freebsd/sys/geom/vinum/geom_vinum_drive.c

/*-
 * Copyright (c) 2004, 2005 Lukas Ertl
 * All rights reserved.
 *
 * 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.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, 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 DAMAGE.
 */

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

#include <sys/param.h>
#include <sys/bio.h>
#include <sys/errno.h>
#include <sys/endian.h>
#include <sys/conf.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/libkern.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/sbuf.h>
#include <sys/systm.h>
#include <sys/time.h>
#include <sys/vimage.h>

#include <geom/geom.h>
#include <geom/vinum/geom_vinum_var.h>
#include <geom/vinum/geom_vinum.h>
#include <geom/vinum/geom_vinum_share.h>

#define GV_LEGACY_I386	0
#define GV_LEGACY_AMD64 1
#define GV_LEGACY_SPARC64 2
#define GV_LEGACY_POWERPC 3

static void	gv_drive_dead(void *, int);
static void	gv_drive_worker(void *);
static int	gv_legacy_header_type(uint8_t *, int);

/*
 * Here are the "offset (size)" for the various struct gv_hdr fields,
 * for the legacy i386 (or 32-bit powerpc), legacy amd64 (or sparc64), and
 * current (cpu & endian agnostic) versions of the on-disk format of the vinum
 * header structure:
 *
 *       i386    amd64   current   field
 *     -------- -------- --------  -----
 *       0 ( 8)   0 ( 8)   0 ( 8)  magic
 *       8 ( 4)   8 ( 8)   8 ( 8)  config_length
 *      12 (32)  16 (32)  16 (32)  label.sysname
 *      44 (32)  48 (32)  48 (32)  label.name
 *      76 ( 4)  80 ( 8)  80 ( 8)  label.date_of_birth.tv_sec
 *      80 ( 4)  88 ( 8)  88 ( 8)  label.date_of_birth.tv_usec
 *      84 ( 4)  96 ( 8)  96 ( 8)  label.last_update.tv_sec
 *      88 ( 4) 104 ( 8) 104 ( 8)  label.last_update.tv_usec
 *      92 ( 8) 112 ( 8) 112 ( 8)  label.drive_size
 *     ======== ======== ========
 *     100      120      120       total size
 *
 * NOTE: i386 and amd64 formats are stored as little-endian; the current
 * format uses big-endian (network order).
 */


/* Checks for legacy format depending on platform. */
static int
gv_legacy_header_type(uint8_t *hdr, int bigendian)
{
	uint32_t *i32;
	int arch_32, arch_64, i;

	/* Set arch according to endianess. */
	if (bigendian) {
		arch_32 = GV_LEGACY_POWERPC;
		arch_64 = GV_LEGACY_SPARC64;
	} else {
		arch_32 = GV_LEGACY_I386;
		arch_64 = GV_LEGACY_AMD64;
	}

	/* if non-empty hostname overlaps 64-bit config_length */
	i32 = (uint32_t *)(hdr + 12);
	if (*i32 != 0)
		return (arch_32);
	/* check for non-empty hostname */
	if (hdr[16] != 0)
		return (arch_64);
	/* check bytes past 32-bit structure */
	for (i = 100; i < 120; i++)
		if (hdr[i] != 0)
			return (arch_32);
	/* check for overlapping timestamp */
	i32 = (uint32_t *)(hdr + 84);

	if (*i32 == 0)
		return (arch_64);
	return (arch_32);
}

/*
 * Read the header while taking magic number into account, and write it to
 * destination pointer.
 */
int
gv_read_header(struct g_consumer *cp, struct gv_hdr *m_hdr)
{
	struct g_provider *pp;
	uint64_t magic_machdep;
	uint8_t *d_hdr;
	int be, off;

#define GV_GET32(endian)					\
		endian##32toh(*((uint32_t *)&d_hdr[off]));	\
		off += 4
#define GV_GET64(endian)					\
		endian##64toh(*((uint64_t *)&d_hdr[off]));	\
		off += 8

	KASSERT(m_hdr != NULL, ("gv_read_header: null m_hdr"));
	KASSERT(cp != NULL, ("gv_read_header: null cp"));
	pp = cp->provider;
	KASSERT(pp != NULL, ("gv_read_header: null pp"));

	d_hdr = g_read_data(cp, GV_HDR_OFFSET, pp->sectorsize, NULL);
	if (d_hdr == NULL)
		return (-1);
	off = 0;
	m_hdr->magic = GV_GET64(be);
	magic_machdep = *((uint64_t *)&d_hdr[0]);
	/*
	 * The big endian machines will have a reverse of GV_OLD_MAGIC, so we
	 * need to decide if we are running on a big endian machine as well as
	 * checking the magic against the reverse of GV_OLD_MAGIC.
	 */
	be = (m_hdr->magic == magic_machdep);
	if (m_hdr->magic == GV_MAGIC) {
		m_hdr->config_length = GV_GET64(be);
		off = 16;
		bcopy(d_hdr + off, m_hdr->label.sysname, GV_HOSTNAME_LEN);
		off += GV_HOSTNAME_LEN;
		bcopy(d_hdr + off, m_hdr->label.name, GV_MAXDRIVENAME);
		off += GV_MAXDRIVENAME;
		m_hdr->label.date_of_birth.tv_sec = GV_GET64(be);
		m_hdr->label.date_of_birth.tv_usec = GV_GET64(be);
		m_hdr->label.last_update.tv_sec = GV_GET64(be);
		m_hdr->label.last_update.tv_usec = GV_GET64(be);
		m_hdr->label.drive_size = GV_GET64(be);
	} else if (m_hdr->magic != GV_OLD_MAGIC &&
	    m_hdr->magic != le64toh(GV_OLD_MAGIC)) {
		/* Not a gvinum drive. */
		g_free(d_hdr);
		return (-1);
	} else if (gv_legacy_header_type(d_hdr, be) == GV_LEGACY_SPARC64) {
		G_VINUM_DEBUG(1, "detected legacy sparc64 header");
		m_hdr->magic = GV_MAGIC;
		/* Legacy sparc64 on-disk header */
		m_hdr->config_length = GV_GET64(be);
		bcopy(d_hdr + 16, m_hdr->label.sysname, GV_HOSTNAME_LEN);
		off += GV_HOSTNAME_LEN;
		bcopy(d_hdr + 48, m_hdr->label.name, GV_MAXDRIVENAME);
		off += GV_MAXDRIVENAME;
		m_hdr->label.date_of_birth.tv_sec = GV_GET64(be);
		m_hdr->label.date_of_birth.tv_usec = GV_GET64(be);
		m_hdr->label.last_update.tv_sec = GV_GET64(be);
		m_hdr->label.last_update.tv_usec = GV_GET64(be);
		m_hdr->label.drive_size = GV_GET64(be);
	} else if (gv_legacy_header_type(d_hdr, be) == GV_LEGACY_POWERPC) {
		G_VINUM_DEBUG(1, "detected legacy PowerPC header");
		m_hdr->magic = GV_MAGIC;
		/* legacy 32-bit big endian on-disk header */
		m_hdr->config_length = GV_GET32(be);
		bcopy(d_hdr + off, m_hdr->label.sysname, GV_HOSTNAME_LEN);
		off += GV_HOSTNAME_LEN;
		bcopy(d_hdr + off, m_hdr->label.name, GV_MAXDRIVENAME);
		off += GV_MAXDRIVENAME;
		m_hdr->label.date_of_birth.tv_sec = GV_GET32(be);
		m_hdr->label.date_of_birth.tv_usec = GV_GET32(be);
		m_hdr->label.last_update.tv_sec = GV_GET32(be);
		m_hdr->label.last_update.tv_usec = GV_GET32(be);
		m_hdr->label.drive_size = GV_GET64(be);
	} else if (gv_legacy_header_type(d_hdr, be) == GV_LEGACY_I386) {
		G_VINUM_DEBUG(1, "detected legacy i386 header");
		m_hdr->magic = GV_MAGIC;
		/* legacy i386 on-disk header */
		m_hdr->config_length = GV_GET32(le);
		bcopy(d_hdr + off, m_hdr->label.sysname, GV_HOSTNAME_LEN);
		off += GV_HOSTNAME_LEN;
		bcopy(d_hdr + off, m_hdr->label.name, GV_MAXDRIVENAME);
		off += GV_MAXDRIVENAME;
		m_hdr->label.date_of_birth.tv_sec = GV_GET32(le);
		m_hdr->label.date_of_birth.tv_usec = GV_GET32(le);
		m_hdr->label.last_update.tv_sec = GV_GET32(le);
		m_hdr->label.last_update.tv_usec = GV_GET32(le);
		m_hdr->label.drive_size = GV_GET64(le);
	} else {
		G_VINUM_DEBUG(1, "detected legacy amd64 header");
		m_hdr->magic = GV_MAGIC;
		/* legacy amd64 on-disk header */
		m_hdr->config_length = GV_GET64(le);
		bcopy(d_hdr + 16, m_hdr->label.sysname, GV_HOSTNAME_LEN);
		off += GV_HOSTNAME_LEN;
		bcopy(d_hdr + 48, m_hdr->label.name, GV_MAXDRIVENAME);
		off += GV_MAXDRIVENAME;
		m_hdr->label.date_of_birth.tv_sec = GV_GET64(le);
		m_hdr->label.date_of_birth.tv_usec = GV_GET64(le);
		m_hdr->label.last_update.tv_sec = GV_GET64(le);
		m_hdr->label.last_update.tv_usec = GV_GET64(le);
		m_hdr->label.drive_size = GV_GET64(le);
	}

	g_free(d_hdr);
	return (0);
}

/* Write out the gvinum header. */
int
gv_write_header(struct g_consumer *cp, struct gv_hdr *m_hdr)
{
	uint8_t d_hdr[GV_HDR_LEN];
	int off, ret;

#define GV_SET64BE(field)					\
	do {							\
		*((uint64_t *)&d_hdr[off]) = htobe64(field);	\
		off += 8;					\
	} while (0)

	KASSERT(m_hdr != NULL, ("gv_write_header: null m_hdr"));

	off = 0;
	memset(d_hdr, 0, GV_HDR_LEN);
	GV_SET64BE(m_hdr->magic);
	GV_SET64BE(m_hdr->config_length);
	off = 16;
	bcopy(m_hdr->label.sysname, d_hdr + off, GV_HOSTNAME_LEN);
	off += GV_HOSTNAME_LEN;
	bcopy(m_hdr->label.name, d_hdr + off, GV_MAXDRIVENAME);
	off += GV_MAXDRIVENAME;
	GV_SET64BE(m_hdr->label.date_of_birth.tv_sec);
	GV_SET64BE(m_hdr->label.date_of_birth.tv_usec);
	GV_SET64BE(m_hdr->label.last_update.tv_sec);
	GV_SET64BE(m_hdr->label.last_update.tv_usec);
	GV_SET64BE(m_hdr->label.drive_size);

	ret = g_write_data(cp, GV_HDR_OFFSET, d_hdr, GV_HDR_LEN);
	return (ret);
}

void
gv_config_new_drive(struct gv_drive *d)
{
	struct gv_hdr *vhdr;
	struct gv_freelist *fl;

	KASSERT(d != NULL, ("config_new_drive: NULL d"));

	vhdr = g_malloc(sizeof(*vhdr), M_WAITOK | M_ZERO);
	vhdr->magic = GV_MAGIC;
	vhdr->config_length = GV_CFG_LEN;

	mtx_lock(&hostname_mtx);
	bcopy(G_hostname, vhdr->label.sysname, GV_HOSTNAME_LEN);
	mtx_unlock(&hostname_mtx);
	strncpy(vhdr->label.name, d->name, GV_MAXDRIVENAME);
	microtime(&vhdr->label.date_of_birth);

	d->hdr = vhdr;

	LIST_INIT(&d->subdisks);
	LIST_INIT(&d->freelist);

	fl = g_malloc(sizeof(struct gv_freelist), M_WAITOK | M_ZERO);
	fl->offset = GV_DATA_START;
	fl->size = d->avail;
	LIST_INSERT_HEAD(&d->freelist, fl, freelist);
	d->freelist_entries = 1;

	d->bqueue = g_malloc(sizeof(struct bio_queue_head), M_WAITOK | M_ZERO);
	bioq_init(d->bqueue);
	mtx_init(&d->bqueue_mtx, "gv_drive", NULL, MTX_DEF);
	kproc_create(gv_drive_worker, d, NULL, 0, 0, "gv_d %s", d->name);
	d->flags |= GV_DRIVE_THREAD_ACTIVE;
}

void
gv_save_config_all(struct gv_softc *sc)
{
	struct gv_drive *d;

	g_topology_assert();

	LIST_FOREACH(d, &sc->drives, drive) {
		if (d->geom == NULL)
			continue;
		gv_save_config(NULL, d, sc);
	}
}

/* Save the vinum configuration back to disk. */
void
gv_save_config(struct g_consumer *cp, struct gv_drive *d, struct gv_softc *sc)
{
	struct g_geom *gp;
	struct g_consumer *cp2;
	struct gv_hdr *vhdr, *hdr;
	struct sbuf *sb;
	int error;

	g_topology_assert();

	KASSERT(d != NULL, ("gv_save_config: null d"));
	KASSERT(sc != NULL, ("gv_save_config: null sc"));

	/*
	 * We can't save the config on a drive that isn't up, but drives that
	 * were just created aren't officially up yet, so we check a special
	 * flag.
	 */
	if ((d->state != GV_DRIVE_UP) && !(d->flags && GV_DRIVE_NEWBORN))
		return;

	if (cp == NULL) {
		gp = d->geom;
		KASSERT(gp != NULL, ("gv_save_config: null gp"));
		cp2 = LIST_FIRST(&gp->consumer);
		KASSERT(cp2 != NULL, ("gv_save_config: null cp2"));
	} else
		cp2 = cp;

	vhdr = g_malloc(GV_HDR_LEN, M_WAITOK | M_ZERO);
	vhdr->magic = GV_MAGIC;
	vhdr->config_length = GV_CFG_LEN;

	hdr = d->hdr;
	if (hdr == NULL) {
		G_VINUM_DEBUG(0, "drive %s has NULL hdr", d->name);
		g_free(vhdr);
		return;
	}
	microtime(&hdr->label.last_update);
	bcopy(&hdr->label, &vhdr->label, sizeof(struct gv_label));

	sb = sbuf_new(NULL, NULL, GV_CFG_LEN, SBUF_FIXEDLEN);
	gv_format_config(sc, sb, 1, NULL);
	sbuf_finish(sb);

	error = g_access(cp2, 0, 1, 0);
	if (error) {
		G_VINUM_DEBUG(0, "g_access failed on drive %s, errno %d",
		    d->name, error);
		sbuf_delete(sb);
		g_free(vhdr);
		return;
	}
	g_topology_unlock();

	do {
		error = gv_write_header(cp2, vhdr);
		if (error) {
			G_VINUM_DEBUG(0, "writing vhdr failed on drive %s, "
			    "errno %d", d->name, error);
			break;
		}

		error = g_write_data(cp2, GV_CFG_OFFSET, sbuf_data(sb),
		    GV_CFG_LEN);
		if (error) {
			G_VINUM_DEBUG(0, "writing first config copy failed "
			    "on drive %s, errno %d", d->name, error);
			break;
		}
		
		error = g_write_data(cp2, GV_CFG_OFFSET + GV_CFG_LEN,
		    sbuf_data(sb), GV_CFG_LEN);
		if (error)
			G_VINUM_DEBUG(0, "writing second config copy failed "
			    "on drive %s, errno %d", d->name, error);
	} while (0);

	g_topology_lock();
	g_access(cp2, 0, -1, 0);
	sbuf_delete(sb);
	g_free(vhdr);

	if (d->geom != NULL)
		gv_drive_modify(d);
}

/* This resembles g_slice_access(). */
static int
gv_drive_access(struct g_provider *pp, int dr, int dw, int de)
{
	struct g_geom *gp;
	struct g_consumer *cp;
	struct g_provider *pp2;
	struct gv_drive *d;
	struct gv_sd *s, *s2;
	int error;

	gp = pp->geom;
	cp = LIST_FIRST(&gp->consumer);
	if (cp == NULL)
		return (0);

	d = gp->softc;
	if (d == NULL)
		return (0);

	s = pp->private;
	KASSERT(s != NULL, ("gv_drive_access: NULL s"));

	LIST_FOREACH(s2, &d->subdisks, from_drive) {
		if (s == s2)
			continue;
		if (s->drive_offset + s->size <= s2->drive_offset)
			continue;
		if (s2->drive_offset + s2->size <= s->drive_offset)
			continue;

		/* Overlap. */
		pp2 = s2->provider;
		KASSERT(s2 != NULL, ("gv_drive_access: NULL s2"));
		if ((pp->acw + dw) > 0 && pp2->ace > 0)
			return (EPERM);
		if ((pp->ace + de) > 0 && pp2->acw > 0)
			return (EPERM);
	}

	error = g_access(cp, dr, dw, de);
	return (error);
}

static void
gv_drive_done(struct bio *bp)
{
	struct gv_drive *d;

	/* Put the BIO on the worker queue again. */
	d = bp->bio_from->geom->softc;
	bp->bio_cflags |= GV_BIO_DONE;
	mtx_lock(&d->bqueue_mtx);
	bioq_insert_tail(d->bqueue, bp);
	wakeup(d);
	mtx_unlock(&d->bqueue_mtx);
}


static void
gv_drive_start(struct bio *bp)
{
	struct gv_drive *d;
	struct gv_sd *s;

	switch (bp->bio_cmd) {
	case BIO_READ:
	case BIO_WRITE:
	case BIO_DELETE:
		break;
	case BIO_GETATTR:
	default:
		g_io_deliver(bp, EOPNOTSUPP);
		return;
	}

	s = bp->bio_to->private;
	if ((s->state == GV_SD_DOWN) || (s->state == GV_SD_STALE)) {
		g_io_deliver(bp, ENXIO);
		return;
	}

	d = bp->bio_to->geom->softc;

	/*
	 * Put the BIO on the worker queue, where the worker thread will pick
	 * it up.
	 */
	mtx_lock(&d->bqueue_mtx);
	bioq_disksort(d->bqueue, bp);
	wakeup(d);
	mtx_unlock(&d->bqueue_mtx);

}

static void
gv_drive_worker(void *arg)
{
	struct bio *bp, *cbp;
	struct g_geom *gp;
	struct g_provider *pp;
	struct gv_drive *d;
	struct gv_sd *s;
	int error;

	d = arg;

	mtx_lock(&d->bqueue_mtx);
	for (;;) {
		/* We were signaled to exit. */
		if (d->flags & GV_DRIVE_THREAD_DIE)
			break;

		/* Take the first BIO from out queue. */
		bp = bioq_takefirst(d->bqueue);
		if (bp == NULL) {
			msleep(d, &d->bqueue_mtx, PRIBIO, "-", hz/10);
			continue;
 		}
		mtx_unlock(&d->bqueue_mtx);
 
		pp = bp->bio_to;
		gp = pp->geom;

		/* Completed request. */
		if (bp->bio_cflags & GV_BIO_DONE) {
			error = bp->bio_error;

			/* Deliver the original request. */
			g_std_done(bp);

			/* The request had an error, we need to clean up. */
			if (error != 0) {
				g_topology_lock();
				gv_set_drive_state(d, GV_DRIVE_DOWN,
				    GV_SETSTATE_FORCE | GV_SETSTATE_CONFIG);
				g_topology_unlock();
				g_post_event(gv_drive_dead, d, M_WAITOK, d,
				    NULL);
			}

		/* New request, needs to be sent downwards. */
		} else {
			s = pp->private;

			if ((s->state == GV_SD_DOWN) ||
			    (s->state == GV_SD_STALE)) {
				g_io_deliver(bp, ENXIO);
				mtx_lock(&d->bqueue_mtx);
				continue;
			}
			if (bp->bio_offset > s->size) {
				g_io_deliver(bp, EINVAL);
				mtx_lock(&d->bqueue_mtx);
				continue;
			}

			cbp = g_clone_bio(bp);
			if (cbp == NULL) {
				g_io_deliver(bp, ENOMEM);
				mtx_lock(&d->bqueue_mtx);
				continue;
			}
			if (cbp->bio_offset + cbp->bio_length > s->size)
				cbp->bio_length = s->size -
				    cbp->bio_offset;
			cbp->bio_done = gv_drive_done;
			cbp->bio_offset += s->drive_offset;
			g_io_request(cbp, LIST_FIRST(&gp->consumer));
		}

		mtx_lock(&d->bqueue_mtx);
	}

	while ((bp = bioq_takefirst(d->bqueue)) != NULL) {
		mtx_unlock(&d->bqueue_mtx);
		if (bp->bio_cflags & GV_BIO_DONE) 
			g_std_done(bp);
		else
			g_io_deliver(bp, ENXIO);
		mtx_lock(&d->bqueue_mtx);
	}
	mtx_unlock(&d->bqueue_mtx);
	d->flags |= GV_DRIVE_THREAD_DEAD;

	kproc_exit(ENXIO);
}


static void
gv_drive_orphan(struct g_consumer *cp)
{
	struct g_geom *gp;
	struct gv_drive *d;

	g_topology_assert();
	gp = cp->geom;
	g_trace(G_T_TOPOLOGY, "gv_drive_orphan(%s)", gp->name);
	d = gp->softc;
	if (d != NULL) {
		gv_set_drive_state(d, GV_DRIVE_DOWN,
		    GV_SETSTATE_FORCE | GV_SETSTATE_CONFIG);
		g_post_event(gv_drive_dead, d, M_WAITOK, d, NULL);
	} else
		g_wither_geom(gp, ENXIO);
}

static struct g_geom *
gv_drive_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
{
	struct g_geom *gp, *gp2;
	struct g_consumer *cp;
	struct gv_drive *d;
	struct gv_sd *s;
	struct gv_softc *sc;
	struct gv_freelist *fl;
	struct gv_hdr *vhdr;
	int error;
	char *buf, errstr[ERRBUFSIZ];

	vhdr = NULL;
	d = NULL;

	g_trace(G_T_TOPOLOGY, "gv_drive_taste(%s, %s)", mp->name, pp->name);
	g_topology_assert();

	/* Find the VINUM class and its associated geom. */
	gp2 = find_vinum_geom();
	if (gp2 == NULL)
		return (NULL);
	sc = gp2->softc;

	gp = g_new_geomf(mp, "%s.vinumdrive", pp->name);
	gp->start = gv_drive_start;
	gp->orphan = gv_drive_orphan;
	gp->access = gv_drive_access;
	gp->start = gv_drive_start;

	cp = g_new_consumer(gp);
	g_attach(cp, pp);
	error = g_access(cp, 1, 0, 0);
	if (error) {
		g_detach(cp);
		g_destroy_consumer(cp);
		g_destroy_geom(gp);
		return (NULL);
	}

	g_topology_unlock();

	/* Now check if the provided slice is a valid vinum drive. */
	do {
		vhdr = g_malloc(GV_HDR_LEN, M_WAITOK | M_ZERO);
		error = gv_read_header(cp, vhdr);
		if (error) {
			g_free(vhdr);
			break;
		}

		/* A valid vinum drive, let's parse the on-disk information. */
		buf = g_read_data(cp, GV_CFG_OFFSET, GV_CFG_LEN, NULL);
		if (buf == NULL) {
			g_free(vhdr);
			break;
		}
		g_topology_lock();
		gv_parse_config(sc, buf, 1);
		g_free(buf);

		/*
		 * Let's see if this drive is already known in the
		 * configuration.
		 */
		d = gv_find_drive(sc, vhdr->label.name);

		/* We already know about this drive. */
		if (d != NULL) {
			/* Check if this drive already has a geom. */
			if (d->geom != NULL) {
				g_topology_unlock();
				g_free(vhdr);
				break;
			}
			bcopy(vhdr, d->hdr, sizeof(*vhdr));
			g_free(vhdr);

		/* This is a new drive. */
		} else {
			d = g_malloc(sizeof(*d), M_WAITOK | M_ZERO);

			/* Initialize all needed variables. */
			d->size = pp->mediasize - GV_DATA_START;
			d->avail = d->size;
			d->hdr = vhdr;
			strncpy(d->name, vhdr->label.name, GV_MAXDRIVENAME);
			LIST_INIT(&d->subdisks);
			LIST_INIT(&d->freelist);

			/* We also need a freelist entry. */
			fl = g_malloc(sizeof(*fl), M_WAITOK | M_ZERO);
			fl->offset = GV_DATA_START;
			fl->size = d->avail;
			LIST_INSERT_HEAD(&d->freelist, fl, freelist);
			d->freelist_entries = 1;

			/* Save it into the main configuration. */
			LIST_INSERT_HEAD(&sc->drives, d, drive);
		}

		/*
		 * Create bio queue, queue mutex and a worker thread, if
		 * necessary.
		 */
		if (d->bqueue == NULL) {
			d->bqueue = g_malloc(sizeof(struct bio_queue_head),
			    M_WAITOK | M_ZERO);
			bioq_init(d->bqueue);
		}
		if (mtx_initialized(&d->bqueue_mtx) == 0)
			mtx_init(&d->bqueue_mtx, "gv_drive", NULL, MTX_DEF);

		if (!(d->flags & GV_DRIVE_THREAD_ACTIVE)) {
			kproc_create(gv_drive_worker, d, NULL, 0, 0,
			    "gv_d %s", d->name);
			d->flags |= GV_DRIVE_THREAD_ACTIVE;
		}

		g_access(cp, -1, 0, 0);

		gp->softc = d;
		d->geom = gp;
		d->vinumconf = sc;
		strncpy(d->device, pp->name, GV_MAXDRIVENAME);

		/*
		 * Find out which subdisks belong to this drive and crosslink
		 * them.
		 */
		LIST_FOREACH(s, &sc->subdisks, sd) {
			if (!strncmp(s->drive, d->name, GV_MAXDRIVENAME))
				/* XXX: errors ignored */
				gv_sd_to_drive(sc, d, s, errstr,
				    sizeof(errstr));
		}

		/* This drive is now up for sure. */
		gv_set_drive_state(d, GV_DRIVE_UP, 0);

		/*
		 * If there are subdisks on this drive, we need to create
		 * providers for them.
		 */ 
		if (d->sdcount)
			gv_drive_modify(d);

		return (gp);

	} while (0);

	g_topology_lock();
	g_access(cp, -1, 0, 0);

	g_detach(cp);
	g_destroy_consumer(cp);
	g_destroy_geom(gp);
	return (NULL);
}

/*
 * Modify the providers for the given drive 'd'.  It is assumed that the
 * subdisk list of 'd' is already correctly set up.
 */
void
gv_drive_modify(struct gv_drive *d)
{
	struct g_geom *gp;
	struct g_consumer *cp;
	struct g_provider *pp, *pp2;
	struct gv_sd *s;

	KASSERT(d != NULL, ("gv_drive_modify: null d"));
	gp = d->geom;
	KASSERT(gp != NULL, ("gv_drive_modify: null gp"));
	cp = LIST_FIRST(&gp->consumer);
	KASSERT(cp != NULL, ("gv_drive_modify: null cp"));
	pp = cp->provider;
	KASSERT(pp != NULL, ("gv_drive_modify: null pp"));

	g_topology_assert();

	LIST_FOREACH(s, &d->subdisks, from_drive) {
		/* This subdisk already has a provider. */
		if (s->provider != NULL)
			continue;
		pp2 = g_new_providerf(gp, "gvinum/sd/%s", s->name);
		pp2->mediasize = s->size;
		pp2->sectorsize = pp->sectorsize;
		g_error_provider(pp2, 0);
		s->provider = pp2;
		pp2->private = s;
	}
}

static void
gv_drive_dead(void *arg, int flag)
{
	struct g_geom *gp;
	struct g_consumer *cp;
	struct gv_drive *d;
	struct gv_sd *s;

	g_topology_assert();
	KASSERT(arg != NULL, ("gv_drive_dead: NULL arg"));

	if (flag == EV_CANCEL)
		return;

	d = arg;
	if (d->state != GV_DRIVE_DOWN)
		return;

	g_trace(G_T_TOPOLOGY, "gv_drive_dead(%s)", d->name);

	gp = d->geom;
	if (gp == NULL)
		return;

	LIST_FOREACH(cp, &gp->consumer, consumer) {
		if (cp->nstart != cp->nend) {
			G_VINUM_DEBUG(0, "dead drive '%s' still has "
			    "active requests, cannot detach consumer",
			    d->name);
			g_post_event(gv_drive_dead, d, M_WAITOK, d,
			    NULL);
			return;
		}
		if (cp->acr != 0 || cp->acw != 0 || cp->ace != 0)
			g_access(cp, -cp->acr, -cp->acw, -cp->ace);
	}

	G_VINUM_DEBUG(1, "lost drive '%s'", d->name);
	d->geom = NULL;
	LIST_FOREACH(s, &d->subdisks, from_drive) {
		s->provider = NULL;
		s->consumer = NULL;
	}
	gv_kill_drive_thread(d);
	gp->softc = NULL;
	g_wither_geom(gp, ENXIO);
}

static int
gv_drive_destroy_geom(struct gctl_req *req, struct g_class *mp,
    struct g_geom *gp)
{
	struct gv_drive *d;

	g_trace(G_T_TOPOLOGY, "gv_drive_destroy_geom: %s", gp->name);
	g_topology_assert();

	d = gp->softc;
	gv_kill_drive_thread(d);

	g_wither_geom(gp, ENXIO);
	return (0);
}

#define	VINUMDRIVE_CLASS_NAME "VINUMDRIVE"

static struct g_class g_vinum_drive_class = {
	.name = VINUMDRIVE_CLASS_NAME,
	.version = G_VERSION,
	.taste = gv_drive_taste,
	.destroy_geom = gv_drive_destroy_geom
};

DECLARE_GEOM_CLASS(g_vinum_drive_class, g_vinum_drive);