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 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	@(#)nfs_syscalls.c	8.5 (Berkeley) 3/30/95
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD$");
37
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/sysproto.h>
41#include <sys/kernel.h>
42#include <sys/sysctl.h>
43#include <sys/file.h>
44#include <sys/filedesc.h>
45#include <sys/vnode.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/proc.h>
49#include <sys/bio.h>
50#include <sys/buf.h>
51#include <sys/mbuf.h>
52#include <sys/socket.h>
53#include <sys/socketvar.h>
54#include <sys/domain.h>
55#include <sys/protosw.h>
56#include <sys/namei.h>
57#include <sys/unistd.h>
58#include <sys/kthread.h>
59#include <sys/fcntl.h>
60#include <sys/lockf.h>
61#include <sys/mutex.h>
62#include <sys/taskqueue.h>
63
64#include <netinet/in.h>
65#include <netinet/tcp.h>
66
67#include <nfs/xdr_subs.h>
68#include <nfs/nfsproto.h>
69#include <nfsclient/nfs.h>
70#include <nfsclient/nfsm_subs.h>
71#include <nfsclient/nfsmount.h>
72#include <nfsclient/nfsnode.h>
73#include <nfs/nfs_lock.h>
74
75static MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure");
76
77static void	nfssvc_iod(void *);
78
79static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
80
81SYSCTL_DECL(_vfs_oldnfs);
82
83/* Maximum number of seconds a nfsiod kthread will sleep before exiting */
84static unsigned int nfs_iodmaxidle = 120;
85SYSCTL_UINT(_vfs_oldnfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0,
86    "Max number of seconds an nfsiod kthread will sleep before exiting");
87
88/* Maximum number of nfsiod kthreads */
89unsigned int nfs_iodmax = 20;
90
91/* Minimum number of nfsiod kthreads to keep as spares */
92static unsigned int nfs_iodmin = 0;
93
94static int nfs_nfsiodnew_sync(void);
95
96static int
97sysctl_iodmin(SYSCTL_HANDLER_ARGS)
98{
99	int error, i;
100	int newmin;
101
102	newmin = nfs_iodmin;
103	error = sysctl_handle_int(oidp, &newmin, 0, req);
104	if (error || (req->newptr == NULL))
105		return (error);
106	mtx_lock(&nfs_iod_mtx);
107	if (newmin > nfs_iodmax) {
108		error = EINVAL;
109		goto out;
110	}
111	nfs_iodmin = newmin;
112	if (nfs_numasync >= nfs_iodmin)
113		goto out;
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_sync();
120out:
121	mtx_unlock(&nfs_iod_mtx);
122	return (0);
123}
124SYSCTL_PROC(_vfs_oldnfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
125    sizeof (nfs_iodmin), sysctl_iodmin, "IU",
126    "Min number of nfsiod kthreads to keep as spares");
127
128
129static int
130sysctl_iodmax(SYSCTL_HANDLER_ARGS)
131{
132	int error, i;
133	int iod, newmax;
134
135	newmax = nfs_iodmax;
136	error = sysctl_handle_int(oidp, &newmax, 0, req);
137	if (error || (req->newptr == NULL))
138		return (error);
139	if (newmax > NFS_MAXASYNCDAEMON)
140		return (EINVAL);
141	mtx_lock(&nfs_iod_mtx);
142	nfs_iodmax = newmax;
143	if (nfs_numasync <= nfs_iodmax)
144		goto out;
145	/*
146	 * If there are some asleep nfsiods that should
147	 * exit, wakeup() them so that they check nfs_iodmax
148	 * and exit.  Those who are active will exit as
149	 * soon as they finish I/O.
150	 */
151	iod = nfs_numasync - 1;
152	for (i = 0; i < nfs_numasync - nfs_iodmax; i++) {
153		if (nfs_iodwant[iod] == NFSIOD_AVAILABLE)
154			wakeup(&nfs_iodwant[iod]);
155		iod--;
156	}
157out:
158	mtx_unlock(&nfs_iod_mtx);
159	return (0);
160}
161SYSCTL_PROC(_vfs_oldnfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
162    sizeof (nfs_iodmax), sysctl_iodmax, "IU",
163    "Max number of nfsiod kthreads");
164
165static int
166nfs_nfsiodnew_sync(void)
167{
168	int error, i;
169
170	mtx_assert(&nfs_iod_mtx, MA_OWNED);
171	for (i = 0; i < nfs_iodmax; i++) {
172		if (nfs_asyncdaemon[i] == 0) {
173			nfs_asyncdaemon[i] = 1;
174			break;
175		}
176	}
177	if (i == nfs_iodmax)
178		return (0);
179	mtx_unlock(&nfs_iod_mtx);
180	error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL,
181	    RFHIGHPID, 0, "nfsiod %d", i);
182	mtx_lock(&nfs_iod_mtx);
183	if (error == 0) {
184		nfs_numasync++;
185		nfs_iodwant[i] = NFSIOD_AVAILABLE;
186	} else
187		nfs_asyncdaemon[i] = 0;
188	return (error);
189}
190
191void
192nfs_nfsiodnew_tq(__unused void *arg, int pending)
193{
194
195	mtx_lock(&nfs_iod_mtx);
196	while (pending > 0) {
197		pending--;
198		nfs_nfsiodnew_sync();
199	}
200	mtx_unlock(&nfs_iod_mtx);
201}
202
203void
204nfs_nfsiodnew(void)
205{
206
207	mtx_assert(&nfs_iod_mtx, MA_OWNED);
208	taskqueue_enqueue(taskqueue_thread, &nfs_nfsiodnew_task);
209}
210
211static void
212nfsiod_setup(void *dummy)
213{
214	int error;
215
216	TUNABLE_INT_FETCH("vfs.oldnfs.iodmin", &nfs_iodmin);
217	mtx_lock(&nfs_iod_mtx);
218	/* Silently limit the start number of nfsiod's */
219	if (nfs_iodmin > NFS_MAXASYNCDAEMON)
220		nfs_iodmin = NFS_MAXASYNCDAEMON;
221
222	while (nfs_numasync < nfs_iodmin) {
223		error = nfs_nfsiodnew_sync();
224		if (error == -1)
225			panic("nfsiod_setup: nfs_nfsiodnew failed");
226	}
227	mtx_unlock(&nfs_iod_mtx);
228}
229SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
230
231static int nfs_defect = 0;
232SYSCTL_INT(_vfs_oldnfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0,
233    "Allow nfsiods to migrate serving different mounts");
234
235/*
236 * Asynchronous I/O daemons for client nfs.
237 * They do read-ahead and write-behind operations on the block I/O cache.
238 * Returns if we hit the timeout defined by the iodmaxidle sysctl.
239 */
240static void
241nfssvc_iod(void *instance)
242{
243	struct buf *bp;
244	struct nfsmount *nmp;
245	int myiod, timo;
246	int error = 0;
247
248	mtx_lock(&nfs_iod_mtx);
249	myiod = (int *)instance - nfs_asyncdaemon;
250	/*
251	 * Main loop
252	 */
253	for (;;) {
254	    while (((nmp = nfs_iodmount[myiod]) == NULL)
255		   || !TAILQ_FIRST(&nmp->nm_bufq)) {
256		if (myiod >= nfs_iodmax)
257			goto finish;
258		if (nmp)
259			nmp->nm_bufqiods--;
260		if (nfs_iodwant[myiod] == NFSIOD_NOT_AVAILABLE)
261			nfs_iodwant[myiod] = NFSIOD_AVAILABLE;
262		nfs_iodmount[myiod] = NULL;
263		/*
264		 * Always keep at least nfs_iodmin kthreads.
265		 */
266		timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
267		error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH,
268		    "-", timo);
269		if (error) {
270			nmp = nfs_iodmount[myiod];
271			/*
272			 * Rechecking the nm_bufq closes a rare race where the
273			 * nfsiod is woken up at the exact time the idle timeout
274			 * fires
275			 */
276			if (nmp && TAILQ_FIRST(&nmp->nm_bufq))
277				error = 0;
278			break;
279		}
280	    }
281	    if (error)
282		    break;
283	    while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
284	        int giant_locked = 0;
285
286		/* Take one off the front of the list */
287		TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
288		nmp->nm_bufqlen--;
289		if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) {
290		    nmp->nm_bufqwant = 0;
291		    wakeup(&nmp->nm_bufq);
292		}
293		mtx_unlock(&nfs_iod_mtx);
294		if (NFS_ISV4(bp->b_vp)) {
295			giant_locked = 1;
296			mtx_lock(&Giant);
297		}
298		if (bp->b_flags & B_DIRECT) {
299			KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
300			(void)nfs_doio_directwrite(bp);
301		} else {
302			if (bp->b_iocmd == BIO_READ)
303				(void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL);
304			else
305				(void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL);
306		}
307		if (giant_locked)
308			mtx_unlock(&Giant);
309		mtx_lock(&nfs_iod_mtx);
310		/*
311		 * Make sure the nmp hasn't been dismounted as soon as
312		 * nfs_doio() completes for the last buffer.
313		 */
314		nmp = nfs_iodmount[myiod];
315		if (nmp == NULL)
316			break;
317
318		/*
319		 * If there are more than one iod on this mount, then defect
320		 * so that the iods can be shared out fairly between the mounts
321		 */
322		if (nfs_defect && nmp->nm_bufqiods > 1) {
323		    NFS_DPF(ASYNCIO,
324			    ("nfssvc_iod: iod %d defecting from mount %p\n",
325			     myiod, nmp));
326		    nfs_iodmount[myiod] = NULL;
327		    nmp->nm_bufqiods--;
328		    break;
329		}
330	    }
331	}
332finish:
333	nfs_asyncdaemon[myiod] = 0;
334	if (nmp)
335	    nmp->nm_bufqiods--;
336	nfs_iodwant[myiod] = NFSIOD_NOT_AVAILABLE;
337	nfs_iodmount[myiod] = NULL;
338	/* Someone may be waiting for the last nfsiod to terminate. */
339	if (--nfs_numasync == 0)
340		wakeup(&nfs_numasync);
341	mtx_unlock(&nfs_iod_mtx);
342	if ((error == 0) || (error == EWOULDBLOCK))
343		kproc_exit(0);
344	/* Abnormal termination */
345	kproc_exit(1);
346}
347