Bring in geom_sched, support for scheduling disk I/O requests

in a device independent manner. Also include an example anticipatory
scheduler, gsched_rr, which gives very nice performance improvements
in presence of competing random access patterns.

This is joint work with Fabio Checconi, developed last year
and presented at BSDCan 2009. You can find details in the
README file or at

http://info.iet.unipi.it/~luigi/geom_sched/

6 files changed, 3330 insertions, 0 deletions

diff --git a/sys/geom/sched/README b/sys/geom/sched/README
new file mode 100644
index 0000000..1b52d90
--- /dev/null
+++ b/sys/geom/sched/README
@@ -0,0 +1,162 @@
+
+	--- GEOM BASED DISK SCHEDULERS FOR FREEBSD ---
+
+This code contains a framework for GEOM-based disk schedulers and a
+couple of sample scheduling algorithms that use the framework and
+implement two forms of "anticipatory scheduling" (see below for more
+details).
+
+As a quick example of what this code can give you, try to run "dd",
+"tar", or some other program with highly SEQUENTIAL access patterns,
+together with "cvs", "cvsup", "svn" or other highly RANDOM access patterns
+(this is not a made-up example: it is pretty common for developers
+to have one or more apps doing random accesses, and others that do
+sequential accesses e.g., loading large binaries from disk, checking
+the integrity of tarballs, watching media streams and so on).
+
+These are the results we get on a local machine (AMD BE2400 dual
+core CPU, SATA 250GB disk):
+
+    /mnt is a partition mounted on /dev/ad0s1f
+
+    cvs: 	cvs -d /mnt/home/ncvs-local update -Pd /mnt/ports
+    dd-read:	dd bs=128k of=/dev/null if=/dev/ad0 (or ad0-sched-)
+    dd-writew	dd bs=128k if=/dev/zero of=/mnt/largefile
+
+			NO SCHEDULER		RR SCHEDULER
+                	dd	cvs		dd	cvs
+
+    dd-read only        72 MB/s	----		72 MB/s	---
+    dd-write only	55 MB/s	---		55 MB/s	---
+    dd-read+cvs		 6 MB/s	ok    		30 MB/s	ok
+    dd-write+cvs	55 MB/s slooow		14 MB/s	ok
+
+As you can see, when a cvs is running concurrently with dd, the
+performance drops dramatically, and depending on read or write mode,
+one of the two is severely penalized.  The use of the RR scheduler
+in this example makes the dd-reader go much faster when competing
+with cvs, and lets cvs progress when competing with a writer.
+
+To try it out:
+
+1. USERS OF FREEBSD 7, PLEASE READ CAREFULLY THE FOLLOWING:
+
+    On loading, this module patches one kernel function (g_io_request())
+    so that I/O requests ("bio's") carry a classification tag, useful
+    for scheduling purposes.
+
+    ON FREEBSD 7, the tag is stored in an existing (though rarely used)
+    field of the "struct bio", a solution which makes this module
+    incompatible with other modules using it, such as ZFS and gjournal.
+    Additionally, g_io_request() is patched in-memory to add a call
+    to the function that initializes this field (i386/amd64 only;
+    for other architectures you need to manually patch sys/geom/geom_io.c).
+    See details in the file g_sched.c.
+
+    On FreeBSD 8.0 and above, the above trick is not necessary,
+    as the struct bio contains dedicated fields for the classifier,
+    and hooks for request classifiers.
+
+    If you don't like the above, don't run this code.
+
+2. PLEASE MAKE SURE THAT THE DISK THAT YOU WILL BE USING FOR TESTS
+   DOES NOT CONTAIN PRECIOUS DATA.
+    This is experimental code, so we make no guarantees, though
+    I am routinely using it on my desktop and laptop.
+
+3. EXTRACT AND BUILD THE PROGRAMS
+    A 'make install' in the directory should work (with root privs),
+    or you can even try the binary modules.
+    If you want to build the modules yourself, look at the Makefile.
+
+4. LOAD THE MODULE, CREATE A GEOM NODE, RUN TESTS
+
+    The scheduler's module must be loaded first:
+
+      # kldload gsched_rr
+
+    substitute with gsched_as to test AS.  Then, supposing that you are
+    using /dev/ad0 for testing, a scheduler can be attached to it with:
+
+      # geom sched insert ad0
+
+    The scheduler is inserted transparently in the geom chain, so
+    mounted partitions and filesystems will keep working, but
+    now requests will go through the scheduler.
+
+    To change scheduler on-the-fly, you can reconfigure the geom:
+
+      # geom sched configure -a as ad0.sched.
+
+    assuming that gsched_as was loaded previously.
+
+5. SCHEDULER REMOVAL
+
+    In principle it is possible to remove the scheduler module
+    even on an active chain by doing
+
+	# geom sched destroy ad0.sched.
+
+    However, there is some race in the geom subsystem which makes
+    the removal unsafe if there are active requests on a chain.
+    So, in order to reduce the risk of data losses, make sure
+    you don't remove a scheduler from a chain with ongoing transactions.
+
+--- NOTES ON THE SCHEDULERS ---
+
+The important contribution of this code is the framework to experiment
+with different scheduling algorithms.  'Anticipatory scheduling'
+is a very powerful technique based on the following reasoning:
+
+    The disk throughput is much better if it serves sequential requests.
+    If we have a mix of sequential and random requests, and we see a
+    non-sequential request, do not serve it immediately but instead wait
+    a little bit (2..5ms) to see if there is another one coming that
+    the disk can serve more efficiently.
+
+There are many details that should be added to make sure that the
+mechanism is effective with different workloads and systems, to
+gain a few extra percent in performance, to improve fairness,
+insulation among processes etc.  A discussion of the vast literature
+on the subject is beyond the purpose of this short note.
+
+--------------------------------------------------------------------------
+
+TRANSPARENT INSERT/DELETE
+
+geom_sched is an ordinary geom module, however it is convenient
+to plug it transparently into the geom graph, so that one can
+enable or disable scheduling on a mounted filesystem, and the
+names in /etc/fstab do not depend on the presence of the scheduler.
+
+To understand how this works in practice, remember that in GEOM
+we have "providers" and "geom" objects.
+Say that we want to hook a scheduler on provider "ad0",
+accessible through pointer 'pp'. Originally, pp is attached to
+geom "ad0" (same name, different object) accessible through pointer old_gp
+
+  BEFORE	---> [ pp    --> old_gp ...]
+
+A normal "geom sched create ad0" call would create a new geom node
+on top of provider ad0/pp, and export a newly created provider
+("ad0.sched." accessible through pointer newpp).
+
+  AFTER create  ---> [ newpp --> gp --> cp ] ---> [ pp    --> old_gp ... ]
+
+On top of newpp, a whole tree will be created automatically, and we
+can e.g. mount partitions on /dev/ad0.sched.s1d, and those requests
+will go through the scheduler, whereas any partition mounted on
+the pre-existing device entries will not go through the scheduler.
+
+With the transparent insert mechanism, the original provider "ad0"/pp
+is hooked to the newly created geom, as follows:
+
+  AFTER insert  ---> [ pp    --> gp --> cp ] ---> [ newpp --> old_gp ... ]
+
+so anything that was previously using provider pp will now have
+the requests routed through the scheduler node.
+
+A removal ("geom sched destroy ad0.sched.") will restore the original
+configuration.
+
+# $FreeBSD$
diff --git a/sys/geom/sched/g_sched.c b/sys/geom/sched/g_sched.c
new file mode 100644
index 0000000..9f8d430
--- /dev/null
+++ b/sys/geom/sched/g_sched.c
@@ -0,0 +1,1901 @@
+/*-
+ * Copyright (c) 2009-2010 Fabio Checconi, Luigi Rizzo
+ * 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 AUTHORS 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 AUTHORS 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.
+ */
+
+/*
+ * $Id$
+ * $FreeBSD$
+ *
+ * Main control module for geom-based disk schedulers ('sched').
+ *
+ * USER VIEW
+ * A 'sched' node is typically inserted transparently between
+ * an existing provider pp and its original geom gp
+ *
+ *	[pp --> gp  ..]
+ *
+ * using the command "geom sched insert <provider>" and
+ * resulting in the following topology
+ *
+ *	[pp --> sched_gp --> cp]   [new_pp --> gp ... ]
+ *
+ * Deletion "geom sched destroy <provider>.sched." restores the
+ * original chain. The normal "geom sched create <provide>"
+ * is also supported.
+ *
+ * INTERNALS
+ * Internally, the 'sched' uses the following data structures
+ *
+ *   geom{}         g_sched_softc{}      g_gsched{}
+ * +----------+    +---------------+   +-------------+
+ * |  softc *-|--->| sc_gsched   *-|-->|  gs_init    |
+ * |  ...     |    |               |   |  gs_fini    |
+ * |          |    | [ hash table] |   |  gs_start   |
+ * +----------+    |               |   |  ...        |
+ *                 |               |   +-------------+
+ *                 |               |
+ *                 |               |     g_*_softc{}
+ *                 |               |   +-------------+
+ *                 | sc_data     *-|-->|             |
+ *                 +---------------+   |  algorithm- |
+ *                                     |  specific   |
+ *                                     +-------------+
+ *
+ * A g_sched_softc{} is created with a "geom sched insert" call.
+ * In turn this instantiates a specific scheduling algorithm,
+ * which sets sc_gsched to point to the algorithm callbacks,
+ * and calls gs_init() to create the g_*_softc{} .
+ * The other callbacks (gs_start, gs_next, ...) are invoked
+ * as needed 
+ *
+ * g_sched_softc{} is defined in g_sched.h and mostly used here;
+ * g_gsched{}, and the gs_callbacks, are documented in gs_scheduler.h;
+ * g_*_softc{} is defined/implemented by each algorithm (gs_*.c)
+ *
+ * DATA MOVING
+ * When a bio is received on the provider, it goes to the
+ * g_sched_start() which calls gs_start() to initially queue it;
+ * then we call g_sched_dispatch() that loops around gs_next()
+ * to select zero or more bio's to be sent downstream.
+ *
+ * g_sched_dispatch() can also be called as a result of a timeout,
+ * e.g. when doing anticipation or pacing requests.
+ *
+ * When a bio comes back, it goes to g_sched_done() which in turn
+ * calls gs_done(). The latter does any necessary housekeeping in
+ * the scheduling algorithm, and may decide to call g_sched_dispatch()
+ * to send more bio's downstream.
+ *
+ * If an algorithm needs per-flow queues, these are created
+ * calling gs_init_class() and destroyed with gs_fini_class(),
+ * and they are also inserted in the hash table implemented in
+ * the g_sched_softc{}
+ *
+ * If an algorithm is replaced, or a transparently-inserted node is
+ * removed with "geom sched destroy", we need to remove all references
+ * to the g_*_softc{} and g_sched_softc from the bio's still in
+ * the scheduler. g_sched_forced_dispatch() helps doing this.
+ * XXX need to explain better.
+ */
+
+#include <sys/cdefs.h>
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/module.h>
+#include <sys/lock.h>
+#include <sys/mutex.h>
+#include <sys/bio.h>
+#include <sys/limits.h>
+#include <sys/hash.h>
+#include <sys/sysctl.h>
+#include <sys/malloc.h>
+#include <sys/proc.h>		/* we access curthread */
+#include <geom/geom.h>
+#include "gs_scheduler.h"
+#include "g_sched.h"		/* geom hooks */
+
+/*
+ * Size of the per-geom hash table storing traffic classes.
+ * We may decide to change it at a later time, it has no ABI
+ * implications as it is only used for run-time allocations.
+ */
+#define G_SCHED_HASH_SIZE	32
+
+static int g_sched_destroy(struct g_geom *gp, boolean_t force);
+static int g_sched_destroy_geom(struct gctl_req *req,
+    struct g_class *mp, struct g_geom *gp);
+static void g_sched_config(struct gctl_req *req, struct g_class *mp,
+    const char *verb);
+static struct g_geom *g_sched_taste(struct g_class *mp,
+    struct g_provider *pp, int flags __unused);
+static void g_sched_dumpconf(struct sbuf *sb, const char *indent,
+    struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
+static void g_sched_init(struct g_class *mp);
+static void g_sched_fini(struct g_class *mp);
+
+struct g_class g_sched_class = {
+	.name = G_SCHED_CLASS_NAME,
+	.version = G_VERSION,
+	.ctlreq = g_sched_config,
+	.taste = g_sched_taste,
+	.destroy_geom = g_sched_destroy_geom,
+	.init = g_sched_init,
+	.fini = g_sched_fini
+};
+
+MALLOC_DEFINE(M_GEOM_SCHED, "GEOM_SCHED", "Geom schedulers data structures");
+
+/*
+ * Global variables describing the state of the geom_sched module.
+ * There is only one static instance of this structure.
+ */
+LIST_HEAD(gs_list, g_gsched);	/* type, link field */
+struct geom_sched_vars {
+	struct mtx	gs_mtx;
+	struct gs_list	gs_scheds;	/* list of algorithms */
+	u_int		gs_debug;
+	u_int		gs_sched_count;	/* how many algorithms ? */
+	u_int 		gs_patched;	/* g_io_request was patched */
+
+	u_int		gs_initialized;
+	u_int		gs_expire_secs;	/* expiration of hash entries */
+
+	struct bio_queue_head gs_pending;
+	u_int		gs_npending;
+
+	/* The following are for stats, usually protected by gs_mtx. */
+	u_long		gs_requests;	/* total requests */
+	u_long		gs_done;	/* total done */
+	u_int 		gs_in_flight;	/* requests in flight */
+	u_int 		gs_writes_in_flight;
+	u_int 		gs_bytes_in_flight;
+	u_int 		gs_write_bytes_in_flight;
+
+	char		gs_names[256];	/* names of schedulers */
+};
+
+static struct geom_sched_vars me = {
+	.gs_expire_secs = 10,
+};
+
+SYSCTL_DECL(_kern_geom);
+SYSCTL_NODE(_kern_geom, OID_AUTO, sched, CTLFLAG_RW, 0,
+    "GEOM_SCHED stuff");
+
+SYSCTL_INT(_kern_geom_sched, OID_AUTO, in_flight_wb, CTLFLAG_RD,
+    &me.gs_write_bytes_in_flight, 0, "Write bytes in flight");
+
+SYSCTL_INT(_kern_geom_sched, OID_AUTO, in_flight_b, CTLFLAG_RD,
+    &me.gs_bytes_in_flight, 0, "Bytes in flight");
+
+SYSCTL_UINT(_kern_geom_sched, OID_AUTO, in_flight_w, CTLFLAG_RD,
+    &me.gs_writes_in_flight, 0, "Write Requests in flight");
+
+SYSCTL_UINT(_kern_geom_sched, OID_AUTO, in_flight, CTLFLAG_RD,
+    &me.gs_in_flight, 0, "Requests in flight");
+
+SYSCTL_ULONG(_kern_geom_sched, OID_AUTO, done, CTLFLAG_RD,
+    &me.gs_done, 0, "Total done");
+
+SYSCTL_ULONG(_kern_geom_sched, OID_AUTO, requests, CTLFLAG_RD,
+    &me.gs_requests, 0, "Total requests");
+
+SYSCTL_STRING(_kern_geom_sched, OID_AUTO, algorithms, CTLFLAG_RD,
+    &me.gs_names, 0, "Algorithm names");
+
+SYSCTL_UINT(_kern_geom_sched, OID_AUTO, alg_count, CTLFLAG_RD,
+    &me.gs_sched_count, 0, "Number of algorithms");
+
+SYSCTL_UINT(_kern_geom_sched, OID_AUTO, debug, CTLFLAG_RW,
+    &me.gs_debug, 0, "Debug level");
+
+SYSCTL_UINT(_kern_geom_sched, OID_AUTO, expire_secs, CTLFLAG_RW,
+    &me.gs_expire_secs, 0, "Expire time in seconds");
+
+/*
+ * g_sched calls the scheduler algorithms with this lock held.
+ * The locking functions are exposed so the scheduler algorithms can also
+ * protect themselves e.g. when running a callout handler.
+ */
+void
+g_sched_lock(struct g_geom *gp)
+{
+	struct g_sched_softc *sc = gp->softc;
+
+	mtx_lock(&sc->sc_mtx);
+}
+
+void
+g_sched_unlock(struct g_geom *gp)
+{
+	struct g_sched_softc *sc = gp->softc;
+
+	mtx_unlock(&sc->sc_mtx);
+}
+
+/*
+ * Support functions to handle references to the module,
+ * which are coming from devices using this scheduler.
+ */
+static inline void
+g_gsched_ref(struct g_gsched *gsp)
+{
+
+	atomic_add_int(&gsp->gs_refs, 1);
+}
+
+static inline void
+g_gsched_unref(struct g_gsched *gsp)
+{
+
+	atomic_add_int(&gsp->gs_refs, -1);
+}
+
+/*
+ * Update the stats when this request is done.
+ */
+static void
+g_sched_update_stats(struct bio *bio)
+{
+
+	me.gs_done++;
+	me.gs_in_flight--;
+	me.gs_bytes_in_flight -= bio->bio_length;
+	if (bio->bio_cmd & BIO_WRITE) {
+		me.gs_writes_in_flight--;
+		me.gs_write_bytes_in_flight -= bio->bio_length;
+	}
+}
+
+/*
+ * Dispatch any pending request.
+ */
+static void
+g_sched_forced_dispatch(struct g_geom *gp)
+{
+	struct g_sched_softc *sc = gp->softc;
+	struct g_gsched *gsp = sc->sc_gsched;
+	struct bio *bp;
+
+	KASSERT(mtx_owned(&sc->sc_mtx),
+	    ("sc_mtx not owned during forced dispatch"));
+
+	while ((bp = gsp->gs_next(sc->sc_data, 1)) != NULL)
+		g_io_request(bp, LIST_FIRST(&gp->consumer));
+}
+
+/*
+ * The main dispatch loop, called either here after the start
+ * routine, or by scheduling algorithms when they receive a timeout
+ * or a 'done' notification.  Does not share code with the forced
+ * dispatch path, since the gs_done() callback can call us.
+ */
+void
+g_sched_dispatch(struct g_geom *gp)
+{
+	struct g_sched_softc *sc = gp->softc;
+	struct g_gsched *gsp = sc->sc_gsched;
+	struct bio *bp;
+
+	KASSERT(mtx_owned(&sc->sc_mtx), ("sc_mtx not owned during dispatch"));
+
+	if ((sc->sc_flags & G_SCHED_FLUSHING))
+		return;
+
+	while ((bp = gsp->gs_next(sc->sc_data, 0)) != NULL)
+		g_io_request(bp, LIST_FIRST(&gp->consumer));
+}
+
+/*
+ * Recent (8.0 and above) versions of FreeBSD have support to
+ * register classifiers of disk requests. The classifier is
+ * invoked by g_io_request(), and stores the information into
+ * bp->bio_classifier1.
+ *
+ * Support for older versions, which is left here only for
+ * documentation purposes, relies on two hacks:
+ * 1. classification info is written into the bio_caller1
+ *    field of the topmost node in the bio chain. This field
+ *    is rarely used, but this module is incompatible with
+ *    those that use bio_caller1 for other purposes,
+ *    such as ZFS and gjournal;
+ * 2. g_io_request() is patched in-memory when the module is
+ *    loaded, so that the function calls a classifier as its
+ *    first thing. g_io_request() is restored when the module
+ *    is unloaded. This functionality is only supported for
+ *    x86 and amd64, other architectures need source code changes.
+ */
+
+/*
+ * Lookup the identity of the issuer of the original request.
+ * In the current implementation we use the curthread of the
+ * issuer, but different mechanisms may be implemented later
+ * so we do not make assumptions on the return value which for
+ * us is just an opaque identifier.
+ */
+
+static inline u_long
+g_sched_classify(struct bio *bp)
+{
+
+#if __FreeBSD_version > 800098
+	/* we have classifier fields in the struct bio */
+#define HAVE_BIO_CLASSIFIER
+	return ((u_long)bp->bio_classifier1);
+#else
+#warning old version!!!
+	while (bp->bio_parent != NULL)
+		bp = bp->bio_parent;
+
+	return ((u_long)bp->bio_caller1);
+#endif
+}
+
+/* Return the hash chain for the given key. */
+static inline struct g_hash *
+g_sched_hash(struct g_sched_softc *sc, u_long key)
+{
+
+	return (&sc->sc_hash[key & sc->sc_mask]);
+}
+
+/*
+ * Helper function for the children classes, which takes
+ * a geom and a bio and returns the private descriptor
+ * associated to the request.  This involves fetching
+ * the classification field and [al]locating the
+ * corresponding entry in the hash table.
+ */
+void *
+g_sched_get_class(struct g_geom *gp, struct bio *bp)
+{
+	struct g_sched_softc *sc;
+	struct g_sched_class *gsc;
+	struct g_gsched *gsp;
+	struct g_hash *bucket;
+	u_long key;
+
+	sc = gp->softc;
+	key = g_sched_classify(bp);
+	bucket = g_sched_hash(sc, key);
+	LIST_FOREACH(gsc, bucket, gsc_clist) {
+		if (key == gsc->gsc_key) {
+			gsc->gsc_refs++;
+			return (gsc->gsc_priv);
+		}
+	}
+
+	gsp = sc->sc_gsched;
+	gsc = malloc(sizeof(*gsc) + gsp->gs_priv_size,
+	    M_GEOM_SCHED, M_NOWAIT | M_ZERO);
+	if (!gsc)
+		return (NULL);
+
+	if (gsp->gs_init_class(sc->sc_data, gsc->gsc_priv)) {
+		free(gsc, M_GEOM_SCHED);
+		return (NULL);
+	}
+
+	gsc->gsc_refs = 2;	/* 1 for the hash table, 1 for the caller. */
+	gsc->gsc_key = key;
+	LIST_INSERT_HEAD(bucket, gsc, gsc_clist);
+
+	gsc->gsc_expire = ticks + me.gs_expire_secs * hz;
+
+	return (gsc->gsc_priv);
+}
+
+/*
+ * Release a reference to the per-client descriptor,
+ */
+void
+g_sched_put_class(struct g_geom *gp, void *priv)
+{
+	struct g_sched_class *gsc;
+	struct g_sched_softc *sc;
+
+	gsc = g_sched_priv2class(priv);
+	gsc->gsc_expire = ticks + me.gs_expire_secs * hz;
+
+	if (--gsc->gsc_refs > 0)
+		return;
+
+	sc = gp->softc;
+	sc->sc_gsched->gs_fini_class(sc->sc_data, priv);
+
+	LIST_REMOVE(gsc, gsc_clist);
+	free(gsc, M_GEOM_SCHED);
+}
+
+static void
+g_sched_hash_fini(struct g_geom *gp, struct g_hash *hp, u_long mask,
+    struct g_gsched *gsp, void *data)
+{
+	struct g_sched_class *cp, *cp2;
+	int i;
+
+	if (!hp)
+		return;
+
+	if (data && gsp->gs_hash_unref)
+		gsp->gs_hash_unref(data);
+
+	for (i = 0; i < G_SCHED_HASH_SIZE; i++) {
+		LIST_FOREACH_SAFE(cp, &hp[i], gsc_clist, cp2)
+			g_sched_put_class(gp, cp->gsc_priv);
+	}
+
+	hashdestroy(hp, M_GEOM_SCHED, mask);
+}
+
+static struct g_hash *
+g_sched_hash_init(struct g_gsched *gsp, u_long *mask, int flags)
+{
+	struct g_hash *hash;
+
+	if (gsp->gs_priv_size == 0)
+		return (NULL);
+
+	hash = hashinit_flags(G_SCHED_HASH_SIZE, M_GEOM_SCHED, mask, flags);
+
+	return (hash);
+}
+
+static void
+g_sched_flush_classes(struct g_geom *gp)
+{
+	struct g_sched_softc *sc;
+	struct g_sched_class *cp, *cp2;
+	int i;
+
+	sc = gp->softc;
+
+	if (!sc->sc_hash || ticks - sc->sc_flush_ticks <= 0)
+		return;
+
+	for (i = 0; i < G_SCHED_HASH_SIZE; i++) {
+		LIST_FOREACH_SAFE(cp, &sc->sc_hash[i], gsc_clist, cp2) {
+			if (cp->gsc_refs == 1 && ticks - cp->gsc_expire > 0)
+				g_sched_put_class(gp, cp->gsc_priv);
+		}
+	}
+
+	sc->sc_flush_ticks = ticks + me.gs_expire_secs * hz;
+}
+
+/*
+ * Wait for the completion of any outstanding request.  To ensure
+ * that this does not take forever the caller has to make sure that
+ * no new request enter the scehduler before calling us.
+ *
+ * Must be called with the gp mutex held and topology locked.
+ */
+static int
+g_sched_wait_pending(struct g_geom *gp)
+{
+	struct g_sched_softc *sc = gp->softc;
+	int endticks = ticks + hz;
+
+	g_topology_assert();
+
+	while (sc->sc_pending && endticks - ticks >= 0)
+		msleep(gp, &sc->sc_mtx, 0, "sched_wait_pending", hz / 4);
+
+	return (sc->sc_pending ? ETIMEDOUT : 0);
+}
+
+static int
+g_sched_remove_locked(struct g_geom *gp, struct g_gsched *gsp)
+{
+	struct g_sched_softc *sc = gp->softc;
+	int error;
+
+	/* Set the flushing flag: new bios will not enter the scheduler. */
+	sc->sc_flags |= G_SCHED_FLUSHING;
+
+	g_sched_forced_dispatch(gp);
+	error = g_sched_wait_pending(gp);
+	if (error)
+		goto failed;
+	
+	/* No more requests pending or in flight from the old gsp. */
+
+	g_sched_hash_fini(gp, sc->sc_hash, sc->sc_mask, gsp, sc->sc_data);
+	sc->sc_hash = NULL;
+
+	/*
+	 * Avoid deadlock here by releasing the gp mutex and reacquiring
+	 * it once done.  It should be safe, since no reconfiguration or
+	 * destruction can take place due to the geom topology lock; no
+	 * new request can use the current sc_data since we flagged the
+	 * geom as being flushed.
+	 */
+	g_sched_unlock(gp);
+	gsp->gs_fini(sc->sc_data);
+	g_sched_lock(gp);
+
+	sc->sc_gsched = NULL;
+	sc->sc_data = NULL;
+	g_gsched_unref(gsp);
+
+failed:
+	sc->sc_flags &= ~G_SCHED_FLUSHING;
+
+	return (error);
+}
+
+static int
+g_sched_remove(struct g_geom *gp, struct g_gsched *gsp)
+{
+	int error;
+
+	g_sched_lock(gp);
+	error = g_sched_remove_locked(gp, gsp); /* gsp is surely non-null */
+	g_sched_unlock(gp);
+
+	return (error);
+}
+
+/*
+ * Support function for create/taste -- locate the desired
+ * algorithm and grab a reference to it.
+ */
+static struct g_gsched *
+g_gsched_find(const char *name)
+{
+	struct g_gsched *gsp = NULL;
+
+	mtx_lock(&me.gs_mtx);
+	LIST_FOREACH(gsp, &me.gs_scheds, glist) {
+		if (strcmp(name, gsp->gs_name) == 0) {
+			g_gsched_ref(gsp);
+			break;
+		}
+	}
+	mtx_unlock(&me.gs_mtx);
+
+	return (gsp);
+}
+
+/*
+ * Rebuild the list of scheduler names.
+ * To be called with me.gs_mtx lock held.
+ */
+static void
+g_gsched_build_names(struct g_gsched *gsp)
+{
+	int pos, l;
+	struct g_gsched *cur;
+
+	pos = 0;
+	LIST_FOREACH(cur, &me.gs_scheds, glist) {
+		l = strlen(cur->gs_name);
+		if (l + pos + 1 + 1 < sizeof(me.gs_names)) {
+			if (pos != 0)
+				me.gs_names[pos++] = ' ';
+			strcpy(me.gs_names + pos, cur->gs_name);
+			pos += l;
+		}
+	}
+	me.gs_names[pos] = '\0';
+}
+
+/*
+ * Register or unregister individual scheduling algorithms.
+ */
+static int
+g_gsched_register(struct g_gsched *gsp)
+{
+	struct g_gsched *cur;
+	int error = 0;
+
+	mtx_lock(&me.gs_mtx);
+	LIST_FOREACH(cur, &me.gs_scheds, glist) {
+		if (strcmp(gsp->gs_name, cur->gs_name) == 0)
+			break;
+	}
+	if (cur != NULL) {
+		G_SCHED_DEBUG(0, "A scheduler named %s already"
+		    "exists.", gsp->gs_name);
+		error = EEXIST;
+	} else {
+		LIST_INSERT_HEAD(&me.gs_scheds, gsp, glist);
+		gsp->gs_refs = 1;
+		me.gs_sched_count++;
+		g_gsched_build_names(gsp);
+	}
+	mtx_unlock(&me.gs_mtx);
+
+	return (error);
+}
+
+struct g_gsched_unregparm {
+	struct g_gsched *gup_gsp;
+	int		gup_error;
+};
+
+static void
+g_gsched_unregister(void *arg, int flag)
+{
+	struct g_gsched_unregparm *parm = arg;
+	struct g_gsched *gsp = parm->gup_gsp, *cur, *tmp;
+	struct g_sched_softc *sc;
+	struct g_geom *gp, *gp_tmp;
+	int error;
+
+	parm->gup_error = 0;
+
+	g_topology_assert();
+
+	if (flag == EV_CANCEL)
+		return;
+
+	mtx_lock(&me.gs_mtx);
+
+	LIST_FOREACH_SAFE(gp, &g_sched_class.geom, geom, gp_tmp) {
+		if (gp->class != &g_sched_class)
+			continue;	/* Should not happen. */
+
+		sc = gp->softc;
+		if (sc->sc_gsched == gsp) {
+			error = g_sched_remove(gp, gsp);
+			if (error)
+				goto failed;
+		}
+	}
+		
+	LIST_FOREACH_SAFE(cur, &me.gs_scheds, glist, tmp) {
+		if (cur != gsp)
+			continue;
+
+		if (gsp->gs_refs != 1) {
+			G_SCHED_DEBUG(0, "%s still in use.",
+			    gsp->gs_name);
+			parm->gup_error = EBUSY;
+		} else {
+			LIST_REMOVE(gsp, glist);
+			me.gs_sched_count--;
+			g_gsched_build_names(gsp);
+		}
+		break;
+	}
+
+	if (cur == NULL) {
+		G_SCHED_DEBUG(0, "%s not registered.", gsp->gs_name);
+		parm->gup_error = ENOENT;
+	}
+
+failed:
+	mtx_unlock(&me.gs_mtx);
+}
+
+static inline void
+g_gsched_global_init(void)
+{
+
+	if (!me.gs_initialized) {
+		G_SCHED_DEBUG(0, "Initializing global data.");
+		mtx_init(&me.gs_mtx, "gsched", NULL, MTX_DEF);
+		LIST_INIT(&me.gs_scheds);
+		gs_bioq_init(&me.gs_pending);
+		me.gs_initialized = 1;
+	}
+}
+
+/*
+ * Module event called when a scheduling algorithm module is loaded or
+ * unloaded.
+ */
+int
+g_gsched_modevent(module_t mod, int cmd, void *arg)
+{
+	struct g_gsched *gsp = arg;
+	struct g_gsched_unregparm parm;
+	int error;
+
+	G_SCHED_DEBUG(0, "Modevent %d.", cmd);
+
+	/*
+	 * If the module is loaded at boot, the geom thread that calls
+	 * g_sched_init() might actually run after g_gsched_modevent(),
+	 * so make sure that the module is properly initialized.
+	 */
+	g_gsched_global_init();
+
+	error = EOPNOTSUPP;
+	switch (cmd) {
+	case MOD_LOAD:
+		error = g_gsched_register(gsp);
+		G_SCHED_DEBUG(0, "Loaded module %s error %d.",
+		    gsp->gs_name, error);
+		if (error == 0)
+			g_retaste(&g_sched_class);
+		break;
+
+	case MOD_UNLOAD:
+		parm.gup_gsp = gsp;
+		parm.gup_error = 0;
+
+		error = g_waitfor_event(g_gsched_unregister,
+		    &parm, M_WAITOK, NULL);
+		if (error == 0)
+			error = parm.gup_error;
+		G_SCHED_DEBUG(0, "Unloaded module %s error %d.",
+		    gsp->gs_name, error);
+		break;
+	};
+
+	return (error);
+}
+
+#ifdef KTR
+#define	TRC_BIO_EVENT(e, bp)	g_sched_trace_bio_ ## e (bp)
+static inline int
+g_sched_issuer_pid(struct bio *bp)
+{
+	struct thread *thread = g_sched_issuer(bp);
+
+	return (thread->td_tid);
+}
+
+static inline char
+g_sched_type(struct bio *bp)
+{
+
+	if (0 != (bp->bio_cmd & BIO_READ))
+		return ('R');
+	else if (0 != (bp->bio_cmd & BIO_WRITE))
+		return ('W');
+	return ('U');
+}
+
+static inline void
+g_sched_trace_bio_START(struct bio *bp)
+{
+
+	CTR5(KTR_GSCHED, "S %d %c %lu/%lu %lu", g_sched_issuer_pid(bp),
+	    g_sched_type(bp), bp->bio_offset / ULONG_MAX,
+	    bp->bio_offset, bp->bio_length);
+}
+
+static inline void
+g_sched_trace_bio_DONE(struct bio *bp)
+{
+
+	CTR5(KTR_GSCHED, "D %d %c %lu/%lu %lu", g_sched_issuer_pid(bp),
+	    g_sched_type(bp), bp->bio_offset / ULONG_MAX,
+	    bp->bio_offset, bp->bio_length);
+}
+#else
+#define	TRC_BIO_EVENT(e, bp)
+#endif
+
+/*
+ * g_sched_done() and g_sched_start() dispatch the geom requests to
+ * the scheduling algorithm in use.
+ */
+static void
+g_sched_done(struct bio *bio)
+{
+	struct g_geom *gp = bio->bio_caller2;
+	struct g_sched_softc *sc = gp->softc;
+
+	TRC_BIO_EVENT(DONE, bio);
+
+	KASSERT(bio->bio_caller1, ("null bio_caller1 in g_sched_done"));
+
+	g_sched_lock(gp);
+
+	g_sched_update_stats(bio);
+	sc->sc_gsched->gs_done(sc->sc_data, bio);
+	if (!--sc->sc_pending)
+		wakeup(gp);
+
+	g_sched_flush_classes(gp);
+	g_sched_unlock(gp);
+
+	g_std_done(bio);
+}
+
+static void
+g_sched_start(struct bio *bp)
+{
+	struct g_geom *gp = bp->bio_to->geom;
+	struct g_sched_softc *sc = gp->softc;
+	struct bio *cbp;
+
+	TRC_BIO_EVENT(START, bp);
+	G_SCHED_LOGREQ(bp, "Request received.");
+
+	cbp = g_clone_bio(bp);
+	if (cbp == NULL) {
+		g_io_deliver(bp, ENOMEM);
+		return;
+	}
+	cbp->bio_done = g_sched_done;
+	cbp->bio_to = LIST_FIRST(&gp->provider);
+	KASSERT(cbp->bio_to != NULL, ("NULL provider"));
+
+	/* We only schedule reads and writes. */
+	if (0 == (bp->bio_cmd & (BIO_READ | BIO_WRITE)))
+		goto bypass;
+
+	G_SCHED_LOGREQ(cbp, "Sending request.");
+
+	g_sched_lock(gp);
+	/*
+	 * Call the algorithm's gs_start to queue the request in the
+	 * scheduler. If gs_start fails then pass the request down,
+	 * otherwise call g_sched_dispatch() which tries to push
+	 * one or more requests down.
+	 */
+	if (!sc->sc_gsched || (sc->sc_flags & G_SCHED_FLUSHING) ||
+	    sc->sc_gsched->gs_start(sc->sc_data, cbp)) {
+		g_sched_unlock(gp);
+		goto bypass;
+	}
+	/*
+	 * We use bio_caller1 to mark requests that are scheduled
+	 * so make sure it is not NULL.
+	 */
+	if (cbp->bio_caller1 == NULL)
+		cbp->bio_caller1 = &me;	/* anything not NULL */
+
+	cbp->bio_caller2 = gp;
+	sc->sc_pending++;
+
+	/* Update general stats. */
+	me.gs_in_flight++;
+	me.gs_requests++;
+	me.gs_bytes_in_flight += bp->bio_length;
+	if (bp->bio_cmd & BIO_WRITE) {
+		me.gs_writes_in_flight++;
+		me.gs_write_bytes_in_flight += bp->bio_length;
+	}
+	g_sched_dispatch(gp);
+	g_sched_unlock(gp);
+	return;
+
+bypass:
+	cbp->bio_done = g_std_done;
+	cbp->bio_caller1 = NULL; /* not scheduled */
+	g_io_request(cbp, LIST_FIRST(&gp->consumer));
+}
+
+/*
+ * The next few functions are the geom glue.
+ */
+static void
+g_sched_orphan(struct g_consumer *cp)
+{
+
+	g_topology_assert();
+	g_sched_destroy(cp->geom, 1);
+}
+
+static int
+g_sched_access(struct g_provider *pp, int dr, int dw, int de)
+{
+	struct g_geom *gp;
+	struct g_consumer *cp;
+	int error;
+
+	gp = pp->geom;
+	cp = LIST_FIRST(&gp->consumer);
+	error = g_access(cp, dr, dw, de);
+
+	return (error);
+}
+
+static void
+g_sched_temporary_start(struct bio *bio)
+{
+
+	mtx_lock(&me.gs_mtx);
+	me.gs_npending++;
+	gs_bioq_disksort(&me.gs_pending, bio);
+	mtx_unlock(&me.gs_mtx);
+}
+
+static void
+g_sched_flush_pending(g_start_t *start)
+{
+	struct bio *bp;
+
+	while ((bp = gs_bioq_takefirst(&me.gs_pending)))
+		start(bp);
+}
+
+static int
+g_insert_proxy(struct g_geom *gp, struct g_provider *newpp,
+    struct g_geom *dstgp, struct g_provider *pp, struct g_consumer *cp)
+{
+	struct g_sched_softc *sc = gp->softc;
+	g_start_t *saved_start, *flush = g_sched_start;
+	int error = 0, endticks = ticks + hz;
+
+	g_cancel_event(newpp);	/* prevent taste() */
+	/* copy private fields */
+	newpp->private = pp->private;
+	newpp->index = pp->index;
+
+	/* Queue all the early requests coming for us. */
+	me.gs_npending = 0;
+	saved_start = pp->geom->start;
+	dstgp->start = g_sched_temporary_start;
+
+	while (pp->nstart - pp->nend != me.gs_npending &&
+	    endticks - ticks >= 0)
+		tsleep(pp, PRIBIO, "-", hz/10);
+
+	if (pp->nstart - pp->nend != me.gs_npending) {
+		flush = saved_start;
+		error = ETIMEDOUT;
+		goto fail;
+	}
+
+	/* link pp to this geom */
+	LIST_REMOVE(pp, provider);
+	pp->geom = gp;
+	LIST_INSERT_HEAD(&gp->provider, pp, provider);
+
+	/*
+	 * replicate the counts from the parent in the
+	 * new provider and consumer nodes
+	 */
+	cp->acr = newpp->acr = pp->acr;
+	cp->acw = newpp->acw = pp->acw;
+	cp->ace = newpp->ace = pp->ace;
+	sc->sc_flags |= G_SCHED_PROXYING;
+
+fail:
+	dstgp->start = saved_start;
+
+	g_sched_flush_pending(flush);
+
+	return (error);
+}
+
+/*
+ * Create a geom node for the device passed as *pp.
+ * If successful, add a reference to this gsp.
+ */
+static int
+g_sched_create(struct gctl_req *req, struct g_class *mp,
+    struct g_provider *pp, struct g_gsched *gsp, int proxy)
+{
+	struct g_sched_softc *sc = NULL;
+	struct g_geom *gp, *dstgp;
+	struct g_provider *newpp = NULL;
+	struct g_consumer *cp = NULL;
+	char name[64];
+	int error;
+
+	g_topology_assert();
+
+	snprintf(name, sizeof(name), "%s%s", pp->name, G_SCHED_SUFFIX);
+	LIST_FOREACH(gp, &mp->geom, geom) {
+		if (strcmp(gp->name, name) == 0) {
+			gctl_error(req, "Geom %s already exists.",
+			    name);
+			return (EEXIST);
+		}
+	}
+
+	gp = g_new_geomf(mp, name);
+	dstgp = proxy ? pp->geom : gp; /* where do we link the provider */
+	if (gp == NULL) {
+		gctl_error(req, "Cannot create geom %s.", name);
+		error = ENOMEM;
+		goto fail;
+	}
+
+	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
+	sc->sc_gsched = gsp;
+	sc->sc_data = gsp->gs_init(gp);
+	if (sc->sc_data == NULL) {
+		error = ENOMEM;
+		goto fail;
+	}
+
+	sc->sc_hash = g_sched_hash_init(gsp, &sc->sc_mask, HASH_WAITOK);
+
+	/*
+	 * Do not initialize the flush mechanism, will be initialized
+	 * on the first insertion on the hash table.
+	 */
+
+	mtx_init(&sc->sc_mtx, "g_sched_mtx", NULL, MTX_DEF);
+
+	gp->softc = sc;
+	gp->start = g_sched_start;
+	gp->orphan = g_sched_orphan;
+	gp->access = g_sched_access;
+	gp->dumpconf = g_sched_dumpconf;
+
+	newpp = g_new_providerf(dstgp, gp->name);
+	if (newpp == NULL) {
+		gctl_error(req, "Cannot create provider %s.", name);
+		error = ENOMEM;
+		goto fail;
+	}
+
+	newpp->mediasize = pp->mediasize;
+	newpp->sectorsize = pp->sectorsize;
+
+	cp = g_new_consumer(gp);
+	if (cp == NULL) {
+		gctl_error(req, "Cannot create consumer for %s.",
+		    gp->name);
+		error = ENOMEM;
+		goto fail;
+	}
+
+	error = g_attach(cp, proxy ? newpp : pp);
+	if (error != 0) {
+		gctl_error(req, "Cannot attach to provider %s.",
+		    pp->name);
+		goto fail;
+	}
+
+	g_error_provider(newpp, 0);
+	if (proxy) {
+		error = g_insert_proxy(gp, newpp, dstgp, pp, cp);
+		if (error)
+			goto fail;
+	}
+	G_SCHED_DEBUG(0, "Device %s created.", gp->name);
+
+	g_gsched_ref(gsp);
+
+	return (0);
+
+fail:
+	if (cp != NULL) {
+		if (cp->provider != NULL)
+			g_detach(cp);
+		g_destroy_consumer(cp);
+	}
+
+	if (newpp != NULL)
+		g_destroy_provider(newpp);
+
+	if (sc && sc->sc_hash) {
+		g_sched_hash_fini(gp, sc->sc_hash, sc->sc_mask,
+		    gsp, sc->sc_data);
+	}
+
+	if (sc && sc->sc_data)
+		gsp->gs_fini(sc->sc_data);
+
+	if (gp != NULL) {
+		if (gp->softc != NULL)
+			g_free(gp->softc);
+		g_destroy_geom(gp);
+	}
+
+	return (error);
+}
+
+/*
+ * Support for dynamic switching of scheduling algorithms.
+ * First initialize the data structures for the new algorithm,
+ * then call g_sched_remove_locked() to flush all references
+ * to the old one, finally link the new algorithm.
+ */
+static int
+g_sched_change_algo(struct gctl_req *req, struct g_class *mp,
+    struct g_provider *pp, struct g_gsched *gsp)
+{
+	struct g_sched_softc *sc;
+	struct g_geom *gp;
+	struct g_hash *newh;
+	void *data;
+	u_long mask;
+	int error = 0;
+
+	gp = pp->geom;
+	sc = gp->softc;
+
+	data = gsp->gs_init(gp);
+	if (data == NULL)
+		return (ENOMEM);
+
+	newh = g_sched_hash_init(gsp, &mask, HASH_WAITOK);
+	if (gsp->gs_priv_size && !newh) {
+		error = ENOMEM;
+		goto fail;
+	}
+
+	g_sched_lock(gp);
+	if (sc->sc_gsched) {	/* can be NULL in some cases */
+		error = g_sched_remove_locked(gp, sc->sc_gsched);
+		if (error)
+			goto fail;
+	}
+
+	g_gsched_ref(gsp);
+	sc->sc_gsched = gsp;
+	sc->sc_data = data;
+	sc->sc_hash = newh;
+	sc->sc_mask = mask;
+
+	g_sched_unlock(gp);
+
+	return (0);
+
+fail:
+	if (newh)
+		g_sched_hash_fini(gp, newh, mask, gsp, data);
+
+	if (data)
+		gsp->gs_fini(data);
+
+	g_sched_unlock(gp);
+
+	return (error);
+}
+
+/*
+ * Stop the request flow directed to the proxy, redirecting the new
+ * requests to the me.gs_pending queue.
+ */
+static struct g_provider *
+g_detach_proxy(struct g_geom *gp)
+{
+	struct g_consumer *cp;
+	struct g_provider *pp, *newpp;
+
+	do {
+		pp = LIST_FIRST(&gp->provider);
+		if (pp == NULL)
+			break;
+		cp = LIST_FIRST(&gp->consumer);
+		if (cp == NULL)
+			break;
+		newpp = cp->provider;
+		if (newpp == NULL)
+			break;
+
+		me.gs_npending = 0;
+		pp->geom->start = g_sched_temporary_start;
+
+		return (pp);
+	} while (0);
+	printf("%s error detaching proxy %s\n", __FUNCTION__, gp->name);
+
+	return (NULL);
+}
+
+static void
+g_sched_blackhole(struct bio *bp)
+{
+
+	g_io_deliver(bp, ENXIO);
+}
+
+static inline void
+g_reparent_provider(struct g_provider *pp, struct g_geom *gp,
+    struct g_provider *newpp)
+{
+
+	LIST_REMOVE(pp, provider);
+	if (newpp) {
+		pp->private = newpp->private;
+		pp->index = newpp->index;
+	}
+	pp->geom = gp;
+	LIST_INSERT_HEAD(&gp->provider, pp, provider);
+}
+
+static inline void
+g_unproxy_provider(struct g_provider *oldpp, struct g_provider *newpp)
+{
+	struct g_geom *gp = oldpp->geom;
+
+	g_reparent_provider(oldpp, newpp->geom, newpp);
+
+	/*
+	 * Hackish: let the system destroy the old provider for us, just
+	 * in case someone attached a consumer to it, in which case a
+	 * direct call to g_destroy_provider() would not work.
+	 */
+	g_reparent_provider(newpp, gp, NULL);
+}
+
+/*
+ * Complete the proxy destruction, linking the old provider to its
+ * original geom, and destroying the proxy provider.  Also take care
+ * of issuing the pending requests collected in me.gs_pending (if any).
+ */
+static int
+g_destroy_proxy(struct g_geom *gp, struct g_provider *oldpp)
+{
+	struct g_consumer *cp;
+	struct g_provider *newpp;
+
+	do {
+		cp = LIST_FIRST(&gp->consumer);
+		if (cp == NULL)
+			break;
+		newpp = cp->provider;
+		if (newpp == NULL)
+			break;
+
+		/* Relink the provider to its original geom. */
+		g_unproxy_provider(oldpp, newpp);
+
+		/* Detach consumer from provider, and destroy provider. */
+		cp->acr = newpp->acr = 0;
+		cp->acw = newpp->acw = 0;
+		cp->ace = newpp->ace = 0;
+		g_detach(cp);
+
+		/* Send the pending bios through the right start function. */
+		g_sched_flush_pending(oldpp->geom->start);
+
+		return (0);
+	} while (0);
+	printf("%s error destroying proxy %s\n", __FUNCTION__, gp->name);
+
+	/* We cannot send the pending bios anywhere... */
+	g_sched_flush_pending(g_sched_blackhole);
+
+	return (EINVAL);
+}
+
+static int
+g_sched_destroy(struct g_geom *gp, boolean_t force)
+{
+	struct g_provider *pp, *oldpp = NULL;
+	struct g_sched_softc *sc;
+	struct g_gsched *gsp;
+	int error;
+
+	g_topology_assert();
+	sc = gp->softc;
+	if (sc == NULL)
+		return (ENXIO);
+	if (!(sc->sc_flags & G_SCHED_PROXYING)) {
+		pp = LIST_FIRST(&gp->provider);
+		if (pp && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
+			const char *msg = force ?
+				"but we force removal" : "cannot remove";
+
+			G_SCHED_DEBUG(!force,
+			    "Device %s is still open (r%dw%de%d), %s.",
+			    pp->name, pp->acr, pp->acw, pp->ace, msg);
+			if (!force)
+				return (EBUSY);
+		} else {
+			G_SCHED_DEBUG(0, "Device %s removed.", gp->name);
+		}
+	} else
+		oldpp = g_detach_proxy(gp);
+
+	gsp = sc->sc_gsched;
+	if (gsp) {
+		/*
+		 * XXX bad hack here: force a dispatch to release
+		 * any reference to the hash table still held by
+		 * the scheduler.
+		 */
+		g_sched_lock(gp);
+		/*
+		 * We are dying here, no new requests should enter
+		 * the scheduler.  This is granted by the topolgy,
+		 * either in case we were proxying (new bios are
+		 * being redirected) or not (see the access check
+		 * above).
+		 */
+		g_sched_forced_dispatch(gp);
+		error = g_sched_wait_pending(gp);
+
+		if (error) {
+			/*
+			 * Not all the requests came home: this might happen
+			 * under heavy load, or if we were waiting for any
+			 * bio which is served in the event path (see
+			 * geom_slice.c for an example of how this can
+			 * happen).  Try to restore a working configuration
+			 * if we can fail.
+			 */
+			if ((sc->sc_flags & G_SCHED_PROXYING) && oldpp) {
+				g_sched_flush_pending(force ?
+				    g_sched_blackhole : g_sched_start);
+			}
+
+			/*
+			 * In the forced destroy case there is not so much
+			 * we can do, we have pending bios that will call
+			 * g_sched_done() somehow, and we don't want them
+			 * to crash the system using freed memory.  We tell
+			 * the user that something went wrong, and leak some
+			 * memory here.
+			 * Note: the callers using force = 1 ignore the
+			 * return value.
+			 */
+			if (force) {
+				G_SCHED_DEBUG(0, "Pending requests while "
+				    " destroying geom, some memory leaked.");
+			}
+
+			return (error);
+		}
+
+		g_sched_unlock(gp);
+		g_sched_hash_fini(gp, sc->sc_hash, sc->sc_mask,
+		    gsp, sc->sc_data);
+		sc->sc_hash = NULL;
+		gsp->gs_fini(sc->sc_data);
+		g_gsched_unref(gsp);
+		sc->sc_gsched = NULL;
+	}
+
+	if ((sc->sc_flags & G_SCHED_PROXYING) && oldpp) {
+		error = g_destroy_proxy(gp, oldpp);
+
+		if (error) {
+			if (force) {
+				G_SCHED_DEBUG(0, "Unrecoverable error while "
+				    "destroying a proxy geom, leaking some "
+				    " memory.");
+			}
+
+			return (error);
+		}
+	}
+
+	mtx_destroy(&sc->sc_mtx);
+
+	g_free(gp->softc);
+	gp->softc = NULL;
+	g_wither_geom(gp, ENXIO);
+
+	return (error);
+}
+
+static int
+g_sched_destroy_geom(struct gctl_req *req, struct g_class *mp,
+    struct g_geom *gp)
+{
+
+	return (g_sched_destroy(gp, 0));
+}
+
+/*
+ * Functions related to the classification of requests.
+ *
+ * On recent FreeBSD versions (8.0 and above), we store a reference
+ * to the issuer of a request in bp->bio_classifier1 as soon
+ * as the bio is posted to the geom queue (and not later, because
+ * requests are managed by the g_down thread afterwards).
+ *
+ * On older versions of the system (but this code is not used
+ * in any existing release), we [ab]use the caller1 field in the
+ * root element of the bio tree to store the classification info.
+ * The marking is done at the beginning of g_io_request()
+ * and only if we find that the field is NULL.
+ *
+ * To avoid rebuilding the kernel, this module will patch the
+ * initial part of g_io_request() so it jumps to some hand-coded
+ * assembly that does the marking and then executes the original
+ * body of g_io_request().
+ *
+ * fake_ioreq[] is architecture-specific machine code
+ * that implements the above. CODE_SIZE, STORE_SIZE etc.
+ * are constants used in the patching routine. Look at the
+ * code in g_ioreq_patch() for the details.
+ */
+
+#ifndef HAVE_BIO_CLASSIFIER
+/*
+ * Support for old FreeBSD versions
+ */
+#if defined(__i386__)
+#define	CODE_SIZE	29
+#define	STORE_SIZE	5
+#define	EPILOGUE	5
+#define	SIZE		(CODE_SIZE + STORE_SIZE + EPILOGUE)
+
+static u_char fake_ioreq[SIZE] = {
+	0x8b, 0x44, 0x24, 0x04,		/* mov bp, %eax */
+	/* 1: */
+	0x89, 0xc2,			/* mov %eax, %edx # edx = bp */
+	0x8b, 0x40, 0x64,		/* mov bp->bio_parent, %eax */
+	0x85, 0xc0,			/* test %eax, %eax */
+	0x75, 0xf7,			/* jne 1b */
+	0x8b, 0x42, 0x30,		/* mov bp->bp_caller1, %eax */
+	0x85, 0xc0,			/* test %eax, %eax */
+	0x75, 0x09,			/* jne 2f */
+	0x64, 0xa1, 0x00, 0x00,		/* mov %fs:0, %eax */
+	0x00, 0x00,
+	0x89, 0x42, 0x30,		/* mov %eax, bp->bio_caller1 */
+	/* 2: */
+        0x55, 0x89, 0xe5, 0x57, 0x56,
+	0xe9, 0x00, 0x00, 0x00, 0x00,	/* jmp back... */
+};
+#elif defined(__amd64)
+#define	CODE_SIZE	38
+#define	STORE_SIZE	6
+#define	EPILOGUE	5
+#define	SIZE		(CODE_SIZE + STORE_SIZE + EPILOGUE)
+
+static u_char fake_ioreq[SIZE] = {
+	0x48, 0x89, 0xf8,		/* mov bp, %rax */
+	/* 1: */
+	0x48, 0x89, 0xc2,		/* mov %rax, %rdx # rdx = bp */
+	0x48, 0x8b, 0x82, 0xa8,		/* mov bp->bio_parent, %rax */
+	0x00, 0x00, 0x00,
+	0x48, 0x85, 0xc0,		/* test %rax, %rax */
+	0x75, 0xf1,			/* jne 1b */
+	0x48, 0x83, 0x7a, 0x58,		/* cmp $0, bp->bp_caller1 */
+	0x00,
+	0x75, 0x0d,			/* jne 2f */
+	0x65, 0x48, 0x8b, 0x04,		/* mov %gs:0, %rax */
+	0x25, 0x00, 0x00, 0x00,
+	0x00,
+	0x48, 0x89, 0x42, 0x58,		/* mov %rax, bp->bio_caller1 */
+	/* 2: */
+	0x55, 0x48, 0x89, 0xe5, 0x41, 0x56,
+	0xe9, 0x00, 0x00, 0x00, 0x00,	/* jmp back... */
+};
+#else /* neither x86 nor amd64 */
+static void
+g_new_io_request(struct bio *bp, struct g_consumer *cp)
+{
+	struct bio *top = bp;
+
+        /*
+         * bio classification: if bio_caller1 is available in the
+         * root of the 'struct bio' tree, store there the thread id
+         * of the thread that originated the request.
+         * More sophisticated classification schemes can be used.
+         */
+	while (top->bio_parent)
+		top = top->bio_parent;
+
+	if (top->bio_caller1 == NULL)
+		top->bio_caller1 = curthread;
+}
+
+#error please add the code above in g_new_io_request() to the beginning of \
+	/sys/geom/geom_io.c::g_io_request(), and remove this line.
+#endif /* end of arch-specific code */
+
+static int
+g_ioreq_patch(void)
+{
+	u_char *original;
+	u_long ofs;
+	int found;
+
+	if (me.gs_patched)
+		return (-1);
+
+	original = (u_char *)g_io_request;
+
+	found = !bcmp(original, fake_ioreq + CODE_SIZE, STORE_SIZE);
+	if (!found)
+		return (-1);
+
+	/* Jump back to the original + STORE_SIZE. */
+	ofs = (original + STORE_SIZE) - (fake_ioreq + SIZE);
+	bcopy(&ofs, fake_ioreq + CODE_SIZE + STORE_SIZE + 1, 4);
+
+	/* Patch the original address with a jump to the trampoline. */
+	*original = 0xe9;     /* jump opcode */
+	ofs = fake_ioreq - (original + 5);
+	bcopy(&ofs, original + 1, 4);
+
+	me.gs_patched = 1;
+
+	return (0);
+}
+
+/*
+ * Restore the original code, this is easy.
+ */
+static void
+g_ioreq_restore(void)
+{
+	u_char *original;
+
+	if (me.gs_patched) {
+		original = (u_char *)g_io_request;
+		bcopy(fake_ioreq + CODE_SIZE, original, STORE_SIZE);
+		me.gs_patched = 0;
+	}
+}
+
+static inline void
+g_classifier_ini(void)
+{
+
+	g_ioreq_patch();
+}
+
+static inline void
+g_classifier_fini(void)
+{
+
+	g_ioreq_restore();
+}
+
+/*--- end of support code for older FreeBSD versions */
+
+#else /* HAVE_BIO_CLASSIFIER */
+
+/*
+ * Classifier support for recent FreeBSD versions: we use
+ * a very simple classifier, only use curthread to tag a request.
+ * The classifier is registered at module load, and unregistered
+ * at module unload.
+ */
+static int
+g_sched_tag(void *arg, struct bio *bp)
+{
+
+	bp->bio_classifier1 = curthread;
+	return (1);
+}
+
+static struct g_classifier_hook g_sched_classifier = {
+	.func =	g_sched_tag,
+};
+
+static inline void
+g_classifier_ini(void)
+{
+
+	g_register_classifier(&g_sched_classifier);
+}
+
+static inline void
+g_classifier_fini(void)
+{
+
+	g_unregister_classifier(&g_sched_classifier);
+}
+#endif /* HAVE_BIO_CLASSIFIER */
+
+static void
+g_sched_init(struct g_class *mp)
+{
+
+	g_gsched_global_init();
+
+	G_SCHED_DEBUG(0, "Loading: mp = %p, g_sched_class = %p.",
+	    mp, &g_sched_class);
+
+	/* Patch g_io_request to store classification info in the bio. */
+	g_classifier_ini();
+}
+
+static void
+g_sched_fini(struct g_class *mp)
+{
+
+	g_classifier_fini();
+
+	G_SCHED_DEBUG(0, "Unloading...");
+
+	KASSERT(LIST_EMPTY(&me.gs_scheds), ("still registered schedulers"));
+	mtx_destroy(&me.gs_mtx);
+}
+
+/*
+ * Read the i-th argument for a request, skipping the /dev/
+ * prefix if present.
+ */
+static const char *
+g_sched_argi(struct gctl_req *req, int i)
+{
+	static const char *dev_prefix = "/dev/";
+	const char *name;
+	char param[16];
+	int l = strlen(dev_prefix);
+
+	snprintf(param, sizeof(param), "arg%d", i);
+	name = gctl_get_asciiparam(req, param);
+	if (name == NULL)
+		gctl_error(req, "No 'arg%d' argument", i);
+	else if (strncmp(name, dev_prefix, l) == 0)
+		name += l;
+	return (name);
+}
+
+/*
+ * Fetch nargs and do appropriate checks.
+ */
+static int
+g_sched_get_nargs(struct gctl_req *req)
+{
+	int *nargs;
+
+	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
+	if (nargs == NULL) {
+		gctl_error(req, "No 'nargs' argument");
+		return (0);
+	}
+	if (*nargs <= 0)
+		gctl_error(req, "Missing device(s).");
+	return (*nargs);
+}
+
+/*
+ * Check whether we should add the class on certain volumes when
+ * this geom is created. Right now this is under control of a kenv
+ * variable containing the names of all devices that we care about.
+ * Probably we should only support transparent insertion as the
+ * preferred mode of operation.
+ */
+static struct g_geom *
+g_sched_taste(struct g_class *mp, struct g_provider *pp,
+		int flags __unused)
+{
+	struct g_gsched *gsp = NULL;	/* the . algorithm we want */
+	const char *s;			/* generic string pointer */
+	const char *taste_names;	/* devices we like */
+	int l;
+    
+        g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__,
+	    mp->name, pp->name);
+        g_topology_assert();
+ 
+        G_SCHED_DEBUG(2, "Tasting %s.", pp->name);
+
+	do {
+		/* do not taste on ourselves */
+		if (pp->geom->class == mp)
+                	break;
+
+		taste_names = getenv("geom.sched.taste");
+		if (taste_names == NULL)
+			break;
+
+		l = strlen(pp->name);
+		for (s = taste_names; *s &&
+		    (s = strstr(s, pp->name)); s++) {
+			/* further checks for an exact match */
+			if ( (s == taste_names || s[-1] == ' ') &&
+			     (s[l] == '\0' || s[l] == ' ') )
+				break;
+		}
+		if (s == NULL)
+			break;
+		G_SCHED_DEBUG(0, "Attach device %s match [%s]\n",
+		    pp->name, s);
+
+		/* look up the provider name in the list */
+		s = getenv("geom.sched.algo");
+		if (s == NULL)
+			s = "rr";
+
+		gsp = g_gsched_find(s);	/* also get a reference */
+		if (gsp == NULL) {
+			G_SCHED_DEBUG(0, "Bad '%s' algorithm.", s);
+			break;
+		}
+
+		/* XXX create with 1 as last argument ? */
+		g_sched_create(NULL, mp, pp, gsp, 0);
+		g_gsched_unref(gsp);
+	} while (0);
+	return NULL;
+}
+
+static void
+g_sched_ctl_create(struct gctl_req *req, struct g_class *mp, int proxy)
+{
+	struct g_provider *pp;
+	struct g_gsched *gsp;
+	const char *name;
+	int i, nargs;
+
+	g_topology_assert();
+
+	name = gctl_get_asciiparam(req, "algo");
+	if (name == NULL) {
+		gctl_error(req, "No '%s' argument", "algo");
+		return;
+	}
+
+	gsp = g_gsched_find(name);	/* also get a reference */
+	if (gsp == NULL) {
+		gctl_error(req, "Bad algorithm '%s'", name);
+		return;
+	}
+
+	nargs = g_sched_get_nargs(req);
+
+	/*
+	 * Run on the arguments, and break on any error.
+	 * We look for a device name, but skip the /dev/ prefix if any.
+	 */
+	for (i = 0; i < nargs; i++) {
+		name = g_sched_argi(req, i);
+		if (name == NULL)
+			break;
+		pp = g_provider_by_name(name);
+		if (pp == NULL) {
+			G_SCHED_DEBUG(1, "Provider %s is invalid.", name);
+			gctl_error(req, "Provider %s is invalid.", name);
+			break;
+		}
+		if (g_sched_create(req, mp, pp, gsp, proxy) != 0)
+			break;
+	}
+
+	g_gsched_unref(gsp);
+}
+
+static void
+g_sched_ctl_configure(struct gctl_req *req, struct g_class *mp)
+{
+	struct g_provider *pp;
+	struct g_gsched *gsp;
+	const char *name;
+	int i, nargs;
+
+	g_topology_assert();
+
+	name = gctl_get_asciiparam(req, "algo");
+	if (name == NULL) {
+		gctl_error(req, "No '%s' argument", "algo");
+		return;
+	}
+
+	gsp = g_gsched_find(name);	/* also get a reference */
+	if (gsp == NULL) {
+		gctl_error(req, "Bad algorithm '%s'", name);
+		return;
+	}
+
+	nargs = g_sched_get_nargs(req);
+
+	/*
+	 * Run on the arguments, and break on any error.
+	 * We look for a device name, but skip the /dev/ prefix if any.
+	 */
+	for (i = 0; i < nargs; i++) {
+		name = g_sched_argi(req, i);
+		if (name == NULL)
+			break;
+		pp = g_provider_by_name(name);
+		if (pp == NULL || pp->geom->class != mp) {
+			G_SCHED_DEBUG(1, "Provider %s is invalid.", name);
+			gctl_error(req, "Provider %s is invalid.", name);
+			break;
+		}
+		if (g_sched_change_algo(req, mp, pp, gsp) != 0)
+			break;
+	}
+
+	g_gsched_unref(gsp);
+}
+
+static struct g_geom *
+g_sched_find_geom(struct g_class *mp, const char *name)
+{
+	struct g_geom *gp;
+
+	LIST_FOREACH(gp, &mp->geom, geom) {
+		if (strcmp(gp->name, name) == 0)
+			return (gp);
+	}
+	return (NULL);
+}
+
+static void
+g_sched_ctl_destroy(struct gctl_req *req, struct g_class *mp)
+{
+	int nargs, *force, error, i;
+	struct g_geom *gp;
+	const char *name;
+
+	g_topology_assert();
+
+	nargs = g_sched_get_nargs(req);
+
+	force = gctl_get_paraml(req, "force", sizeof(*force));
+	if (force == NULL) {
+		gctl_error(req, "No 'force' argument");
+		return;
+	}
+
+	for (i = 0; i < nargs; i++) {
+		name = g_sched_argi(req, i);
+		if (name == NULL)
+			break;
+
+		gp = g_sched_find_geom(mp, name);
+		if (gp == NULL) {
+			G_SCHED_DEBUG(1, "Device %s is invalid.", name);
+			gctl_error(req, "Device %s is invalid.", name);
+			break;
+		}
+
+		error = g_sched_destroy(gp, *force);
+		if (error != 0) {
+			gctl_error(req, "Cannot destroy device %s (error=%d).",
+			    gp->name, error);
+			break;
+		}
+	}
+}
+
+static void
+g_sched_config(struct gctl_req *req, struct g_class *mp, const char *verb)
+{
+	uint32_t *version;
+
+	g_topology_assert();
+
+	version = gctl_get_paraml(req, "version", sizeof(*version));
+	if (version == NULL) {
+		gctl_error(req, "No '%s' argument.", "version");
+		return;
+	}
+
+	if (*version != G_SCHED_VERSION) {
+		gctl_error(req, "Userland and kernel parts are "
+		    "out of sync.");
+		return;
+	}
+
+	if (strcmp(verb, "create") == 0) {
+		g_sched_ctl_create(req, mp, 0);
+		return;
+	} else if (strcmp(verb, "insert") == 0) {
+		g_sched_ctl_create(req, mp, 1);
+		return;
+	} else if (strcmp(verb, "configure") == 0) {
+		g_sched_ctl_configure(req, mp);
+		return;
+	} else if (strcmp(verb, "destroy") == 0) {
+		g_sched_ctl_destroy(req, mp);
+		return;
+	}
+
+	gctl_error(req, "Unknown verb.");
+}
+
+static void
+g_sched_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
+    struct g_consumer *cp, struct g_provider *pp)
+{
+	struct g_sched_softc *sc = gp->softc;
+	struct g_gsched *gsp = sc->sc_gsched;
+	if (indent == NULL) {	/* plaintext */
+		sbuf_printf(sb, " algo %s", gsp ? gsp->gs_name : "--");
+	}
+	if (gsp->gs_dumpconf)
+		gsp->gs_dumpconf(sb, indent, gp, cp, pp);
+}
+
+DECLARE_GEOM_CLASS(g_sched_class, g_sched);
+MODULE_VERSION(geom_sched, 0);
diff --git a/sys/geom/sched/g_sched.h b/sys/geom/sched/g_sched.h
new file mode 100644
index 0000000..86d1023
--- /dev/null
+++ b/sys/geom/sched/g_sched.h
@@ -0,0 +1,137 @@
+/*-
+ * Copyright (c) 2009-2010 Fabio Checconi, Luigi Rizzo
+ * 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 AUTHORS 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 AUTHORS 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.
+ */
+
+#ifndef	_G_SCHED_H_
+#define	_G_SCHED_H_
+
+/*
+ * $Id$
+ * $FreeBSD$
+ *
+ * Header for the geom_sched class (userland library and kernel part).
+ * See g_sched.c for documentation.
+ * The userland code only needs the three G_SCHED_* values below.
+ */
+
+#define	G_SCHED_CLASS_NAME	"SCHED"
+#define	G_SCHED_VERSION		0
+#define	G_SCHED_SUFFIX		".sched."
+
+#ifdef _KERNEL
+#define	G_SCHED_DEBUG(lvl, ...)	do {				\
+	if (me.gs_debug >= (lvl)) {				\
+		printf("GEOM_SCHED");				\
+		if (me.gs_debug > 0)				\
+			printf("[%u]", lvl);			\
+		printf(": ");					\
+		printf(__VA_ARGS__);				\
+		printf("\n");					\
+	}							\
+} while (0)
+
+#define	G_SCHED_LOGREQ(bp, ...)	do {				\
+	if (me.gs_debug >= 2) {					\
+		printf("GEOM_SCHED[2]: ");			\
+		printf(__VA_ARGS__);				\
+		printf(" ");					\
+		g_print_bio(bp);				\
+		printf("\n");					\
+	}							\
+} while (0)
+
+LIST_HEAD(g_hash, g_sched_class);
+
+/*
+ * Descriptor of a scheduler.
+ * In addition to the obvious fields, sc_flushing and sc_pending
+ * support dynamic switching of scheduling algorithm.
+ * Normally, sc_flushing is 0, and requests that are scheduled are
+ * also added to the sc_pending queue, and removed when we receive
+ * the 'done' event.
+ *
+ * When we are transparently inserted on an existing provider,
+ * sc_proxying is set. The detach procedure is slightly different.
+ *
+ * When switching schedulers, sc_flushing is set so requests bypass us,
+ * and at the same time we update the pointer in the pending bios
+ * to ignore us when they return up.
+ * XXX it would be more efficient to implement sc_pending with
+ * a generation number: the softc generation is increased when
+ * we change scheduling algorithm, we store the current generation
+ * number in the pending bios, and when they come back we ignore
+ * the done() call if the generation number do not match.
+ */
+struct g_sched_softc {
+	/*
+	 * Generic fields used by any scheduling algorithm:
+	 * a mutex, the class descriptor, flags, list of pending
+	 * requests (used when flushing the module) and support
+	 * for hash tables where we store per-flow queues.
+	 */
+	struct mtx	sc_mtx;
+	struct g_gsched	*sc_gsched;	/* Scheduler descriptor. */
+	int		sc_pending;	/* Pending requests. */
+	int		sc_flags;	/* Various flags. */
+
+	/*
+	 * Hash tables to store per-flow queues are generally useful
+	 * so we handle them in the common code.
+	 * sc_hash and sc_mask are parameters of the hash table,
+	 * the last two fields are used to periodically remove
+	 * expired items from the hash table.
+	 */
+	struct g_hash	*sc_hash;
+	u_long		sc_mask;
+	int		sc_flush_ticks;	/* Next tick for a flush. */
+	int		sc_flush_bucket; /* Next bucket to flush. */
+
+	/*
+	 * Pointer to the algorithm's private data, which is the value
+	 * returned by sc_gsched->gs_init() . A NULL here means failure.
+	 * XXX intptr_t might be more appropriate.
+	 */
+	void		*sc_data;
+};
+
+#define	G_SCHED_PROXYING	1
+#define	G_SCHED_FLUSHING	2
+
+/*
+ * Temporary- our own version of the disksort, because the
+ * version in 7.x and 8.x before march 2009 is buggy.
+ */
+void gs_bioq_init(struct bio_queue_head *);
+void gs_bioq_remove(struct bio_queue_head *, struct bio *);
+void gs_bioq_flush(struct bio_queue_head *, struct devstat *, int);
+void gs_bioq_insert_head(struct bio_queue_head *, struct bio *);
+void gs_bioq_insert_tail(struct bio_queue_head *, struct bio *);
+struct bio *gs_bioq_first(struct bio_queue_head *);
+struct bio *gs_bioq_takefirst(struct bio_queue_head *);
+void gs_bioq_disksort(struct bio_queue_head *, struct bio *);
+
+#endif	/* _KERNEL */
+
+#endif	/* _G_SCHED_H_ */
diff --git a/sys/geom/sched/gs_rr.c b/sys/geom/sched/gs_rr.c
new file mode 100644
index 0000000..185dc71
--- /dev/null
+++ b/sys/geom/sched/gs_rr.c
@@ -0,0 +1,685 @@
+/*-
+ * Copyright (c) 2009-2010 Fabio Checconi, Luigi Rizzo
+ * 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.
+ */
+
+/*
+ * $Id$
+ * $FreeBSD$
+ *
+ * A round-robin (RR) anticipatory scheduler, with per-client queues.
+ *
+ * The goal of this implementation is to improve throughput compared
+ * to the pure elevator algorithm, and insure some fairness among
+ * clients.
+ * 
+ * Requests coming from the same client are put in the same queue.
+ * We use anticipation to help reducing seeks, and each queue
+ * is never served continuously for more than a given amount of
+ * time or data. Queues are then served in a round-robin fashion.
+ *
+ * Each queue can be in any of the following states:
+ *     READY	immediately serve the first pending request;
+ *     BUSY	one request is under service, wait for completion;
+ *     IDLING	do not serve incoming requests immediately, unless
+ * 		they are "eligible" as defined later.
+ *
+ * Scheduling is made looking at the status of all queues,
+ * and the first one in round-robin order is privileged.
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/bio.h>
+#include <sys/callout.h>
+#include <sys/malloc.h>
+#include <sys/module.h>
+#include <sys/proc.h>
+#include <sys/queue.h>
+#include <sys/sysctl.h>
+#include "gs_scheduler.h"
+
+/* possible states of the scheduler */
+enum g_rr_state {
+	G_QUEUE_READY = 0,	/* Ready to dispatch. */
+	G_QUEUE_BUSY,		/* Waiting for a completion. */
+	G_QUEUE_IDLING		/* Waiting for a new request. */
+};
+
+/* possible queue flags */
+enum g_rr_flags {
+	G_FLAG_COMPLETED = 1,	/* Completed a req. in the current budget. */
+};
+
+struct g_rr_softc;
+
+/*
+ * Queue descriptor, containing reference count, scheduling
+ * state, a queue of pending requests, configuration parameters.
+ * Queues with pending request(s) and not under service are also
+ * stored in a Round Robin (RR) list.
+ */
+struct g_rr_queue {
+	struct g_rr_softc *q_sc;	/* link to the parent */
+
+	enum g_rr_state	q_status;
+	unsigned int	q_service;	/* service received so far */
+	int		q_slice_end;	/* actual slice end in ticks */
+	enum g_rr_flags	q_flags;	/* queue flags */
+	struct bio_queue_head q_bioq;
+
+	/* Scheduling parameters */
+	unsigned int	q_budget;	/* slice size in bytes */
+	unsigned int	q_slice_duration; /* slice size in ticks */
+	unsigned int	q_wait_ticks;	/* wait time for anticipation */
+
+	/* Stats to drive the various heuristics. */
+	struct g_savg	q_thinktime;	/* Thinktime average. */
+	struct g_savg	q_seekdist;	/* Seek distance average. */
+
+	int		q_bionum;	/* Number of requests. */
+
+	off_t		q_lastoff;	/* Last submitted req. offset. */
+	int		q_lastsub;	/* Last submitted req. time. */
+
+	/* Expiration deadline for an empty queue. */
+	int		q_expire;
+
+	TAILQ_ENTRY(g_rr_queue) q_tailq; /* RR list link field */
+};
+
+/* List types. */
+TAILQ_HEAD(g_rr_tailq, g_rr_queue);
+
+/* list of scheduler instances */
+LIST_HEAD(g_scheds, g_rr_softc);
+
+/* Default quantum for RR between queues. */
+#define	G_RR_DEFAULT_BUDGET	0x00800000
+
+/*
+ * Per device descriptor, holding the Round Robin list of queues
+ * accessing the disk, a reference to the geom, and the timer.
+ */
+struct g_rr_softc {
+	struct g_geom	*sc_geom;
+
+	/*
+	 * sc_active is the queue we are anticipating for.
+	 * It is set only in gs_rr_next(), and possibly cleared
+	 * only in gs_rr_next() or on a timeout.
+	 * The active queue is never in the Round Robin list
+	 * even if it has requests queued.
+	 */
+	struct g_rr_queue *sc_active;
+	struct callout	sc_wait;	/* timer for sc_active */
+
+	struct g_rr_tailq sc_rr_tailq;	/* the round-robin list */
+	int		sc_nqueues;	/* number of queues */
+
+	/* Statistics */
+	int		sc_in_flight;	/* requests in the driver */
+
+	LIST_ENTRY(g_rr_softc)	sc_next;
+};
+
+/* Descriptor for bounded values, min and max are constant. */
+struct x_bound {		
+	const int	x_min;
+	int		x_cur;
+	const int	x_max;
+};
+
+/*
+ * parameters, config and stats
+ */
+struct g_rr_params {
+	int	queues;			/* total number of queues */
+	int	w_anticipate;		/* anticipate writes */
+	int	bypass;			/* bypass scheduling writes */
+
+	int	units;			/* how many instances */
+	/* sc_head is used for debugging */
+	struct g_scheds	sc_head;	/* first scheduler instance */
+
+	struct x_bound queue_depth;	/* max parallel requests */
+	struct x_bound wait_ms;		/* wait time, milliseconds */
+	struct x_bound quantum_ms;	/* quantum size, milliseconds */
+	struct x_bound quantum_kb;	/* quantum size, Kb (1024 bytes) */
+
+	/* statistics */
+	int	wait_hit;		/* success in anticipation */
+	int	wait_miss;		/* failure in anticipation */
+};
+
+/*
+ * Default parameters for the scheduler.  The quantum sizes target
+ * a 80MB/s disk; if the hw is faster or slower the minimum of the
+ * two will have effect: the clients will still be isolated but
+ * the fairness may be limited.  A complete solution would involve
+ * the on-line measurement of the actual disk throughput to derive
+ * these parameters.  Or we may just choose to ignore service domain
+ * fairness and accept what can be achieved with time-only budgets.
+ */
+static struct g_rr_params me = {
+	.sc_head = LIST_HEAD_INITIALIZER(&me.sc_head),
+	.w_anticipate =	1,
+	.queue_depth =	{ 1,	1,	50 },
+	.wait_ms =	{ 1, 	10,	30 },
+	.quantum_ms =	{ 1, 	100,	500 },
+	.quantum_kb =	{ 16, 	8192,	65536 },
+};
+
+struct g_rr_params *gs_rr_me = &me;
+
+SYSCTL_DECL(_kern_geom_sched);
+SYSCTL_NODE(_kern_geom_sched, OID_AUTO, rr, CTLFLAG_RW, 0,
+    "GEOM_SCHED ROUND ROBIN stuff");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, units, CTLFLAG_RD,
+    &me.units, 0, "Scheduler instances");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, queues, CTLFLAG_RD,
+    &me.queues, 0, "Total rr queues");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, wait_ms, CTLFLAG_RW,
+    &me.wait_ms.x_cur, 0, "Wait time milliseconds");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, quantum_ms, CTLFLAG_RW,
+    &me.quantum_ms.x_cur, 0, "Quantum size milliseconds");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, bypass, CTLFLAG_RW,
+    &me.bypass, 0, "Bypass scheduler");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, w_anticipate, CTLFLAG_RW,
+    &me.w_anticipate, 0, "Do anticipation on writes");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, quantum_kb, CTLFLAG_RW,
+    &me.quantum_kb.x_cur, 0, "Quantum size Kbytes");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, queue_depth, CTLFLAG_RW,
+    &me.queue_depth.x_cur, 0, "Maximum simultaneous requests");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, wait_hit, CTLFLAG_RW,
+    &me.wait_hit, 0, "Hits in anticipation");
+SYSCTL_UINT(_kern_geom_sched_rr, OID_AUTO, wait_miss, CTLFLAG_RW,
+    &me.wait_miss, 0, "Misses in anticipation");
+
+#ifdef DEBUG_QUEUES
+/* print the status of a queue */
+static void
+gs_rr_dump_q(struct g_rr_queue *qp, int index)
+{
+	int l = 0;
+	struct bio *bp;
+
+	TAILQ_FOREACH(bp, &(qp->q_bioq.queue), bio_queue) {
+		l++;
+	}
+	printf("--- rr queue %d %p status %d len %d ---\n",
+	    index, qp, qp->q_status, l);
+}
+
+/*
+ * Dump the scheduler status when writing to this sysctl variable.
+ * XXX right now we only dump the status of the last instance created.
+ * not a severe issue because this is only for debugging
+ */
+static int
+gs_rr_sysctl_status(SYSCTL_HANDLER_ARGS)
+{
+        int error, val = 0;
+	struct g_rr_softc *sc;
+
+        error = sysctl_handle_int(oidp, &val, 0, req);
+        if (error || !req->newptr )
+                return (error);
+
+        printf("called %s\n", __FUNCTION__);
+
+	LIST_FOREACH(sc, &me.sc_head, sc_next) {
+		int i, tot = 0;
+		printf("--- sc %p active %p nqueues %d "
+		    "callout %d in_flight %d ---\n",
+		    sc, sc->sc_active, sc->sc_nqueues,
+		    callout_active(&sc->sc_wait),
+		    sc->sc_in_flight);
+		for (i = 0; i < G_RR_HASH_SIZE; i++) {
+			struct g_rr_queue *qp;
+			LIST_FOREACH(qp, &sc->sc_hash[i], q_hash) {
+				gs_rr_dump_q(qp, tot);
+				tot++;
+			}
+		}
+	}
+        return (0);
+}
+
+SYSCTL_PROC(_kern_geom_sched_rr, OID_AUTO, status,
+	CTLTYPE_UINT | CTLFLAG_RW,
+    0, sizeof(int), gs_rr_sysctl_status, "I", "status");
+
+#endif	/* DEBUG_QUEUES */
+
+/*
+ * Get a bounded value, optionally convert to a min of t_min ticks.
+ */
+static int
+get_bounded(struct x_bound *v, int t_min)
+{
+	int x;
+
+	x = v->x_cur;
+	if (x < v->x_min)
+		x = v->x_min;
+	else if (x > v->x_max)
+		x = v->x_max;
+	if (t_min) {
+		x = x * hz / 1000;	/* convert to ticks */
+		if (x < t_min)
+			x = t_min;
+	}
+	return x;
+}
+
+/*
+ * Get a reference to the queue for bp, using the generic
+ * classification mechanism.
+ */
+static struct g_rr_queue *
+g_rr_queue_get(struct g_rr_softc *sc, struct bio *bp)
+{
+
+	return (g_sched_get_class(sc->sc_geom, bp));
+}
+
+static int
+g_rr_init_class(void *data, void *priv)
+{
+	struct g_rr_softc *sc = data;
+	struct g_rr_queue *qp = priv;
+
+	gs_bioq_init(&qp->q_bioq);
+
+	/*
+	 * Set the initial parameters for the client:
+	 * slice size in bytes and ticks, and wait ticks.
+	 * Right now these are constant, but we could have
+	 * autoconfiguration code to adjust the values based on
+	 * the actual workload.
+	 */
+	qp->q_budget = 1024 * get_bounded(&me.quantum_kb, 0);
+	qp->q_slice_duration = get_bounded(&me.quantum_ms, 2);
+	qp->q_wait_ticks = get_bounded(&me.wait_ms, 2);
+
+	qp->q_sc = sc;		/* link to the parent */
+	qp->q_sc->sc_nqueues++;
+	me.queues++;
+
+	return (0);
+}
+
+/*
+ * Release a reference to the queue.
+ */
+static void
+g_rr_queue_put(struct g_rr_queue *qp)
+{
+
+	g_sched_put_class(qp->q_sc->sc_geom, qp);
+}
+
+static void
+g_rr_fini_class(void *data, void *priv)
+{
+	struct g_rr_queue *qp = priv;
+
+	KASSERT(gs_bioq_first(&qp->q_bioq) == NULL,
+			("released nonempty queue"));
+	qp->q_sc->sc_nqueues--;
+	me.queues--;
+}
+
+static inline int
+g_rr_queue_expired(struct g_rr_queue *qp)
+{
+
+	if (qp->q_service >= qp->q_budget)
+		return (1);
+
+	if ((qp->q_flags & G_FLAG_COMPLETED) &&
+	    ticks - qp->q_slice_end >= 0)
+		return (1);
+
+	return (0);
+}
+
+static inline int
+g_rr_should_anticipate(struct g_rr_queue *qp, struct bio *bp)
+{
+	int wait = get_bounded(&me.wait_ms, 2);
+
+	if (!me.w_anticipate && (bp->bio_cmd & BIO_WRITE))
+		return (0);
+
+	if (g_savg_valid(&qp->q_thinktime) &&
+	    g_savg_read(&qp->q_thinktime) > wait)
+		return (0);
+
+	if (g_savg_valid(&qp->q_seekdist) &&
+	    g_savg_read(&qp->q_seekdist) > 8192)
+		return (0);
+
+	return (1);
+}
+
+/*
+ * Called on a request arrival, timeout or completion.
+ * Try to serve a request among those queued.
+ */
+static struct bio *
+g_rr_next(void *data, int force)
+{
+	struct g_rr_softc *sc = data;
+	struct g_rr_queue *qp;
+	struct bio *bp, *next;
+	int expired;
+
+	qp = sc->sc_active;
+	if (me.bypass == 0 && !force) {
+		if (sc->sc_in_flight >= get_bounded(&me.queue_depth, 0))
+			return (NULL);
+
+		/* Try with the queue under service first. */
+		if (qp != NULL && qp->q_status != G_QUEUE_READY) {
+			/*
+			 * Queue is anticipating, ignore request.
+			 * We should check that we are not past
+			 * the timeout, but in that case the timeout
+			 * will fire immediately afterwards so we
+			 * don't bother.
+			 */
+			return (NULL);
+		}
+	} else if (qp != NULL && qp->q_status != G_QUEUE_READY) {
+		g_rr_queue_put(qp);
+		sc->sc_active = qp = NULL;
+	}
+
+	/*
+	 * No queue under service, look for the first in RR order.
+	 * If we find it, select if as sc_active, clear service
+	 * and record the end time of the slice.
+	 */
+	if (qp == NULL) {
+		qp = TAILQ_FIRST(&sc->sc_rr_tailq);
+		if (qp == NULL)
+			return (NULL); /* no queues at all, return */
+		/* otherwise select the new queue for service. */
+		TAILQ_REMOVE(&sc->sc_rr_tailq, qp, q_tailq);
+		sc->sc_active = qp;
+		qp->q_service = 0;
+		qp->q_flags &= ~G_FLAG_COMPLETED;
+	}
+
+	bp = gs_bioq_takefirst(&qp->q_bioq);	/* surely not NULL */
+	qp->q_service += bp->bio_length;	/* charge the service */
+
+	/*
+	 * The request at the head of the active queue is always
+	 * dispatched, and gs_rr_next() will be called again
+	 * immediately.
+	 * We need to prepare for what to do next:
+	 *
+	 * 1. have we reached the end of the (time or service) slice ?
+	 *    If so, clear sc_active and possibly requeue the previous
+	 *    active queue if it has more requests pending;
+	 * 2. do we have more requests in sc_active ?
+	 *    If yes, do not anticipate, as gs_rr_next() will run again;
+	 *    if no, decide whether or not to anticipate depending
+	 *    on read or writes (e.g., anticipate only on reads).
+	 */
+	expired = g_rr_queue_expired(qp);	/* are we expired ? */
+	next = gs_bioq_first(&qp->q_bioq);	/* do we have one more ? */
+ 	if (expired) {
+		sc->sc_active = NULL;
+		/* Either requeue or release reference. */
+		if (next != NULL)
+			TAILQ_INSERT_TAIL(&sc->sc_rr_tailq, qp, q_tailq);
+		else
+			g_rr_queue_put(qp);
+	} else if (next != NULL) {
+		qp->q_status = G_QUEUE_READY;
+	} else {
+		if (!force && g_rr_should_anticipate(qp, bp)) {
+			/* anticipate */
+			qp->q_status = G_QUEUE_BUSY;
+		} else {
+			/* do not anticipate, release reference */
+			g_rr_queue_put(qp);
+			sc->sc_active = NULL;
+		}
+	}
+	/* If sc_active != NULL, its q_status is always correct. */
+
+	sc->sc_in_flight++;
+
+	return (bp);
+}
+
+static inline void
+g_rr_update_thinktime(struct g_rr_queue *qp)
+{
+	int delta = ticks - qp->q_lastsub, wait = get_bounded(&me.wait_ms, 2);
+
+	if (qp->q_sc->sc_active != qp)
+		return;
+
+	qp->q_lastsub = ticks;
+	delta = (delta > 2 * wait) ? 2 * wait : delta;
+	if (qp->q_bionum > 7)
+		g_savg_add_sample(&qp->q_thinktime, delta);
+}
+
+static inline void
+g_rr_update_seekdist(struct g_rr_queue *qp, struct bio *bp)
+{
+	off_t dist;
+
+	if (qp->q_lastoff > bp->bio_offset)
+		dist = qp->q_lastoff - bp->bio_offset;
+	else
+		dist = bp->bio_offset - qp->q_lastoff;
+
+	if (dist > (8192 * 8))
+		dist = 8192 * 8;
+
+	qp->q_lastoff = bp->bio_offset + bp->bio_length;
+
+	if (qp->q_bionum > 7)
+		g_savg_add_sample(&qp->q_seekdist, dist);
+}
+
+/*
+ * Called when a real request for disk I/O arrives.
+ * Locate the queue associated with the client.
+ * If the queue is the one we are anticipating for, reset its timeout;
+ * if the queue is not in the round robin list, insert it in the list.
+ * On any error, do not queue the request and return -1, the caller
+ * will take care of this request.
+ */
+static int
+g_rr_start(void *data, struct bio *bp)
+{
+	struct g_rr_softc *sc = data;
+	struct g_rr_queue *qp;
+
+	if (me.bypass)
+		return (-1);	/* bypass the scheduler */
+
+	/* Get the queue for the request. */
+	qp = g_rr_queue_get(sc, bp);
+	if (qp == NULL)
+		return (-1); /* allocation failed, tell upstream */
+
+	if (gs_bioq_first(&qp->q_bioq) == NULL) {
+		/*
+		 * We are inserting into an empty queue.
+		 * Reset its state if it is sc_active,
+		 * otherwise insert it in the RR list.
+		 */
+		if (qp == sc->sc_active) {
+			qp->q_status = G_QUEUE_READY;
+			callout_stop(&sc->sc_wait);
+		} else {
+			g_sched_priv_ref(qp);
+			TAILQ_INSERT_TAIL(&sc->sc_rr_tailq, qp, q_tailq);
+		}
+	}
+
+	qp->q_bionum = 1 + qp->q_bionum - (qp->q_bionum >> 3);
+
+	g_rr_update_thinktime(qp);
+	g_rr_update_seekdist(qp, bp);
+
+	/* Inherit the reference returned by g_rr_queue_get(). */
+	bp->bio_caller1 = qp;
+	gs_bioq_disksort(&qp->q_bioq, bp);
+
+	return (0);
+}
+
+/*
+ * Callout executed when a queue times out anticipating a new request.
+ */
+static void
+g_rr_wait_timeout(void *data)
+{
+	struct g_rr_softc *sc = data;
+	struct g_geom *geom = sc->sc_geom;
+
+	g_sched_lock(geom);
+	/*
+	 * We can race with other events, so check if
+	 * sc_active is still valid.
+	 */
+	if (sc->sc_active != NULL) {
+		/* Release the reference to the queue. */
+		g_rr_queue_put(sc->sc_active);
+		sc->sc_active = NULL;
+		me.wait_hit--;
+		me.wait_miss++;	/* record the miss */
+	}
+	g_sched_dispatch(geom);
+	g_sched_unlock(geom);
+}
+
+/*
+ * Module glue: allocate descriptor, initialize its fields.
+ */
+static void *
+g_rr_init(struct g_geom *geom)
+{
+	struct g_rr_softc *sc;
+
+	/* XXX check whether we can sleep */
+	sc = malloc(sizeof *sc, M_GEOM_SCHED, M_NOWAIT | M_ZERO);
+	sc->sc_geom = geom;
+	TAILQ_INIT(&sc->sc_rr_tailq);
+	callout_init(&sc->sc_wait, CALLOUT_MPSAFE);
+	LIST_INSERT_HEAD(&me.sc_head, sc, sc_next);
+	me.units++;
+
+	return (sc);
+}
+
+/*
+ * Module glue -- drain the callout structure, destroy the
+ * hash table and its element, and free the descriptor.
+ */
+static void
+g_rr_fini(void *data)
+{
+	struct g_rr_softc *sc = data;
+
+	callout_drain(&sc->sc_wait);
+	KASSERT(sc->sc_active == NULL, ("still a queue under service"));
+	KASSERT(TAILQ_EMPTY(&sc->sc_rr_tailq), ("still scheduled queues"));
+
+	LIST_REMOVE(sc, sc_next);
+	me.units--;
+	free(sc, M_GEOM_SCHED);
+}
+
+/*
+ * Called when the request under service terminates.
+ * Start the anticipation timer if needed.
+ */
+static void
+g_rr_done(void *data, struct bio *bp)
+{
+	struct g_rr_softc *sc = data;
+	struct g_rr_queue *qp;
+
+	sc->sc_in_flight--;
+
+	qp = bp->bio_caller1;
+	if (qp == sc->sc_active && qp->q_status == G_QUEUE_BUSY) {
+		if (!(qp->q_flags & G_FLAG_COMPLETED)) {
+			qp->q_flags |= G_FLAG_COMPLETED;
+			/* in case we want to make the slice adaptive */
+			qp->q_slice_duration = get_bounded(&me.quantum_ms, 2);
+			qp->q_slice_end = ticks + qp->q_slice_duration;
+		}
+
+		/* The queue is trying anticipation, start the timer. */
+		qp->q_status = G_QUEUE_IDLING;
+		/* may make this adaptive */
+		qp->q_wait_ticks = get_bounded(&me.wait_ms, 2);
+		me.wait_hit++;
+		callout_reset(&sc->sc_wait, qp->q_wait_ticks,
+		    g_rr_wait_timeout, sc);
+	} else
+		g_sched_dispatch(sc->sc_geom);
+
+	/* Release a reference to the queue. */
+	g_rr_queue_put(qp);
+}
+
+static void
+g_rr_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
+    struct g_consumer *cp, struct g_provider *pp)
+{
+	if (indent == NULL) {   /* plaintext */
+		sbuf_printf(sb, " units %d queues %d",
+			me.units, me.queues);
+        }
+}
+
+static struct g_gsched g_rr = {
+	.gs_name = "rr",
+	.gs_priv_size = sizeof(struct g_rr_queue),
+	.gs_init = g_rr_init,
+	.gs_fini = g_rr_fini,
+	.gs_start = g_rr_start,
+	.gs_done = g_rr_done,
+	.gs_next = g_rr_next,
+	.gs_dumpconf = g_rr_dumpconf,
+	.gs_init_class = g_rr_init_class,
+	.gs_fini_class = g_rr_fini_class,
+};
+
+DECLARE_GSCHED_MODULE(rr, &g_rr);
diff --git a/sys/geom/sched/gs_scheduler.h b/sys/geom/sched/gs_scheduler.h
new file mode 100644
index 0000000..fa9cb14
--- /dev/null
+++ b/sys/geom/sched/gs_scheduler.h
@@ -0,0 +1,236 @@
+/*-
+ * Copyright (c) 2009-2010 Fabio Checconi, Luigi Rizzo
+ * 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 AUTHORS 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 AUTHORS 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.
+ */
+
+/*
+ * $Id$
+ * $FreeBSD$
+ *
+ * Prototypes for GEOM-based disk scheduling algorithms.
+ * See g_sched.c for generic documentation.
+ *
+ * This file is used by the kernel modules implementing the various
+ * scheduling algorithms. They should provide all the methods
+ * defined in struct g_gsched, and also invoke the macro
+ *	DECLARE_GSCHED_MODULE
+ * which registers the scheduling algorithm with the geom_sched module.
+ *
+ * The various scheduling algorithms do not need to know anything
+ * about geom, they only need to handle the 'bio' requests they
+ * receive, pass them down when needed, and use the locking interface
+ * defined below.
+ */
+
+#ifndef	_G_GSCHED_H_
+#define	_G_GSCHED_H_
+
+#ifdef _KERNEL
+#include <sys/param.h>
+#include <sys/kernel.h>
+#include <sys/ktr.h>
+#include <sys/module.h>
+#include <sys/queue.h>
+#include <geom/geom.h>
+#include "g_sched.h"
+
+/*
+ * This is the interface exported to scheduling modules.
+ *
+ * gs_init() is called when our scheduling algorithm
+ *    starts being used by a geom 'sched'
+ *
+ * gs_fini() is called when the algorithm is released.
+ *
+ * gs_start() is called when a new request comes in. It should
+ *    enqueue the request and return 0 if success, or return non-zero
+ *    in case of failure (meaning the request is passed down).
+ *    The scheduler can use bio->bio_caller1 to store a non-null
+ *    pointer meaning the request is under its control.
+ *
+ * gs_next() is called in a loop by g_sched_dispatch(), right after
+ *    gs_start(), or on timeouts or 'done' events. It should return
+ *    immediately, either a pointer to the bio to be served or NULL
+ *    if no bio should be served now.  If force is specified, a
+ *    work-conserving behavior is expected.
+ *
+ * gs_done() is called when a request under service completes.
+ *    In turn the scheduler may decide to call the dispatch loop
+ *    to serve other pending requests (or make sure there is a pending
+ *    timeout to avoid stalls).
+ *
+ * gs_init_class() is called when a new client (as determined by
+ *    the classifier) starts being used.
+ *
+ * gs_hash_unref() is called right before the class hashtable is
+ *    destroyed; after this call, the scheduler is supposed to hold no
+ *    more references to the elements in the table.
+ */
+
+/* Forward declarations for prototypes. */
+struct g_geom;
+struct g_sched_class;
+
+typedef void *gs_init_t (struct g_geom *geom);
+typedef void gs_fini_t (void *data);
+typedef int gs_start_t (void *data, struct bio *bio);
+typedef void gs_done_t (void *data, struct bio *bio);
+typedef struct bio *gs_next_t (void *data, int force);
+typedef int gs_init_class_t (void *data, void *priv);
+typedef void gs_fini_class_t (void *data, void *priv);
+typedef void gs_hash_unref_t (void *data);
+
+struct g_gsched {
+	const char	*gs_name;
+	int		gs_refs;
+	int		gs_priv_size;
+
+	gs_init_t	*gs_init;
+	gs_fini_t	*gs_fini;
+	gs_start_t	*gs_start;
+	gs_done_t	*gs_done;
+	gs_next_t	*gs_next;
+	g_dumpconf_t	*gs_dumpconf;
+
+	gs_init_class_t	*gs_init_class;
+	gs_fini_class_t	*gs_fini_class;
+	gs_hash_unref_t *gs_hash_unref;
+
+	LIST_ENTRY(g_gsched) glist;
+};
+
+#define	KTR_GSCHED	KTR_SPARE4
+
+MALLOC_DECLARE(M_GEOM_SCHED);
+
+/*
+ * Basic classification mechanism.  Each request is associated to
+ * a g_sched_class, and each scheduler has the opportunity to set
+ * its own private data for the given (class, geom) pair.  The
+ * private data have a base type of g_sched_private, and are
+ * extended at the end with the actual private fields of each
+ * scheduler.
+ */
+struct g_sched_class {
+	int	gsc_refs;
+	int	gsc_expire;
+	u_long	gsc_key;
+	LIST_ENTRY(g_sched_class) gsc_clist;
+
+	void	*gsc_priv[0];
+};
+
+/*
+ * Manipulate the classifier's data.  g_sched_get_class() gets a reference
+ * to the the class corresponding to bp in gp, allocating and initializing
+ * it if necessary.  g_sched_put_class() releases the reference.
+ * The returned value points to the private data for the class.
+ */
+void *g_sched_get_class(struct g_geom *gp, struct bio *bp);
+void g_sched_put_class(struct g_geom *gp, void *priv);
+
+static inline struct g_sched_class *
+g_sched_priv2class(void *priv)
+{
+
+	return ((struct g_sched_class *)((u_long)priv -
+	    offsetof(struct g_sched_class, gsc_priv)));
+}
+
+static inline void
+g_sched_priv_ref(void *priv)
+{
+	struct g_sched_class *gsc;
+
+	gsc = g_sched_priv2class(priv);
+	gsc->gsc_refs++;
+}
+
+/*
+ * Locking interface.  When each operation registered with the
+ * scheduler is invoked, a per-instance lock is taken to protect
+ * the data associated with it.  If the scheduler needs something
+ * else to access the same data (e.g., a callout) it must use
+ * these functions.
+ */
+void g_sched_lock(struct g_geom *gp);
+void g_sched_unlock(struct g_geom *gp);
+
+/*
+ * Restart request dispatching.  Must be called with the per-instance
+ * mutex held.
+ */
+void g_sched_dispatch(struct g_geom *geom);
+
+/*
+ * Simple gathering of statistical data, used by schedulers to collect
+ * info on process history.  Just keep an exponential average of the
+ * samples, with some extra bits of precision.
+ */
+struct g_savg {
+	uint64_t	gs_avg;
+	unsigned int	gs_smpl;
+};
+
+static inline void
+g_savg_add_sample(struct g_savg *ss, uint64_t sample)
+{
+
+	/* EMA with alpha = 0.125, fixed point, 3 bits of precision. */
+	ss->gs_avg = sample + ss->gs_avg - (ss->gs_avg >> 3);
+	ss->gs_smpl = 1 + ss->gs_smpl - (ss->gs_smpl >> 3);
+}
+
+static inline int
+g_savg_valid(struct g_savg *ss)
+{
+
+	/* We want at least 8 samples to deem an average as valid. */
+	return (ss->gs_smpl > 7);
+}
+
+static inline uint64_t
+g_savg_read(struct g_savg *ss)
+{
+
+	return (ss->gs_avg / ss->gs_smpl);
+}
+
+/*
+ * Declaration of a scheduler module.
+ */
+int g_gsched_modevent(module_t mod, int cmd, void *arg);
+
+#define	DECLARE_GSCHED_MODULE(name, gsched)			\
+	static moduledata_t name##_mod = {			\
+		#name,						\
+		g_gsched_modevent,				\
+		gsched,						\
+	};							\
+	DECLARE_MODULE(name, name##_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE); \
+	MODULE_DEPEND(name, geom_sched, 0, 0, 0);
+
+#endif	/* _KERNEL */
+
+#endif	/* _G_GSCHED_H_ */
diff --git a/sys/geom/sched/subr_disk.c b/sys/geom/sched/subr_disk.c
new file mode 100644
index 0000000..008eaab
--- /dev/null
+++ b/sys/geom/sched/subr_disk.c
@@ -0,0 +1,209 @@
+/*-
+ * ----------------------------------------------------------------------------
+ * "THE BEER-WARE LICENSE" (Revision 42):
+ * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
+ * can do whatever you want with this stuff. If we meet some day, and you think
+ * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
+ * ----------------------------------------------------------------------------
+ *
+ * The bioq_disksort() (and the specification of the bioq API)
+ * have been written by Luigi Rizzo and Fabio Checconi under the same
+ * license as above.
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+//#include "opt_geom.h"
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/bio.h>
+#include <sys/conf.h>
+#include <sys/disk.h>
+#include <geom/geom_disk.h>
+#include "g_sched.h"
+
+/*
+ * BIO queue implementation
+ *
+ * Please read carefully the description below before making any change
+ * to the code, or you might change the behaviour of the data structure
+ * in undesirable ways.
+ *
+ * A bioq stores disk I/O request (bio), normally sorted according to
+ * the distance of the requested position (bio->bio_offset) from the
+ * current head position (bioq->last_offset) in the scan direction, i.e.
+ *
+ * 	(uoff_t)(bio_offset - last_offset)
+ *
+ * Note that the cast to unsigned (uoff_t) is fundamental to insure
+ * that the distance is computed in the scan direction.
+ *
+ * The main methods for manipulating the bioq are:
+ *
+ *   bioq_disksort()	performs an ordered insertion;
+ *
+ *   bioq_first()	return the head of the queue, without removing;
+ *
+ *   bioq_takefirst()	return and remove the head of the queue,
+ *		updating the 'current head position' as
+ *		bioq->last_offset = bio->bio_offset + bio->bio_length;
+ *
+ * When updating the 'current head position', we assume that the result of
+ * bioq_takefirst() is dispatched to the device, so bioq->last_offset
+ * represents the head position once the request is complete.
+ *
+ * If the bioq is manipulated using only the above calls, it starts
+ * with a sorted sequence of requests with bio_offset >= last_offset,
+ * possibly followed by another sorted sequence of requests with
+ * 0 <= bio_offset < bioq->last_offset 
+ *
+ * NOTE: historical behaviour was to ignore bio->bio_length in the
+ *	update, but its use tracks the head position in a better way.
+ *	Historical behaviour was also to update the head position when
+ *	the request under service is complete, rather than when the
+ *	request is extracted from the queue. However, the current API
+ *	has no method to update the head position; secondly, once
+ *	a request has been submitted to the disk, we have no idea of
+ *	the actual head position, so the final one is our best guess.
+ *
+ * --- Direct queue manipulation ---
+ *
+ * A bioq uses an underlying TAILQ to store requests, so we also
+ * export methods to manipulate the TAILQ, in particular:
+ *
+ * bioq_insert_tail()	insert an entry at the end.
+ *		It also creates a 'barrier' so all subsequent
+ *		insertions through bioq_disksort() will end up
+ *		after this entry;
+ *
+ * bioq_insert_head()	insert an entry at the head, update
+ *		bioq->last_offset = bio->bio_offset so that
+ *		all subsequent insertions through bioq_disksort()
+ *		will end up after this entry;
+ *
+ * bioq_remove()	remove a generic element from the queue, act as
+ *		bioq_takefirst() if invoked on the head of the queue.
+ *
+ * The semantic of these methods is the same of the operations
+ * on the underlying TAILQ, but with additional guarantees on
+ * subsequent bioq_disksort() calls. E.g. bioq_insert_tail()
+ * can be useful for making sure that all previous ops are flushed
+ * to disk before continuing.
+ *
+ * Updating bioq->last_offset on a bioq_insert_head() guarantees
+ * that the bio inserted with the last bioq_insert_head() will stay
+ * at the head of the queue even after subsequent bioq_disksort().
+ *
+ * Note that when the direct queue manipulation functions are used,
+ * the queue may contain multiple inversion points (i.e. more than
+ * two sorted sequences of requests).
+ *
+ */
+
+void
+gs_bioq_init(struct bio_queue_head *head)
+{
+
+	TAILQ_INIT(&head->queue);
+	head->last_offset = 0;
+	head->insert_point = NULL;
+}
+
+void
+gs_bioq_remove(struct bio_queue_head *head, struct bio *bp)
+{
+
+	if (bp == TAILQ_FIRST(&head->queue))
+		head->last_offset = bp->bio_offset + bp->bio_length;
+
+	if (bp == head->insert_point)
+		head->insert_point = NULL;
+
+	TAILQ_REMOVE(&head->queue, bp, bio_queue);
+}
+
+void
+gs_bioq_flush(struct bio_queue_head *head, struct devstat *stp, int error)
+{
+	struct bio *bp;
+
+	while ((bp = gs_bioq_takefirst(head)) != NULL)
+		biofinish(bp, stp, error);
+}
+
+void
+gs_bioq_insert_head(struct bio_queue_head *head, struct bio *bp)
+{
+
+	head->last_offset = bp->bio_offset;
+	TAILQ_INSERT_HEAD(&head->queue, bp, bio_queue);
+}
+
+void
+gs_bioq_insert_tail(struct bio_queue_head *head, struct bio *bp)
+{
+
+	TAILQ_INSERT_TAIL(&head->queue, bp, bio_queue);
+	head->insert_point = bp;
+}
+
+struct bio *
+gs_bioq_first(struct bio_queue_head *head)
+{
+
+	return (TAILQ_FIRST(&head->queue));
+}
+
+struct bio *
+gs_bioq_takefirst(struct bio_queue_head *head)
+{
+	struct bio *bp;
+
+	bp = TAILQ_FIRST(&head->queue);
+	if (bp != NULL)
+		gs_bioq_remove(head, bp);
+	return (bp);
+}
+
+/*
+ * Compute the sorting key. The cast to unsigned is
+ * fundamental for correctness, see the description
+ * near the beginning of the file.
+ */
+static inline uoff_t
+gs_bioq_bio_key(struct bio_queue_head *head, struct bio *bp)
+{
+
+	return ((uoff_t)(bp->bio_offset - head->last_offset));
+}
+
+/*
+ * Seek sort for disks.
+ *
+ * Sort all requests in a single queue while keeping
+ * track of the current position of the disk with last_offset.
+ * See above for details.
+ */
+void
+gs_bioq_disksort(struct bio_queue_head *head, struct bio *bp)
+{
+	struct bio *cur, *prev = NULL;
+	uoff_t key = gs_bioq_bio_key(head, bp);
+
+	cur = TAILQ_FIRST(&head->queue);
+
+	if (head->insert_point)
+		cur = head->insert_point;
+
+	while (cur != NULL && key >= gs_bioq_bio_key(head, cur)) {
+		prev = cur;
+		cur = TAILQ_NEXT(cur, bio_queue);
+	}
+
+	if (prev == NULL)
+		TAILQ_INSERT_HEAD(&head->queue, bp, bio_queue);
+	else
+		TAILQ_INSERT_AFTER(&head->queue, prev, bp, bio_queue);
+}