nfs_nfsiod.c revision 210455
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 210455 2010-07-24 22:11:11Z 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 <nfs/xdr_subs.h> 67#include <nfs/nfsproto.h> 68#include <nfsclient/nfs.h> 69#include <nfsclient/nfsm_subs.h> 70#include <nfsclient/nfsmount.h> 71#include <nfsclient/nfsnode.h> 72#include <nfs/nfs_lock.h> 73 74static MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure"); 75 76static void nfssvc_iod(void *); 77 78static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 79 80SYSCTL_DECL(_vfs_nfs); 81 82/* Maximum number of seconds a nfsiod kthread will sleep before exiting */ 83static unsigned int nfs_iodmaxidle = 120; 84SYSCTL_UINT(_vfs_nfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0, 85 "Max number of seconds an nfsiod kthread will sleep before exiting"); 86 87/* Maximum number of nfsiod kthreads */ 88unsigned int nfs_iodmax = 20; 89 90/* Minimum number of nfsiod kthreads to keep as spares */ 91static unsigned int nfs_iodmin = 0; 92 93static int 94sysctl_iodmin(SYSCTL_HANDLER_ARGS) 95{ 96 int error, i; 97 int newmin; 98 99 newmin = nfs_iodmin; 100 error = sysctl_handle_int(oidp, &newmin, 0, req); 101 if (error || (req->newptr == NULL)) 102 return (error); 103 mtx_lock(&nfs_iod_mtx); 104 if (newmin > nfs_iodmax) { 105 error = EINVAL; 106 goto out; 107 } 108 nfs_iodmin = newmin; 109 if (nfs_numasync >= nfs_iodmin) 110 goto out; 111 /* 112 * If the current number of nfsiod is lower 113 * than the new minimum, create some more. 114 */ 115 for (i = nfs_iodmin - nfs_numasync; i > 0; i--) 116 nfs_nfsiodnew(0); 117out: 118 mtx_unlock(&nfs_iod_mtx); 119 return (0); 120} 121SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, 122 sizeof (nfs_iodmin), sysctl_iodmin, "IU", 123 "Min number of nfsiod kthreads to keep as spares"); 124 125 126static int 127sysctl_iodmax(SYSCTL_HANDLER_ARGS) 128{ 129 int error, i; 130 int iod, newmax; 131 132 newmax = nfs_iodmax; 133 error = sysctl_handle_int(oidp, &newmax, 0, req); 134 if (error || (req->newptr == NULL)) 135 return (error); 136 if (newmax > NFS_MAXASYNCDAEMON) 137 return (EINVAL); 138 mtx_lock(&nfs_iod_mtx); 139 nfs_iodmax = newmax; 140 if (nfs_numasync <= nfs_iodmax) 141 goto out; 142 /* 143 * If there are some asleep nfsiods that should 144 * exit, wakeup() them so that they check nfs_iodmax 145 * and exit. Those who are active will exit as 146 * soon as they finish I/O. 147 */ 148 iod = nfs_numasync - 1; 149 for (i = 0; i < nfs_numasync - nfs_iodmax; i++) { 150 if (nfs_iodwant[iod] == NFSIOD_AVAILABLE) 151 wakeup(&nfs_iodwant[iod]); 152 iod--; 153 } 154out: 155 mtx_unlock(&nfs_iod_mtx); 156 return (0); 157} 158SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0, 159 sizeof (nfs_iodmax), sysctl_iodmax, "IU", 160 "Max number of nfsiod kthreads"); 161 162int 163nfs_nfsiodnew(int set_iodwant) 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 if (set_iodwant > 0) 180 nfs_iodwant[i] = NFSIOD_CREATED_FOR_NFS_ASYNCIO; 181 mtx_unlock(&nfs_iod_mtx); 182 error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID, 183 0, "nfsiod %d", newiod); 184 mtx_lock(&nfs_iod_mtx); 185 if (error) { 186 if (set_iodwant > 0) 187 nfs_iodwant[i] = NFSIOD_NOT_AVAILABLE; 188 return (-1); 189 } 190 nfs_numasync++; 191 return (newiod); 192} 193 194static void 195nfsiod_setup(void *dummy) 196{ 197 int i; 198 int error; 199 200 TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin); 201 mtx_lock(&nfs_iod_mtx); 202 /* Silently limit the start number of nfsiod's */ 203 if (nfs_iodmin > NFS_MAXASYNCDAEMON) 204 nfs_iodmin = NFS_MAXASYNCDAEMON; 205 206 for (i = 0; i < nfs_iodmin; i++) { 207 error = nfs_nfsiodnew(0); 208 if (error == -1) 209 panic("nfsiod_setup: nfs_nfsiodnew failed"); 210 } 211 mtx_unlock(&nfs_iod_mtx); 212} 213SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); 214 215static int nfs_defect = 0; 216SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, 217 "Allow nfsiods to migrate serving different mounts"); 218 219/* 220 * Asynchronous I/O daemons for client nfs. 221 * They do read-ahead and write-behind operations on the block I/O cache. 222 * Returns if we hit the timeout defined by the iodmaxidle sysctl. 223 */ 224static void 225nfssvc_iod(void *instance) 226{ 227 struct buf *bp; 228 struct nfsmount *nmp; 229 int myiod, timo; 230 int error = 0; 231 232 mtx_lock(&nfs_iod_mtx); 233 myiod = (int *)instance - nfs_asyncdaemon; 234 /* 235 * Main loop 236 */ 237 for (;;) { 238 while (((nmp = nfs_iodmount[myiod]) == NULL) 239 || !TAILQ_FIRST(&nmp->nm_bufq)) { 240 if (myiod >= nfs_iodmax) 241 goto finish; 242 if (nmp) 243 nmp->nm_bufqiods--; 244 if (nfs_iodwant[myiod] == NFSIOD_NOT_AVAILABLE) 245 nfs_iodwant[myiod] = NFSIOD_AVAILABLE; 246 nfs_iodmount[myiod] = NULL; 247 /* 248 * Always keep at least nfs_iodmin kthreads. 249 */ 250 timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz; 251 error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH, 252 "-", timo); 253 if (error) { 254 nmp = nfs_iodmount[myiod]; 255 /* 256 * Rechecking the nm_bufq closes a rare race where the 257 * nfsiod is woken up at the exact time the idle timeout 258 * fires 259 */ 260 if (nmp && TAILQ_FIRST(&nmp->nm_bufq)) 261 error = 0; 262 break; 263 } 264 } 265 if (error) 266 break; 267 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 268 int giant_locked = 0; 269 270 /* Take one off the front of the list */ 271 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 272 nmp->nm_bufqlen--; 273 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 274 nmp->nm_bufqwant = 0; 275 wakeup(&nmp->nm_bufq); 276 } 277 mtx_unlock(&nfs_iod_mtx); 278 if (NFS_ISV4(bp->b_vp)) { 279 giant_locked = 1; 280 mtx_lock(&Giant); 281 } 282 if (bp->b_flags & B_DIRECT) { 283 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); 284 (void)nfs_doio_directwrite(bp); 285 } else { 286 if (bp->b_iocmd == BIO_READ) 287 (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL); 288 else 289 (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL); 290 } 291 if (giant_locked) 292 mtx_unlock(&Giant); 293 mtx_lock(&nfs_iod_mtx); 294 /* 295 * If there are more than one iod on this mount, then defect 296 * so that the iods can be shared out fairly between the mounts 297 */ 298 if (nfs_defect && nmp->nm_bufqiods > 1) { 299 NFS_DPF(ASYNCIO, 300 ("nfssvc_iod: iod %d defecting from mount %p\n", 301 myiod, nmp)); 302 nfs_iodmount[myiod] = NULL; 303 nmp->nm_bufqiods--; 304 break; 305 } 306 } 307 } 308finish: 309 nfs_asyncdaemon[myiod] = 0; 310 if (nmp) 311 nmp->nm_bufqiods--; 312 nfs_iodwant[myiod] = NFSIOD_NOT_AVAILABLE; 313 nfs_iodmount[myiod] = NULL; 314 /* Someone may be waiting for the last nfsiod to terminate. */ 315 if (--nfs_numasync == 0) 316 wakeup(&nfs_numasync); 317 mtx_unlock(&nfs_iod_mtx); 318 if ((error == 0) || (error == EWOULDBLOCK)) 319 kproc_exit(0); 320 /* Abnormal termination */ 321 kproc_exit(1); 322} 323