2002-03-11 21:42:35 +00:00
|
|
|
/*-
|
|
|
|
|
* 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.
|
|
|
|
|
*/
|
|
|
|
|
|
2003-06-11 06:49:16 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
|
2002-10-05 16:35:33 +00:00
|
|
|
#include "opt_geom.h"
|
|
|
|
|
|
2002-03-11 21:42:35 +00:00
|
|
|
#include <sys/param.h>
|
|
|
|
|
#include <sys/systm.h>
|
|
|
|
|
#include <sys/kernel.h>
|
|
|
|
|
#include <sys/sysctl.h>
|
|
|
|
|
#include <sys/bio.h>
|
2009-09-04 09:39:06 +00:00
|
|
|
#include <sys/ctype.h>
|
2003-01-30 20:34:23 +00:00
|
|
|
#include <sys/fcntl.h>
|
2002-03-11 21:42:35 +00:00
|
|
|
#include <sys/malloc.h>
|
2011-07-11 05:22:31 +00:00
|
|
|
#include <sys/sbuf.h>
|
2002-03-11 21:42:35 +00:00
|
|
|
#include <sys/sysctl.h>
|
2003-03-08 08:01:31 +00:00
|
|
|
#include <sys/devicestat.h>
|
2002-03-11 21:42:35 +00:00
|
|
|
#include <machine/md_var.h>
|
|
|
|
|
|
|
|
|
|
#include <sys/lock.h>
|
|
|
|
|
#include <sys/mutex.h>
|
|
|
|
|
#include <geom/geom.h>
|
2003-04-01 18:57:14 +00:00
|
|
|
#include <geom/geom_disk.h>
|
2003-03-21 22:05:33 +00:00
|
|
|
#include <geom/geom_int.h>
|
2002-03-11 21:42:35 +00:00
|
|
|
|
2011-03-24 19:23:42 +00:00
|
|
|
#include <dev/led/led.h>
|
|
|
|
|
|
|
|
|
|
struct g_disk_softc {
|
2013-03-25 05:45:24 +00:00
|
|
|
struct mtx done_mtx;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct disk *dp;
|
|
|
|
|
struct sysctl_ctx_list sysctl_ctx;
|
|
|
|
|
struct sysctl_oid *sysctl_tree;
|
|
|
|
|
char led[64];
|
|
|
|
|
uint32_t state;
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
struct mtx start_mtx;
|
2011-03-24 19:23:42 +00:00
|
|
|
};
|
|
|
|
|
|
2002-03-11 21:42:35 +00:00
|
|
|
static g_access_t g_disk_access;
|
2004-08-08 06:49:07 +00:00
|
|
|
static g_start_t g_disk_start;
|
|
|
|
|
static g_ioctl_t g_disk_ioctl;
|
|
|
|
|
static g_dumpconf_t g_disk_dumpconf;
|
Fix a bug which causes a panic in daopen(). The panic is caused by
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
2012-06-24 04:29:03 +00:00
|
|
|
static g_provgone_t g_disk_providergone;
|
2002-03-11 21:42:35 +00:00
|
|
|
|
2005-02-10 12:10:35 +00:00
|
|
|
static struct g_class g_disk_class = {
|
2013-04-15 15:55:40 +00:00
|
|
|
.name = G_DISK_CLASS_NAME,
|
2004-08-08 07:57:53 +00:00
|
|
|
.version = G_VERSION,
|
2004-08-08 06:49:07 +00:00
|
|
|
.start = g_disk_start,
|
|
|
|
|
.access = g_disk_access,
|
|
|
|
|
.ioctl = g_disk_ioctl,
|
Fix a bug which causes a panic in daopen(). The panic is caused by
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
2012-06-24 04:29:03 +00:00
|
|
|
.providergone = g_disk_providergone,
|
2004-08-08 06:49:07 +00:00
|
|
|
.dumpconf = g_disk_dumpconf,
|
2002-03-11 21:42:35 +00:00
|
|
|
};
|
|
|
|
|
|
2011-03-24 19:23:42 +00:00
|
|
|
SYSCTL_DECL(_kern_geom);
|
2011-11-07 15:43:11 +00:00
|
|
|
static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0,
|
|
|
|
|
"GEOM_DISK stuff");
|
2011-03-24 19:23:42 +00:00
|
|
|
|
2003-05-31 18:13:07 +00:00
|
|
|
DECLARE_GEOM_CLASS(g_disk_class, g_disk);
|
2002-10-11 20:52:44 +00:00
|
|
|
|
2002-03-11 21:42:35 +00:00
|
|
|
static int
|
|
|
|
|
g_disk_access(struct g_provider *pp, int r, int w, int e)
|
|
|
|
|
{
|
|
|
|
|
struct disk *dp;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
2002-03-11 21:42:35 +00:00
|
|
|
int error;
|
|
|
|
|
|
|
|
|
|
g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
|
|
|
|
|
pp->name, r, w, e);
|
|
|
|
|
g_topology_assert();
|
2013-03-25 05:45:24 +00:00
|
|
|
sc = pp->private;
|
2011-03-24 19:23:42 +00:00
|
|
|
if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
|
2004-02-06 23:10:49 +00:00
|
|
|
/*
|
|
|
|
|
* Allow decreasing access count even if disk is not
|
|
|
|
|
* avaliable anymore.
|
|
|
|
|
*/
|
|
|
|
|
if (r <= 0 && w <= 0 && e <= 0)
|
|
|
|
|
return (0);
|
2004-02-18 21:36:53 +00:00
|
|
|
return (ENXIO);
|
2004-02-06 23:10:49 +00:00
|
|
|
}
|
2002-03-11 21:42:35 +00:00
|
|
|
r += pp->acr;
|
|
|
|
|
w += pp->acw;
|
|
|
|
|
e += pp->ace;
|
2003-01-30 20:34:23 +00:00
|
|
|
error = 0;
|
2002-03-11 21:42:35 +00:00
|
|
|
if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
|
2003-02-28 10:02:02 +00:00
|
|
|
if (dp->d_open != NULL) {
|
|
|
|
|
error = dp->d_open(dp);
|
2007-02-21 07:45:02 +00:00
|
|
|
if (bootverbose && error != 0)
|
2003-01-30 20:34:23 +00:00
|
|
|
printf("Opened disk %s -> %d\n",
|
|
|
|
|
pp->name, error);
|
2013-06-11 10:06:07 +00:00
|
|
|
if (error != 0)
|
|
|
|
|
return (error);
|
2003-01-30 20:34:23 +00:00
|
|
|
}
|
2002-10-20 20:28:24 +00:00
|
|
|
pp->mediasize = dp->d_mediasize;
|
|
|
|
|
pp->sectorsize = dp->d_sectorsize;
|
2003-02-11 21:23:34 +00:00
|
|
|
if (dp->d_maxsize == 0) {
|
|
|
|
|
printf("WARNING: Disk drive %s%d has no d_maxsize\n",
|
|
|
|
|
dp->d_name, dp->d_unit);
|
|
|
|
|
dp->d_maxsize = DFLTPHYS;
|
|
|
|
|
}
|
2013-07-03 23:46:30 +00:00
|
|
|
if (dp->d_delmaxsize == 0) {
|
|
|
|
|
if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
|
|
|
|
|
printf("WARNING: Disk drive %s%d has no "
|
|
|
|
|
"d_delmaxsize\n", dp->d_name, dp->d_unit);
|
2013-04-26 16:22:54 +00:00
|
|
|
}
|
2013-07-03 23:46:30 +00:00
|
|
|
dp->d_delmaxsize = dp->d_maxsize;
|
2013-04-26 16:22:54 +00:00
|
|
|
}
|
|
|
|
|
pp->stripeoffset = dp->d_stripeoffset;
|
|
|
|
|
pp->stripesize = dp->d_stripesize;
|
|
|
|
|
dp->d_flags |= DISKFLAG_OPEN;
|
2002-03-11 21:42:35 +00:00
|
|
|
} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
|
2003-02-28 10:02:02 +00:00
|
|
|
if (dp->d_close != NULL) {
|
|
|
|
|
error = dp->d_close(dp);
|
2003-01-30 20:34:23 +00:00
|
|
|
if (error != 0)
|
|
|
|
|
printf("Closed disk %s -> %d\n",
|
|
|
|
|
pp->name, error);
|
|
|
|
|
}
|
2011-03-24 19:23:42 +00:00
|
|
|
sc->state = G_STATE_ACTIVE;
|
|
|
|
|
if (sc->led[0] != 0)
|
|
|
|
|
led_set(sc->led, "0");
|
2003-01-30 20:34:23 +00:00
|
|
|
dp->d_flags &= ~DISKFLAG_OPEN;
|
2002-03-11 21:42:35 +00:00
|
|
|
}
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
2002-04-19 09:24:12 +00:00
|
|
|
static void
|
|
|
|
|
g_disk_kerneldump(struct bio *bp, struct disk *dp)
|
2011-03-24 08:37:48 +00:00
|
|
|
{
|
2002-04-19 09:24:12 +00:00
|
|
|
struct g_kerneldump *gkd;
|
2002-10-04 10:06:19 +00:00
|
|
|
struct g_geom *gp;
|
2002-04-19 09:24:12 +00:00
|
|
|
|
|
|
|
|
gkd = (struct g_kerneldump*)bp->bio_data;
|
2002-10-04 10:06:19 +00:00
|
|
|
gp = bp->bio_to->geom;
|
|
|
|
|
g_trace(G_T_TOPOLOGY, "g_disk_kernedump(%s, %jd, %jd)",
|
|
|
|
|
gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
|
2003-09-29 07:44:23 +00:00
|
|
|
if (dp->d_dump == NULL) {
|
|
|
|
|
g_io_deliver(bp, ENODEV);
|
|
|
|
|
return;
|
|
|
|
|
}
|
2011-03-24 08:37:48 +00:00
|
|
|
gkd->di.dumper = dp->d_dump;
|
|
|
|
|
gkd->di.priv = dp;
|
|
|
|
|
gkd->di.blocksize = dp->d_sectorsize;
|
|
|
|
|
gkd->di.maxiosize = dp->d_maxsize;
|
|
|
|
|
gkd->di.mediaoffset = gkd->offset;
|
2005-01-29 16:49:43 +00:00
|
|
|
if ((gkd->offset + gkd->length) > dp->d_mediasize)
|
|
|
|
|
gkd->length = dp->d_mediasize - gkd->offset;
|
2011-03-24 08:37:48 +00:00
|
|
|
gkd->di.mediasize = gkd->length;
|
|
|
|
|
g_io_deliver(bp, 0);
|
2002-04-19 09:24:12 +00:00
|
|
|
}
|
|
|
|
|
|
2011-03-24 19:23:42 +00:00
|
|
|
static void
|
|
|
|
|
g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
|
|
|
|
|
{
|
|
|
|
|
const char *cmd;
|
|
|
|
|
|
|
|
|
|
memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
|
|
|
|
|
if (sc->led[0] != 0) {
|
|
|
|
|
switch (sc->state) {
|
|
|
|
|
case G_STATE_FAILED:
|
|
|
|
|
cmd = "1";
|
|
|
|
|
break;
|
|
|
|
|
case G_STATE_REBUILD:
|
|
|
|
|
cmd = "f5";
|
|
|
|
|
break;
|
|
|
|
|
case G_STATE_RESYNC:
|
|
|
|
|
cmd = "f1";
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
cmd = "0";
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
led_set(sc->led, cmd);
|
|
|
|
|
}
|
|
|
|
|
g_io_deliver(bp, 0);
|
|
|
|
|
}
|
|
|
|
|
|
2002-03-11 21:42:35 +00:00
|
|
|
static void
|
|
|
|
|
g_disk_done(struct bio *bp)
|
|
|
|
|
{
|
2013-10-16 09:12:40 +00:00
|
|
|
struct bintime now;
|
2003-03-08 08:01:31 +00:00
|
|
|
struct bio *bp2;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
2002-03-11 21:42:35 +00:00
|
|
|
|
2003-02-11 18:32:31 +00:00
|
|
|
/* See "notes" for why we need a mutex here */
|
|
|
|
|
/* XXX: will witness accept a mix of Giant/unGiant drivers here ? */
|
2003-03-08 08:01:31 +00:00
|
|
|
bp2 = bp->bio_parent;
|
2013-03-25 05:45:24 +00:00
|
|
|
sc = bp2->bio_to->private;
|
|
|
|
|
bp->bio_completed = bp->bio_length - bp->bio_resid;
|
2013-10-16 09:12:40 +00:00
|
|
|
binuptime(&now);
|
2013-03-25 05:45:24 +00:00
|
|
|
mtx_lock(&sc->done_mtx);
|
2003-03-08 08:01:31 +00:00
|
|
|
if (bp2->bio_error == 0)
|
|
|
|
|
bp2->bio_error = bp->bio_error;
|
|
|
|
|
bp2->bio_completed += bp->bio_completed;
|
2013-03-25 05:45:24 +00:00
|
|
|
if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) != 0)
|
2013-10-16 09:12:40 +00:00
|
|
|
devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
|
2003-03-08 08:01:31 +00:00
|
|
|
bp2->bio_inbed++;
|
|
|
|
|
if (bp2->bio_children == bp2->bio_inbed) {
|
2013-10-16 09:18:01 +00:00
|
|
|
mtx_unlock(&sc->done_mtx);
|
2003-03-15 10:49:26 +00:00
|
|
|
bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
|
2003-03-08 08:01:31 +00:00
|
|
|
g_io_deliver(bp2, bp2->bio_error);
|
2013-10-16 09:18:01 +00:00
|
|
|
} else
|
|
|
|
|
mtx_unlock(&sc->done_mtx);
|
|
|
|
|
g_destroy_bio(bp);
|
2002-03-11 21:42:35 +00:00
|
|
|
}
|
|
|
|
|
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
static void
|
|
|
|
|
g_disk_done_single(struct bio *bp)
|
|
|
|
|
{
|
|
|
|
|
struct bintime now;
|
|
|
|
|
struct g_disk_softc *sc;
|
|
|
|
|
|
|
|
|
|
bp->bio_completed = bp->bio_length - bp->bio_resid;
|
|
|
|
|
bp->bio_done = (void *)bp->bio_to;
|
|
|
|
|
bp->bio_to = LIST_FIRST(&bp->bio_disk->d_geom->provider);
|
|
|
|
|
if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) != 0) {
|
|
|
|
|
binuptime(&now);
|
|
|
|
|
sc = bp->bio_to->private;
|
|
|
|
|
mtx_lock(&sc->done_mtx);
|
|
|
|
|
devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
|
|
|
|
|
mtx_unlock(&sc->done_mtx);
|
|
|
|
|
}
|
|
|
|
|
g_io_deliver(bp, bp->bio_error);
|
|
|
|
|
}
|
|
|
|
|
|
2003-09-01 20:45:32 +00:00
|
|
|
static int
|
2004-12-12 10:09:05 +00:00
|
|
|
g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
|
2003-09-01 20:45:32 +00:00
|
|
|
{
|
|
|
|
|
struct disk *dp;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
2003-09-01 20:45:32 +00:00
|
|
|
int error;
|
|
|
|
|
|
2013-03-25 05:45:24 +00:00
|
|
|
sc = pp->private;
|
2011-03-24 19:23:42 +00:00
|
|
|
dp = sc->dp;
|
2003-09-01 20:45:32 +00:00
|
|
|
|
|
|
|
|
if (dp->d_ioctl == NULL)
|
|
|
|
|
return (ENOIOCTL);
|
2004-12-12 10:09:05 +00:00
|
|
|
error = dp->d_ioctl(dp, cmd, data, fflag, td);
|
2011-10-25 14:05:39 +00:00
|
|
|
return (error);
|
2003-09-01 20:45:32 +00:00
|
|
|
}
|
|
|
|
|
|
2002-03-11 21:42:35 +00:00
|
|
|
static void
|
|
|
|
|
g_disk_start(struct bio *bp)
|
|
|
|
|
{
|
2003-02-11 18:32:31 +00:00
|
|
|
struct bio *bp2, *bp3;
|
2002-03-11 21:42:35 +00:00
|
|
|
struct disk *dp;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
2002-03-16 09:24:19 +00:00
|
|
|
int error;
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
off_t d_maxsize, off;
|
2002-03-11 21:42:35 +00:00
|
|
|
|
2013-03-25 05:45:24 +00:00
|
|
|
sc = bp->bio_to->private;
|
2011-03-24 19:23:42 +00:00
|
|
|
if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
|
2003-05-21 18:52:29 +00:00
|
|
|
g_io_deliver(bp, ENXIO);
|
2005-03-18 07:01:31 +00:00
|
|
|
return;
|
|
|
|
|
}
|
2002-10-07 07:15:37 +00:00
|
|
|
error = EJUSTRETURN;
|
2002-03-11 21:42:35 +00:00
|
|
|
switch(bp->bio_cmd) {
|
2002-10-07 07:15:37 +00:00
|
|
|
case BIO_DELETE:
|
2003-01-30 20:34:23 +00:00
|
|
|
if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
|
2011-10-25 14:07:17 +00:00
|
|
|
error = EOPNOTSUPP;
|
2002-10-07 07:15:37 +00:00
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
/* fall-through */
|
2002-03-11 21:42:35 +00:00
|
|
|
case BIO_READ:
|
|
|
|
|
case BIO_WRITE:
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
d_maxsize = (bp->bio_cmd == BIO_DELETE) ?
|
|
|
|
|
dp->d_delmaxsize : dp->d_maxsize;
|
|
|
|
|
if (bp->bio_length <= d_maxsize) {
|
|
|
|
|
bp->bio_disk = dp;
|
|
|
|
|
bp->bio_to = (void *)bp->bio_done;
|
|
|
|
|
bp->bio_done = g_disk_done_single;
|
|
|
|
|
bp->bio_pblkno = bp->bio_offset / dp->d_sectorsize;
|
|
|
|
|
bp->bio_bcount = bp->bio_length;
|
|
|
|
|
mtx_lock(&sc->start_mtx);
|
|
|
|
|
devstat_start_transaction_bio(dp->d_devstat, bp);
|
|
|
|
|
mtx_unlock(&sc->start_mtx);
|
|
|
|
|
dp->d_strategy(bp);
|
|
|
|
|
break;
|
|
|
|
|
}
|
2003-02-11 18:32:31 +00:00
|
|
|
off = 0;
|
|
|
|
|
bp3 = NULL;
|
2002-03-11 21:42:35 +00:00
|
|
|
bp2 = g_clone_bio(bp);
|
2003-02-06 22:00:47 +00:00
|
|
|
if (bp2 == NULL) {
|
|
|
|
|
error = ENOMEM;
|
|
|
|
|
break;
|
|
|
|
|
}
|
2003-02-11 18:32:31 +00:00
|
|
|
do {
|
|
|
|
|
bp2->bio_offset += off;
|
|
|
|
|
bp2->bio_length -= off;
|
2013-03-19 14:49:15 +00:00
|
|
|
if ((bp->bio_flags & BIO_UNMAPPED) == 0) {
|
|
|
|
|
bp2->bio_data += off;
|
|
|
|
|
} else {
|
|
|
|
|
KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO)
|
|
|
|
|
!= 0,
|
|
|
|
|
("unmapped bio not supported by disk %s",
|
|
|
|
|
dp->d_name));
|
|
|
|
|
bp2->bio_ma += off / PAGE_SIZE;
|
|
|
|
|
bp2->bio_ma_offset += off;
|
|
|
|
|
bp2->bio_ma_offset %= PAGE_SIZE;
|
|
|
|
|
bp2->bio_ma_n -= off / PAGE_SIZE;
|
|
|
|
|
}
|
2013-04-26 16:22:54 +00:00
|
|
|
if (bp2->bio_length > d_maxsize) {
|
2003-02-11 18:32:31 +00:00
|
|
|
/*
|
|
|
|
|
* XXX: If we have a stripesize we should really
|
2013-04-26 16:22:54 +00:00
|
|
|
* use it here. Care should be taken in the delete
|
|
|
|
|
* case if this is done as deletes can be very
|
|
|
|
|
* sensitive to size given how they are processed.
|
2003-02-11 18:32:31 +00:00
|
|
|
*/
|
2013-04-26 16:22:54 +00:00
|
|
|
bp2->bio_length = d_maxsize;
|
2013-03-19 14:49:15 +00:00
|
|
|
if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
|
|
|
|
|
bp2->bio_ma_n = howmany(
|
|
|
|
|
bp2->bio_ma_offset +
|
|
|
|
|
bp2->bio_length, PAGE_SIZE);
|
|
|
|
|
}
|
2013-04-26 16:22:54 +00:00
|
|
|
off += d_maxsize;
|
2003-02-11 18:32:31 +00:00
|
|
|
/*
|
|
|
|
|
* To avoid a race, we need to grab the next bio
|
|
|
|
|
* before we schedule this one. See "notes".
|
|
|
|
|
*/
|
|
|
|
|
bp3 = g_clone_bio(bp);
|
|
|
|
|
if (bp3 == NULL)
|
|
|
|
|
bp->bio_error = ENOMEM;
|
|
|
|
|
}
|
|
|
|
|
bp2->bio_done = g_disk_done;
|
|
|
|
|
bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
|
|
|
|
|
bp2->bio_bcount = bp2->bio_length;
|
|
|
|
|
bp2->bio_disk = dp;
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
mtx_lock(&sc->start_mtx);
|
2005-09-30 17:32:08 +00:00
|
|
|
devstat_start_transaction_bio(dp->d_devstat, bp2);
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
mtx_unlock(&sc->start_mtx);
|
2003-02-11 18:32:31 +00:00
|
|
|
dp->d_strategy(bp2);
|
|
|
|
|
bp2 = bp3;
|
|
|
|
|
bp3 = NULL;
|
|
|
|
|
} while (bp2 != NULL);
|
2002-03-11 21:42:35 +00:00
|
|
|
break;
|
|
|
|
|
case BIO_GETATTR:
|
Plumb device physical path reporting from CAM devices, through GEOM and
DEVFS, and make it accessible via the diskinfo utility.
Extend GEOM's generic attribute query mechanism into generic disk consumers.
sys/geom/geom_disk.c:
sys/geom/geom_disk.h:
sys/cam/scsi/scsi_da.c:
sys/cam/ata/ata_da.c:
- Allow disk providers to implement a new method which can override
the default BIO_GETATTR response, d_getattr(struct bio *). This
function returns -1 if not handled, otherwise it returns 0 or an
errno to be passed to g_io_deliver().
sys/cam/scsi/scsi_da.c:
sys/cam/ata/ata_da.c:
- Don't copy the serial number to dp->d_ident anymore, as the CAM XPT
is now responsible for returning this information via
d_getattr()->(a)dagetattr()->xpt_getatr().
sys/geom/geom_dev.c:
- Implement a new ioctl, DIOCGPHYSPATH, which returns the GEOM
attribute "GEOM::physpath", if possible. If the attribute request
returns a zero-length string, ENOENT is returned.
usr.sbin/diskinfo/diskinfo.c:
- If the DIOCGPHYSPATH ioctl is successful, report physical path
data when diskinfo is executed with the '-v' option.
Submitted by: will
Reviewed by: gibbs
Sponsored by: Spectra Logic Corporation
Add generic attribute change notification support to GEOM.
sys/sys/geom/geom.h:
Add a new attrchanged method field to both g_class
and g_geom.
sys/sys/geom/geom.h:
sys/geom/geom_event.c:
- Provide the g_attr_changed() function that providers
can use to advertise attribute changes.
- Perform delivery of attribute change notifications
from a thread context via the standard GEOM event
mechanism.
sys/geom/geom_subr.c:
Inherit the attrchanged method from class to geom (class instance).
sys/geom/geom_disk.c:
Provide disk_attr_changed() to provide g_attr_changed() access
to consumers of the disk API.
sys/cam/scsi/scsi_pass.c:
sys/cam/scsi/scsi_da.c:
sys/geom/geom_dev.c:
sys/geom/geom_disk.c:
Use attribute changed events to track updates to physical path
information.
sys/cam/scsi/scsi_da.c:
Add AC_ADVINFO_CHANGED to the registered asynchronous CAM
events for this driver. When this event occurs, and
the updated buffer type references our physical path
attribute, emit a GEOM attribute changed event via the
disk_attr_changed() API.
sys/cam/scsi/scsi_pass.c:
Add AC_ADVINFO_CHANGED to the registered asynchronous CAM
events for this driver. When this event occurs, update
the physical patch devfs alias for this pass instance.
Submitted by: gibbs
Sponsored by: Spectra Logic Corporation
2011-06-14 17:10:32 +00:00
|
|
|
/* Give the driver a chance to override */
|
|
|
|
|
if (dp->d_getattr != NULL) {
|
|
|
|
|
if (bp->bio_disk == NULL)
|
|
|
|
|
bp->bio_disk = dp;
|
|
|
|
|
error = dp->d_getattr(bp);
|
|
|
|
|
if (error != -1)
|
|
|
|
|
break;
|
|
|
|
|
error = EJUSTRETURN;
|
|
|
|
|
}
|
2010-12-29 12:11:07 +00:00
|
|
|
if (g_handleattr_int(bp, "GEOM::candelete",
|
|
|
|
|
(dp->d_flags & DISKFLAG_CANDELETE) != 0))
|
|
|
|
|
break;
|
|
|
|
|
else if (g_handleattr_int(bp, "GEOM::fwsectors",
|
|
|
|
|
dp->d_fwsectors))
|
2002-03-16 09:24:19 +00:00
|
|
|
break;
|
2002-09-20 19:36:05 +00:00
|
|
|
else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
|
2002-03-16 09:24:19 +00:00
|
|
|
break;
|
2002-06-09 10:57:34 +00:00
|
|
|
else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
|
2002-04-09 15:43:32 +00:00
|
|
|
break;
|
2007-05-05 17:12:15 +00:00
|
|
|
else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
|
|
|
|
|
break;
|
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
|
|
|
else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
|
|
|
|
|
dp->d_hba_vendor))
|
2010-07-25 15:43:52 +00:00
|
|
|
break;
|
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
|
|
|
else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
|
|
|
|
|
dp->d_hba_device))
|
2010-07-25 15:43:52 +00:00
|
|
|
break;
|
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
|
|
|
else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
|
|
|
|
|
dp->d_hba_subvendor))
|
2010-07-25 15:43:52 +00:00
|
|
|
break;
|
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
|
|
|
else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
|
|
|
|
|
dp->d_hba_subdevice))
|
2010-07-25 15:43:52 +00:00
|
|
|
break;
|
2002-04-19 09:24:12 +00:00
|
|
|
else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
|
|
|
|
|
g_disk_kerneldump(bp, dp);
|
2011-03-24 19:23:42 +00:00
|
|
|
else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
|
|
|
|
|
g_disk_setstate(bp, sc);
|
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
|
|
|
else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
|
|
|
|
|
dp->d_rotation_rate))
|
|
|
|
|
break;
|
2003-09-01 20:45:32 +00:00
|
|
|
else
|
2002-10-05 21:55:31 +00:00
|
|
|
error = ENOIOCTL;
|
|
|
|
|
break;
|
2006-10-31 21:12:43 +00:00
|
|
|
case BIO_FLUSH:
|
2012-09-18 07:57:34 +00:00
|
|
|
g_trace(G_T_BIO, "g_disk_flushcache(%s)",
|
2006-10-31 21:12:43 +00:00
|
|
|
bp->bio_to->name);
|
|
|
|
|
if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
|
2011-10-25 14:07:17 +00:00
|
|
|
error = EOPNOTSUPP;
|
|
|
|
|
break;
|
2006-10-31 21:12:43 +00:00
|
|
|
}
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
bp->bio_disk = dp;
|
|
|
|
|
bp->bio_to = (void *)bp->bio_done;
|
|
|
|
|
bp->bio_done = g_disk_done_single;
|
|
|
|
|
dp->d_strategy(bp);
|
2006-10-31 21:12:43 +00:00
|
|
|
break;
|
2002-03-11 21:42:35 +00:00
|
|
|
default:
|
2002-03-16 09:24:19 +00:00
|
|
|
error = EOPNOTSUPP;
|
|
|
|
|
break;
|
|
|
|
|
}
|
2002-10-07 07:15:37 +00:00
|
|
|
if (error != EJUSTRETURN)
|
2002-09-30 08:54:46 +00:00
|
|
|
g_io_deliver(bp, error);
|
2002-03-16 09:24:19 +00:00
|
|
|
return;
|
2002-03-11 21:42:35 +00:00
|
|
|
}
|
|
|
|
|
|
2002-10-20 18:09:01 +00:00
|
|
|
static void
|
2002-12-16 22:33:27 +00:00
|
|
|
g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
|
2002-10-20 18:09:01 +00:00
|
|
|
{
|
2013-06-12 13:36:20 +00:00
|
|
|
struct bio *bp;
|
2002-10-20 18:09:01 +00:00
|
|
|
struct disk *dp;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
2013-06-12 13:36:20 +00:00
|
|
|
char *buf;
|
|
|
|
|
int res = 0;
|
2002-10-20 18:09:01 +00:00
|
|
|
|
2011-03-24 19:23:42 +00:00
|
|
|
sc = gp->softc;
|
|
|
|
|
if (sc == NULL || (dp = sc->dp) == NULL)
|
2003-09-23 07:53:59 +00:00
|
|
|
return;
|
2002-10-28 22:43:54 +00:00
|
|
|
if (indent == NULL) {
|
|
|
|
|
sbuf_printf(sb, " hd %u", dp->d_fwheads);
|
|
|
|
|
sbuf_printf(sb, " sc %u", dp->d_fwsectors);
|
|
|
|
|
return;
|
|
|
|
|
}
|
2002-10-20 18:46:25 +00:00
|
|
|
if (pp != NULL) {
|
2002-10-20 18:09:01 +00:00
|
|
|
sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
|
|
|
|
|
indent, dp->d_fwheads);
|
|
|
|
|
sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
|
|
|
|
|
indent, dp->d_fwsectors);
|
2013-06-12 13:36:20 +00:00
|
|
|
if (dp->d_getattr != NULL) {
|
|
|
|
|
buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
|
|
|
|
|
bp = g_alloc_bio();
|
|
|
|
|
bp->bio_disk = dp;
|
|
|
|
|
bp->bio_attribute = "GEOM::ident";
|
|
|
|
|
bp->bio_length = DISK_IDENT_SIZE;
|
|
|
|
|
bp->bio_data = buf;
|
|
|
|
|
res = dp->d_getattr(bp);
|
|
|
|
|
sbuf_printf(sb, "%s<ident>%s</ident>\n", indent,
|
|
|
|
|
res == 0 ? buf: dp->d_ident);
|
|
|
|
|
bp->bio_attribute = "GEOM::lunid";
|
|
|
|
|
bp->bio_length = DISK_IDENT_SIZE;
|
|
|
|
|
bp->bio_data = buf;
|
|
|
|
|
if (dp->d_getattr(bp) == 0)
|
|
|
|
|
sbuf_printf(sb, "%s<lunid>%s</lunid>\n",
|
|
|
|
|
indent, buf);
|
2013-08-24 09:42:14 +00:00
|
|
|
bp->bio_attribute = "GEOM::lunname";
|
|
|
|
|
bp->bio_length = DISK_IDENT_SIZE;
|
|
|
|
|
bp->bio_data = buf;
|
|
|
|
|
if (dp->d_getattr(bp) == 0)
|
|
|
|
|
sbuf_printf(sb, "%s<lunname>%s</lunname>\n",
|
|
|
|
|
indent, buf);
|
2013-06-12 13:36:20 +00:00
|
|
|
g_destroy_bio(bp);
|
|
|
|
|
g_free(buf);
|
|
|
|
|
} else
|
|
|
|
|
sbuf_printf(sb, "%s<ident>%s</ident>\n", indent,
|
|
|
|
|
dp->d_ident);
|
2011-02-26 14:58:54 +00:00
|
|
|
sbuf_printf(sb, "%s<descr>%s</descr>\n", indent, dp->d_descr);
|
2002-10-20 18:09:01 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2012-07-07 21:28:31 +00:00
|
|
|
static void
|
2012-10-29 17:52:43 +00:00
|
|
|
g_disk_resize(void *ptr, int flag)
|
2012-07-07 21:28:31 +00:00
|
|
|
{
|
2012-10-29 17:52:43 +00:00
|
|
|
struct disk *dp;
|
2012-07-07 21:28:31 +00:00
|
|
|
struct g_geom *gp;
|
|
|
|
|
struct g_provider *pp;
|
|
|
|
|
|
2012-10-29 17:52:43 +00:00
|
|
|
if (flag == EV_CANCEL)
|
|
|
|
|
return;
|
|
|
|
|
g_topology_assert();
|
|
|
|
|
|
|
|
|
|
dp = ptr;
|
2012-07-07 21:28:31 +00:00
|
|
|
gp = dp->d_geom;
|
|
|
|
|
|
2012-10-29 17:52:43 +00:00
|
|
|
if (dp->d_destroyed || gp == NULL)
|
|
|
|
|
return;
|
|
|
|
|
|
2012-07-07 21:28:31 +00:00
|
|
|
LIST_FOREACH(pp, &gp->provider, provider) {
|
|
|
|
|
if (pp->sectorsize != 0 &&
|
|
|
|
|
pp->sectorsize != dp->d_sectorsize)
|
|
|
|
|
g_wither_provider(pp, ENXIO);
|
|
|
|
|
else
|
|
|
|
|
g_resize_provider(pp, dp->d_mediasize);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2002-10-11 20:52:44 +00:00
|
|
|
static void
|
2003-05-25 16:57:10 +00:00
|
|
|
g_disk_create(void *arg, int flag)
|
2002-03-11 21:42:35 +00:00
|
|
|
{
|
|
|
|
|
struct g_geom *gp;
|
|
|
|
|
struct g_provider *pp;
|
2003-02-11 14:12:06 +00:00
|
|
|
struct disk *dp;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
|
|
|
|
char tmpstr[80];
|
2002-03-11 21:42:35 +00:00
|
|
|
|
2003-05-25 16:57:10 +00:00
|
|
|
if (flag == EV_CANCEL)
|
|
|
|
|
return;
|
2002-10-11 20:52:44 +00:00
|
|
|
g_topology_assert();
|
2003-02-11 14:12:06 +00:00
|
|
|
dp = arg;
|
2011-03-24 19:23:42 +00:00
|
|
|
sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF);
|
2013-03-25 05:45:24 +00:00
|
|
|
mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
|
2011-03-24 19:23:42 +00:00
|
|
|
sc->dp = dp;
|
2003-02-11 14:12:06 +00:00
|
|
|
gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
|
2011-03-24 19:23:42 +00:00
|
|
|
gp->softc = sc;
|
2002-10-11 20:52:44 +00:00
|
|
|
pp = g_new_providerf(gp, "%s", gp->name);
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
devstat_remove_entry(pp->stat);
|
|
|
|
|
pp->stat = NULL;
|
|
|
|
|
dp->d_devstat->id = pp;
|
2003-02-11 14:12:06 +00:00
|
|
|
pp->mediasize = dp->d_mediasize;
|
|
|
|
|
pp->sectorsize = dp->d_sectorsize;
|
2003-02-11 14:57:34 +00:00
|
|
|
pp->stripeoffset = dp->d_stripeoffset;
|
|
|
|
|
pp->stripesize = dp->d_stripesize;
|
2013-03-19 14:49:15 +00:00
|
|
|
if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
|
|
|
|
|
pp->flags |= G_PF_ACCEPT_UNMAPPED;
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
|
|
|
|
|
pp->flags |= G_PF_DIRECT_SEND;
|
|
|
|
|
pp->flags |= G_PF_DIRECT_RECEIVE;
|
2002-10-25 20:09:45 +00:00
|
|
|
if (bootverbose)
|
|
|
|
|
printf("GEOM: new disk %s\n", gp->name);
|
2011-03-24 19:23:42 +00:00
|
|
|
sysctl_ctx_init(&sc->sysctl_ctx);
|
|
|
|
|
snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
|
|
|
|
|
sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
|
|
|
|
|
SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
|
|
|
|
|
CTLFLAG_RD, 0, tmpstr);
|
|
|
|
|
if (sc->sysctl_tree != NULL) {
|
|
|
|
|
snprintf(tmpstr, sizeof(tmpstr),
|
|
|
|
|
"kern.geom.disk.%s.led", gp->name);
|
|
|
|
|
TUNABLE_STR_FETCH(tmpstr, sc->led, sizeof(sc->led));
|
|
|
|
|
SYSCTL_ADD_STRING(&sc->sysctl_ctx,
|
|
|
|
|
SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
|
|
|
|
|
CTLFLAG_RW | CTLFLAG_TUN, sc->led, sizeof(sc->led),
|
|
|
|
|
"LED name");
|
|
|
|
|
}
|
|
|
|
|
pp->private = sc;
|
2003-04-02 21:10:04 +00:00
|
|
|
dp->d_geom = gp;
|
|
|
|
|
g_error_provider(pp, 0);
|
2002-10-11 20:52:44 +00:00
|
|
|
}
|
|
|
|
|
|
Fix a bug which causes a panic in daopen(). The panic is caused by
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
2012-06-24 04:29:03 +00:00
|
|
|
/*
|
|
|
|
|
* We get this callback after all of the consumers have gone away, and just
|
|
|
|
|
* before the provider is freed. If the disk driver provided a d_gone
|
|
|
|
|
* callback, let them know that it is okay to free resources -- they won't
|
|
|
|
|
* be getting any more accesses from GEOM.
|
|
|
|
|
*/
|
|
|
|
|
static void
|
|
|
|
|
g_disk_providergone(struct g_provider *pp)
|
|
|
|
|
{
|
|
|
|
|
struct disk *dp;
|
|
|
|
|
struct g_disk_softc *sc;
|
|
|
|
|
|
2013-03-25 05:45:24 +00:00
|
|
|
sc = (struct g_disk_softc *)pp->private;
|
Fix a bug which causes a panic in daopen(). The panic is caused by
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
2012-06-24 04:29:03 +00:00
|
|
|
dp = sc->dp;
|
2013-03-25 05:45:24 +00:00
|
|
|
if (dp != NULL && dp->d_gone != NULL)
|
Fix a bug which causes a panic in daopen(). The panic is caused by
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
2012-06-24 04:29:03 +00:00
|
|
|
dp->d_gone(dp);
|
2013-03-25 05:45:24 +00:00
|
|
|
if (sc->sysctl_tree != NULL) {
|
|
|
|
|
sysctl_ctx_free(&sc->sysctl_ctx);
|
|
|
|
|
sc->sysctl_tree = NULL;
|
|
|
|
|
}
|
|
|
|
|
if (sc->led[0] != 0) {
|
|
|
|
|
led_set(sc->led, "0");
|
|
|
|
|
sc->led[0] = 0;
|
|
|
|
|
}
|
|
|
|
|
pp->private = NULL;
|
|
|
|
|
pp->geom->softc = NULL;
|
|
|
|
|
mtx_destroy(&sc->done_mtx);
|
Merge GEOM direct dispatch changes from the projects/camlock branch.
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
2013-10-22 08:22:19 +00:00
|
|
|
mtx_destroy(&sc->start_mtx);
|
2013-03-25 05:45:24 +00:00
|
|
|
g_free(sc);
|
Fix a bug which causes a panic in daopen(). The panic is caused by
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
2012-06-24 04:29:03 +00:00
|
|
|
}
|
|
|
|
|
|
2003-09-01 12:03:13 +00:00
|
|
|
static void
|
|
|
|
|
g_disk_destroy(void *ptr, int flag)
|
|
|
|
|
{
|
2004-02-18 21:36:53 +00:00
|
|
|
struct disk *dp;
|
2003-09-01 12:03:13 +00:00
|
|
|
struct g_geom *gp;
|
2011-03-24 19:23:42 +00:00
|
|
|
struct g_disk_softc *sc;
|
2002-10-11 20:52:44 +00:00
|
|
|
|
2003-09-01 12:03:13 +00:00
|
|
|
g_topology_assert();
|
2004-02-18 21:36:53 +00:00
|
|
|
dp = ptr;
|
|
|
|
|
gp = dp->d_geom;
|
2005-01-14 21:05:35 +00:00
|
|
|
if (gp != NULL) {
|
2011-03-24 19:23:42 +00:00
|
|
|
sc = gp->softc;
|
2013-03-25 05:45:24 +00:00
|
|
|
if (sc != NULL)
|
|
|
|
|
sc->dp = NULL;
|
|
|
|
|
dp->d_geom = NULL;
|
2005-01-14 21:05:35 +00:00
|
|
|
g_wither_geom(gp, ENXIO);
|
|
|
|
|
}
|
2004-02-18 21:36:53 +00:00
|
|
|
g_free(dp);
|
2003-09-01 12:03:13 +00:00
|
|
|
}
|
2002-10-11 20:52:44 +00:00
|
|
|
|
2007-05-05 17:47:20 +00:00
|
|
|
/*
|
2009-09-04 09:39:06 +00:00
|
|
|
* We only allow printable characters in disk ident,
|
|
|
|
|
* the rest is converted to 'x<HH>'.
|
2007-05-05 17:47:20 +00:00
|
|
|
*/
|
|
|
|
|
static void
|
|
|
|
|
g_disk_ident_adjust(char *ident, size_t size)
|
|
|
|
|
{
|
2009-09-04 09:39:06 +00:00
|
|
|
char *p, tmp[4], newid[DISK_IDENT_SIZE];
|
|
|
|
|
|
|
|
|
|
newid[0] = '\0';
|
|
|
|
|
for (p = ident; *p != '\0'; p++) {
|
|
|
|
|
if (isprint(*p)) {
|
|
|
|
|
tmp[0] = *p;
|
|
|
|
|
tmp[1] = '\0';
|
|
|
|
|
} else {
|
|
|
|
|
snprintf(tmp, sizeof(tmp), "x%02hhx",
|
|
|
|
|
*(unsigned char *)p);
|
2007-05-05 17:47:20 +00:00
|
|
|
}
|
2009-09-04 09:39:06 +00:00
|
|
|
if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
|
|
|
|
|
break;
|
2007-05-05 17:47:20 +00:00
|
|
|
}
|
|
|
|
|
bzero(ident, size);
|
|
|
|
|
strlcpy(ident, newid, size);
|
|
|
|
|
}
|
|
|
|
|
|
2004-02-18 21:36:53 +00:00
|
|
|
struct disk *
|
2011-10-25 14:04:59 +00:00
|
|
|
disk_alloc(void)
|
2002-10-11 20:52:44 +00:00
|
|
|
{
|
2004-02-18 21:36:53 +00:00
|
|
|
|
2011-10-25 14:04:59 +00:00
|
|
|
return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO));
|
2004-02-18 21:36:53 +00:00
|
|
|
}
|
2002-10-11 20:52:44 +00:00
|
|
|
|
2004-02-18 21:36:53 +00:00
|
|
|
void
|
|
|
|
|
disk_create(struct disk *dp, int version)
|
|
|
|
|
{
|
2011-10-25 14:05:39 +00:00
|
|
|
|
2013-07-03 23:46:30 +00:00
|
|
|
if (version != DISK_VERSION) {
|
2004-02-18 21:36:53 +00:00
|
|
|
printf("WARNING: Attempt to add disk %s%d %s",
|
|
|
|
|
dp->d_name, dp->d_unit,
|
|
|
|
|
" using incompatible ABI version of disk(9)\n");
|
|
|
|
|
printf("WARNING: Ignoring disk %s%d\n",
|
|
|
|
|
dp->d_name, dp->d_unit);
|
|
|
|
|
return;
|
|
|
|
|
}
|
2013-10-25 19:19:12 +00:00
|
|
|
if (dp->d_flags & DISKFLAG_RESERVED) {
|
|
|
|
|
printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
|
|
|
|
|
dp->d_name, dp->d_unit);
|
|
|
|
|
printf("WARNING: Ignoring disk %s%d\n",
|
|
|
|
|
dp->d_name, dp->d_unit);
|
|
|
|
|
return;
|
|
|
|
|
}
|
2003-01-30 20:34:23 +00:00
|
|
|
KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
|
|
|
|
|
KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
|
|
|
|
|
KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
|
2003-02-04 10:32:40 +00:00
|
|
|
KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
|
2003-09-23 07:53:59 +00:00
|
|
|
if (dp->d_devstat == NULL)
|
|
|
|
|
dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
|
|
|
|
|
dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
|
|
|
|
|
DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
|
2003-04-02 21:10:04 +00:00
|
|
|
dp->d_geom = NULL;
|
2007-05-05 17:47:20 +00:00
|
|
|
g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
|
2003-04-23 20:46:12 +00:00
|
|
|
g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
|
2002-03-11 21:42:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void
|
2003-02-21 15:13:26 +00:00
|
|
|
disk_destroy(struct disk *dp)
|
2002-03-11 21:42:35 +00:00
|
|
|
{
|
|
|
|
|
|
2003-04-02 21:10:04 +00:00
|
|
|
g_cancel_event(dp);
|
2004-02-18 21:36:53 +00:00
|
|
|
dp->d_destroyed = 1;
|
2004-06-27 20:53:20 +00:00
|
|
|
if (dp->d_devstat != NULL)
|
|
|
|
|
devstat_remove_entry(dp->d_devstat);
|
2004-06-29 08:33:58 +00:00
|
|
|
g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
|
2002-03-11 21:42:35 +00:00
|
|
|
}
|
|
|
|
|
|
Fix a bug that caused some /dev entries to continue to exist after
the underlying drive had been hot-unplugged from the system. Here
is a specific example. Filesystem code had opened /dev/da1s1e.
Subsequently, the drive was hot-unplugged. This (correctly) caused
all of the associated /dev/da1* entries to be deleted. When the
filesystem later realized that the drive was gone it closed the
device, reducing the write-access counts to 0 on the geom providers
for da1s1e, da1s1, and da1. This caused geom to re-taste the
providers, resulting in the devices being created again. When the
drive was hot-plugged back in, it resulted in duplicate /dev entries
for da1s1e, da1s1, and da1.
This fix adds a new disk_gone() function which is called by CAM when a
drive goes away. It orphans all of the providers associated with the
drive, setting an error condition of ENXIO in each one. In addition,
we prevent a re-taste on last close for writing if an error condition
has been set in the provider.
Sponsored by: Isilon Systems
Reviewed by: phk
MFC after: 1 week
2005-11-18 02:43:49 +00:00
|
|
|
void
|
|
|
|
|
disk_gone(struct disk *dp)
|
|
|
|
|
{
|
|
|
|
|
struct g_geom *gp;
|
|
|
|
|
struct g_provider *pp;
|
|
|
|
|
|
|
|
|
|
gp = dp->d_geom;
|
2012-09-22 12:41:49 +00:00
|
|
|
if (gp != NULL) {
|
2012-09-28 08:22:51 +00:00
|
|
|
pp = LIST_FIRST(&gp->provider);
|
|
|
|
|
if (pp != NULL) {
|
|
|
|
|
KASSERT(LIST_NEXT(pp, provider) == NULL,
|
|
|
|
|
("geom %p has more than one provider", gp));
|
2006-04-10 03:55:13 +00:00
|
|
|
g_wither_provider(pp, ENXIO);
|
2012-09-28 08:22:51 +00:00
|
|
|
}
|
2012-09-22 12:41:49 +00:00
|
|
|
}
|
Fix a bug that caused some /dev entries to continue to exist after
the underlying drive had been hot-unplugged from the system. Here
is a specific example. Filesystem code had opened /dev/da1s1e.
Subsequently, the drive was hot-unplugged. This (correctly) caused
all of the associated /dev/da1* entries to be deleted. When the
filesystem later realized that the drive was gone it closed the
device, reducing the write-access counts to 0 on the geom providers
for da1s1e, da1s1, and da1. This caused geom to re-taste the
providers, resulting in the devices being created again. When the
drive was hot-plugged back in, it resulted in duplicate /dev entries
for da1s1e, da1s1, and da1.
This fix adds a new disk_gone() function which is called by CAM when a
drive goes away. It orphans all of the providers associated with the
drive, setting an error condition of ENXIO in each one. In addition,
we prevent a re-taste on last close for writing if an error condition
has been set in the provider.
Sponsored by: Isilon Systems
Reviewed by: phk
MFC after: 1 week
2005-11-18 02:43:49 +00:00
|
|
|
}
|
|
|
|
|
|
Plumb device physical path reporting from CAM devices, through GEOM and
DEVFS, and make it accessible via the diskinfo utility.
Extend GEOM's generic attribute query mechanism into generic disk consumers.
sys/geom/geom_disk.c:
sys/geom/geom_disk.h:
sys/cam/scsi/scsi_da.c:
sys/cam/ata/ata_da.c:
- Allow disk providers to implement a new method which can override
the default BIO_GETATTR response, d_getattr(struct bio *). This
function returns -1 if not handled, otherwise it returns 0 or an
errno to be passed to g_io_deliver().
sys/cam/scsi/scsi_da.c:
sys/cam/ata/ata_da.c:
- Don't copy the serial number to dp->d_ident anymore, as the CAM XPT
is now responsible for returning this information via
d_getattr()->(a)dagetattr()->xpt_getatr().
sys/geom/geom_dev.c:
- Implement a new ioctl, DIOCGPHYSPATH, which returns the GEOM
attribute "GEOM::physpath", if possible. If the attribute request
returns a zero-length string, ENOENT is returned.
usr.sbin/diskinfo/diskinfo.c:
- If the DIOCGPHYSPATH ioctl is successful, report physical path
data when diskinfo is executed with the '-v' option.
Submitted by: will
Reviewed by: gibbs
Sponsored by: Spectra Logic Corporation
Add generic attribute change notification support to GEOM.
sys/sys/geom/geom.h:
Add a new attrchanged method field to both g_class
and g_geom.
sys/sys/geom/geom.h:
sys/geom/geom_event.c:
- Provide the g_attr_changed() function that providers
can use to advertise attribute changes.
- Perform delivery of attribute change notifications
from a thread context via the standard GEOM event
mechanism.
sys/geom/geom_subr.c:
Inherit the attrchanged method from class to geom (class instance).
sys/geom/geom_disk.c:
Provide disk_attr_changed() to provide g_attr_changed() access
to consumers of the disk API.
sys/cam/scsi/scsi_pass.c:
sys/cam/scsi/scsi_da.c:
sys/geom/geom_dev.c:
sys/geom/geom_disk.c:
Use attribute changed events to track updates to physical path
information.
sys/cam/scsi/scsi_da.c:
Add AC_ADVINFO_CHANGED to the registered asynchronous CAM
events for this driver. When this event occurs, and
the updated buffer type references our physical path
attribute, emit a GEOM attribute changed event via the
disk_attr_changed() API.
sys/cam/scsi/scsi_pass.c:
Add AC_ADVINFO_CHANGED to the registered asynchronous CAM
events for this driver. When this event occurs, update
the physical patch devfs alias for this pass instance.
Submitted by: gibbs
Sponsored by: Spectra Logic Corporation
2011-06-14 17:10:32 +00:00
|
|
|
void
|
|
|
|
|
disk_attr_changed(struct disk *dp, const char *attr, int flag)
|
|
|
|
|
{
|
|
|
|
|
struct g_geom *gp;
|
|
|
|
|
struct g_provider *pp;
|
|
|
|
|
|
|
|
|
|
gp = dp->d_geom;
|
|
|
|
|
if (gp != NULL)
|
|
|
|
|
LIST_FOREACH(pp, &gp->provider, provider)
|
|
|
|
|
(void)g_attr_changed(pp, attr, flag);
|
|
|
|
|
}
|
|
|
|
|
|
2012-07-29 11:51:48 +00:00
|
|
|
void
|
|
|
|
|
disk_media_changed(struct disk *dp, int flag)
|
|
|
|
|
{
|
|
|
|
|
struct g_geom *gp;
|
|
|
|
|
struct g_provider *pp;
|
|
|
|
|
|
|
|
|
|
gp = dp->d_geom;
|
|
|
|
|
if (gp != NULL) {
|
2013-04-05 13:11:28 +00:00
|
|
|
pp = LIST_FIRST(&gp->provider);
|
|
|
|
|
if (pp != NULL) {
|
|
|
|
|
KASSERT(LIST_NEXT(pp, provider) == NULL,
|
|
|
|
|
("geom %p has more than one provider", gp));
|
2012-07-29 11:51:48 +00:00
|
|
|
g_media_changed(pp, flag);
|
2013-04-05 13:11:28 +00:00
|
|
|
}
|
2012-07-29 11:51:48 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
|
disk_media_gone(struct disk *dp, int flag)
|
|
|
|
|
{
|
|
|
|
|
struct g_geom *gp;
|
|
|
|
|
struct g_provider *pp;
|
|
|
|
|
|
|
|
|
|
gp = dp->d_geom;
|
|
|
|
|
if (gp != NULL) {
|
2013-04-05 13:11:28 +00:00
|
|
|
pp = LIST_FIRST(&gp->provider);
|
|
|
|
|
if (pp != NULL) {
|
|
|
|
|
KASSERT(LIST_NEXT(pp, provider) == NULL,
|
|
|
|
|
("geom %p has more than one provider", gp));
|
2012-07-29 11:51:48 +00:00
|
|
|
g_media_gone(pp, flag);
|
2013-04-05 13:11:28 +00:00
|
|
|
}
|
2012-07-29 11:51:48 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2012-10-29 17:52:43 +00:00
|
|
|
int
|
|
|
|
|
disk_resize(struct disk *dp, int flag)
|
2012-07-07 21:28:31 +00:00
|
|
|
{
|
|
|
|
|
|
2012-10-29 17:52:43 +00:00
|
|
|
if (dp->d_destroyed || dp->d_geom == NULL)
|
|
|
|
|
return (0);
|
2012-07-07 21:28:31 +00:00
|
|
|
|
2012-10-29 17:52:43 +00:00
|
|
|
return (g_post_event(g_disk_resize, dp, flag, NULL));
|
2012-07-07 21:28:31 +00:00
|
|
|
}
|
|
|
|
|
|
2002-10-04 10:15:26 +00:00
|
|
|
static void
|
2003-04-02 20:41:18 +00:00
|
|
|
g_kern_disks(void *p, int flag __unused)
|
2002-10-04 10:15:26 +00:00
|
|
|
{
|
|
|
|
|
struct sbuf *sb;
|
|
|
|
|
struct g_geom *gp;
|
|
|
|
|
char *sp;
|
|
|
|
|
|
|
|
|
|
sb = p;
|
|
|
|
|
sp = "";
|
|
|
|
|
g_topology_assert();
|
|
|
|
|
LIST_FOREACH(gp, &g_disk_class.geom, geom) {
|
|
|
|
|
sbuf_printf(sb, "%s%s", sp, gp->name);
|
|
|
|
|
sp = " ";
|
|
|
|
|
}
|
|
|
|
|
sbuf_finish(sb);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
sysctl_disks(SYSCTL_HANDLER_ARGS)
|
|
|
|
|
{
|
|
|
|
|
int error;
|
|
|
|
|
struct sbuf *sb;
|
|
|
|
|
|
2008-08-09 11:14:05 +00:00
|
|
|
sb = sbuf_new_auto();
|
2003-04-23 21:28:27 +00:00
|
|
|
g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
|
2003-04-23 20:46:12 +00:00
|
|
|
error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
|
2002-10-04 10:15:26 +00:00
|
|
|
sbuf_delete(sb);
|
|
|
|
|
return error;
|
|
|
|
|
}
|
|
|
|
|
|
2011-01-26 22:48:09 +00:00
|
|
|
SYSCTL_PROC(_kern, OID_AUTO, disks,
|
|
|
|
|
CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
|
2002-10-04 10:15:26 +00:00
|
|
|
sysctl_disks, "A", "names of available disks");
|