1139823Simp/*- 21541Srgrimes * Copyright (c) 1989, 1993 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * This code is derived from software contributed to Berkeley by 61541Srgrimes * Rick Macklem at The University of Guelph. 71541Srgrimes * 81541Srgrimes * Redistribution and use in source and binary forms, with or without 91541Srgrimes * modification, are permitted provided that the following conditions 101541Srgrimes * are met: 111541Srgrimes * 1. Redistributions of source code must retain the above copyright 121541Srgrimes * notice, this list of conditions and the following disclaimer. 131541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 141541Srgrimes * notice, this list of conditions and the following disclaimer in the 151541Srgrimes * documentation and/or other materials provided with the distribution. 161541Srgrimes * 4. Neither the name of the University nor the names of its contributors 171541Srgrimes * may be used to endorse or promote products derived from this software 181541Srgrimes * without specific prior written permission. 191541Srgrimes * 201541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 211541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 221541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 231541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 241541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 251541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 261541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 271541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 281541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 291541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 301541Srgrimes * SUCH DAMAGE. 311541Srgrimes * 3222521Sdyson * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 331541Srgrimes */ 341541Srgrimes 3583651Speter#include <sys/cdefs.h> 3683651Speter__FBSDID("$FreeBSD$"); 3783651Speter 381541Srgrimes#include <sys/param.h> 391541Srgrimes#include <sys/systm.h> 4012274Sbde#include <sys/sysproto.h> 411541Srgrimes#include <sys/kernel.h> 4219449Sdfr#include <sys/sysctl.h> 431541Srgrimes#include <sys/file.h> 4415480Sbde#include <sys/filedesc.h> 451541Srgrimes#include <sys/vnode.h> 4630354Sphk#include <sys/malloc.h> 471541Srgrimes#include <sys/mount.h> 481541Srgrimes#include <sys/proc.h> 4960041Sphk#include <sys/bio.h> 501541Srgrimes#include <sys/buf.h> 511541Srgrimes#include <sys/mbuf.h> 521541Srgrimes#include <sys/socket.h> 531541Srgrimes#include <sys/socketvar.h> 541541Srgrimes#include <sys/domain.h> 551541Srgrimes#include <sys/protosw.h> 561541Srgrimes#include <sys/namei.h> 5783651Speter#include <sys/unistd.h> 5883651Speter#include <sys/kthread.h> 5975631Salfred#include <sys/fcntl.h> 6075631Salfred#include <sys/lockf.h> 6183651Speter#include <sys/mutex.h> 62212506Skib#include <sys/taskqueue.h> 631541Srgrimes 641541Srgrimes#include <netinet/in.h> 651541Srgrimes#include <netinet/tcp.h> 66122698Salfred 679336Sdfr#include <nfs/xdr_subs.h> 689336Sdfr#include <nfs/nfsproto.h> 6983651Speter#include <nfsclient/nfs.h> 7083651Speter#include <nfsclient/nfsm_subs.h> 7183651Speter#include <nfsclient/nfsmount.h> 7283651Speter#include <nfsclient/nfsnode.h> 73210455Srmacklem#include <nfs/nfs_lock.h> 741541Srgrimes 75151897Srwatsonstatic MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure"); 7630309Sphk 7783651Speterstatic void nfssvc_iod(void *); 781541Srgrimes 791541Srgrimesstatic int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 8012457Sbde 81221973SrmacklemSYSCTL_DECL(_vfs_oldnfs); 8244112Sdfr 8389407Speter/* Maximum number of seconds a nfsiod kthread will sleep before exiting */ 8489407Speterstatic unsigned int nfs_iodmaxidle = 120; 85221973SrmacklemSYSCTL_UINT(_vfs_oldnfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0, 86184561Strhodes "Max number of seconds an nfsiod kthread will sleep before exiting"); 8789407Speter 8889407Speter/* Maximum number of nfsiod kthreads */ 89128111Speadarunsigned int nfs_iodmax = 20; 9089407Speter 9189324Speter/* Minimum number of nfsiod kthreads to keep as spares */ 92158903Srwatsonstatic unsigned int nfs_iodmin = 0; 9389324Speter 94214026Skibstatic int nfs_nfsiodnew_sync(void); 95214026Skib 9689407Speterstatic int 9789407Spetersysctl_iodmin(SYSCTL_HANDLER_ARGS) 9889407Speter{ 9989407Speter int error, i; 10089407Speter int newmin; 10189324Speter 10289407Speter newmin = nfs_iodmin; 10389407Speter error = sysctl_handle_int(oidp, &newmin, 0, req); 10489407Speter if (error || (req->newptr == NULL)) 10589407Speter return (error); 106158739Smohans mtx_lock(&nfs_iod_mtx); 107158739Smohans if (newmin > nfs_iodmax) { 108158739Smohans error = EINVAL; 109158739Smohans goto out; 110158739Smohans } 11189407Speter nfs_iodmin = newmin; 11289407Speter if (nfs_numasync >= nfs_iodmin) 113158739Smohans goto out; 11489407Speter /* 11589407Speter * If the current number of nfsiod is lower 11689407Speter * than the new minimum, create some more. 11789407Speter */ 11889407Speter for (i = nfs_iodmin - nfs_numasync; i > 0; i--) 119214026Skib nfs_nfsiodnew_sync(); 120158739Smohansout: 121158739Smohans mtx_unlock(&nfs_iod_mtx); 12289407Speter return (0); 12389407Speter} 124221973SrmacklemSYSCTL_PROC(_vfs_oldnfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, 125184561Strhodes sizeof (nfs_iodmin), sysctl_iodmin, "IU", 126184561Strhodes "Min number of nfsiod kthreads to keep as spares"); 12789407Speter 12889407Speter 12989407Speterstatic int 13089407Spetersysctl_iodmax(SYSCTL_HANDLER_ARGS) 13189407Speter{ 13289407Speter int error, i; 13389407Speter int iod, newmax; 13489407Speter 13589407Speter newmax = nfs_iodmax; 13689407Speter error = sysctl_handle_int(oidp, &newmax, 0, req); 13789407Speter if (error || (req->newptr == NULL)) 13889407Speter return (error); 13989407Speter if (newmax > NFS_MAXASYNCDAEMON) 14089407Speter return (EINVAL); 141158739Smohans mtx_lock(&nfs_iod_mtx); 14289407Speter nfs_iodmax = newmax; 14389407Speter if (nfs_numasync <= nfs_iodmax) 144158739Smohans goto out; 14589407Speter /* 14689407Speter * If there are some asleep nfsiods that should 14789407Speter * exit, wakeup() them so that they check nfs_iodmax 14889407Speter * and exit. Those who are active will exit as 14989407Speter * soon as they finish I/O. 15089407Speter */ 15189407Speter iod = nfs_numasync - 1; 15289407Speter for (i = 0; i < nfs_numasync - nfs_iodmax; i++) { 153203072Srmacklem if (nfs_iodwant[iod] == NFSIOD_AVAILABLE) 154111748Sdes wakeup(&nfs_iodwant[iod]); 15589407Speter iod--; 15689407Speter } 157158739Smohansout: 158158739Smohans mtx_unlock(&nfs_iod_mtx); 15989407Speter return (0); 16089407Speter} 161221973SrmacklemSYSCTL_PROC(_vfs_oldnfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0, 162184561Strhodes sizeof (nfs_iodmax), sysctl_iodmax, "IU", 163184561Strhodes "Max number of nfsiod kthreads"); 16489407Speter 165214026Skibstatic int 166214026Skibnfs_nfsiodnew_sync(void) 167214026Skib{ 168214026Skib int error, i; 169214026Skib 170214026Skib mtx_assert(&nfs_iod_mtx, MA_OWNED); 171214026Skib for (i = 0; i < nfs_iodmax; i++) { 172214026Skib if (nfs_asyncdaemon[i] == 0) { 173214026Skib nfs_asyncdaemon[i] = 1; 174214026Skib break; 175214026Skib } 176214026Skib } 177214026Skib if (i == nfs_iodmax) 178214026Skib return (0); 179214026Skib mtx_unlock(&nfs_iod_mtx); 180214026Skib error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, 181214026Skib RFHIGHPID, 0, "nfsiod %d", i); 182214026Skib mtx_lock(&nfs_iod_mtx); 183214026Skib if (error == 0) { 184214026Skib nfs_numasync++; 185214026Skib nfs_iodwant[i] = NFSIOD_AVAILABLE; 186214026Skib } else 187214026Skib nfs_asyncdaemon[i] = 0; 188214026Skib return (error); 189214026Skib} 190214026Skib 191212506Skibvoid 192212506Skibnfs_nfsiodnew_tq(__unused void *arg, int pending) 193212506Skib{ 194212506Skib 195212506Skib mtx_lock(&nfs_iod_mtx); 196214026Skib while (pending > 0) { 197214026Skib pending--; 198214026Skib nfs_nfsiodnew_sync(); 199212506Skib } 200212506Skib mtx_unlock(&nfs_iod_mtx); 201212506Skib} 202212506Skib 203214026Skibvoid 204214026Skibnfs_nfsiodnew(void) 20589324Speter{ 20689324Speter 207214026Skib mtx_assert(&nfs_iod_mtx, MA_OWNED); 208212506Skib taskqueue_enqueue(taskqueue_thread, &nfs_nfsiodnew_task); 20989324Speter} 21089324Speter 21183651Speterstatic void 21283651Speternfsiod_setup(void *dummy) 2131541Srgrimes{ 2141541Srgrimes int error; 2151541Srgrimes 216221973Srmacklem TUNABLE_INT_FETCH("vfs.oldnfs.iodmin", &nfs_iodmin); 217158739Smohans mtx_lock(&nfs_iod_mtx); 21889324Speter /* Silently limit the start number of nfsiod's */ 21989324Speter if (nfs_iodmin > NFS_MAXASYNCDAEMON) 22089324Speter nfs_iodmin = NFS_MAXASYNCDAEMON; 22189324Speter 222214026Skib while (nfs_numasync < nfs_iodmin) { 223214026Skib error = nfs_nfsiodnew_sync(); 22489324Speter if (error == -1) 22589324Speter panic("nfsiod_setup: nfs_nfsiodnew failed"); 2261541Srgrimes } 227158739Smohans mtx_unlock(&nfs_iod_mtx); 2281541Srgrimes} 22983651SpeterSYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); 2301541Srgrimes 23183651Speterstatic int nfs_defect = 0; 232221973SrmacklemSYSCTL_INT(_vfs_oldnfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, 233184561Strhodes "Allow nfsiods to migrate serving different mounts"); 2341541Srgrimes 23544246Speter/* 2361541Srgrimes * Asynchronous I/O daemons for client nfs. 2371541Srgrimes * They do read-ahead and write-behind operations on the block I/O cache. 23889324Speter * Returns if we hit the timeout defined by the iodmaxidle sysctl. 2391541Srgrimes */ 24083651Speterstatic void 24189324Speternfssvc_iod(void *instance) 2421541Srgrimes{ 24383651Speter struct buf *bp; 24419449Sdfr struct nfsmount *nmp; 24589324Speter int myiod, timo; 24631016Sphk int error = 0; 2471541Srgrimes 248158739Smohans mtx_lock(&nfs_iod_mtx); 24989324Speter myiod = (int *)instance - nfs_asyncdaemon; 2501541Srgrimes /* 25189324Speter * Main loop 2521541Srgrimes */ 2531541Srgrimes for (;;) { 254172324Smohans while (((nmp = nfs_iodmount[myiod]) == NULL) 255172324Smohans || !TAILQ_FIRST(&nmp->nm_bufq)) { 25689407Speter if (myiod >= nfs_iodmax) 25789407Speter goto finish; 25819449Sdfr if (nmp) 25989324Speter nmp->nm_bufqiods--; 260203072Srmacklem if (nfs_iodwant[myiod] == NFSIOD_NOT_AVAILABLE) 261203072Srmacklem nfs_iodwant[myiod] = NFSIOD_AVAILABLE; 26219449Sdfr nfs_iodmount[myiod] = NULL; 26389324Speter /* 26489324Speter * Always keep at least nfs_iodmin kthreads. 26589324Speter */ 26689324Speter timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz; 267158739Smohans error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH, 268117152Sphk "-", timo); 269172324Smohans if (error) { 270172324Smohans nmp = nfs_iodmount[myiod]; 271172324Smohans /* 272172324Smohans * Rechecking the nm_bufq closes a rare race where the 273172324Smohans * nfsiod is woken up at the exact time the idle timeout 274172324Smohans * fires 275172324Smohans */ 276172324Smohans if (nmp && TAILQ_FIRST(&nmp->nm_bufq)) 277172324Smohans error = 0; 278172324Smohans break; 279172324Smohans } 2809336Sdfr } 28189324Speter if (error) 28289324Speter break; 28383651Speter while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 284158905Smohans int giant_locked = 0; 285158905Smohans 2869336Sdfr /* Take one off the front of the list */ 28719449Sdfr TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 28819449Sdfr nmp->nm_bufqlen--; 28955431Sdillon if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 29089324Speter nmp->nm_bufqwant = 0; 29119449Sdfr wakeup(&nmp->nm_bufq); 29219449Sdfr } 293158739Smohans mtx_unlock(&nfs_iod_mtx); 294158905Smohans if (NFS_ISV4(bp->b_vp)) { 295158905Smohans giant_locked = 1; 296158905Smohans mtx_lock(&Giant); 297158905Smohans } 298138899Sps if (bp->b_flags & B_DIRECT) { 299138899Sps KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); 300138899Sps (void)nfs_doio_directwrite(bp); 301138899Sps } else { 302138899Sps if (bp->b_iocmd == BIO_READ) 303138899Sps (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL); 304138899Sps else 305138899Sps (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL); 306138899Sps } 307158905Smohans if (giant_locked) 308158905Smohans mtx_unlock(&Giant); 309158739Smohans mtx_lock(&nfs_iod_mtx); 31019449Sdfr /* 311249630Srmacklem * Make sure the nmp hasn't been dismounted as soon as 312249630Srmacklem * nfs_doio() completes for the last buffer. 313249630Srmacklem */ 314249630Srmacklem nmp = nfs_iodmount[myiod]; 315249630Srmacklem if (nmp == NULL) 316249630Srmacklem break; 317249630Srmacklem 318249630Srmacklem /* 31919449Sdfr * If there are more than one iod on this mount, then defect 32019449Sdfr * so that the iods can be shared out fairly between the mounts 32119449Sdfr */ 32219449Sdfr if (nfs_defect && nmp->nm_bufqiods > 1) { 32319449Sdfr NFS_DPF(ASYNCIO, 32419449Sdfr ("nfssvc_iod: iod %d defecting from mount %p\n", 32519449Sdfr myiod, nmp)); 32619449Sdfr nfs_iodmount[myiod] = NULL; 32719449Sdfr nmp->nm_bufqiods--; 32819449Sdfr break; 32919449Sdfr } 3309336Sdfr } 3311541Srgrimes } 33289407Speterfinish: 33389324Speter nfs_asyncdaemon[myiod] = 0; 33489324Speter if (nmp) 33589324Speter nmp->nm_bufqiods--; 336203072Srmacklem nfs_iodwant[myiod] = NFSIOD_NOT_AVAILABLE; 33789324Speter nfs_iodmount[myiod] = NULL; 338128111Speadar /* Someone may be waiting for the last nfsiod to terminate. */ 339128111Speadar if (--nfs_numasync == 0) 340128111Speadar wakeup(&nfs_numasync); 341158739Smohans mtx_unlock(&nfs_iod_mtx); 34289407Speter if ((error == 0) || (error == EWOULDBLOCK)) 343172836Sjulian kproc_exit(0); 34489324Speter /* Abnormal termination */ 345172836Sjulian kproc_exit(1); 3461541Srgrimes} 347