nfs_nfsiod.c revision 184561
119304Speter/*- 219304Speter * Copyright (c) 1989, 1993 319304Speter * The Regents of the University of California. All rights reserved. 419304Speter * 519304Speter * This code is derived from software contributed to Berkeley by 619304Speter * Rick Macklem at The University of Guelph. 719304Speter * 819304Speter * Redistribution and use in source and binary forms, with or without 919304Speter * modification, are permitted provided that the following conditions 1019304Speter * are met: 1119304Speter * 1. Redistributions of source code must retain the above copyright 1219304Speter * notice, this list of conditions and the following disclaimer. 13254225Speter * 2. Redistributions in binary form must reproduce the above copyright 1419304Speter * notice, this list of conditions and the following disclaimer in the 1519304Speter * documentation and/or other materials provided with the distribution. 1619304Speter * 4. Neither the name of the University nor the names of its contributors 1719304Speter * may be used to endorse or promote products derived from this software 1819304Speter * without specific prior written permission. 1919304Speter * 2019304Speter * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2119304Speter * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2219304Speter * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2319304Speter * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2419304Speter * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2519304Speter * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2619304Speter * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2719304Speter * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2819304Speter * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2919304Speter * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3019304Speter * SUCH DAMAGE. 3119304Speter * 3219304Speter * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 3319304Speter */ 3419304Speter 3519304Speter#include <sys/cdefs.h> 36254225Speter__FBSDID("$FreeBSD: head/sys/nfsclient/nfs_nfsiod.c 184561 2008-11-02 17:00:23Z trhodes $"); 3719304Speter 3819304Speter#include <sys/param.h> 3919304Speter#include <sys/systm.h> 4019304Speter#include <sys/sysproto.h> 4119304Speter#include <sys/kernel.h> 4219304Speter#include <sys/sysctl.h> 4319304Speter#include <sys/file.h> 4419304Speter#include <sys/filedesc.h> 4519304Speter#include <sys/vnode.h> 4619304Speter#include <sys/malloc.h> 4719304Speter#include <sys/mount.h> 48254225Speter#include <sys/proc.h> 4919304Speter#include <sys/bio.h> 5019304Speter#include <sys/buf.h> 5119304Speter#include <sys/mbuf.h> 5219304Speter#include <sys/socket.h> 5319304Speter#include <sys/socketvar.h> 5419304Speter#include <sys/domain.h> 5519304Speter#include <sys/protosw.h> 5619304Speter#include <sys/namei.h> 5719304Speter#include <sys/unistd.h> 5819304Speter#include <sys/kthread.h> 5919304Speter#include <sys/fcntl.h> 6019304Speter#include <sys/lockf.h> 6119304Speter#include <sys/mutex.h> 6219304Speter 6319304Speter#include <netinet/in.h> 6419304Speter#include <netinet/tcp.h> 6519304Speter 6619304Speter#include <rpc/rpcclnt.h> 6719304Speter 6819304Speter#include <nfs/xdr_subs.h> 6919304Speter#include <nfs/rpcv2.h> 7019304Speter#include <nfs/nfsproto.h> 7119304Speter#include <nfsclient/nfs.h> 72254225Speter#include <nfsclient/nfsm_subs.h> 7319304Speter#include <nfsclient/nfsmount.h> 7419304Speter#include <nfsclient/nfsnode.h> 7519304Speter#include <nfsclient/nfs_lock.h> 7619304Speter 7719304Speterstatic MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure"); 7819304Speter 7919304Speterstatic void nfssvc_iod(void *); 8019304Speter 8119304Speterstatic int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 8219304Speter 8319304SpeterSYSCTL_DECL(_vfs_nfs); 8419304Speter 8519304Speter/* Maximum number of seconds a nfsiod kthread will sleep before exiting */ 8619304Speterstatic unsigned int nfs_iodmaxidle = 120; 8719304SpeterSYSCTL_UINT(_vfs_nfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0, 8819304Speter "Max number of seconds an nfsiod kthread will sleep before exiting"); 8919304Speter 9019304Speter/* Maximum number of nfsiod kthreads */ 9119304Speterunsigned int nfs_iodmax = 20; 9219304Speter 9319304Speter/* Minimum number of nfsiod kthreads to keep as spares */ 9419304Speterstatic unsigned int nfs_iodmin = 0; 9519304Speter 9619304Speterstatic int 9719304Spetersysctl_iodmin(SYSCTL_HANDLER_ARGS) 9819304Speter{ 9919304Speter int error, i; 10019304Speter int newmin; 101254225Speter 10219304Speter newmin = nfs_iodmin; 10319304Speter error = sysctl_handle_int(oidp, &newmin, 0, req); 10419304Speter if (error || (req->newptr == NULL)) 10519304Speter return (error); 10619304Speter mtx_lock(&nfs_iod_mtx); 10719304Speter if (newmin > nfs_iodmax) { 108254225Speter error = EINVAL; 10919304Speter goto out; 11019304Speter } 11119304Speter nfs_iodmin = newmin; 11219304Speter if (nfs_numasync >= nfs_iodmin) 11319304Speter goto out; 114254225Speter /* 115254225Speter * If the current number of nfsiod is lower 11619304Speter * than the new minimum, create some more. 11719304Speter */ 11819304Speter for (i = nfs_iodmin - nfs_numasync; i > 0; i--) 11919304Speter nfs_nfsiodnew(); 12019304Speterout: 12119304Speter mtx_unlock(&nfs_iod_mtx); 12219304Speter return (0); 12319304Speter} 12419304SpeterSYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, 12519304Speter sizeof (nfs_iodmin), sysctl_iodmin, "IU", 12619304Speter "Min number of nfsiod kthreads to keep as spares"); 12719304Speter 12819304Speter 12919304Speterstatic int 13019304Spetersysctl_iodmax(SYSCTL_HANDLER_ARGS) 13119304Speter{ 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 } 157out: 158 mtx_unlock(&nfs_iod_mtx); 159 return (0); 160} 161SYSCTL_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 165int 166nfs_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 192static void 193nfsiod_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} 211SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); 212 213static int nfs_defect = 0; 214SYSCTL_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 */ 222static void 223nfssvc_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 } 305finish: 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