY_TYPE_ONESHOT:
				break;
			default:
				delay_info->status =
					CTL_DELAY_STATUS_INVALID_TYPE;
				break;
			}

			switch (delay_info->delay_loc) {
			case CTL_DELAY_LOC_DATAMOVE:
				lun->delay_info.datamove_type =
					delay_info->delay_type;
				lun->delay_info.datamove_delay =
					delay_info->delay_secs;
				break;
			case CTL_DELAY_LOC_DONE:
				lun->delay_info.done_type =
					delay_info->delay_type;
				lun->delay_info.done_delay =
					delay_info->delay_secs;
				break;
			default:
				delay_info->status =
					CTL_DELAY_STATUS_INVALID_LOC;
				break;
			}
			mtx_unlock(&lun->lun_lock);
		}

		mtx_unlock(&softc->ctl_lock);
#else
		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
#endif /* CTL_IO_DELAY */
		break;
	}
	case CTL_REALSYNC_SET: {
		int *syncstate;

		syncstate = (int *)addr;

		mtx_lock(&softc->ctl_lock);
		switch (*syncstate) {
		case 0:
			softc->flags &= ~CTL_FLAG_REAL_SYNC;
			break;
		case 1:
			softc->flags |= CTL_FLAG_REAL_SYNC;
			break;
		default:
			retval = EINVAL;
			break;
		}
		mtx_unlock(&softc->ctl_lock);
		break;
	}
	case CTL_REALSYNC_GET: {
		int *syncstate;

		syncstate = (int*)addr;

		mtx_lock(&softc->ctl_lock);
		if (softc->flags & CTL_FLAG_REAL_SYNC)
			*syncstate = 1;
		else
			*syncstate = 0;
		mtx_unlock(&softc->ctl_lock);

		break;
	}
	case CTL_SETSYNC:
	case CTL_GETSYNC: {
		struct ctl_sync_info *sync_info;

		sync_info = (struct ctl_sync_info *)addr;

		mtx_lock(&softc->ctl_lock);
		lun = softc->ctl_luns[sync_info->lun_id];
		if (lun == NULL) {
			mtx_unlock(&softc->ctl_lock);
			sync_info->status = CTL_GS_SYNC_NO_LUN;
			break;
		}
		/*
		 * Get or set the sync interval.  We're not bounds checking
		 * in the set case, hopefully the user won't do something
		 * silly.
		 */
		mtx_lock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);
		if (cmd == CTL_GETSYNC)
			sync_info->sync_interval = lun->sync_interval;
		else
			lun->sync_interval = sync_info->sync_interval;
		mtx_unlock(&lun->lun_lock);

		sync_info->status = CTL_GS_SYNC_OK;

		break;
	}
	case CTL_GETSTATS: {
		struct ctl_stats *stats;
		int i;

		stats = (struct ctl_stats *)addr;

		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
		     stats->alloc_len) {
			stats->status = CTL_SS_NEED_MORE_SPACE;
			stats->num_luns = softc->num_luns;
			break;
		}
		/*
		 * XXX KDM no locking here.  If the LUN list changes,
		 * things can blow up.
		 */
		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
		     i++, lun = STAILQ_NEXT(lun, links)) {
			retval = copyout(&lun->stats, &stats->lun_stats[i],
					 sizeof(lun->stats));
			if (retval != 0)
				break;
		}
		stats->num_luns = softc->num_luns;
		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
				 softc->num_luns;
		stats->status = CTL_SS_OK;
#ifdef CTL_TIME_IO
		stats->flags = CTL_STATS_FLAG_TIME_VALID;
#else
		stats->flags = CTL_STATS_FLAG_NONE;
#endif
		getnanouptime(&stats->timestamp);
		break;
	}
	case CTL_ERROR_INJECT: {
		struct ctl_error_desc *err_desc, *new_err_desc;

		err_desc = (struct ctl_error_desc *)addr;

		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
				      M_WAITOK | M_ZERO);
		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));

		mtx_lock(&softc->ctl_lock);
		lun = softc->ctl_luns[err_desc->lun_id];
		if (lun == NULL) {
			mtx_unlock(&softc->ctl_lock);
			free(new_err_desc, M_CTL);
			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
			       __func__, (uintmax_t)err_desc->lun_id);
			retval = EINVAL;
			break;
		}
		mtx_lock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);

		/*
		 * We could do some checking here to verify the validity
		 * of the request, but given the complexity of error
		 * injection requests, the checking logic would be fairly
		 * complex.
		 *
		 * For now, if the request is invalid, it just won't get
		 * executed and might get deleted.
		 */
		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);

		/*
		 * XXX KDM check to make sure the serial number is unique,
		 * in case we somehow manage to wrap.  That shouldn't
		 * happen for a very long time, but it's the right thing to
		 * do.
		 */
		new_err_desc->serial = lun->error_serial;
		err_desc->serial = lun->error_serial;
		lun->error_serial++;

		mtx_unlock(&lun->lun_lock);
		break;
	}
	case CTL_ERROR_INJECT_DELETE: {
		struct ctl_error_desc *delete_desc, *desc, *desc2;
		int delete_done;

		delete_desc = (struct ctl_error_desc *)addr;
		delete_done = 0;

		mtx_lock(&softc->ctl_lock);
		lun = softc->ctl_luns[delete_desc->lun_id];
		if (lun == NULL) {
			mtx_unlock(&softc->ctl_lock);
			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
			       __func__, (uintmax_t)delete_desc->lun_id);
			retval = EINVAL;
			break;
		}
		mtx_lock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);
		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
			if (desc->serial != delete_desc->serial)
				continue;

			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
				      links);
			free(desc, M_CTL);
			delete_done = 1;
		}
		mtx_unlock(&lun->lun_lock);
		if (delete_done == 0) {
			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
			       "error serial %ju on LUN %u\n", __func__, 
			       delete_desc->serial, delete_desc->lun_id);
			retval = EINVAL;
			break;
		}
		break;
	}
	case CTL_DUMP_STRUCTS: {
		int i, j, k;
		struct ctl_port *port;
		struct ctl_frontend *fe;

		mtx_lock(&softc->ctl_lock);
		printf("CTL Persistent Reservation information start:\n");
		for (i = 0; i < CTL_MAX_LUNS; i++) {
			lun = softc->ctl_luns[i];

			if ((lun == NULL)
			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
				continue;

			for (j = 0; j < CTL_MAX_PORTS; j++) {
				if (lun->pr_keys[j] == NULL)
					continue;
				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
					if (lun->pr_keys[j][k] == 0)
						continue;
					printf("  LUN %d port %d iid %d key "
					       "%#jx\n", i, j, k,
					       (uintmax_t)lun->pr_keys[j][k]);
				}
			}
		}
		printf("CTL Persistent Reservation information end\n");
		printf("CTL Ports:\n");
		STAILQ_FOREACH(port, &softc->port_list, links) {
			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
			       port->frontend->name, port->port_type,
			       port->physical_port, port->virtual_port,
			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
				if (port->wwpn_iid[j].in_use == 0 &&
				    port->wwpn_iid[j].wwpn == 0 &&
				    port->wwpn_iid[j].name == NULL)
					continue;

				printf("    iid %u use %d WWPN %#jx '%s'\n",
				    j, port->wwpn_iid[j].in_use,
				    (uintmax_t)port->wwpn_iid[j].wwpn,
				    port->wwpn_iid[j].name);
			}
		}
		printf("CTL Port information end\n");
		mtx_unlock(&softc->ctl_lock);
		/*
		 * XXX KDM calling this without a lock.  We'd likely want
		 * to drop the lock before calling the frontend's dump
		 * routine anyway.
		 */
		printf("CTL Frontends:\n");
		STAILQ_FOREACH(fe, &softc->fe_list, links) {
			printf("  Frontend '%s'\n", fe->name);
			if (fe->fe_dump != NULL)
				fe->fe_dump();
		}
		printf("CTL Frontend information end\n");
		break;
	}
	case CTL_LUN_REQ: {
		struct ctl_lun_req *lun_req;
		struct ctl_backend_driver *backend;

		lun_req = (struct ctl_lun_req *)addr;

		backend = ctl_backend_find(lun_req->backend);
		if (backend == NULL) {
			lun_req->status = CTL_LUN_ERROR;
			snprintf(lun_req->error_str,
				 sizeof(lun_req->error_str),
				 "Backend \"%s\" not found.",
				 lun_req->backend);
			break;
		}
		if (lun_req->num_be_args > 0) {
			lun_req->kern_be_args = ctl_copyin_args(
				lun_req->num_be_args,
				lun_req->be_args,
				lun_req->error_str,
				sizeof(lun_req->error_str));
			if (lun_req->kern_be_args == NULL) {
				lun_req->status = CTL_LUN_ERROR;
				break;
			}
		}

		retval = backend->ioctl(dev, cmd, addr, flag, td);

		if (lun_req->num_be_args > 0) {
			ctl_copyout_args(lun_req->num_be_args,
				      lun_req->kern_be_args);
			ctl_free_args(lun_req->num_be_args,
				      lun_req->kern_be_args);
		}
		break;
	}
	case CTL_LUN_LIST: {
		struct sbuf *sb;
		struct ctl_lun_list *list;
		struct ctl_option *opt;

		list = (struct ctl_lun_list *)addr;

		/*
		 * Allocate a fixed length sbuf here, based on the length
		 * of the user's buffer.  We could allocate an auto-extending
		 * buffer, and then tell the user how much larger our
		 * amount of data is than his buffer, but that presents
		 * some problems:
		 *
		 * 1.  The sbuf(9) routines use a blocking malloc, and so
		 *     we can't hold a lock while calling them with an
		 *     auto-extending buffer.
 		 *
		 * 2.  There is not currently a LUN reference counting
		 *     mechanism, outside of outstanding transactions on
		 *     the LUN's OOA queue.  So a LUN could go away on us
		 *     while we're getting the LUN number, backend-specific
		 *     information, etc.  Thus, given the way things
		 *     currently work, we need to hold the CTL lock while
		 *     grabbing LUN information.
		 *
		 * So, from the user's standpoint, the best thing to do is
		 * allocate what he thinks is a reasonable buffer length,
		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
		 * double the buffer length and try again.  (And repeat
		 * that until he succeeds.)
		 */
		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
		if (sb == NULL) {
			list->status = CTL_LUN_LIST_ERROR;
			snprintf(list->error_str, sizeof(list->error_str),
				 "Unable to allocate %d bytes for LUN list",
				 list->alloc_len);
			break;
		}

		sbuf_printf(sb, "<ctllunlist>\n");

		mtx_lock(&softc->ctl_lock);
		STAILQ_FOREACH(lun, &softc->lun_list, links) {
			mtx_lock(&lun->lun_lock);
			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
					     (uintmax_t)lun->lun);

			/*
			 * Bail out as soon as we see that we've overfilled
			 * the buffer.
			 */
			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "\t<backend_type>%s"
					     "</backend_type>\n",
					     (lun->backend == NULL) ?  "none" :
					     lun->backend->name);

			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
					     lun->be_lun->lun_type);

			if (retval != 0)
				break;

			if (lun->backend == NULL) {
				retval = sbuf_printf(sb, "</lun>\n");
				if (retval != 0)
					break;
				continue;
			}

			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
					     (lun->be_lun->maxlba > 0) ?
					     lun->be_lun->maxlba + 1 : 0);

			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
					     lun->be_lun->blocksize);

			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "\t<serial_number>");

			if (retval != 0)
				break;

			retval = ctl_sbuf_printf_esc(sb,
			    lun->be_lun->serial_num,
			    sizeof(lun->be_lun->serial_num));

			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "</serial_number>\n");
		
			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "\t<device_id>");

			if (retval != 0)
				break;

			retval = ctl_sbuf_printf_esc(sb,
			    lun->be_lun->device_id,
			    sizeof(lun->be_lun->device_id));

			if (retval != 0)
				break;

			retval = sbuf_printf(sb, "</device_id>\n");

			if (retval != 0)
				break;

			if (lun->backend->lun_info != NULL) {
				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
				if (retval != 0)
					break;
			}
			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
				    opt->name, opt->value, opt->name);
				if (retval != 0)
					break;
			}

			retval = sbuf_printf(sb, "</lun>\n");

			if (retval != 0)
				break;
			mtx_unlock(&lun->lun_lock);
		}
		if (lun != NULL)
			mtx_unlock(&lun->lun_lock);
		mtx_unlock(&softc->ctl_lock);

		if ((retval != 0)
		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
			retval = 0;
			sbuf_delete(sb);
			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
			snprintf(list->error_str, sizeof(list->error_str),
				 "Out of space,