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 *	from 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 <fs/nfs/nfsport.h>
68#include <fs/nfsclient/nfsmount.h>
69#include <fs/nfsclient/nfs.h>
70#include <fs/nfsclient/nfsnode.h>
71
72extern struct mtx	ncl_iod_mutex;
73extern struct task	ncl_nfsiodnew_task;
74
75int ncl_numasync;
76enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
77struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
78
79static void	nfssvc_iod(void *);
80
81static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
82
83SYSCTL_DECL(_vfs_nfs);
84
85/* Maximum number of seconds a nfsiod kthread will sleep before exiting */
86static unsigned int nfs_iodmaxidle = 120;
87SYSCTL_UINT(_vfs_nfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0,
88    "Max number of seconds an nfsiod kthread will sleep before exiting");
89
90/* Maximum number of nfsiod kthreads */
91unsigned int ncl_iodmax = 20;
92
93/* Minimum number of nfsiod kthreads to keep as spares */
94static unsigned int nfs_iodmin = 0;
95
96static int nfs_nfsiodnew_sync(void);
97
98static int
99sysctl_iodmin(SYSCTL_HANDLER_ARGS)
100{
101	int error, i;
102	int newmin;
103
104	newmin = nfs_iodmin;
105	error = sysctl_handle_int(oidp, &newmin, 0, req);
106	if (error || (req->newptr == NULL))
107		return (error);
108	mtx_lock(&ncl_iod_mutex);
109	if (newmin > ncl_iodmax) {
110		error = EINVAL;
111		goto out;
112	}
113	nfs_iodmin = newmin;
114	if (ncl_numasync >= nfs_iodmin)
115		goto out;
116	/*
117	 * If the current number of nfsiod is lower
118	 * than the new minimum, create some more.
119	 */
120	for (i = nfs_iodmin - ncl_numasync; i > 0; i--)
121		nfs_nfsiodnew_sync();
122out:
123	mtx_unlock(&ncl_iod_mutex);
124	return (0);
125}
126SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
127    sizeof (nfs_iodmin), sysctl_iodmin, "IU",
128    "Min number of nfsiod kthreads to keep as spares");
129
130static int
131sysctl_iodmax(SYSCTL_HANDLER_ARGS)
132{
133	int error, i;
134	int iod, newmax;
135
136	newmax = ncl_iodmax;
137	error = sysctl_handle_int(oidp, &newmax, 0, req);
138	if (error || (req->newptr == NULL))
139		return (error);
140	if (newmax > NFS_MAXASYNCDAEMON)
141		return (EINVAL);
142	mtx_lock(&ncl_iod_mutex);
143	ncl_iodmax = newmax;
144	if (ncl_numasync <= ncl_iodmax)
145		goto out;
146	/*
147	 * If there are some asleep nfsiods that should
148	 * exit, wakeup() them so that they check ncl_iodmax
149	 * and exit.  Those who are active will exit as
150	 * soon as they finish I/O.
151	 */
152	iod = ncl_numasync - 1;
153	for (i = 0; i < ncl_numasync - ncl_iodmax; i++) {
154		if (ncl_iodwant[iod] == NFSIOD_AVAILABLE)
155			wakeup(&ncl_iodwant[iod]);
156		iod--;
157	}
158out:
159	mtx_unlock(&ncl_iod_mutex);
160	return (0);
161}
162SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
163    sizeof (ncl_iodmax), sysctl_iodmax, "IU",
164    "Max number of nfsiod kthreads");
165
166static int
167nfs_nfsiodnew_sync(void)
168{
169	int error, i;
170
171	mtx_assert(&ncl_iod_mutex, MA_OWNED);
172	for (i = 0; i < ncl_iodmax; i++) {
173		if (nfs_asyncdaemon[i] == 0) {
174			nfs_asyncdaemon[i] = 1;
175			break;
176		}
177	}
178	if (i == ncl_iodmax)
179		return (0);
180	mtx_unlock(&ncl_iod_mutex);
181	error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL,
182	    RFHIGHPID, 0, "newnfs %d", i);
183	mtx_lock(&ncl_iod_mutex);
184	if (error == 0) {
185		ncl_numasync++;
186		ncl_iodwant[i] = NFSIOD_AVAILABLE;
187	} else
188		nfs_asyncdaemon[i] = 0;
189	return (error);
190}
191
192void
193ncl_nfsiodnew_tq(__unused void *arg, int pending)
194{
195
196	mtx_lock(&ncl_iod_mutex);
197	while (pending > 0) {
198		pending--;
199		nfs_nfsiodnew_sync();
200	}
201	mtx_unlock(&ncl_iod_mutex);
202}
203
204void
205ncl_nfsiodnew(void)
206{
207
208	mtx_assert(&ncl_iod_mutex, MA_OWNED);
209	taskqueue_enqueue(taskqueue_thread, &ncl_nfsiodnew_task);
210}
211
212static void
213nfsiod_setup(void *dummy)
214{
215	int error;
216
217	TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin);
218	nfscl_init();
219	mtx_lock(&ncl_iod_mutex);
220	/* Silently limit the start number of nfsiod's */
221	if (nfs_iodmin > NFS_MAXASYNCDAEMON)
222		nfs_iodmin = NFS_MAXASYNCDAEMON;
223
224	while (ncl_numasync < nfs_iodmin) {
225		error = nfs_nfsiodnew_sync();
226		if (error == -1)
227			panic("nfsiod_setup: nfs_nfsiodnew failed");
228	}
229	mtx_unlock(&ncl_iod_mutex);
230}
231SYSINIT(newnfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
232
233static int nfs_defect = 0;
234SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0,
235    "Allow nfsiods to migrate serving different mounts");
236
237/*
238 * Asynchronous I/O daemons for client nfs.
239 * They do read-ahead and write-behind operations on the block I/O cache.
240 * Returns if we hit the timeout defined by the iodmaxidle sysctl.
241 */
242static void
243nfssvc_iod(void *instance)
244{
245	struct buf *bp;
246	struct nfsmount *nmp;
247	int myiod, timo;
248	int error = 0;
249
250	mtx_lock(&ncl_iod_mutex);
251	myiod = (int *)instance - nfs_asyncdaemon;
252	/*
253	 * Main loop
254	 */
255	for (;;) {
256	    while (((nmp = ncl_iodmount[myiod]) == NULL)
257		   || !TAILQ_FIRST(&nmp->nm_bufq)) {
258		if (myiod >= ncl_iodmax)
259			goto finish;
260		if (nmp)
261			nmp->nm_bufqiods--;
262		if (ncl_iodwant[myiod] == NFSIOD_NOT_AVAILABLE)
263			ncl_iodwant[myiod] = NFSIOD_AVAILABLE;
264		ncl_iodmount[myiod] = NULL;
265		/*
266		 * Always keep at least nfs_iodmin kthreads.
267		 */
268		timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
269		error = msleep(&ncl_iodwant[myiod], &ncl_iod_mutex, PWAIT | PCATCH,
270		    "-", timo);
271		if (error) {
272			nmp = ncl_iodmount[myiod];
273			/*
274			 * Rechecking the nm_bufq closes a rare race where the
275			 * nfsiod is woken up at the exact time the idle timeout
276			 * fires
277			 */
278			if (nmp && TAILQ_FIRST(&nmp->nm_bufq))
279				error = 0;
280			break;
281		}
282	    }
283	    if (error)
284		    break;
285	    while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
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 <= ncl_numasync) {
290		    nmp->nm_bufqwant = 0;
291		    wakeup(&nmp->nm_bufq);
292		}
293		mtx_unlock(&ncl_iod_mutex);
294		if (bp->b_flags & B_DIRECT) {
295			KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
296			(void)ncl_doio_directwrite(bp);
297		} else {
298			if (bp->b_iocmd == BIO_READ)
299				(void) ncl_doio(bp->b_vp, bp, bp->b_rcred,
300				    NULL, 0);
301			else
302				(void) ncl_doio(bp->b_vp, bp, bp->b_wcred,
303				    NULL, 0);
304		}
305		mtx_lock(&ncl_iod_mutex);
306		/*
307		 * Make sure the nmp hasn't been dismounted as soon as
308		 * ncl_doio() completes for the last buffer.
309		 */
310		nmp = ncl_iodmount[myiod];
311		if (nmp == NULL)
312			break;
313
314		/*
315		 * If there are more than one iod on this mount, then defect
316		 * so that the iods can be shared out fairly between the mounts
317		 */
318		if (nfs_defect && nmp->nm_bufqiods > 1) {
319		    NFS_DPF(ASYNCIO,
320			    ("nfssvc_iod: iod %d defecting from mount %p\n",
321			     myiod, nmp));
322		    ncl_iodmount[myiod] = NULL;
323		    nmp->nm_bufqiods--;
324		    break;
325		}
326	    }
327	}
328finish:
329	nfs_asyncdaemon[myiod] = 0;
330	if (nmp)
331	    nmp->nm_bufqiods--;
332	ncl_iodwant[myiod] = NFSIOD_NOT_AVAILABLE;
333	ncl_iodmount[myiod] = NULL;
334	/* Someone may be waiting for the last nfsiod to terminate. */
335	if (--ncl_numasync == 0)
336		wakeup(&ncl_numasync);
337	mtx_unlock(&ncl_iod_mutex);
338	if ((error == 0) || (error == EWOULDBLOCK))
339		kproc_exit(0);
340	/* Abnormal termination */
341	kproc_exit(1);
342}
343