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: releng/8.0/sys/fs/nfsclient/nfs_clnfsiod.c 191783 2009-05-04 15:23:58Z rmacklem $");
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
63 #include <netinet/in.h>
64 #include <netinet/tcp.h>
65
66 #include <fs/nfs/nfsport.h>
67 #include <fs/nfsclient/nfsmount.h>
68 #include <fs/nfsclient/nfs.h>
69 #include <fs/nfsclient/nfsnode.h>
70 #include <fs/nfsclient/nfs_lock.h>
71
72 extern struct mtx ncl_iod_mutex;
73
74 int ncl_numasync;
75 struct proc *ncl_iodwant[NFS_MAXRAHEAD];
76 struct nfsmount *ncl_iodmount[NFS_MAXRAHEAD];
77
78 static void nfssvc_iod(void *);
79
80 static int nfs_asyncdaemon[NFS_MAXRAHEAD];
81
82 SYSCTL_DECL(_vfs_newnfs);
83
84 /* Maximum number of seconds a nfsiod kthread will sleep before exiting */
85 static unsigned int ncl_iodmaxidle = 120;
86 SYSCTL_UINT(_vfs_newnfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &ncl_iodmaxidle, 0, "");
87
88 /* Maximum number of nfsiod kthreads */
89 unsigned int ncl_iodmax = NFS_MAXRAHEAD;
90
91 /* Minimum number of nfsiod kthreads to keep as spares */
92 static unsigned int nfs_iodmin = 0;
93
94 static int
95 sysctl_iodmin(SYSCTL_HANDLER_ARGS)
96 {
97 int error, i;
98 int newmin;
99
100 newmin = nfs_iodmin;
101 error = sysctl_handle_int(oidp, &newmin, 0, req);
102 if (error || (req->newptr == NULL))
103 return (error);
104 mtx_lock(&ncl_iod_mutex);
105 if (newmin > ncl_iodmax) {
106 error = EINVAL;
107 goto out;
108 }
109 nfs_iodmin = newmin;
110 if (ncl_numasync >= nfs_iodmin)
111 goto out;
112 /*
113 * If the current number of nfsiod is lower
114 * than the new minimum, create some more.
115 */
116 for (i = nfs_iodmin - ncl_numasync; i > 0; i--)
117 ncl_nfsiodnew();
118 out:
119 mtx_unlock(&ncl_iod_mutex);
120 return (0);
121 }
122 SYSCTL_PROC(_vfs_newnfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
123 sizeof (nfs_iodmin), sysctl_iodmin, "IU", "");
124
125
126 static int
127 sysctl_iodmax(SYSCTL_HANDLER_ARGS)
128 {
129 int error, i;
130 int iod, newmax;
131
132 newmax = ncl_iodmax;
133 error = sysctl_handle_int(oidp, &newmax, 0, req);
134 if (error || (req->newptr == NULL))
135 return (error);
136 if (newmax > NFS_MAXRAHEAD)
137 return (EINVAL);
138 mtx_lock(&ncl_iod_mutex);
139 ncl_iodmax = newmax;
140 if (ncl_numasync <= ncl_iodmax)
141 goto out;
142 /*
143 * If there are some asleep nfsiods that should
144 * exit, wakeup() them so that they check ncl_iodmax
145 * and exit. Those who are active will exit as
146 * soon as they finish I/O.
147 */
148 iod = ncl_numasync - 1;
149 for (i = 0; i < ncl_numasync - ncl_iodmax; i++) {
150 if (ncl_iodwant[iod])
151 wakeup(&ncl_iodwant[iod]);
152 iod--;
153 }
154 out:
155 mtx_unlock(&ncl_iod_mutex);
156 return (0);
157 }
158 SYSCTL_PROC(_vfs_newnfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
159 sizeof (ncl_iodmax), sysctl_iodmax, "IU", "");
160
161 int
162 ncl_nfsiodnew(void)
163 {
164 int error, i;
165 int newiod;
166
167 if (ncl_numasync >= ncl_iodmax)
168 return (-1);
169 newiod = -1;
170 for (i = 0; i < ncl_iodmax; i++)
171 if (nfs_asyncdaemon[i] == 0) {
172 nfs_asyncdaemon[i]++;
173 newiod = i;
174 break;
175 }
176 if (newiod == -1)
177 return (-1);
178 mtx_unlock(&ncl_iod_mutex);
179 error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID,
180 0, "nfsiod %d", newiod);
181 mtx_lock(&ncl_iod_mutex);
182 if (error)
183 return (-1);
184 ncl_numasync++;
185 return (newiod);
186 }
187
188 static void
189 nfsiod_setup(void *dummy)
190 {
191 int i;
192 int error;
193
194 TUNABLE_INT_FETCH("vfs.newnfs.iodmin", &nfs_iodmin);
195 nfscl_init();
196 mtx_lock(&ncl_iod_mutex);
197 /* Silently limit the start number of nfsiod's */
198 if (nfs_iodmin > NFS_MAXRAHEAD)
199 nfs_iodmin = NFS_MAXRAHEAD;
200
201 for (i = 0; i < nfs_iodmin; i++) {
202 error = ncl_nfsiodnew();
203 if (error == -1)
204 panic("newnfsiod_setup: ncl_nfsiodnew failed");
205 }
206 mtx_unlock(&ncl_iod_mutex);
207 }
208 SYSINIT(newnfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
209
210 static int nfs_defect = 0;
211 SYSCTL_INT(_vfs_newnfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
212
213 /*
214 * Asynchronous I/O daemons for client nfs.
215 * They do read-ahead and write-behind operations on the block I/O cache.
216 * Returns if we hit the timeout defined by the iodmaxidle sysctl.
217 */
218 static void
219 nfssvc_iod(void *instance)
220 {
221 struct buf *bp;
222 struct nfsmount *nmp;
223 int myiod, timo;
224 int error = 0;
225
226 mtx_lock(&ncl_iod_mutex);
227 myiod = (int *)instance - nfs_asyncdaemon;
228 /*
229 * Main loop
230 */
231 for (;;) {
232 while (((nmp = ncl_iodmount[myiod]) == NULL)
233 || !TAILQ_FIRST(&nmp->nm_bufq)) {
234 if (myiod >= ncl_iodmax)
235 goto finish;
236 if (nmp)
237 nmp->nm_bufqiods--;
238 ncl_iodwant[myiod] = curthread->td_proc;
239 ncl_iodmount[myiod] = NULL;
240 /*
241 * Always keep at least nfs_iodmin kthreads.
242 */
243 timo = (myiod < nfs_iodmin) ? 0 : ncl_iodmaxidle * hz;
244 error = msleep(&ncl_iodwant[myiod], &ncl_iod_mutex, PWAIT | PCATCH,
245 "-", timo);
246 if (error) {
247 nmp = ncl_iodmount[myiod];
248 /*
249 * Rechecking the nm_bufq closes a rare race where the
250 * nfsiod is woken up at the exact time the idle timeout
251 * fires
252 */
253 if (nmp && TAILQ_FIRST(&nmp->nm_bufq))
254 error = 0;
255 break;
256 }
257 }
258 if (error)
259 break;
260 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
261
262 /* Take one off the front of the list */
263 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
264 nmp->nm_bufqlen--;
265 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= ncl_numasync) {
266 nmp->nm_bufqwant = 0;
267 wakeup(&nmp->nm_bufq);
268 }
269 mtx_unlock(&ncl_iod_mutex);
270 if (bp->b_flags & B_DIRECT) {
271 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
272 (void)ncl_doio_directwrite(bp);
273 } else {
274 if (bp->b_iocmd == BIO_READ)
275 (void) ncl_doio(bp->b_vp, bp, bp->b_rcred, NULL);
276 else
277 (void) ncl_doio(bp->b_vp, bp, bp->b_wcred, NULL);
278 }
279 mtx_lock(&ncl_iod_mutex);
280 /*
281 * If there are more than one iod on this mount, then defect
282 * so that the iods can be shared out fairly between the mounts
283 */
284 if (nfs_defect && nmp->nm_bufqiods > 1) {
285 NFS_DPF(ASYNCIO,
286 ("nfssvc_iod: iod %d defecting from mount %p\n",
287 myiod, nmp));
288 ncl_iodmount[myiod] = NULL;
289 nmp->nm_bufqiods--;
290 break;
291 }
292 }
293 }
294 finish:
295 nfs_asyncdaemon[myiod] = 0;
296 if (nmp)
297 nmp->nm_bufqiods--;
298 ncl_iodwant[myiod] = NULL;
299 ncl_iodmount[myiod] = NULL;
300 /* Someone may be waiting for the last nfsiod to terminate. */
301 if (--ncl_numasync == 0)
302 wakeup(&ncl_numasync);
303 mtx_unlock(&ncl_iod_mutex);
304 if ((error == 0) || (error == EWOULDBLOCK))
305 kproc_exit(0);
306 /* Abnormal termination */
307 kproc_exit(1);
308 }
Cache object: d78c9f8eb8c45ac2b6c1ce514f328037
|