nfs_nfsiod.c revision 151897
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 151897 2005-10-31 15:41:29Z rwatson $"); 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 if (newmin > nfs_iodmax) 106 return (EINVAL); 107 nfs_iodmin = newmin; 108 if (nfs_numasync >= nfs_iodmin) 109 return (0); 110 /* 111 * If the current number of nfsiod is lower 112 * than the new minimum, create some more. 113 */ 114 for (i = nfs_iodmin - nfs_numasync; i > 0; i--) 115 nfs_nfsiodnew(); 116 return (0); 117} 118SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, 119 sizeof (nfs_iodmin), sysctl_iodmin, "IU", ""); 120 121 122static int 123sysctl_iodmax(SYSCTL_HANDLER_ARGS) 124{ 125 int error, i; 126 int iod, newmax; 127 128 newmax = nfs_iodmax; 129 error = sysctl_handle_int(oidp, &newmax, 0, req); 130 if (error || (req->newptr == NULL)) 131 return (error); 132 if (newmax > NFS_MAXASYNCDAEMON) 133 return (EINVAL); 134 nfs_iodmax = newmax; 135 if (nfs_numasync <= nfs_iodmax) 136 return (0); 137 /* 138 * If there are some asleep nfsiods that should 139 * exit, wakeup() them so that they check nfs_iodmax 140 * and exit. Those who are active will exit as 141 * soon as they finish I/O. 142 */ 143 iod = nfs_numasync - 1; 144 for (i = 0; i < nfs_numasync - nfs_iodmax; i++) { 145 if (nfs_iodwant[iod]) 146 wakeup(&nfs_iodwant[iod]); 147 iod--; 148 } 149 return (0); 150} 151SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0, 152 sizeof (nfs_iodmax), sysctl_iodmax, "IU", ""); 153 154int 155nfs_nfsiodnew(void) 156{ 157 int error, i; 158 int newiod; 159 160 if (nfs_numasync >= nfs_iodmax) 161 return (-1); 162 newiod = -1; 163 for (i = 0; i < nfs_iodmax; i++) 164 if (nfs_asyncdaemon[i] == 0) { 165 nfs_asyncdaemon[i]++; 166 newiod = i; 167 break; 168 } 169 if (newiod == -1) 170 return (-1); 171 error = kthread_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID, 172 0, "nfsiod %d", newiod); 173 if (error) 174 return (-1); 175 nfs_numasync++; 176 return (newiod); 177} 178 179static void 180nfsiod_setup(void *dummy) 181{ 182 int i; 183 int error; 184 185 TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin); 186 /* Silently limit the start number of nfsiod's */ 187 if (nfs_iodmin > NFS_MAXASYNCDAEMON) 188 nfs_iodmin = NFS_MAXASYNCDAEMON; 189 190 for (i = 0; i < nfs_iodmin; i++) { 191 error = nfs_nfsiodnew(); 192 if (error == -1) 193 panic("nfsiod_setup: nfs_nfsiodnew failed"); 194 } 195} 196SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); 197 198static int nfs_defect = 0; 199SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 200 201/* 202 * Asynchronous I/O daemons for client nfs. 203 * They do read-ahead and write-behind operations on the block I/O cache. 204 * Returns if we hit the timeout defined by the iodmaxidle sysctl. 205 */ 206static void 207nfssvc_iod(void *instance) 208{ 209 struct buf *bp; 210 struct nfsmount *nmp; 211 int myiod, timo; 212 int error = 0; 213 214 mtx_lock(&Giant); 215 myiod = (int *)instance - nfs_asyncdaemon; 216 /* 217 * Main loop 218 */ 219 for (;;) { 220 while (((nmp = nfs_iodmount[myiod]) == NULL 221 || !TAILQ_FIRST(&nmp->nm_bufq)) 222 && error == 0) { 223 if (myiod >= nfs_iodmax) 224 goto finish; 225 if (nmp) 226 nmp->nm_bufqiods--; 227 nfs_iodwant[myiod] = curthread->td_proc; 228 nfs_iodmount[myiod] = NULL; 229 /* 230 * Always keep at least nfs_iodmin kthreads. 231 */ 232 timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz; 233 error = tsleep(&nfs_iodwant[myiod], PWAIT | PCATCH, 234 "-", timo); 235 } 236 if (error) 237 break; 238 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 239 /* Take one off the front of the list */ 240 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 241 nmp->nm_bufqlen--; 242 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 243 nmp->nm_bufqwant = 0; 244 wakeup(&nmp->nm_bufq); 245 } 246 if (bp->b_flags & B_DIRECT) { 247 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); 248 (void)nfs_doio_directwrite(bp); 249 } else { 250 if (bp->b_iocmd == BIO_READ) 251 (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL); 252 else 253 (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL); 254 } 255 256 /* 257 * If there are more than one iod on this mount, then defect 258 * so that the iods can be shared out fairly between the mounts 259 */ 260 if (nfs_defect && nmp->nm_bufqiods > 1) { 261 NFS_DPF(ASYNCIO, 262 ("nfssvc_iod: iod %d defecting from mount %p\n", 263 myiod, nmp)); 264 nfs_iodmount[myiod] = NULL; 265 nmp->nm_bufqiods--; 266 break; 267 } 268 } 269 } 270finish: 271 nfs_asyncdaemon[myiod] = 0; 272 if (nmp) 273 nmp->nm_bufqiods--; 274 nfs_iodwant[myiod] = NULL; 275 nfs_iodmount[myiod] = NULL; 276 /* Someone may be waiting for the last nfsiod to terminate. */ 277 if (--nfs_numasync == 0) 278 wakeup(&nfs_numasync); 279 mtx_unlock(&Giant); 280 if ((error == 0) || (error == EWOULDBLOCK)) 281 kthread_exit(0); 282 /* Abnormal termination */ 283 kthread_exit(1); 284} 285