xref: /freebsd/sys/fs/nfsclient/nfs_clnfsiod.c (revision d8a0fe102c0cfdfcd5b818f850eff09d8536c9bc)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_syscalls.c	8.5 (Berkeley) 3/30/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/file.h>
46 #include <sys/vnode.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/proc.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/mbuf.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/domain.h>
56 #include <sys/protosw.h>
57 #include <sys/namei.h>
58 #include <sys/unistd.h>
59 #include <sys/kthread.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 #include <sys/mutex.h>
63 #include <sys/taskqueue.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/tcp.h>
67 
68 #include <fs/nfs/nfsport.h>
69 #include <fs/nfsclient/nfsmount.h>
70 #include <fs/nfsclient/nfs.h>
71 #include <fs/nfsclient/nfsnode.h>
72 
73 extern struct mtx	ncl_iod_mutex;
74 extern struct task	ncl_nfsiodnew_task;
75 
76 int ncl_numasync;
77 enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
78 struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
79 
80 static void	nfssvc_iod(void *);
81 
82 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
83 
84 SYSCTL_DECL(_vfs_nfs);
85 
86 /* Maximum number of seconds a nfsiod kthread will sleep before exiting */
87 static unsigned int nfs_iodmaxidle = 120;
88 SYSCTL_UINT(_vfs_nfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0,
89     "Max number of seconds an nfsiod kthread will sleep before exiting");
90 
91 /* Maximum number of nfsiod kthreads */
92 unsigned int ncl_iodmax = 20;
93 
94 /* Minimum number of nfsiod kthreads to keep as spares */
95 static unsigned int nfs_iodmin = 0;
96 
97 static int nfs_nfsiodnew_sync(void);
98 
99 static int
100 sysctl_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 	mtx_lock(&ncl_iod_mutex);
110 	if (newmin > ncl_iodmax) {
111 		error = EINVAL;
112 		goto out;
113 	}
114 	nfs_iodmin = newmin;
115 	if (ncl_numasync >= nfs_iodmin)
116 		goto out;
117 	/*
118 	 * If the current number of nfsiod is lower
119 	 * than the new minimum, create some more.
120 	 */
121 	for (i = nfs_iodmin - ncl_numasync; i > 0; i--)
122 		nfs_nfsiodnew_sync();
123 out:
124 	mtx_unlock(&ncl_iod_mutex);
125 	return (0);
126 }
127 SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
128     sizeof (nfs_iodmin), sysctl_iodmin, "IU",
129     "Min number of nfsiod kthreads to keep as spares");
130 
131 static int
132 sysctl_iodmax(SYSCTL_HANDLER_ARGS)
133 {
134 	int error, i;
135 	int iod, newmax;
136 
137 	newmax = ncl_iodmax;
138 	error = sysctl_handle_int(oidp, &newmax, 0, req);
139 	if (error || (req->newptr == NULL))
140 		return (error);
141 	if (newmax > NFS_MAXASYNCDAEMON)
142 		return (EINVAL);
143 	mtx_lock(&ncl_iod_mutex);
144 	ncl_iodmax = newmax;
145 	if (ncl_numasync <= ncl_iodmax)
146 		goto out;
147 	/*
148 	 * If there are some asleep nfsiods that should
149 	 * exit, wakeup() them so that they check ncl_iodmax
150 	 * and exit.  Those who are active will exit as
151 	 * soon as they finish I/O.
152 	 */
153 	iod = ncl_numasync - 1;
154 	for (i = 0; i < ncl_numasync - ncl_iodmax; i++) {
155 		if (ncl_iodwant[iod] == NFSIOD_AVAILABLE)
156 			wakeup(&ncl_iodwant[iod]);
157 		iod--;
158 	}
159 out:
160 	mtx_unlock(&ncl_iod_mutex);
161 	return (0);
162 }
163 SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
164     sizeof (ncl_iodmax), sysctl_iodmax, "IU",
165     "Max number of nfsiod kthreads");
166 
167 static int
168 nfs_nfsiodnew_sync(void)
169 {
170 	int error, i;
171 
172 	mtx_assert(&ncl_iod_mutex, MA_OWNED);
173 	for (i = 0; i < ncl_iodmax; i++) {
174 		if (nfs_asyncdaemon[i] == 0) {
175 			nfs_asyncdaemon[i] = 1;
176 			break;
177 		}
178 	}
179 	if (i == ncl_iodmax)
180 		return (0);
181 	mtx_unlock(&ncl_iod_mutex);
182 	error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL,
183 	    RFHIGHPID, 0, "newnfs %d", i);
184 	mtx_lock(&ncl_iod_mutex);
185 	if (error == 0) {
186 		ncl_numasync++;
187 		ncl_iodwant[i] = NFSIOD_AVAILABLE;
188 	} else
189 		nfs_asyncdaemon[i] = 0;
190 	return (error);
191 }
192 
193 void
194 ncl_nfsiodnew_tq(__unused void *arg, int pending)
195 {
196 
197 	mtx_lock(&ncl_iod_mutex);
198 	while (pending > 0) {
199 		pending--;
200 		nfs_nfsiodnew_sync();
201 	}
202 	mtx_unlock(&ncl_iod_mutex);
203 }
204 
205 void
206 ncl_nfsiodnew(void)
207 {
208 
209 	mtx_assert(&ncl_iod_mutex, MA_OWNED);
210 	taskqueue_enqueue(taskqueue_thread, &ncl_nfsiodnew_task);
211 }
212 
213 static void
214 nfsiod_setup(void *dummy)
215 {
216 	int error;
217 
218 	TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin);
219 	nfscl_init();
220 	mtx_lock(&ncl_iod_mutex);
221 	/* Silently limit the start number of nfsiod's */
222 	if (nfs_iodmin > NFS_MAXASYNCDAEMON)
223 		nfs_iodmin = NFS_MAXASYNCDAEMON;
224 
225 	while (ncl_numasync < nfs_iodmin) {
226 		error = nfs_nfsiodnew_sync();
227 		if (error == -1)
228 			panic("nfsiod_setup: nfs_nfsiodnew failed");
229 	}
230 	mtx_unlock(&ncl_iod_mutex);
231 }
232 SYSINIT(newnfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
233 
234 static int nfs_defect = 0;
235 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0,
236     "Allow nfsiods to migrate serving different mounts");
237 
238 /*
239  * Asynchronous I/O daemons for client nfs.
240  * They do read-ahead and write-behind operations on the block I/O cache.
241  * Returns if we hit the timeout defined by the iodmaxidle sysctl.
242  */
243 static void
244 nfssvc_iod(void *instance)
245 {
246 	struct buf *bp;
247 	struct nfsmount *nmp;
248 	int myiod, timo;
249 	int error = 0;
250 
251 	mtx_lock(&ncl_iod_mutex);
252 	myiod = (int *)instance - nfs_asyncdaemon;
253 	/*
254 	 * Main loop
255 	 */
256 	for (;;) {
257 	    while (((nmp = ncl_iodmount[myiod]) == NULL)
258 		   || !TAILQ_FIRST(&nmp->nm_bufq)) {
259 		if (myiod >= ncl_iodmax)
260 			goto finish;
261 		if (nmp)
262 			nmp->nm_bufqiods--;
263 		if (ncl_iodwant[myiod] == NFSIOD_NOT_AVAILABLE)
264 			ncl_iodwant[myiod] = NFSIOD_AVAILABLE;
265 		ncl_iodmount[myiod] = NULL;
266 		/*
267 		 * Always keep at least nfs_iodmin kthreads.
268 		 */
269 		timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
270 		error = msleep(&ncl_iodwant[myiod], &ncl_iod_mutex, PWAIT | PCATCH,
271 		    "-", timo);
272 		if (error) {
273 			nmp = ncl_iodmount[myiod];
274 			/*
275 			 * Rechecking the nm_bufq closes a rare race where the
276 			 * nfsiod is woken up at the exact time the idle timeout
277 			 * fires
278 			 */
279 			if (nmp && TAILQ_FIRST(&nmp->nm_bufq))
280 				error = 0;
281 			break;
282 		}
283 	    }
284 	    if (error)
285 		    break;
286 	    while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
287 		/* Take one off the front of the list */
288 		TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
289 		nmp->nm_bufqlen--;
290 		if (nmp->nm_bufqwant && nmp->nm_bufqlen <= ncl_numasync) {
291 		    nmp->nm_bufqwant = 0;
292 		    wakeup(&nmp->nm_bufq);
293 		}
294 		mtx_unlock(&ncl_iod_mutex);
295 		if (bp->b_flags & B_DIRECT) {
296 			KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
297 			(void)ncl_doio_directwrite(bp);
298 		} else {
299 			if (bp->b_iocmd == BIO_READ)
300 				(void) ncl_doio(bp->b_vp, bp, bp->b_rcred,
301 				    NULL, 0);
302 			else
303 				(void) ncl_doio(bp->b_vp, bp, bp->b_wcred,
304 				    NULL, 0);
305 		}
306 		mtx_lock(&ncl_iod_mutex);
307 		/*
308 		 * Make sure the nmp hasn't been dismounted as soon as
309 		 * ncl_doio() completes for the last buffer.
310 		 */
311 		nmp = ncl_iodmount[myiod];
312 		if (nmp == NULL)
313 			break;
314 
315 		/*
316 		 * If there are more than one iod on this mount, then defect
317 		 * so that the iods can be shared out fairly between the mounts
318 		 */
319 		if (nfs_defect && nmp->nm_bufqiods > 1) {
320 		    NFS_DPF(ASYNCIO,
321 			    ("nfssvc_iod: iod %d defecting from mount %p\n",
322 			     myiod, nmp));
323 		    ncl_iodmount[myiod] = NULL;
324 		    nmp->nm_bufqiods--;
325 		    break;
326 		}
327 	    }
328 	}
329 finish:
330 	nfs_asyncdaemon[myiod] = 0;
331 	if (nmp)
332 	    nmp->nm_bufqiods--;
333 	ncl_iodwant[myiod] = NFSIOD_NOT_AVAILABLE;
334 	ncl_iodmount[myiod] = NULL;
335 	/* Someone may be waiting for the last nfsiod to terminate. */
336 	if (--ncl_numasync == 0)
337 		wakeup(&ncl_numasync);
338 	mtx_unlock(&ncl_iod_mutex);
339 	if ((error == 0) || (error == EWOULDBLOCK))
340 		kproc_exit(0);
341 	/* Abnormal termination */
342 	kproc_exit(1);
343 }
344