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 * from nfs_syscalls.c 8.5 (Berkeley) 3/30/95 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 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 <fs/nfs/nfsport.h> 67 #include <fs/nfsclient/nfsmount.h> 68 #include <fs/nfsclient/nfs.h> 69 #include <fs/nfsclient/nfsnode.h> 70 #include <fs/nfsclient/nfs_lock.h> 71 72 extern struct mtx ncl_iod_mutex; 73 74 int ncl_numasync; 75 enum nfsiod_state ncl_iodwant[NFS_MAXRAHEAD]; 76 struct nfsmount *ncl_iodmount[NFS_MAXRAHEAD]; 77 78 static void nfssvc_iod(void *); 79 80 static int nfs_asyncdaemon[NFS_MAXRAHEAD]; 81 82 SYSCTL_DECL(_vfs_newnfs); 83 84 /* Maximum number of seconds a nfsiod kthread will sleep before exiting */ 85 static unsigned int ncl_iodmaxidle = 120; 86 SYSCTL_UINT(_vfs_newnfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &ncl_iodmaxidle, 0, ""); 87 88 /* Maximum number of nfsiod kthreads */ 89 unsigned int ncl_iodmax = NFS_MAXRAHEAD; 90 91 /* Minimum number of nfsiod kthreads to keep as spares */ 92 static unsigned int nfs_iodmin = 0; 93 94 static int 95 sysctl_iodmin(SYSCTL_HANDLER_ARGS) 96 { 97 int error, i; 98 int newmin; 99 100 newmin = nfs_iodmin; 101 error = sysctl_handle_int(oidp, &newmin, 0, req); 102 if (error || (req->newptr == NULL)) 103 return (error); 104 mtx_lock(&ncl_iod_mutex); 105 if (newmin > ncl_iodmax) { 106 error = EINVAL; 107 goto out; 108 } 109 nfs_iodmin = newmin; 110 if (ncl_numasync >= nfs_iodmin) 111 goto out; 112 /* 113 * If the current number of nfsiod is lower 114 * than the new minimum, create some more. 115 */ 116 for (i = nfs_iodmin - ncl_numasync; i > 0; i--) 117 ncl_nfsiodnew(0); 118 out: 119 mtx_unlock(&ncl_iod_mutex); 120 return (0); 121 } 122 SYSCTL_PROC(_vfs_newnfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, 123 sizeof (nfs_iodmin), sysctl_iodmin, "IU", ""); 124 125 126 static int 127 sysctl_iodmax(SYSCTL_HANDLER_ARGS) 128 { 129 int error, i; 130 int iod, newmax; 131 132 newmax = ncl_iodmax; 133 error = sysctl_handle_int(oidp, &newmax, 0, req); 134 if (error || (req->newptr == NULL)) 135 return (error); 136 if (newmax > NFS_MAXRAHEAD) 137 return (EINVAL); 138 mtx_lock(&ncl_iod_mutex); 139 ncl_iodmax = newmax; 140 if (ncl_numasync <= ncl_iodmax) 141 goto out; 142 /* 143 * If there are some asleep nfsiods that should 144 * exit, wakeup() them so that they check ncl_iodmax 145 * and exit. Those who are active will exit as 146 * soon as they finish I/O. 147 */ 148 iod = ncl_numasync - 1; 149 for (i = 0; i < ncl_numasync - ncl_iodmax; i++) { 150 if (ncl_iodwant[iod] == NFSIOD_AVAILABLE) 151 wakeup(&ncl_iodwant[iod]); 152 iod--; 153 } 154 out: 155 mtx_unlock(&ncl_iod_mutex); 156 return (0); 157 } 158 SYSCTL_PROC(_vfs_newnfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0, 159 sizeof (ncl_iodmax), sysctl_iodmax, "IU", ""); 160 161 int 162 ncl_nfsiodnew(int set_iodwant) 163 { 164 int error, i; 165 int newiod; 166 167 if (ncl_numasync >= ncl_iodmax) 168 return (-1); 169 newiod = -1; 170 for (i = 0; i < ncl_iodmax; i++) 171 if (nfs_asyncdaemon[i] == 0) { 172 nfs_asyncdaemon[i]++; 173 newiod = i; 174 break; 175 } 176 if (newiod == -1) 177 return (-1); 178 if (set_iodwant > 0) 179 ncl_iodwant[i] = NFSIOD_CREATED_FOR_NFS_ASYNCIO; 180 mtx_unlock(&ncl_iod_mutex); 181 error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID, 182 0, "nfsiod %d", newiod); 183 mtx_lock(&ncl_iod_mutex); 184 if (error) { 185 if (set_iodwant > 0) 186 ncl_iodwant[i] = NFSIOD_NOT_AVAILABLE; 187 return (-1); 188 } 189 ncl_numasync++; 190 return (newiod); 191 } 192 193 static void 194 nfsiod_setup(void *dummy) 195 { 196 int i; 197 int error; 198 199 TUNABLE_INT_FETCH("vfs.newnfs.iodmin", &nfs_iodmin); 200 nfscl_init(); 201 mtx_lock(&ncl_iod_mutex); 202 /* Silently limit the start number of nfsiod's */ 203 if (nfs_iodmin > NFS_MAXRAHEAD) 204 nfs_iodmin = NFS_MAXRAHEAD; 205 206 for (i = 0; i < nfs_iodmin; i++) { 207 error = ncl_nfsiodnew(0); 208 if (error == -1) 209 panic("newnfsiod_setup: ncl_nfsiodnew failed"); 210 } 211 mtx_unlock(&ncl_iod_mutex); 212 } 213 SYSINIT(newnfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); 214 215 static int nfs_defect = 0; 216 SYSCTL_INT(_vfs_newnfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 217 218 /* 219 * Asynchronous I/O daemons for client nfs. 220 * They do read-ahead and write-behind operations on the block I/O cache. 221 * Returns if we hit the timeout defined by the iodmaxidle sysctl. 222 */ 223 static void 224 nfssvc_iod(void *instance) 225 { 226 struct buf *bp; 227 struct nfsmount *nmp; 228 int myiod, timo; 229 int error = 0; 230 231 mtx_lock(&ncl_iod_mutex); 232 myiod = (int *)instance - nfs_asyncdaemon; 233 /* 234 * Main loop 235 */ 236 for (;;) { 237 while (((nmp = ncl_iodmount[myiod]) == NULL) 238 || !TAILQ_FIRST(&nmp->nm_bufq)) { 239 if (myiod >= ncl_iodmax) 240 goto finish; 241 if (nmp) 242 nmp->nm_bufqiods--; 243 if (ncl_iodwant[myiod] == NFSIOD_NOT_AVAILABLE) 244 ncl_iodwant[myiod] = NFSIOD_AVAILABLE; 245 ncl_iodmount[myiod] = NULL; 246 /* 247 * Always keep at least nfs_iodmin kthreads. 248 */ 249 timo = (myiod < nfs_iodmin) ? 0 : ncl_iodmaxidle * hz; 250 error = msleep(&ncl_iodwant[myiod], &ncl_iod_mutex, PWAIT | PCATCH, 251 "-", timo); 252 if (error) { 253 nmp = ncl_iodmount[myiod]; 254 /* 255 * Rechecking the nm_bufq closes a rare race where the 256 * nfsiod is woken up at the exact time the idle timeout 257 * fires 258 */ 259 if (nmp && TAILQ_FIRST(&nmp->nm_bufq)) 260 error = 0; 261 break; 262 } 263 } 264 if (error) 265 break; 266 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 267 268 /* Take one off the front of the list */ 269 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 270 nmp->nm_bufqlen--; 271 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= ncl_numasync) { 272 nmp->nm_bufqwant = 0; 273 wakeup(&nmp->nm_bufq); 274 } 275 mtx_unlock(&ncl_iod_mutex); 276 if (bp->b_flags & B_DIRECT) { 277 KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); 278 (void)ncl_doio_directwrite(bp); 279 } else { 280 if (bp->b_iocmd == BIO_READ) 281 (void) ncl_doio(bp->b_vp, bp, bp->b_rcred, 282 NULL, 0); 283 else 284 (void) ncl_doio(bp->b_vp, bp, bp->b_wcred, 285 NULL, 0); 286 } 287 mtx_lock(&ncl_iod_mutex); 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 ncl_iodmount[myiod] = NULL; 297 nmp->nm_bufqiods--; 298 break; 299 } 300 } 301 } 302 finish: 303 nfs_asyncdaemon[myiod] = 0; 304 if (nmp) 305 nmp->nm_bufqiods--; 306 ncl_iodwant[myiod] = NFSIOD_NOT_AVAILABLE; 307 ncl_iodmount[myiod] = NULL; 308 /* Someone may be waiting for the last nfsiod to terminate. */ 309 if (--ncl_numasync == 0) 310 wakeup(&ncl_numasync); 311 mtx_unlock(&ncl_iod_mutex); 312 if ((error == 0) || (error == EWOULDBLOCK)) 313 kproc_exit(0); 314 /* Abnormal termination */ 315 kproc_exit(1); 316 } 317