freebsd-skq/sys/geom/geom_dev.c
Steven Hartland 6f926c0b82 Added a sysctl (kern.geom.dev.delete_max_sectors) to control the maximum
size of a delete request sent to the providing device performed by g_dev_ioctl.

This allows the kernel and apps via ioctl e.g. newfs -E to request large LBA
deletes which siginificantly improves performance.

Previously this was hard coded to 65536 sectors, the new default is 262144
which doubles the throughput of deletes on commonly available SSD's.

In tests on a Intel 520 120GB FW: 400i disk it improved the delete throughput
from 1.6GB/s to over 2.6GB/s on a full disk delete such as that done via
newfs -E

For some SSD's where delete time is pretty much constant, no matter what
the request, setting this to 0 will provide significantly better throughput
e.g. Samsung 840 240GB FW DXT07B0Q @ 262144 = 79G/s, @ 0 = 2259G/s

Reviewed by:	mav
Approved by:	pjd (mentor)
MFC after:	2 weeks
2013-04-26 15:43:24 +00:00

623 lines
16 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 <sys/param.h>
#include <sys/systm.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
#include <sys/conf.h>
#include <sys/ctype.h>
#include <sys/bio.h>
#include <sys/bus.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/errno.h>
#include <sys/time.h>
#include <sys/disk.h>
#include <sys/fcntl.h>
#include <sys/limits.h>
#include <sys/sysctl.h>
#include <geom/geom.h>
#include <geom/geom_int.h>
#include <machine/stdarg.h>
struct g_dev_softc {
struct mtx sc_mtx;
struct cdev *sc_dev;
struct cdev *sc_alias;
int sc_open;
int sc_active;
};
static d_open_t g_dev_open;
static d_close_t g_dev_close;
static d_strategy_t g_dev_strategy;
static d_ioctl_t g_dev_ioctl;
static struct cdevsw g_dev_cdevsw = {
.d_version = D_VERSION,
.d_open = g_dev_open,
.d_close = g_dev_close,
.d_read = physread,
.d_write = physwrite,
.d_ioctl = g_dev_ioctl,
.d_strategy = g_dev_strategy,
.d_name = "g_dev",
.d_flags = D_DISK | D_TRACKCLOSE | D_UNMAPPED_IO,
};
static g_taste_t g_dev_taste;
static g_orphan_t g_dev_orphan;
static g_attrchanged_t g_dev_attrchanged;
static struct g_class g_dev_class = {
.name = "DEV",
.version = G_VERSION,
.taste = g_dev_taste,
.orphan = g_dev_orphan,
.attrchanged = g_dev_attrchanged
};
/*
* We target 262144 (8 x 32768) sectors by default as this significantly
* increases the throughput on commonly used SSD's with a marginal
* increase in non-interruptible request latency.
*/
static uint64_t g_dev_del_max_sectors = 262144;
SYSCTL_DECL(_kern_geom);
SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
&g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
"delete request sent to the provider. Larger requests are chunked "
"so they can be interrupted. (0 = disable chunking)");
static void
g_dev_destroy(void *arg, int flags __unused)
{
struct g_consumer *cp;
struct g_geom *gp;
struct g_dev_softc *sc;
g_topology_assert();
cp = arg;
gp = cp->geom;
sc = cp->private;
g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
g_access(cp, -cp->acr, -cp->acw, -cp->ace);
g_detach(cp);
g_destroy_consumer(cp);
g_destroy_geom(gp);
mtx_destroy(&sc->sc_mtx);
g_free(sc);
}
void
g_dev_print(void)
{
struct g_geom *gp;
char const *p = "";
LIST_FOREACH(gp, &g_dev_class.geom, geom) {
printf("%s%s", p, gp->name);
p = " ";
}
printf("\n");
}
static void
g_dev_attrchanged(struct g_consumer *cp, const char *attr)
{
struct g_dev_softc *sc;
struct cdev *dev;
char buf[SPECNAMELEN + 6];
sc = cp->private;
if (strcmp(attr, "GEOM::media") == 0) {
dev = sc->sc_dev;
snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
dev = sc->sc_alias;
if (dev != NULL) {
snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
M_WAITOK);
}
return;
}
if (strcmp(attr, "GEOM::physpath") != 0)
return;
if (g_access(cp, 1, 0, 0) == 0) {
char *physpath;
int error, physpath_len;
physpath_len = MAXPATHLEN;
physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
error =
g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
g_access(cp, -1, 0, 0);
if (error == 0 && strlen(physpath) != 0) {
struct cdev *old_alias_dev;
struct cdev **alias_devp;
dev = sc->sc_dev;
old_alias_dev = sc->sc_alias;
alias_devp = (struct cdev **)&sc->sc_alias;
make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp,
dev, old_alias_dev, physpath);
} else if (sc->sc_alias) {
destroy_dev((struct cdev *)sc->sc_alias);
sc->sc_alias = NULL;
}
g_free(physpath);
}
}
struct g_provider *
g_dev_getprovider(struct cdev *dev)
{
struct g_consumer *cp;
g_topology_assert();
if (dev == NULL)
return (NULL);
if (dev->si_devsw != &g_dev_cdevsw)
return (NULL);
cp = dev->si_drv2;
return (cp->provider);
}
static struct g_geom *
g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
{
struct g_geom *gp;
struct g_consumer *cp;
struct g_dev_softc *sc;
int error, len;
struct cdev *dev, *adev;
char buf[64], *val;
g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
g_topology_assert();
gp = g_new_geomf(mp, "%s", pp->name);
sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
cp = g_new_consumer(gp);
cp->private = sc;
error = g_attach(cp, pp);
KASSERT(error == 0,
("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
&g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
if (error != 0) {
printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
__func__, gp->name, error);
g_detach(cp);
g_destroy_consumer(cp);
g_destroy_geom(gp);
mtx_destroy(&sc->sc_mtx);
g_free(sc);
return (NULL);
}
sc->sc_dev = dev;
/* Search for device alias name and create it if found. */
adev = NULL;
for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
buf[14 + len] = 0;
val = getenv(buf);
if (val != NULL) {
snprintf(buf, sizeof(buf), "%s%s",
val, gp->name + len);
freeenv(val);
make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
&adev, dev, "%s", buf);
break;
}
}
dev->si_iosize_max = MAXPHYS;
dev->si_drv2 = cp;
if (adev != NULL) {
adev->si_iosize_max = MAXPHYS;
adev->si_drv2 = cp;
}
g_dev_attrchanged(cp, "GEOM::physpath");
return (gp);
}
static int
g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
{
struct g_consumer *cp;
struct g_dev_softc *sc;
int error, r, w, e;
cp = dev->si_drv2;
if (cp == NULL)
return(ENXIO); /* g_dev_taste() not done yet */
g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
cp->geom->name, flags, fmt, td);
r = flags & FREAD ? 1 : 0;
w = flags & FWRITE ? 1 : 0;
#ifdef notyet
e = flags & O_EXCL ? 1 : 0;
#else
e = 0;
#endif
if (w) {
/*
* When running in very secure mode, do not allow
* opens for writing of any disks.
*/
error = securelevel_ge(td->td_ucred, 2);
if (error)
return (error);
}
g_topology_lock();
error = g_access(cp, r, w, e);
g_topology_unlock();
if (error == 0) {
sc = cp->private;
mtx_lock(&sc->sc_mtx);
if (sc->sc_open == 0 && sc->sc_active != 0)
wakeup(&sc->sc_active);
sc->sc_open += r + w + e;
mtx_unlock(&sc->sc_mtx);
}
return(error);
}
static int
g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
{
struct g_consumer *cp;
struct g_dev_softc *sc;
int error, r, w, e;
cp = dev->si_drv2;
if (cp == NULL)
return(ENXIO);
g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
cp->geom->name, flags, fmt, td);
r = flags & FREAD ? -1 : 0;
w = flags & FWRITE ? -1 : 0;
#ifdef notyet
e = flags & O_EXCL ? -1 : 0;
#else
e = 0;
#endif
sc = cp->private;
mtx_lock(&sc->sc_mtx);
sc->sc_open += r + w + e;
while (sc->sc_open == 0 && sc->sc_active != 0)
msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
mtx_unlock(&sc->sc_mtx);
g_topology_lock();
error = g_access(cp, r, w, e);
g_topology_unlock();
return (error);
}
/*
* XXX: Until we have unmessed the ioctl situation, there is a race against
* XXX: a concurrent orphanization. We cannot close it by holding topology
* XXX: since that would prevent us from doing our job, and stalling events
* XXX: will break (actually: stall) the BSD disklabel hacks.
*/
static int
g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
{
struct g_consumer *cp;
struct g_provider *pp;
struct g_kerneldump kd;
off_t offset, length, chunk;
int i, error;
u_int u;
cp = dev->si_drv2;
pp = cp->provider;
error = 0;
KASSERT(cp->acr || cp->acw,
("Consumer with zero access count in g_dev_ioctl"));
i = IOCPARM_LEN(cmd);
switch (cmd) {
case DIOCGSECTORSIZE:
*(u_int *)data = cp->provider->sectorsize;
if (*(u_int *)data == 0)
error = ENOENT;
break;
case DIOCGMEDIASIZE:
*(off_t *)data = cp->provider->mediasize;
if (*(off_t *)data == 0)
error = ENOENT;
break;
case DIOCGFWSECTORS:
error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
if (error == 0 && *(u_int *)data == 0)
error = ENOENT;
break;
case DIOCGFWHEADS:
error = g_io_getattr("GEOM::fwheads", cp, &i, data);
if (error == 0 && *(u_int *)data == 0)
error = ENOENT;
break;
case DIOCGFRONTSTUFF:
error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
break;
case DIOCSKERNELDUMP:
u = *((u_int *)data);
if (!u) {
set_dumper(NULL, NULL);
error = 0;
break;
}
kd.offset = 0;
kd.length = OFF_MAX;
i = sizeof kd;
error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
if (!error) {
error = set_dumper(&kd.di, devtoname(dev));
if (!error)
dev->si_flags |= SI_DUMPDEV;
}
break;
case DIOCGFLUSH:
error = g_io_flush(cp);
break;
case DIOCGDELETE:
offset = ((off_t *)data)[0];
length = ((off_t *)data)[1];
if ((offset % cp->provider->sectorsize) != 0 ||
(length % cp->provider->sectorsize) != 0 || length <= 0) {
printf("%s: offset=%jd length=%jd\n", __func__, offset,
length);
error = EINVAL;
break;
}
while (length > 0) {
chunk = length;
if (g_dev_del_max_sectors != 0 && chunk >
g_dev_del_max_sectors * cp->provider->sectorsize) {
chunk = g_dev_del_max_sectors *
cp->provider->sectorsize;
}
error = g_delete_data(cp, offset, chunk);
length -= chunk;
offset += chunk;
if (error)
break;
/*
* Since the request size can be large, the service
* time can be is likewise. We make this ioctl
* interruptible by checking for signals for each bio.
*/
if (SIGPENDING(td))
break;
}
break;
case DIOCGIDENT:
error = g_io_getattr("GEOM::ident", cp, &i, data);
break;
case DIOCGPROVIDERNAME:
if (pp == NULL)
return (ENOENT);
strlcpy(data, pp->name, i);
break;
case DIOCGSTRIPESIZE:
*(off_t *)data = cp->provider->stripesize;
break;
case DIOCGSTRIPEOFFSET:
*(off_t *)data = cp->provider->stripeoffset;
break;
case DIOCGPHYSPATH:
error = g_io_getattr("GEOM::physpath", cp, &i, data);
if (error == 0 && *(char *)data == '\0')
error = ENOENT;
break;
default:
if (cp->provider->geom->ioctl != NULL) {
error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
} else {
error = ENOIOCTL;
}
}
return (error);
}
static void
g_dev_done(struct bio *bp2)
{
struct g_consumer *cp;
struct g_dev_softc *sc;
struct bio *bp;
int destroy;
cp = bp2->bio_from;
sc = cp->private;
bp = bp2->bio_parent;
bp->bio_error = bp2->bio_error;
if (bp->bio_error != 0) {
g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
bp2, bp->bio_error);
bp->bio_flags |= BIO_ERROR;
} else {
g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed);
}
bp->bio_resid = bp->bio_length - bp2->bio_completed;
bp->bio_completed = bp2->bio_completed;
g_destroy_bio(bp2);
destroy = 0;
mtx_lock(&sc->sc_mtx);
if ((--sc->sc_active) == 0) {
if (sc->sc_open == 0)
wakeup(&sc->sc_active);
if (sc->sc_dev == NULL)
destroy = 1;
}
mtx_unlock(&sc->sc_mtx);
if (destroy)
g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
biodone(bp);
}
static void
g_dev_strategy(struct bio *bp)
{
struct g_consumer *cp;
struct bio *bp2;
struct cdev *dev;
struct g_dev_softc *sc;
KASSERT(bp->bio_cmd == BIO_READ ||
bp->bio_cmd == BIO_WRITE ||
bp->bio_cmd == BIO_DELETE ||
bp->bio_cmd == BIO_FLUSH,
("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
dev = bp->bio_dev;
cp = dev->si_drv2;
sc = cp->private;
KASSERT(cp->acr || cp->acw,
("Consumer with zero access count in g_dev_strategy"));
#ifdef INVARIANTS
if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
(bp->bio_bcount % cp->provider->sectorsize) != 0) {
bp->bio_resid = bp->bio_bcount;
biofinish(bp, NULL, EINVAL);
return;
}
#endif
mtx_lock(&sc->sc_mtx);
KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
sc->sc_active++;
mtx_unlock(&sc->sc_mtx);
for (;;) {
/*
* XXX: This is not an ideal solution, but I belive it to
* XXX: deadlock safe, all things considered.
*/
bp2 = g_clone_bio(bp);
if (bp2 != NULL)
break;
pause("gdstrat", hz / 10);
}
KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
bp2->bio_done = g_dev_done;
g_trace(G_T_BIO,
"g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
bp2->bio_data, bp2->bio_cmd);
g_io_request(bp2, cp);
KASSERT(cp->acr || cp->acw,
("g_dev_strategy raced with g_dev_close and lost"));
}
/*
* g_dev_callback()
*
* Called by devfs when asynchronous device destruction is completed.
* - Mark that we have no attached device any more.
* - If there are no outstanding requests, schedule geom destruction.
* Otherwise destruction will be scheduled later by g_dev_done().
*/
static void
g_dev_callback(void *arg)
{
struct g_consumer *cp;
struct g_dev_softc *sc;
int destroy;
cp = arg;
sc = cp->private;
g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
mtx_lock(&sc->sc_mtx);
sc->sc_dev = NULL;
sc->sc_alias = NULL;
destroy = (sc->sc_active == 0);
mtx_unlock(&sc->sc_mtx);
if (destroy)
g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
}
/*
* g_dev_orphan()
*
* Called from below when the provider orphaned us.
* - Clear any dump settings.
* - Request asynchronous device destruction to prevent any more requests
* from coming in. The provider is already marked with an error, so
* anything which comes in in the interrim will be returned immediately.
*/
static void
g_dev_orphan(struct g_consumer *cp)
{
struct cdev *dev;
struct g_dev_softc *sc;
g_topology_assert();
sc = cp->private;
dev = sc->sc_dev;
g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
/* Reset any dump-area set on this device */
if (dev->si_flags & SI_DUMPDEV)
set_dumper(NULL, NULL);
/* Destroy the struct cdev *so we get no more requests */
destroy_dev_sched_cb(dev, g_dev_callback, cp);
}
DECLARE_GEOM_CLASS(g_dev_class, g_dev);