nfs_nfsiod.c revision 158739
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: head/sys/nfsclient/nfs_nfsiod.c 158739 2006-05-19 00:04:24Z mohans $"); 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 77static MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure"); 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 89/* Maximum number of nfsiod kthreads */ 90unsigned int nfs_iodmax = 20; 91 92/* Minimum number of nfsiod kthreads to keep as spares */ 93static unsigned int nfs_iodmin = 4; 94 95static int 96sysctl_iodmin(SYSCTL_HANDLER_ARGS) 97{ 98 int error, i; 99 int newmin; 100 101 newmin = nfs_iodmin; 102 error = sysctl_handle_int(oidp, &newmin, 0, req); 103 if (error || (req->newptr == NULL)) 104 return (error); 105 mtx_lock(&nfs_iod_mtx); 106 if (newmin > nfs_iodmax) { 107 error = EINVAL; 108 goto out; 109 } 110 nfs_iodmin = newmin; 111 if (nfs_numasync >= nfs_iodmin) 112 goto out; 113 /* 114 * If the current number of nfsiod is lower 115 * than the new minimum, create some more. 116 */ 117 for (i = nfs_iodmin - nfs_numasync; i > 0; i--) 118 nfs_nfsiodnew(); 119out: 120 mtx_unlock(&nfs_iod_mtx); 121 return (0); 122} 123SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, 124 sizeof (nfs_iodmin), sysctl_iodmin, "IU", ""); 125 126 127static int 128sysctl_iodmax(SYSCTL_HANDLER_ARGS) 129{ 130 int error, i; 131 int iod, newmax; 132 133 newmax = nfs_iodmax; 134 error = sysctl_handle_int(oidp, &newmax, 0, req); 135 if (error || (req->newptr == NULL)) 136 return (error); 137 if (newmax > NFS_MAXASYNCDAEMON) 138 return (EINVAL); 139 mtx_lock(&nfs_iod_mtx); 140 nfs_iodmax = newmax; 141 if (nfs_numasync <= nfs_iodmax) 142 goto out; 143 /* 144 * If there are some asleep nfsiods that should 145 * exit, wakeup() them so that they check nfs_iodmax 146 * and exit. Those who are active will exit as 147 * soon as they finish I/O. 148 */ 149 iod = nfs_numasync - 1; 150 for (i = 0; i < nfs_numasync - nfs_iodmax; i++) { 151 if (nfs_iodwant[iod]) 152 wakeup(&nfs_iodwant[iod]); 153 iod--; 154 } 155out: 156 mtx_unlock(&nfs_iod_mtx); 157 return (0); 158} 159SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0, 160 sizeof (nfs_iodmax), sysctl_iodmax, "IU", ""); 161 162int 163nfs_nfsiodnew(void) 164{ 165 int error, i; 166 int newiod; 167 168 if (nfs_numasync >= nfs_iodmax) 169 return (-1); 170 newiod = -1; 171 for (i = 0; i < nfs_iodmax; i++) 172 if (nfs_asyncdaemon[i] == 0) { 173 nfs_asyncdaemon[i]++; 174 newiod = i; 175 break; 176 } 177 if (newiod == -1) 178 return (-1); 179 mtx_unlock(&nfs_iod_mtx); 180 error = kthread_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID, 181 0, "nfsiod %d", newiod); 182 mtx_lock(&nfs_iod_mtx); 183 if (error) 184 return (-1); 185 nfs_numasync++; 186 return (newiod); 187} 188 189static void 190nfsiod_setup(void *dummy) 191{ 192 int i; 193 int error; 194 195 TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin); 196 mtx_lock(&nfs_iod_mtx); 197 /* Silently limit the start number of nfsiod's */ 198 if (nfs_iodmin > NFS_MAXASYNCDAEMON) 199 nfs_iodmin = NFS_MAXASYNCDAEMON; 200 201 for (i = 0; i < nfs_iodmin; i++) { 202 error = nfs_nfsiodnew(); 203 if (error == -1) 204 panic("nfsiod_setup: nfs_nfsiodnew failed"); 205 } 206 mtx_unlock(&nfs_iod_mtx); 207} 208SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); 209 210static int nfs_defect = 0; 211SYSCTL_INT(_vfs_nfs, 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 */ 218static void 219nfssvc_iod(void *instance) 220{ 221 struct buf *bp; 222 struct nfsmount *nmp; 223 int myiod, timo; 224 int error = 0; 225 226 mtx_lock(&nfs_iod_mtx); 227 myiod = (int *)instance - nfs_asyncdaemon; 228 /* 229 * Main loop 230 */ 231 for (;;) { 232 while (((nmp = nfs_iodmount[myiod]) == NULL 233 || !TAILQ_FIRST(&nmp->nm_bufq)) 234 && error == 0) { 235 if (myiod >= nfs_iodmax) 236 goto finish; 237 if (nmp) 238 nmp->nm_bufqiods--; 239 nfs_iodwant[myiod] = curthread->td_proc; 240 nfs_iodmount[myiod] = NULL; 241 /* 242 * Always keep at least nfs_iodmin kthreads. 243 */ 244 timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz; 245 error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH, 246 "-", timo); 247 } 248 if (error) 249 break; 250 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 251 /* Take one off the front of the list */ 252 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 253 nmp->nm_bufqlen--; 254 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 255 nmp->nm_bufqwant = 0; 256 wakeup(&nmp->nm_bufq); 257 } 258 mtx_unlock(&nfs_iod_mtx); 259 if (bp->b_flags & B_DIRECT) { 260 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); 261 (void)nfs_doio_directwrite(bp); 262 } else { 263 if (bp->b_iocmd == BIO_READ) 264 (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL); 265 else 266 (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL); 267 } 268 mtx_lock(&nfs_iod_mtx); 269 /* 270 * If there are more than one iod on this mount, then defect 271 * so that the iods can be shared out fairly between the mounts 272 */ 273 if (nfs_defect && nmp->nm_bufqiods > 1) { 274 NFS_DPF(ASYNCIO, 275 ("nfssvc_iod: iod %d defecting from mount %p\n", 276 myiod, nmp)); 277 nfs_iodmount[myiod] = NULL; 278 nmp->nm_bufqiods--; 279 break; 280 } 281 } 282 } 283finish: 284 nfs_asyncdaemon[myiod] = 0; 285 if (nmp) 286 nmp->nm_bufqiods--; 287 nfs_iodwant[myiod] = NULL; 288 nfs_iodmount[myiod] = NULL; 289 /* Someone may be waiting for the last nfsiod to terminate. */ 290 if (--nfs_numasync == 0) 291 wakeup(&nfs_numasync); 292 mtx_unlock(&nfs_iod_mtx); 293 if ((error == 0) || (error == EWOULDBLOCK)) 294 kthread_exit(0); 295 /* Abnormal termination */ 296 kthread_exit(1); 297} 298