nfs_nfsiod.c revision 122698
1/*
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 *    must display the following acknowledgement:
18 *	This product includes software developed by the University of
19 *	California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 *    may be used to endorse or promote products derived from this software
22 *    without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 *	@(#)nfs_syscalls.c	8.5 (Berkeley) 3/30/95
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD: head/sys/nfsclient/nfs_nfsiod.c 122698 2003-11-14 20:54:10Z alfred $");
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/sysproto.h>
45#include <sys/kernel.h>
46#include <sys/sysctl.h>
47#include <sys/file.h>
48#include <sys/filedesc.h>
49#include <sys/vnode.h>
50#include <sys/malloc.h>
51#include <sys/mount.h>
52#include <sys/proc.h>
53#include <sys/bio.h>
54#include <sys/buf.h>
55#include <sys/mbuf.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/domain.h>
59#include <sys/protosw.h>
60#include <sys/namei.h>
61#include <sys/unistd.h>
62#include <sys/kthread.h>
63#include <sys/fcntl.h>
64#include <sys/lockf.h>
65#include <sys/mutex.h>
66
67#include <netinet/in.h>
68#include <netinet/tcp.h>
69
70#include <rpc/rpcclnt.h>
71
72#include <nfs/xdr_subs.h>
73#include <nfs/rpcv2.h>
74#include <nfs/nfsproto.h>
75#include <nfsclient/nfs.h>
76#include <nfsclient/nfsm_subs.h>
77#include <nfsclient/nfsmount.h>
78#include <nfsclient/nfsnode.h>
79#include <nfsclient/nfs_lock.h>
80
81static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
82
83static void	nfssvc_iod(void *);
84
85static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
86
87SYSCTL_DECL(_vfs_nfs);
88
89/* Maximum number of seconds a nfsiod kthread will sleep before exiting */
90static unsigned int nfs_iodmaxidle = 120;
91SYSCTL_UINT(_vfs_nfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0, "");
92
93/* Maximum number of nfsiod kthreads */
94static unsigned int nfs_iodmax = 20;
95
96/* Minimum number of nfsiod kthreads to keep as spares */
97static unsigned int nfs_iodmin = 4;
98
99static int
100sysctl_iodmin(SYSCTL_HANDLER_ARGS)
101{
102	int error, i;
103	int newmin;
104
105	newmin = nfs_iodmin;
106	error = sysctl_handle_int(oidp, &newmin, 0, req);
107	if (error || (req->newptr == NULL))
108		return (error);
109	if (newmin > nfs_iodmax)
110		return (EINVAL);
111	nfs_iodmin = newmin;
112	if (nfs_numasync >= nfs_iodmin)
113		return (0);
114	/*
115	 * If the current number of nfsiod is lower
116	 * than the new minimum, create some more.
117	 */
118	for (i = nfs_iodmin - nfs_numasync; i > 0; i--)
119		nfs_nfsiodnew();
120	return (0);
121}
122SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
123    sizeof (nfs_iodmin), sysctl_iodmin, "IU", "");
124
125
126static int
127sysctl_iodmax(SYSCTL_HANDLER_ARGS)
128{
129	int error, i;
130	int iod, newmax;
131
132	newmax = nfs_iodmax;
133	error = sysctl_handle_int(oidp, &newmax, 0, req);
134	if (error || (req->newptr == NULL))
135		return (error);
136	if (newmax > NFS_MAXASYNCDAEMON)
137		return (EINVAL);
138	nfs_iodmax = newmax;
139	if (nfs_numasync <= nfs_iodmax)
140		return (0);
141	/*
142	 * If there are some asleep nfsiods that should
143	 * exit, wakeup() them so that they check nfs_iodmax
144	 * and exit.  Those who are active will exit as
145	 * soon as they finish I/O.
146	 */
147	iod = nfs_numasync - 1;
148	for (i = 0; i < nfs_numasync - nfs_iodmax; i++) {
149		if (nfs_iodwant[iod])
150			wakeup(&nfs_iodwant[iod]);
151		iod--;
152	}
153	return (0);
154}
155SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
156    sizeof (nfs_iodmax), sysctl_iodmax, "IU", "");
157
158int
159nfs_nfsiodnew(void)
160{
161	int error, i;
162	int newiod;
163
164	if (nfs_numasync >= nfs_iodmax)
165		return (-1);
166	newiod = -1;
167	for (i = 0; i < nfs_iodmax; i++)
168		if (nfs_asyncdaemon[i] == 0) {
169			nfs_asyncdaemon[i]++;
170			newiod = i;
171			break;
172		}
173	if (newiod == -1)
174		return (-1);
175	error = kthread_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID,
176	    0, "nfsiod %d", newiod);
177	if (error)
178		return (-1);
179	nfs_numasync++;
180	return (newiod);
181}
182
183static void
184nfsiod_setup(void *dummy)
185{
186	int i;
187	int error;
188
189	TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin);
190	/* Silently limit the start number of nfsiod's */
191	if (nfs_iodmin > NFS_MAXASYNCDAEMON)
192		nfs_iodmin = NFS_MAXASYNCDAEMON;
193
194	for (i = 0; i < nfs_iodmin; i++) {
195		error = nfs_nfsiodnew();
196		if (error == -1)
197			panic("nfsiod_setup: nfs_nfsiodnew failed");
198	}
199}
200SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
201
202static int nfs_defect = 0;
203SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
204
205int
206nfsclnt(struct thread *td, struct nfsclnt_args *uap)
207{
208	struct lockd_ans la;
209	int error;
210
211	if ((uap->flag & NFSCLNT_LOCKDANS) != 0) {
212		error = copyin(uap->argp, &la, sizeof(la));
213		return (error != 0 ? error : nfslockdans(td, &la));
214	}
215	return EINVAL;
216}
217
218/*
219 * Asynchronous I/O daemons for client nfs.
220 * They do read-ahead and write-behind operations on the block I/O cache.
221 * Returns if we hit the timeout defined by the iodmaxidle sysctl.
222 */
223static void
224nfssvc_iod(void *instance)
225{
226	struct buf *bp;
227	struct nfsmount *nmp;
228	int myiod, timo;
229	int error = 0;
230
231	mtx_lock(&Giant);
232	myiod = (int *)instance - nfs_asyncdaemon;
233	/*
234	 * Main loop
235	 */
236	for (;;) {
237	    while (((nmp = nfs_iodmount[myiod]) == NULL
238		   || !TAILQ_FIRST(&nmp->nm_bufq))
239		   && error == 0) {
240		if (myiod >= nfs_iodmax)
241			goto finish;
242		if (nmp)
243			nmp->nm_bufqiods--;
244		nfs_iodwant[myiod] = curthread->td_proc;
245		nfs_iodmount[myiod] = NULL;
246		/*
247		 * Always keep at least nfs_iodmin kthreads.
248		 */
249		timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
250		error = tsleep(&nfs_iodwant[myiod], PWAIT | PCATCH,
251		    "-", timo);
252	    }
253	    if (error)
254		    break;
255	    while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
256		/* Take one off the front of the list */
257		TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
258		nmp->nm_bufqlen--;
259		if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) {
260		    nmp->nm_bufqwant = 0;
261		    wakeup(&nmp->nm_bufq);
262		}
263		if (bp->b_iocmd == BIO_READ)
264		    (void) nfs_doio(bp, bp->b_rcred, NULL);
265		else
266		    (void) nfs_doio(bp, bp->b_wcred, NULL);
267		/*
268		 * If there are more than one iod on this mount, then defect
269		 * so that the iods can be shared out fairly between the mounts
270		 */
271		if (nfs_defect && nmp->nm_bufqiods > 1) {
272		    NFS_DPF(ASYNCIO,
273			    ("nfssvc_iod: iod %d defecting from mount %p\n",
274			     myiod, nmp));
275		    nfs_iodmount[myiod] = NULL;
276		    nmp->nm_bufqiods--;
277		    break;
278		}
279	    }
280	}
281finish:
282	nfs_asyncdaemon[myiod] = 0;
283	if (nmp)
284	    nmp->nm_bufqiods--;
285	nfs_iodwant[myiod] = NULL;
286	nfs_iodmount[myiod] = NULL;
287	nfs_numasync--;
288	if ((error == 0) || (error == EWOULDBLOCK))
289		kthread_exit(0);
290	/* Abnormal termination */
291	kthread_exit(1);
292}
293