geom_kern.c revision 139451
161452Sdfr/*-
261452Sdfr * Copyright (c) 2002 Poul-Henning Kamp
361452Sdfr * Copyright (c) 2002 Networks Associates Technology, Inc.
461452Sdfr * All rights reserved.
561452Sdfr *
661452Sdfr * This software was developed for the FreeBSD Project by Poul-Henning Kamp
761452Sdfr * and NAI Labs, the Security Research Division of Network Associates, Inc.
861452Sdfr * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
961452Sdfr * DARPA CHATS research program.
1061452Sdfr *
1161452Sdfr * Redistribution and use in source and binary forms, with or without
1261452Sdfr * modification, are permitted provided that the following conditions
1361452Sdfr * are met:
1461452Sdfr * 1. Redistributions of source code must retain the above copyright
1561452Sdfr *    notice, this list of conditions and the following disclaimer.
1661452Sdfr * 2. Redistributions in binary form must reproduce the above copyright
1761452Sdfr *    notice, this list of conditions and the following disclaimer in the
1861452Sdfr *    documentation and/or other materials provided with the distribution.
1961452Sdfr * 3. The names of the authors may not be used to endorse or promote
2061452Sdfr *    products derived from this software without specific prior written
2161452Sdfr *    permission.
2261452Sdfr *
2361452Sdfr * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
2461452Sdfr * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
2561452Sdfr * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
2661452Sdfr * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
2761452Sdfr * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
2861452Sdfr * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
2961452Sdfr * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
3061452Sdfr * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
3161452Sdfr * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
3261452Sdfr * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
3361452Sdfr * SUCH DAMAGE.
3461452Sdfr */
3561452Sdfr
3661452Sdfr#include <sys/cdefs.h>
3761452Sdfr__FBSDID("$FreeBSD: head/sys/geom/geom_kern.c 139451 2004-12-30 20:29:58Z jhb $");
3876827Salfred
3979339Sjhb#include <sys/param.h>
4061452Sdfr#include <sys/systm.h>
4161452Sdfr#include <sys/kernel.h>
4261452Sdfr#include <sys/eventhandler.h>
4361452Sdfr#include <sys/malloc.h>
4461452Sdfr#include <sys/bio.h>
4561452Sdfr#include <sys/sysctl.h>
4661452Sdfr#include <sys/proc.h>
4761452Sdfr#include <sys/kthread.h>
4861452Sdfr#include <sys/lock.h>
4961452Sdfr#include <sys/mutex.h>
5061452Sdfr#include <sys/sbuf.h>
5161452Sdfr#include <sys/sched.h>
5261452Sdfr#include <sys/sx.h>
5361452Sdfr#include <geom/geom.h>
5461452Sdfr#include <geom/geom_int.h>
5561452Sdfr
5661452SdfrMALLOC_DEFINE(M_GEOM, "GEOM", "Geom data structures");
5761452Sdfr
5861452Sdfrstruct sx topology_lock;
5961452Sdfr
6061452Sdfrstatic struct proc *g_up_proc;
6161452Sdfr
6261452Sdfrint g_debugflags;
6361452Sdfrint g_collectstats = 1;
6461452Sdfrint g_shutdown;
6561452Sdfr
6661452Sdfr/*
6761452Sdfr * G_UP and G_DOWN are the two threads which push I/O through the
6861452Sdfr * stack.
6961452Sdfr *
7061452Sdfr * Things are procesed in a FIFO order, but these threads could be
7161452Sdfr * part of I/O prioritization by deciding which bios/bioqs to service
7261452Sdfr * in what order.
7361452Sdfr *
7461452Sdfr * We have only one thread in each direction, it is belived that until
7561452Sdfr * a very non-trivial workload in the UP/DOWN path this will be enough,
7661452Sdfr * but more than one can actually be run without problems.
7761452Sdfr *
7861452Sdfr * Holding the "mymutex" is a debugging feature:  It prevents people
7961452Sdfr * from sleeping in the UP/DOWN I/O path by mistake or design (doing
8061452Sdfr * so almost invariably result in deadlocks since it stalls all I/O
8161452Sdfr * processing in the given direction.
8261452Sdfr */
8361452Sdfr
8461452Sdfrstatic void
8561452Sdfrg_up_procbody(void)
8661452Sdfr{
8761452Sdfr	struct proc *p = g_up_proc;
8861452Sdfr	struct thread *tp = FIRST_THREAD_IN_PROC(p);
8961452Sdfr
9061452Sdfr	mtx_assert(&Giant, MA_NOTOWNED);
9161452Sdfr	mtx_lock_spin(&sched_lock);
9261452Sdfr	sched_prio(tp, PRIBIO);
9361452Sdfr	mtx_unlock_spin(&sched_lock);
9461452Sdfr	for(;;) {
9561452Sdfr		g_io_schedule_up(tp);
9661452Sdfr	}
9761452Sdfr}
9861452Sdfr
9961452Sdfrstruct kproc_desc g_up_kp = {
10061452Sdfr	"g_up",
10161452Sdfr	g_up_procbody,
10261452Sdfr	&g_up_proc,
10361452Sdfr};
10461452Sdfr
10561452Sdfrstatic struct proc *g_down_proc;
10661452Sdfr
10761452Sdfrstatic void
10861452Sdfrg_down_procbody(void)
10961452Sdfr{
11061452Sdfr	struct proc *p = g_down_proc;
11161452Sdfr	struct thread *tp = FIRST_THREAD_IN_PROC(p);
11261452Sdfr
11361452Sdfr	mtx_assert(&Giant, MA_NOTOWNED);
11461452Sdfr	mtx_lock_spin(&sched_lock);
11561452Sdfr	sched_prio(tp, PRIBIO);
11661452Sdfr	mtx_unlock_spin(&sched_lock);
11761452Sdfr	for(;;) {
11861452Sdfr		g_io_schedule_down(tp);
11961452Sdfr	}
12061452Sdfr}
12161452Sdfr
12261452Sdfrstruct kproc_desc g_down_kp = {
12361452Sdfr	"g_down",
12461452Sdfr	g_down_procbody,
12561452Sdfr	&g_down_proc,
12661452Sdfr};
12761452Sdfr
12861452Sdfrstatic struct proc *g_event_proc;
12961452Sdfr
13061452Sdfrstatic void
13161452Sdfrg_event_procbody(void)
13261452Sdfr{
13361452Sdfr	struct proc *p = g_event_proc;
13461452Sdfr	struct thread *tp = FIRST_THREAD_IN_PROC(p);
13561452Sdfr
13661452Sdfr	mtx_assert(&Giant, MA_NOTOWNED);
13761452Sdfr	mtx_lock_spin(&sched_lock);
13861452Sdfr	sched_prio(tp, PRIBIO);
13961452Sdfr	mtx_unlock_spin(&sched_lock);
14061452Sdfr	for(;;) {
14161452Sdfr		g_run_events();
14261452Sdfr		tsleep(&g_wait_event, PRIBIO, "-", hz/10);
14361452Sdfr	}
14461452Sdfr}
14561452Sdfr
14661452Sdfrstatic struct kproc_desc g_event_kp = {
14761452Sdfr	"g_event",
14861452Sdfr	g_event_procbody,
14961452Sdfr	&g_event_proc,
15061452Sdfr};
15161452Sdfr
15261452Sdfrstatic void
15361452Sdfrgeom_shutdown(void *foo __unused)
15461452Sdfr{
15561452Sdfr
15661452Sdfr	g_shutdown = 1;
15761452Sdfr}
15861452Sdfr
15961452Sdfrvoid
16061452Sdfrg_init(void)
16161452Sdfr{
16261452Sdfr
16361452Sdfr	g_trace(G_T_TOPOLOGY, "g_ignition");
16461452Sdfr	sx_init(&topology_lock, "GEOM topology");
16561452Sdfr	g_io_init();
16661452Sdfr	g_event_init();
16761452Sdfr	g_ctl_init();
16861452Sdfr	mtx_lock(&Giant);
16961452Sdfr	kproc_start(&g_event_kp);
17061452Sdfr	kproc_start(&g_up_kp);
17161452Sdfr	kproc_start(&g_down_kp);
17261452Sdfr	mtx_unlock(&Giant);
17361452Sdfr	EVENTHANDLER_REGISTER(shutdown_pre_sync, geom_shutdown, NULL,
17461452Sdfr		SHUTDOWN_PRI_FIRST);
17561452Sdfr}
17661452Sdfr
17761452Sdfrstatic int
17861452Sdfrsysctl_kern_geom_conftxt(SYSCTL_HANDLER_ARGS)
17961452Sdfr{
18061452Sdfr	int error;
18161452Sdfr	struct sbuf *sb;
18261452Sdfr
18361452Sdfr	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
18461452Sdfr	g_waitfor_event(g_conftxt, sb, M_WAITOK, NULL);
18561452Sdfr	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
18661452Sdfr	sbuf_delete(sb);
18761452Sdfr	return error;
18861452Sdfr}
18961452Sdfr
19061452Sdfrstatic int
19161452Sdfrsysctl_kern_geom_confdot(SYSCTL_HANDLER_ARGS)
19261452Sdfr{
19361452Sdfr	int error;
19461452Sdfr	struct sbuf *sb;
19561452Sdfr
19661452Sdfr	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
19761452Sdfr	g_waitfor_event(g_confdot, sb, M_WAITOK, NULL);
19861452Sdfr	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
19961452Sdfr	sbuf_delete(sb);
20061452Sdfr	return error;
20161452Sdfr}
20261452Sdfr
20361452Sdfrstatic int
20461452Sdfrsysctl_kern_geom_confxml(SYSCTL_HANDLER_ARGS)
20561452Sdfr{
20661452Sdfr	int error;
20761452Sdfr	struct sbuf *sb;
20861452Sdfr
20961452Sdfr	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
21061452Sdfr	g_waitfor_event(g_confxml, sb, M_WAITOK, NULL);
21161452Sdfr	error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
21261452Sdfr	sbuf_delete(sb);
21361452Sdfr	return error;
21461452Sdfr}
21561452Sdfr
21661452SdfrSYSCTL_NODE(_kern, OID_AUTO, geom, CTLFLAG_RW, 0, "GEOMetry management");
21761452Sdfr
21861452SdfrSYSCTL_PROC(_kern_geom, OID_AUTO, confxml, CTLTYPE_STRING|CTLFLAG_RD,
21961452Sdfr	0, 0, sysctl_kern_geom_confxml, "",
22061452Sdfr	"Dump the GEOM config in XML");
22161452Sdfr
22261452SdfrSYSCTL_PROC(_kern_geom, OID_AUTO, confdot, CTLTYPE_STRING|CTLFLAG_RD,
22361452Sdfr	0, 0, sysctl_kern_geom_confdot, "",
22461452Sdfr	"Dump the GEOM config in dot");
22561452Sdfr
22661452SdfrSYSCTL_PROC(_kern_geom, OID_AUTO, conftxt, CTLTYPE_STRING|CTLFLAG_RD,
22761452Sdfr	0, 0, sysctl_kern_geom_conftxt, "",
22861452Sdfr	"Dump the GEOM config in txt");
22961452Sdfr
23061452SdfrTUNABLE_INT("kern.geom.debugflags", &g_debugflags);
23161452SdfrSYSCTL_INT(_kern_geom, OID_AUTO, debugflags, CTLFLAG_RW,
23261452Sdfr	&g_debugflags, 0, "");
23361452Sdfr
23461452SdfrSYSCTL_INT(_kern_geom, OID_AUTO, collectstats, CTLFLAG_RW,
23561452Sdfr	&g_collectstats, 0, "");
23661452Sdfr
23761452SdfrSYSCTL_INT(_debug_sizeof, OID_AUTO, g_class, CTLFLAG_RD,
23861452Sdfr	0, sizeof(struct g_class), "");
23961452SdfrSYSCTL_INT(_debug_sizeof, OID_AUTO, g_geom, CTLFLAG_RD,
24061452Sdfr	0, sizeof(struct g_geom), "");
24161452SdfrSYSCTL_INT(_debug_sizeof, OID_AUTO, g_provider, CTLFLAG_RD,
24261452Sdfr	0, sizeof(struct g_provider), "");
24361452SdfrSYSCTL_INT(_debug_sizeof, OID_AUTO, g_consumer, CTLFLAG_RD,
24461452Sdfr	0, sizeof(struct g_consumer), "");
24561452SdfrSYSCTL_INT(_debug_sizeof, OID_AUTO, g_bioq, CTLFLAG_RD,
24661452Sdfr	0, sizeof(struct g_bioq), "");
24761452Sdfr