nfs_nfsiod.c revision 195202
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 195202 2009-06-30 19:03:27Z dfr $"); 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 <nfsclient/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(); 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]) 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(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 = kproc_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 "Allow nfsiods to migrate serving different mounts"); 213 214/* 215 * Asynchronous I/O daemons for client nfs. 216 * They do read-ahead and write-behind operations on the block I/O cache. 217 * Returns if we hit the timeout defined by the iodmaxidle sysctl. 218 */ 219static void 220nfssvc_iod(void *instance) 221{ 222 struct buf *bp; 223 struct nfsmount *nmp; 224 int myiod, timo; 225 int error = 0; 226 227 mtx_lock(&nfs_iod_mtx); 228 myiod = (int *)instance - nfs_asyncdaemon; 229 /* 230 * Main loop 231 */ 232 for (;;) { 233 while (((nmp = nfs_iodmount[myiod]) == NULL) 234 || !TAILQ_FIRST(&nmp->nm_bufq)) { 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 if (error) { 248 nmp = nfs_iodmount[myiod]; 249 /* 250 * Rechecking the nm_bufq closes a rare race where the 251 * nfsiod is woken up at the exact time the idle timeout 252 * fires 253 */ 254 if (nmp && TAILQ_FIRST(&nmp->nm_bufq)) 255 error = 0; 256 break; 257 } 258 } 259 if (error) 260 break; 261 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 262 int giant_locked = 0; 263 264 /* Take one off the front of the list */ 265 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 266 nmp->nm_bufqlen--; 267 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 268 nmp->nm_bufqwant = 0; 269 wakeup(&nmp->nm_bufq); 270 } 271 mtx_unlock(&nfs_iod_mtx); 272 if (NFS_ISV4(bp->b_vp)) { 273 giant_locked = 1; 274 mtx_lock(&Giant); 275 } 276 if (bp->b_flags & B_DIRECT) { 277 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); 278 (void)nfs_doio_directwrite(bp); 279 } else { 280 if (bp->b_iocmd == BIO_READ) 281 (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL); 282 else 283 (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL); 284 } 285 if (giant_locked) 286 mtx_unlock(&Giant); 287 mtx_lock(&nfs_iod_mtx); 288 /* 289 * If there are more than one iod on this mount, then defect 290 * so that the iods can be shared out fairly between the mounts 291 */ 292 if (nfs_defect && nmp->nm_bufqiods > 1) { 293 NFS_DPF(ASYNCIO, 294 ("nfssvc_iod: iod %d defecting from mount %p\n", 295 myiod, nmp)); 296 nfs_iodmount[myiod] = NULL; 297 nmp->nm_bufqiods--; 298 break; 299 } 300 } 301 } 302finish: 303 nfs_asyncdaemon[myiod] = 0; 304 if (nmp) 305 nmp->nm_bufqiods--; 306 nfs_iodwant[myiod] = NULL; 307 nfs_iodmount[myiod] = NULL; 308 /* Someone may be waiting for the last nfsiod to terminate. */ 309 if (--nfs_numasync == 0) 310 wakeup(&nfs_numasync); 311 mtx_unlock(&nfs_iod_mtx); 312 if ((error == 0) || (error == EWOULDBLOCK)) 313 kproc_exit(0); 314 /* Abnormal termination */ 315 kproc_exit(1); 316} 317