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: src/sys/nfsclient/nfs_nfsiod.c,v 1.93 2008/11/02 17:00:23 trhodes Exp $");
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 <rpc/rpcclnt.h>
67
68 #include <nfs/xdr_subs.h>
69 #include <nfs/rpcv2.h>
70 #include <nfs/nfsproto.h>
71 #include <nfsclient/nfs.h>
72 #include <nfsclient/nfsm_subs.h>
73 #include <nfsclient/nfsmount.h>
74 #include <nfsclient/nfsnode.h>
75 #include <nfsclient/nfs_lock.h>
76
77 static MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure");
78
79 static void nfssvc_iod(void *);
80
81 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
82
83 SYSCTL_DECL(_vfs_nfs);
84
85 /* Maximum number of seconds a nfsiod kthread will sleep before exiting */
86 static unsigned int nfs_iodmaxidle = 120;
87 SYSCTL_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 */
91 unsigned int nfs_iodmax = 20;
92
93 /* Minimum number of nfsiod kthreads to keep as spares */
94 static unsigned int nfs_iodmin = 0;
95
96 static int
97 sysctl_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();
120 out:
121 mtx_unlock(&nfs_iod_mtx);
122 return (0);
123 }
124 SYSCTL_PROC(_vfs_nfs, 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
129 static int
130 sysctl_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])
154 wakeup(&nfs_iodwant[iod]);
155 iod--;
156 }
157 out:
158 mtx_unlock(&nfs_iod_mtx);
159 return (0);
160 }
161 SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
162 sizeof (nfs_iodmax), sysctl_iodmax, "IU",
163 "Max number of nfsiod kthreads");
164
165 int
166 nfs_nfsiodnew(void)
167 {
168 int error, i;
169 int newiod;
170
171 if (nfs_numasync >= nfs_iodmax)
172 return (-1);
173 newiod = -1;
174 for (i = 0; i < nfs_iodmax; i++)
175 if (nfs_asyncdaemon[i] == 0) {
176 nfs_asyncdaemon[i]++;
177 newiod = i;
178 break;
179 }
180 if (newiod == -1)
181 return (-1);
182 mtx_unlock(&nfs_iod_mtx);
183 error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID,
184 0, "nfsiod %d", newiod);
185 mtx_lock(&nfs_iod_mtx);
186 if (error)
187 return (-1);
188 nfs_numasync++;
189 return (newiod);
190 }
191
192 static void
193 nfsiod_setup(void *dummy)
194 {
195 int i;
196 int error;
197
198 TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin);
199 mtx_lock(&nfs_iod_mtx);
200 /* Silently limit the start number of nfsiod's */
201 if (nfs_iodmin > NFS_MAXASYNCDAEMON)
202 nfs_iodmin = NFS_MAXASYNCDAEMON;
203
204 for (i = 0; i < nfs_iodmin; i++) {
205 error = nfs_nfsiodnew();
206 if (error == -1)
207 panic("nfsiod_setup: nfs_nfsiodnew failed");
208 }
209 mtx_unlock(&nfs_iod_mtx);
210 }
211 SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
212
213 static int nfs_defect = 0;
214 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0,
215 "Allow nfsiods to migrate serving different mounts");
216
217 /*
218 * Asynchronous I/O daemons for client nfs.
219 * They do read-ahead and write-behind operations on the block I/O cache.
220 * Returns if we hit the timeout defined by the iodmaxidle sysctl.
221 */
222 static void
223 nfssvc_iod(void *instance)
224 {
225 struct buf *bp;
226 struct nfsmount *nmp;
227 int myiod, timo;
228 int error = 0;
229
230 mtx_lock(&nfs_iod_mtx);
231 myiod = (int *)instance - nfs_asyncdaemon;
232 /*
233 * Main loop
234 */
235 for (;;) {
236 while (((nmp = nfs_iodmount[myiod]) == NULL)
237 || !TAILQ_FIRST(&nmp->nm_bufq)) {
238 if (myiod >= nfs_iodmax)
239 goto finish;
240 if (nmp)
241 nmp->nm_bufqiods--;
242 nfs_iodwant[myiod] = curthread->td_proc;
243 nfs_iodmount[myiod] = NULL;
244 /*
245 * Always keep at least nfs_iodmin kthreads.
246 */
247 timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
248 error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH,
249 "-", timo);
250 if (error) {
251 nmp = nfs_iodmount[myiod];
252 /*
253 * Rechecking the nm_bufq closes a rare race where the
254 * nfsiod is woken up at the exact time the idle timeout
255 * fires
256 */
257 if (nmp && TAILQ_FIRST(&nmp->nm_bufq))
258 error = 0;
259 break;
260 }
261 }
262 if (error)
263 break;
264 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
265 int giant_locked = 0;
266
267 /* Take one off the front of the list */
268 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
269 nmp->nm_bufqlen--;
270 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) {
271 nmp->nm_bufqwant = 0;
272 wakeup(&nmp->nm_bufq);
273 }
274 mtx_unlock(&nfs_iod_mtx);
275 if (NFS_ISV4(bp->b_vp)) {
276 giant_locked = 1;
277 mtx_lock(&Giant);
278 }
279 if (bp->b_flags & B_DIRECT) {
280 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
281 (void)nfs_doio_directwrite(bp);
282 } else {
283 if (bp->b_iocmd == BIO_READ)
284 (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL);
285 else
286 (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL);
287 }
288 if (giant_locked)
289 mtx_unlock(&Giant);
290 mtx_lock(&nfs_iod_mtx);
291 /*
292 * If there are more than one iod on this mount, then defect
293 * so that the iods can be shared out fairly between the mounts
294 */
295 if (nfs_defect && nmp->nm_bufqiods > 1) {
296 NFS_DPF(ASYNCIO,
297 ("nfssvc_iod: iod %d defecting from mount %p\n",
298 myiod, nmp));
299 nfs_iodmount[myiod] = NULL;
300 nmp->nm_bufqiods--;
301 break;
302 }
303 }
304 }
305 finish:
306 nfs_asyncdaemon[myiod] = 0;
307 if (nmp)
308 nmp->nm_bufqiods--;
309 nfs_iodwant[myiod] = NULL;
310 nfs_iodmount[myiod] = NULL;
311 /* Someone may be waiting for the last nfsiod to terminate. */
312 if (--nfs_numasync == 0)
313 wakeup(&nfs_numasync);
314 mtx_unlock(&nfs_iod_mtx);
315 if ((error == 0) || (error == EWOULDBLOCK))
316 kproc_exit(0);
317 /* Abnormal termination */
318 kproc_exit(1);
319 }
320
|