freebsd-skq/sys/geom/geom_subr.c
smh 5c7a6f5d92 Improve ZFS N-way mirror read performance by using load and locality
information.

The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.

The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.

Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.

This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.

The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.

With pre-fetch disabled (vfs.zfs.prefetch_disable=1):

== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s

With pre-fetch enabled (vfs.zfs.prefetch_disable=0):

== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s

In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.

The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc

These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487

Reviewed by:	gibbs, mav, will
MFC after:	2 weeks
Sponsored by:	Multiplay
2013-10-23 09:54:58 +00:00

1527 lines
36 KiB
C

/*-
* Copyright (c) 2002 Poul-Henning Kamp
* Copyright (c) 2002 Networks Associates Technology, Inc.
* All rights reserved.
*
* This software was developed for the FreeBSD Project by Poul-Henning Kamp
* and NAI Labs, the Security Research Division of Network Associates, Inc.
* under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
* DARPA CHATS research program.
*
* 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.
* 3. The names of the authors may not be used to endorse or promote
* products derived from this software without specific prior written
* permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ddb.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/devicestat.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/bio.h>
#include <sys/sysctl.h>
#include <sys/proc.h>
#include <sys/kthread.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/errno.h>
#include <sys/sbuf.h>
#include <geom/geom.h>
#include <geom/geom_int.h>
#include <machine/stdarg.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
#ifdef KDB
#include <sys/kdb.h>
#endif
struct class_list_head g_classes = LIST_HEAD_INITIALIZER(g_classes);
static struct g_tailq_head geoms = TAILQ_HEAD_INITIALIZER(geoms);
char *g_wait_event, *g_wait_up, *g_wait_down, *g_wait_sim;
struct g_hh00 {
struct g_class *mp;
struct g_provider *pp;
off_t size;
int error;
int post;
};
/*
* This event offers a new class a chance to taste all preexisting providers.
*/
static void
g_load_class(void *arg, int flag)
{
struct g_hh00 *hh;
struct g_class *mp2, *mp;
struct g_geom *gp;
struct g_provider *pp;
g_topology_assert();
if (flag == EV_CANCEL) /* XXX: can't happen ? */
return;
if (g_shutdown)
return;
hh = arg;
mp = hh->mp;
hh->error = 0;
if (hh->post) {
g_free(hh);
hh = NULL;
}
g_trace(G_T_TOPOLOGY, "g_load_class(%s)", mp->name);
KASSERT(mp->name != NULL && *mp->name != '\0',
("GEOM class has no name"));
LIST_FOREACH(mp2, &g_classes, class) {
if (mp2 == mp) {
printf("The GEOM class %s is already loaded.\n",
mp2->name);
if (hh != NULL)
hh->error = EEXIST;
return;
} else if (strcmp(mp2->name, mp->name) == 0) {
printf("A GEOM class %s is already loaded.\n",
mp2->name);
if (hh != NULL)
hh->error = EEXIST;
return;
}
}
LIST_INIT(&mp->geom);
LIST_INSERT_HEAD(&g_classes, mp, class);
if (mp->init != NULL)
mp->init(mp);
if (mp->taste == NULL)
return;
LIST_FOREACH(mp2, &g_classes, class) {
if (mp == mp2)
continue;
LIST_FOREACH(gp, &mp2->geom, geom) {
LIST_FOREACH(pp, &gp->provider, provider) {
mp->taste(mp, pp, 0);
g_topology_assert();
}
}
}
}
static int
g_unload_class(struct g_class *mp)
{
struct g_geom *gp;
struct g_provider *pp;
struct g_consumer *cp;
int error;
g_topology_lock();
g_trace(G_T_TOPOLOGY, "g_unload_class(%s)", mp->name);
retry:
G_VALID_CLASS(mp);
LIST_FOREACH(gp, &mp->geom, geom) {
/* We refuse to unload if anything is open */
LIST_FOREACH(pp, &gp->provider, provider)
if (pp->acr || pp->acw || pp->ace) {
g_topology_unlock();
return (EBUSY);
}
LIST_FOREACH(cp, &gp->consumer, consumer)
if (cp->acr || cp->acw || cp->ace) {
g_topology_unlock();
return (EBUSY);
}
/* If the geom is withering, wait for it to finish. */
if (gp->flags & G_GEOM_WITHER) {
g_topology_sleep(mp, 1);
goto retry;
}
}
/*
* We allow unloading if we have no geoms, or a class
* method we can use to get rid of them.
*/
if (!LIST_EMPTY(&mp->geom) && mp->destroy_geom == NULL) {
g_topology_unlock();
return (EOPNOTSUPP);
}
/* Bar new entries */
mp->taste = NULL;
mp->config = NULL;
LIST_FOREACH(gp, &mp->geom, geom) {
error = mp->destroy_geom(NULL, mp, gp);
if (error != 0) {
g_topology_unlock();
return (error);
}
}
/* Wait for withering to finish. */
for (;;) {
gp = LIST_FIRST(&mp->geom);
if (gp == NULL)
break;
KASSERT(gp->flags & G_GEOM_WITHER,
("Non-withering geom in class %s", mp->name));
g_topology_sleep(mp, 1);
}
G_VALID_CLASS(mp);
if (mp->fini != NULL)
mp->fini(mp);
LIST_REMOVE(mp, class);
g_topology_unlock();
return (0);
}
int
g_modevent(module_t mod, int type, void *data)
{
struct g_hh00 *hh;
int error;
static int g_ignition;
struct g_class *mp;
mp = data;
if (mp->version != G_VERSION) {
printf("GEOM class %s has Wrong version %x\n",
mp->name, mp->version);
return (EINVAL);
}
if (!g_ignition) {
g_ignition++;
g_init();
}
error = EOPNOTSUPP;
switch (type) {
case MOD_LOAD:
g_trace(G_T_TOPOLOGY, "g_modevent(%s, LOAD)", mp->name);
hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
hh->mp = mp;
/*
* Once the system is not cold, MOD_LOAD calls will be
* from the userland and the g_event thread will be able
* to acknowledge their completion.
*/
if (cold) {
hh->post = 1;
error = g_post_event(g_load_class, hh, M_WAITOK, NULL);
} else {
error = g_waitfor_event(g_load_class, hh, M_WAITOK,
NULL);
if (error == 0)
error = hh->error;
g_free(hh);
}
break;
case MOD_UNLOAD:
g_trace(G_T_TOPOLOGY, "g_modevent(%s, UNLOAD)", mp->name);
DROP_GIANT();
error = g_unload_class(mp);
PICKUP_GIANT();
if (error == 0) {
KASSERT(LIST_EMPTY(&mp->geom),
("Unloaded class (%s) still has geom", mp->name));
}
break;
}
return (error);
}
static void
g_retaste_event(void *arg, int flag)
{
struct g_class *mp, *mp2;
struct g_geom *gp;
struct g_hh00 *hh;
struct g_provider *pp;
struct g_consumer *cp;
g_topology_assert();
if (flag == EV_CANCEL) /* XXX: can't happen ? */
return;
if (g_shutdown || g_notaste)
return;
hh = arg;
mp = hh->mp;
hh->error = 0;
if (hh->post) {
g_free(hh);
hh = NULL;
}
g_trace(G_T_TOPOLOGY, "g_retaste(%s)", mp->name);
LIST_FOREACH(mp2, &g_classes, class) {
LIST_FOREACH(gp, &mp2->geom, geom) {
LIST_FOREACH(pp, &gp->provider, provider) {
if (pp->acr || pp->acw || pp->ace)
continue;
LIST_FOREACH(cp, &pp->consumers, consumers) {
if (cp->geom->class == mp &&
(cp->flags & G_CF_ORPHAN) == 0)
break;
}
if (cp != NULL) {
cp->flags |= G_CF_ORPHAN;
g_wither_geom(cp->geom, ENXIO);
}
mp->taste(mp, pp, 0);
g_topology_assert();
}
}
}
}
int
g_retaste(struct g_class *mp)
{
struct g_hh00 *hh;
int error;
if (mp->taste == NULL)
return (EINVAL);
hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
hh->mp = mp;
if (cold) {
hh->post = 1;
error = g_post_event(g_retaste_event, hh, M_WAITOK, NULL);
} else {
error = g_waitfor_event(g_retaste_event, hh, M_WAITOK, NULL);
if (error == 0)
error = hh->error;
g_free(hh);
}
return (error);
}
struct g_geom *
g_new_geomf(struct g_class *mp, const char *fmt, ...)
{
struct g_geom *gp;
va_list ap;
struct sbuf *sb;
g_topology_assert();
G_VALID_CLASS(mp);
sb = sbuf_new_auto();
va_start(ap, fmt);
sbuf_vprintf(sb, fmt, ap);
va_end(ap);
sbuf_finish(sb);
gp = g_malloc(sizeof *gp, M_WAITOK | M_ZERO);
gp->name = g_malloc(sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
gp->class = mp;
gp->rank = 1;
LIST_INIT(&gp->consumer);
LIST_INIT(&gp->provider);
LIST_INSERT_HEAD(&mp->geom, gp, geom);
TAILQ_INSERT_HEAD(&geoms, gp, geoms);
strcpy(gp->name, sbuf_data(sb));
sbuf_delete(sb);
/* Fill in defaults from class */
gp->start = mp->start;
gp->spoiled = mp->spoiled;
gp->attrchanged = mp->attrchanged;
gp->providergone = mp->providergone;
gp->dumpconf = mp->dumpconf;
gp->access = mp->access;
gp->orphan = mp->orphan;
gp->ioctl = mp->ioctl;
gp->resize = mp->resize;
return (gp);
}
void
g_destroy_geom(struct g_geom *gp)
{
g_topology_assert();
G_VALID_GEOM(gp);
g_trace(G_T_TOPOLOGY, "g_destroy_geom(%p(%s))", gp, gp->name);
KASSERT(LIST_EMPTY(&gp->consumer),
("g_destroy_geom(%s) with consumer(s) [%p]",
gp->name, LIST_FIRST(&gp->consumer)));
KASSERT(LIST_EMPTY(&gp->provider),
("g_destroy_geom(%s) with provider(s) [%p]",
gp->name, LIST_FIRST(&gp->provider)));
g_cancel_event(gp);
LIST_REMOVE(gp, geom);
TAILQ_REMOVE(&geoms, gp, geoms);
g_free(gp->name);
g_free(gp);
}
/*
* This function is called (repeatedly) until the geom has withered away.
*/
void
g_wither_geom(struct g_geom *gp, int error)
{
struct g_provider *pp;
g_topology_assert();
G_VALID_GEOM(gp);
g_trace(G_T_TOPOLOGY, "g_wither_geom(%p(%s))", gp, gp->name);
if (!(gp->flags & G_GEOM_WITHER)) {
gp->flags |= G_GEOM_WITHER;
LIST_FOREACH(pp, &gp->provider, provider)
if (!(pp->flags & G_PF_ORPHAN))
g_orphan_provider(pp, error);
}
g_do_wither();
}
/*
* Convenience function to destroy a particular provider.
*/
void
g_wither_provider(struct g_provider *pp, int error)
{
pp->flags |= G_PF_WITHER;
if (!(pp->flags & G_PF_ORPHAN))
g_orphan_provider(pp, error);
}
/*
* This function is called (repeatedly) until the has withered away.
*/
void
g_wither_geom_close(struct g_geom *gp, int error)
{
struct g_consumer *cp;
g_topology_assert();
G_VALID_GEOM(gp);
g_trace(G_T_TOPOLOGY, "g_wither_geom_close(%p(%s))", gp, gp->name);
LIST_FOREACH(cp, &gp->consumer, consumer)
if (cp->acr || cp->acw || cp->ace)
g_access(cp, -cp->acr, -cp->acw, -cp->ace);
g_wither_geom(gp, error);
}
/*
* This function is called (repeatedly) until we cant wash away more
* withered bits at present.
*/
void
g_wither_washer()
{
struct g_class *mp;
struct g_geom *gp, *gp2;
struct g_provider *pp, *pp2;
struct g_consumer *cp, *cp2;
g_topology_assert();
LIST_FOREACH(mp, &g_classes, class) {
LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
LIST_FOREACH_SAFE(pp, &gp->provider, provider, pp2) {
if (!(pp->flags & G_PF_WITHER))
continue;
if (LIST_EMPTY(&pp->consumers))
g_destroy_provider(pp);
}
if (!(gp->flags & G_GEOM_WITHER))
continue;
LIST_FOREACH_SAFE(pp, &gp->provider, provider, pp2) {
if (LIST_EMPTY(&pp->consumers))
g_destroy_provider(pp);
}
LIST_FOREACH_SAFE(cp, &gp->consumer, consumer, cp2) {
if (cp->acr || cp->acw || cp->ace)
continue;
if (cp->provider != NULL)
g_detach(cp);
g_destroy_consumer(cp);
}
if (LIST_EMPTY(&gp->provider) &&
LIST_EMPTY(&gp->consumer))
g_destroy_geom(gp);
}
}
}
struct g_consumer *
g_new_consumer(struct g_geom *gp)
{
struct g_consumer *cp;
g_topology_assert();
G_VALID_GEOM(gp);
KASSERT(!(gp->flags & G_GEOM_WITHER),
("g_new_consumer on WITHERing geom(%s) (class %s)",
gp->name, gp->class->name));
KASSERT(gp->orphan != NULL,
("g_new_consumer on geom(%s) (class %s) without orphan",
gp->name, gp->class->name));
cp = g_malloc(sizeof *cp, M_WAITOK | M_ZERO);
cp->geom = gp;
cp->stat = devstat_new_entry(cp, -1, 0, DEVSTAT_ALL_SUPPORTED,
DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
LIST_INSERT_HEAD(&gp->consumer, cp, consumer);
return(cp);
}
void
g_destroy_consumer(struct g_consumer *cp)
{
struct g_geom *gp;
g_topology_assert();
G_VALID_CONSUMER(cp);
g_trace(G_T_TOPOLOGY, "g_destroy_consumer(%p)", cp);
KASSERT (cp->provider == NULL, ("g_destroy_consumer but attached"));
KASSERT (cp->acr == 0, ("g_destroy_consumer with acr"));
KASSERT (cp->acw == 0, ("g_destroy_consumer with acw"));
KASSERT (cp->ace == 0, ("g_destroy_consumer with ace"));
g_cancel_event(cp);
gp = cp->geom;
LIST_REMOVE(cp, consumer);
devstat_remove_entry(cp->stat);
g_free(cp);
if (gp->flags & G_GEOM_WITHER)
g_do_wither();
}
static void
g_new_provider_event(void *arg, int flag)
{
struct g_class *mp;
struct g_provider *pp;
struct g_consumer *cp, *next_cp;
g_topology_assert();
if (flag == EV_CANCEL)
return;
if (g_shutdown)
return;
pp = arg;
G_VALID_PROVIDER(pp);
KASSERT(!(pp->flags & G_PF_WITHER),
("g_new_provider_event but withered"));
LIST_FOREACH_SAFE(cp, &pp->consumers, consumers, next_cp) {
if ((cp->flags & G_CF_ORPHAN) == 0 &&
cp->geom->attrchanged != NULL)
cp->geom->attrchanged(cp, "GEOM::media");
}
if (g_notaste)
return;
LIST_FOREACH(mp, &g_classes, class) {
if (mp->taste == NULL)
continue;
LIST_FOREACH(cp, &pp->consumers, consumers)
if (cp->geom->class == mp &&
(cp->flags & G_CF_ORPHAN) == 0)
break;
if (cp != NULL)
continue;
mp->taste(mp, pp, 0);
g_topology_assert();
}
}
struct g_provider *
g_new_providerf(struct g_geom *gp, const char *fmt, ...)
{
struct g_provider *pp;
struct sbuf *sb;
va_list ap;
g_topology_assert();
G_VALID_GEOM(gp);
KASSERT(gp->access != NULL,
("new provider on geom(%s) without ->access (class %s)",
gp->name, gp->class->name));
KASSERT(gp->start != NULL,
("new provider on geom(%s) without ->start (class %s)",
gp->name, gp->class->name));
KASSERT(!(gp->flags & G_GEOM_WITHER),
("new provider on WITHERing geom(%s) (class %s)",
gp->name, gp->class->name));
sb = sbuf_new_auto();
va_start(ap, fmt);
sbuf_vprintf(sb, fmt, ap);
va_end(ap);
sbuf_finish(sb);
pp = g_malloc(sizeof *pp + sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
pp->name = (char *)(pp + 1);
strcpy(pp->name, sbuf_data(sb));
sbuf_delete(sb);
LIST_INIT(&pp->consumers);
pp->error = ENXIO;
pp->geom = gp;
pp->stat = devstat_new_entry(pp, -1, 0, DEVSTAT_ALL_SUPPORTED,
DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
LIST_INSERT_HEAD(&gp->provider, pp, provider);
g_post_event(g_new_provider_event, pp, M_WAITOK, pp, gp, NULL);
return (pp);
}
void
g_error_provider(struct g_provider *pp, int error)
{
/* G_VALID_PROVIDER(pp); We may not have g_topology */
pp->error = error;
}
static void
g_resize_provider_event(void *arg, int flag)
{
struct g_hh00 *hh;
struct g_class *mp;
struct g_geom *gp;
struct g_provider *pp;
struct g_consumer *cp, *cp2;
off_t size;
g_topology_assert();
if (g_shutdown)
return;
hh = arg;
pp = hh->pp;
size = hh->size;
g_free(hh);
G_VALID_PROVIDER(pp);
g_trace(G_T_TOPOLOGY, "g_resize_provider_event(%p)", pp);
LIST_FOREACH_SAFE(cp, &pp->consumers, consumers, cp2) {
gp = cp->geom;
if (gp->resize == NULL && size < pp->mediasize) {
cp->flags |= G_CF_ORPHAN;
cp->geom->orphan(cp);
}
}
pp->mediasize = size;
LIST_FOREACH_SAFE(cp, &pp->consumers, consumers, cp2) {
gp = cp->geom;
if (gp->resize != NULL)
gp->resize(cp);
}
/*
* After resizing, the previously invalid GEOM class metadata
* might become valid. This means we should retaste.
*/
LIST_FOREACH(mp, &g_classes, class) {
if (mp->taste == NULL)
continue;
LIST_FOREACH(cp, &pp->consumers, consumers)
if (cp->geom->class == mp &&
(cp->flags & G_CF_ORPHAN) == 0)
break;
if (cp != NULL)
continue;
mp->taste(mp, pp, 0);
g_topology_assert();
}
}
void
g_resize_provider(struct g_provider *pp, off_t size)
{
struct g_hh00 *hh;
G_VALID_PROVIDER(pp);
if (size == pp->mediasize)
return;
hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
hh->pp = pp;
hh->size = size;
g_post_event(g_resize_provider_event, hh, M_WAITOK, NULL);
}
#ifndef _PATH_DEV
#define _PATH_DEV "/dev/"
#endif
struct g_provider *
g_provider_by_name(char const *arg)
{
struct g_class *cp;
struct g_geom *gp;
struct g_provider *pp;
if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
arg += sizeof(_PATH_DEV) - 1;
LIST_FOREACH(cp, &g_classes, class) {
LIST_FOREACH(gp, &cp->geom, geom) {
LIST_FOREACH(pp, &gp->provider, provider) {
if (!strcmp(arg, pp->name))
return (pp);
}
}
}
return (NULL);
}
void
g_destroy_provider(struct g_provider *pp)
{
struct g_geom *gp;
g_topology_assert();
G_VALID_PROVIDER(pp);
KASSERT(LIST_EMPTY(&pp->consumers),
("g_destroy_provider but attached"));
KASSERT (pp->acr == 0, ("g_destroy_provider with acr"));
KASSERT (pp->acw == 0, ("g_destroy_provider with acw"));
KASSERT (pp->ace == 0, ("g_destroy_provider with ace"));
g_cancel_event(pp);
LIST_REMOVE(pp, provider);
gp = pp->geom;
devstat_remove_entry(pp->stat);
/*
* If a callback was provided, send notification that the provider
* is now gone.
*/
if (gp->providergone != NULL)
gp->providergone(pp);
g_free(pp);
if ((gp->flags & G_GEOM_WITHER))
g_do_wither();
}
/*
* We keep the "geoms" list sorted by topological order (== increasing
* numerical rank) at all times.
* When an attach is done, the attaching geoms rank is invalidated
* and it is moved to the tail of the list.
* All geoms later in the sequence has their ranks reevaluated in
* sequence. If we cannot assign rank to a geom because it's
* prerequisites do not have rank, we move that element to the tail
* of the sequence with invalid rank as well.
* At some point we encounter our original geom and if we stil fail
* to assign it a rank, there must be a loop and we fail back to
* g_attach() which detach again and calls redo_rank again
* to fix up the damage.
* It would be much simpler code wise to do it recursively, but we
* can't risk that on the kernel stack.
*/
static int
redo_rank(struct g_geom *gp)
{
struct g_consumer *cp;
struct g_geom *gp1, *gp2;
int n, m;
g_topology_assert();
G_VALID_GEOM(gp);
/* Invalidate this geoms rank and move it to the tail */
gp1 = TAILQ_NEXT(gp, geoms);
if (gp1 != NULL) {
gp->rank = 0;
TAILQ_REMOVE(&geoms, gp, geoms);
TAILQ_INSERT_TAIL(&geoms, gp, geoms);
} else {
gp1 = gp;
}
/* re-rank the rest of the sequence */
for (; gp1 != NULL; gp1 = gp2) {
gp1->rank = 0;
m = 1;
LIST_FOREACH(cp, &gp1->consumer, consumer) {
if (cp->provider == NULL)
continue;
n = cp->provider->geom->rank;
if (n == 0) {
m = 0;
break;
} else if (n >= m)
m = n + 1;
}
gp1->rank = m;
gp2 = TAILQ_NEXT(gp1, geoms);
/* got a rank, moving on */
if (m != 0)
continue;
/* no rank to original geom means loop */
if (gp == gp1)
return (ELOOP);
/* no rank, put it at the end move on */
TAILQ_REMOVE(&geoms, gp1, geoms);
TAILQ_INSERT_TAIL(&geoms, gp1, geoms);
}
return (0);
}
int
g_attach(struct g_consumer *cp, struct g_provider *pp)
{
int error;
g_topology_assert();
G_VALID_CONSUMER(cp);
G_VALID_PROVIDER(pp);
g_trace(G_T_TOPOLOGY, "g_attach(%p, %p)", cp, pp);
KASSERT(cp->provider == NULL, ("attach but attached"));
cp->provider = pp;
LIST_INSERT_HEAD(&pp->consumers, cp, consumers);
error = redo_rank(cp->geom);
if (error) {
LIST_REMOVE(cp, consumers);
cp->provider = NULL;
redo_rank(cp->geom);
}
return (error);
}
void
g_detach(struct g_consumer *cp)
{
struct g_provider *pp;
g_topology_assert();
G_VALID_CONSUMER(cp);
g_trace(G_T_TOPOLOGY, "g_detach(%p)", cp);
KASSERT(cp->provider != NULL, ("detach but not attached"));
KASSERT(cp->acr == 0, ("detach but nonzero acr"));
KASSERT(cp->acw == 0, ("detach but nonzero acw"));
KASSERT(cp->ace == 0, ("detach but nonzero ace"));
KASSERT(cp->nstart == cp->nend,
("detach with active requests"));
pp = cp->provider;
LIST_REMOVE(cp, consumers);
cp->provider = NULL;
if ((cp->geom->flags & G_GEOM_WITHER) ||
(pp->geom->flags & G_GEOM_WITHER) ||
(pp->flags & G_PF_WITHER))
g_do_wither();
redo_rank(cp->geom);
}
/*
* g_access()
*
* Access-check with delta values. The question asked is "can provider
* "cp" change the access counters by the relative amounts dc[rwe] ?"
*/
int
g_access(struct g_consumer *cp, int dcr, int dcw, int dce)
{
struct g_provider *pp;
int pr,pw,pe;
int error;
g_topology_assert();
G_VALID_CONSUMER(cp);
pp = cp->provider;
KASSERT(pp != NULL, ("access but not attached"));
G_VALID_PROVIDER(pp);
g_trace(G_T_ACCESS, "g_access(%p(%s), %d, %d, %d)",
cp, pp->name, dcr, dcw, dce);
KASSERT(cp->acr + dcr >= 0, ("access resulting in negative acr"));
KASSERT(cp->acw + dcw >= 0, ("access resulting in negative acw"));
KASSERT(cp->ace + dce >= 0, ("access resulting in negative ace"));
KASSERT(dcr != 0 || dcw != 0 || dce != 0, ("NOP access request"));
KASSERT(pp->geom->access != NULL, ("NULL geom->access"));
/*
* If our class cares about being spoiled, and we have been, we
* are probably just ahead of the event telling us that. Fail
* now rather than having to unravel this later.
*/
if (cp->geom->spoiled != NULL && (cp->flags & G_CF_SPOILED) &&
(dcr > 0 || dcw > 0 || dce > 0))
return (ENXIO);
/*
* Figure out what counts the provider would have had, if this
* consumer had (r0w0e0) at this time.
*/
pr = pp->acr - cp->acr;
pw = pp->acw - cp->acw;
pe = pp->ace - cp->ace;
g_trace(G_T_ACCESS,
"open delta:[r%dw%de%d] old:[r%dw%de%d] provider:[r%dw%de%d] %p(%s)",
dcr, dcw, dce,
cp->acr, cp->acw, cp->ace,
pp->acr, pp->acw, pp->ace,
pp, pp->name);
/* If foot-shooting is enabled, any open on rank#1 is OK */
if ((g_debugflags & 16) && pp->geom->rank == 1)
;
/* If we try exclusive but already write: fail */
else if (dce > 0 && pw > 0)
return (EPERM);
/* If we try write but already exclusive: fail */
else if (dcw > 0 && pe > 0)
return (EPERM);
/* If we try to open more but provider is error'ed: fail */
else if ((dcr > 0 || dcw > 0 || dce > 0) && pp->error != 0)
return (pp->error);
/* Ok then... */
error = pp->geom->access(pp, dcr, dcw, dce);
KASSERT(dcr > 0 || dcw > 0 || dce > 0 || error == 0,
("Geom provider %s::%s failed closing ->access()",
pp->geom->class->name, pp->name));
if (!error) {
/*
* If we open first write, spoil any partner consumers.
* If we close last write and provider is not errored,
* trigger re-taste.
*/
if (pp->acw == 0 && dcw != 0)
g_spoil(pp, cp);
else if (pp->acw != 0 && pp->acw == -dcw && pp->error == 0 &&
!(pp->geom->flags & G_GEOM_WITHER))
g_post_event(g_new_provider_event, pp, M_WAITOK,
pp, NULL);
pp->acr += dcr;
pp->acw += dcw;
pp->ace += dce;
cp->acr += dcr;
cp->acw += dcw;
cp->ace += dce;
if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)
KASSERT(pp->sectorsize > 0,
("Provider %s lacks sectorsize", pp->name));
if ((cp->geom->flags & G_GEOM_WITHER) &&
cp->acr == 0 && cp->acw == 0 && cp->ace == 0)
g_do_wither();
}
return (error);
}
int
g_handleattr_int(struct bio *bp, const char *attribute, int val)
{
return (g_handleattr(bp, attribute, &val, sizeof val));
}
int
g_handleattr_uint16_t(struct bio *bp, const char *attribute, uint16_t val)
{
return (g_handleattr(bp, attribute, &val, sizeof val));
}
int
g_handleattr_off_t(struct bio *bp, const char *attribute, off_t val)
{
return (g_handleattr(bp, attribute, &val, sizeof val));
}
int
g_handleattr_str(struct bio *bp, const char *attribute, const char *str)
{
return (g_handleattr(bp, attribute, str, 0));
}
int
g_handleattr(struct bio *bp, const char *attribute, const void *val, int len)
{
int error = 0;
if (strcmp(bp->bio_attribute, attribute))
return (0);
if (len == 0) {
bzero(bp->bio_data, bp->bio_length);
if (strlcpy(bp->bio_data, val, bp->bio_length) >=
bp->bio_length) {
printf("%s: %s bio_length %jd len %zu -> EFAULT\n",
__func__, bp->bio_to->name,
(intmax_t)bp->bio_length, strlen(val));
error = EFAULT;
}
} else if (bp->bio_length == len) {
bcopy(val, bp->bio_data, len);
} else {
printf("%s: %s bio_length %jd len %d -> EFAULT\n", __func__,
bp->bio_to->name, (intmax_t)bp->bio_length, len);
error = EFAULT;
}
if (error == 0)
bp->bio_completed = bp->bio_length;
g_io_deliver(bp, error);
return (1);
}
int
g_std_access(struct g_provider *pp,
int dr __unused, int dw __unused, int de __unused)
{
g_topology_assert();
G_VALID_PROVIDER(pp);
return (0);
}
void
g_std_done(struct bio *bp)
{
struct bio *bp2;
bp2 = bp->bio_parent;
if (bp2->bio_error == 0)
bp2->bio_error = bp->bio_error;
bp2->bio_completed += bp->bio_completed;
g_destroy_bio(bp);
bp2->bio_inbed++;
if (bp2->bio_children == bp2->bio_inbed)
g_io_deliver(bp2, bp2->bio_error);
}
/* XXX: maybe this is only g_slice_spoiled */
void
g_std_spoiled(struct g_consumer *cp)
{
struct g_geom *gp;
struct g_provider *pp;
g_topology_assert();
G_VALID_CONSUMER(cp);
g_trace(G_T_TOPOLOGY, "g_std_spoiled(%p)", cp);
cp->flags |= G_CF_ORPHAN;
g_detach(cp);
gp = cp->geom;
LIST_FOREACH(pp, &gp->provider, provider)
g_orphan_provider(pp, ENXIO);
g_destroy_consumer(cp);
if (LIST_EMPTY(&gp->provider) && LIST_EMPTY(&gp->consumer))
g_destroy_geom(gp);
else
gp->flags |= G_GEOM_WITHER;
}
/*
* Spoiling happens when a provider is opened for writing, but consumers
* which are configured by in-band data are attached (slicers for instance).
* Since the write might potentially change the in-band data, such consumers
* need to re-evaluate their existence after the writing session closes.
* We do this by (offering to) tear them down when the open for write happens
* in return for a re-taste when it closes again.
* Together with the fact that such consumers grab an 'e' bit whenever they
* are open, regardless of mode, this ends up DTRT.
*/
static void
g_spoil_event(void *arg, int flag)
{
struct g_provider *pp;
struct g_consumer *cp, *cp2;
g_topology_assert();
if (flag == EV_CANCEL)
return;
pp = arg;
G_VALID_PROVIDER(pp);
for (cp = LIST_FIRST(&pp->consumers); cp != NULL; cp = cp2) {
cp2 = LIST_NEXT(cp, consumers);
if ((cp->flags & G_CF_SPOILED) == 0)
continue;
cp->flags &= ~G_CF_SPOILED;
if (cp->geom->spoiled == NULL)
continue;
cp->geom->spoiled(cp);
g_topology_assert();
}
}
void
g_spoil(struct g_provider *pp, struct g_consumer *cp)
{
struct g_consumer *cp2;
g_topology_assert();
G_VALID_PROVIDER(pp);
G_VALID_CONSUMER(cp);
LIST_FOREACH(cp2, &pp->consumers, consumers) {
if (cp2 == cp)
continue;
/*
KASSERT(cp2->acr == 0, ("spoiling cp->acr = %d", cp2->acr));
KASSERT(cp2->acw == 0, ("spoiling cp->acw = %d", cp2->acw));
*/
KASSERT(cp2->ace == 0, ("spoiling cp->ace = %d", cp2->ace));
cp2->flags |= G_CF_SPOILED;
}
g_post_event(g_spoil_event, pp, M_WAITOK, pp, NULL);
}
static void
g_media_changed_event(void *arg, int flag)
{
struct g_provider *pp;
int retaste;
g_topology_assert();
if (flag == EV_CANCEL)
return;
pp = arg;
G_VALID_PROVIDER(pp);
/*
* If provider was not open for writing, queue retaste after spoiling.
* If it was, retaste will happen automatically on close.
*/
retaste = (pp->acw == 0 && pp->error == 0 &&
!(pp->geom->flags & G_GEOM_WITHER));
g_spoil_event(arg, flag);
if (retaste)
g_post_event(g_new_provider_event, pp, M_WAITOK, pp, NULL);
}
int
g_media_changed(struct g_provider *pp, int flag)
{
struct g_consumer *cp;
LIST_FOREACH(cp, &pp->consumers, consumers)
cp->flags |= G_CF_SPOILED;
return (g_post_event(g_media_changed_event, pp, flag, pp, NULL));
}
int
g_media_gone(struct g_provider *pp, int flag)
{
struct g_consumer *cp;
LIST_FOREACH(cp, &pp->consumers, consumers)
cp->flags |= G_CF_SPOILED;
return (g_post_event(g_spoil_event, pp, flag, pp, NULL));
}
int
g_getattr__(const char *attr, struct g_consumer *cp, void *var, int len)
{
int error, i;
i = len;
error = g_io_getattr(attr, cp, &i, var);
if (error)
return (error);
if (i != len)
return (EINVAL);
return (0);
}
static int
g_get_device_prefix_len(const char *name)
{
int len;
if (strncmp(name, "ada", 3) == 0)
len = 3;
else if (strncmp(name, "ad", 2) == 0)
len = 2;
else
return (0);
if (name[len] < '0' || name[len] > '9')
return (0);
do {
len++;
} while (name[len] >= '0' && name[len] <= '9');
return (len);
}
int
g_compare_names(const char *namea, const char *nameb)
{
int deva, devb;
if (strcmp(namea, nameb) == 0)
return (1);
deva = g_get_device_prefix_len(namea);
if (deva == 0)
return (0);
devb = g_get_device_prefix_len(nameb);
if (devb == 0)
return (0);
if (strcmp(namea + deva, nameb + devb) == 0)
return (1);
return (0);
}
#if defined(DIAGNOSTIC) || defined(DDB)
/*
* This function walks the mesh and returns a non-zero integer if it
* finds the argument pointer is an object. The return value indicates
* which type of object it is believed to be. If topology is not locked,
* this function is potentially dangerous, but we don't assert that the
* topology lock is held when called from debugger.
*/
int
g_valid_obj(void const *ptr)
{
struct g_class *mp;
struct g_geom *gp;
struct g_consumer *cp;
struct g_provider *pp;
#ifdef KDB
if (kdb_active == 0)
#endif
g_topology_assert();
LIST_FOREACH(mp, &g_classes, class) {
if (ptr == mp)
return (1);
LIST_FOREACH(gp, &mp->geom, geom) {
if (ptr == gp)
return (2);
LIST_FOREACH(cp, &gp->consumer, consumer)
if (ptr == cp)
return (3);
LIST_FOREACH(pp, &gp->provider, provider)
if (ptr == pp)
return (4);
}
}
return(0);
}
#endif
#ifdef DDB
#define gprintf(...) do { \
db_printf("%*s", indent, ""); \
db_printf(__VA_ARGS__); \
} while (0)
#define gprintln(...) do { \
gprintf(__VA_ARGS__); \
db_printf("\n"); \
} while (0)
#define ADDFLAG(obj, flag, sflag) do { \
if ((obj)->flags & (flag)) { \
if (comma) \
strlcat(str, ",", size); \
strlcat(str, (sflag), size); \
comma = 1; \
} \
} while (0)
static char *
provider_flags_to_string(struct g_provider *pp, char *str, size_t size)
{
int comma = 0;
bzero(str, size);
if (pp->flags == 0) {
strlcpy(str, "NONE", size);
return (str);
}
ADDFLAG(pp, G_PF_WITHER, "G_PF_WITHER");
ADDFLAG(pp, G_PF_ORPHAN, "G_PF_ORPHAN");
return (str);
}
static char *
geom_flags_to_string(struct g_geom *gp, char *str, size_t size)
{
int comma = 0;
bzero(str, size);
if (gp->flags == 0) {
strlcpy(str, "NONE", size);
return (str);
}
ADDFLAG(gp, G_GEOM_WITHER, "G_GEOM_WITHER");
return (str);
}
static void
db_show_geom_consumer(int indent, struct g_consumer *cp)
{
if (indent == 0) {
gprintln("consumer: %p", cp);
gprintln(" class: %s (%p)", cp->geom->class->name,
cp->geom->class);
gprintln(" geom: %s (%p)", cp->geom->name, cp->geom);
if (cp->provider == NULL)
gprintln(" provider: none");
else {
gprintln(" provider: %s (%p)", cp->provider->name,
cp->provider);
}
gprintln(" access: r%dw%de%d", cp->acr, cp->acw, cp->ace);
gprintln(" flags: 0x%04x", cp->flags);
gprintln(" nstart: %u", cp->nstart);
gprintln(" nend: %u", cp->nend);
} else {
gprintf("consumer: %p (%s), access=r%dw%de%d", cp,
cp->provider != NULL ? cp->provider->name : "none",
cp->acr, cp->acw, cp->ace);
if (cp->flags)
db_printf(", flags=0x%04x", cp->flags);
db_printf("\n");
}
}
static void
db_show_geom_provider(int indent, struct g_provider *pp)
{
struct g_consumer *cp;
char flags[64];
if (indent == 0) {
gprintln("provider: %s (%p)", pp->name, pp);
gprintln(" class: %s (%p)", pp->geom->class->name,
pp->geom->class);
gprintln(" geom: %s (%p)", pp->geom->name, pp->geom);
gprintln(" mediasize: %jd", (intmax_t)pp->mediasize);
gprintln(" sectorsize: %u", pp->sectorsize);
gprintln(" stripesize: %u", pp->stripesize);
gprintln(" stripeoffset: %u", pp->stripeoffset);
gprintln(" access: r%dw%de%d", pp->acr, pp->acw,
pp->ace);
gprintln(" flags: %s (0x%04x)",
provider_flags_to_string(pp, flags, sizeof(flags)),
pp->flags);
gprintln(" error: %d", pp->error);
gprintln(" nstart: %u", pp->nstart);
gprintln(" nend: %u", pp->nend);
if (LIST_EMPTY(&pp->consumers))
gprintln(" consumers: none");
} else {
gprintf("provider: %s (%p), access=r%dw%de%d",
pp->name, pp, pp->acr, pp->acw, pp->ace);
if (pp->flags != 0) {
db_printf(", flags=%s (0x%04x)",
provider_flags_to_string(pp, flags, sizeof(flags)),
pp->flags);
}
db_printf("\n");
}
if (!LIST_EMPTY(&pp->consumers)) {
LIST_FOREACH(cp, &pp->consumers, consumers) {
db_show_geom_consumer(indent + 2, cp);
if (db_pager_quit)
break;
}
}
}
static void
db_show_geom_geom(int indent, struct g_geom *gp)
{
struct g_provider *pp;
struct g_consumer *cp;
char flags[64];
if (indent == 0) {
gprintln("geom: %s (%p)", gp->name, gp);
gprintln(" class: %s (%p)", gp->class->name, gp->class);
gprintln(" flags: %s (0x%04x)",
geom_flags_to_string(gp, flags, sizeof(flags)), gp->flags);
gprintln(" rank: %d", gp->rank);
if (LIST_EMPTY(&gp->provider))
gprintln(" providers: none");
if (LIST_EMPTY(&gp->consumer))
gprintln(" consumers: none");
} else {
gprintf("geom: %s (%p), rank=%d", gp->name, gp, gp->rank);
if (gp->flags != 0) {
db_printf(", flags=%s (0x%04x)",
geom_flags_to_string(gp, flags, sizeof(flags)),
gp->flags);
}
db_printf("\n");
}
if (!LIST_EMPTY(&gp->provider)) {
LIST_FOREACH(pp, &gp->provider, provider) {
db_show_geom_provider(indent + 2, pp);
if (db_pager_quit)
break;
}
}
if (!LIST_EMPTY(&gp->consumer)) {
LIST_FOREACH(cp, &gp->consumer, consumer) {
db_show_geom_consumer(indent + 2, cp);
if (db_pager_quit)
break;
}
}
}
static void
db_show_geom_class(struct g_class *mp)
{
struct g_geom *gp;
db_printf("class: %s (%p)\n", mp->name, mp);
LIST_FOREACH(gp, &mp->geom, geom) {
db_show_geom_geom(2, gp);
if (db_pager_quit)
break;
}
}
/*
* Print the GEOM topology or the given object.
*/
DB_SHOW_COMMAND(geom, db_show_geom)
{
struct g_class *mp;
if (!have_addr) {
/* No address given, print the entire topology. */
LIST_FOREACH(mp, &g_classes, class) {
db_show_geom_class(mp);
db_printf("\n");
if (db_pager_quit)
break;
}
} else {
switch (g_valid_obj((void *)addr)) {
case 1:
db_show_geom_class((struct g_class *)addr);
break;
case 2:
db_show_geom_geom(0, (struct g_geom *)addr);
break;
case 3:
db_show_geom_consumer(0, (struct g_consumer *)addr);
break;
case 4:
db_show_geom_provider(0, (struct g_provider *)addr);
break;
default:
db_printf("Not a GEOM object.\n");
break;
}
}
}
static void
db_print_bio_cmd(struct bio *bp)
{
db_printf(" cmd: ");
switch (bp->bio_cmd) {
case BIO_READ: db_printf("BIO_READ"); break;
case BIO_WRITE: db_printf("BIO_WRITE"); break;
case BIO_DELETE: db_printf("BIO_DELETE"); break;
case BIO_GETATTR: db_printf("BIO_GETATTR"); break;
case BIO_FLUSH: db_printf("BIO_FLUSH"); break;
case BIO_CMD0: db_printf("BIO_CMD0"); break;
case BIO_CMD1: db_printf("BIO_CMD1"); break;
case BIO_CMD2: db_printf("BIO_CMD2"); break;
default: db_printf("UNKNOWN"); break;
}
db_printf("\n");
}
static void
db_print_bio_flags(struct bio *bp)
{
int comma;
comma = 0;
db_printf(" flags: ");
if (bp->bio_flags & BIO_ERROR) {
db_printf("BIO_ERROR");
comma = 1;
}
if (bp->bio_flags & BIO_DONE) {
db_printf("%sBIO_DONE", (comma ? ", " : ""));
comma = 1;
}
if (bp->bio_flags & BIO_ONQUEUE)
db_printf("%sBIO_ONQUEUE", (comma ? ", " : ""));
db_printf("\n");
}
/*
* Print useful information in a BIO
*/
DB_SHOW_COMMAND(bio, db_show_bio)
{
struct bio *bp;
if (have_addr) {
bp = (struct bio *)addr;
db_printf("BIO %p\n", bp);
db_print_bio_cmd(bp);
db_print_bio_flags(bp);
db_printf(" cflags: 0x%hhx\n", bp->bio_cflags);
db_printf(" pflags: 0x%hhx\n", bp->bio_pflags);
db_printf(" offset: %jd\n", (intmax_t)bp->bio_offset);
db_printf(" length: %jd\n", (intmax_t)bp->bio_length);
db_printf(" bcount: %ld\n", bp->bio_bcount);
db_printf(" resid: %ld\n", bp->bio_resid);
db_printf(" completed: %jd\n", (intmax_t)bp->bio_completed);
db_printf(" children: %u\n", bp->bio_children);
db_printf(" inbed: %u\n", bp->bio_inbed);
db_printf(" error: %d\n", bp->bio_error);
db_printf(" parent: %p\n", bp->bio_parent);
db_printf(" driver1: %p\n", bp->bio_driver1);
db_printf(" driver2: %p\n", bp->bio_driver2);
db_printf(" caller1: %p\n", bp->bio_caller1);
db_printf(" caller2: %p\n", bp->bio_caller2);
db_printf(" bio_from: %p\n", bp->bio_from);
db_printf(" bio_to: %p\n", bp->bio_to);
}
}
#undef gprintf
#undef gprintln
#undef ADDFLAG
#endif /* DDB */