nfs_bio.c revision 12662
11541Srgrimes/* 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 * 3. All advertising materials mentioning features or use of this software 171541Srgrimes * must display the following acknowledgement: 181541Srgrimes * This product includes software developed by the University of 191541Srgrimes * California, Berkeley and its contributors. 201541Srgrimes * 4. Neither the name of the University nor the names of its contributors 211541Srgrimes * may be used to endorse or promote products derived from this software 221541Srgrimes * without specific prior written permission. 231541Srgrimes * 241541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 251541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 261541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 271541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 281541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 291541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 301541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 311541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 321541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 331541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 341541Srgrimes * SUCH DAMAGE. 351541Srgrimes * 361541Srgrimes * @(#)nfs_bio.c 8.5 (Berkeley) 1/4/94 3712662Sdg * $Id: nfs_bio.c,v 1.19 1995/12/03 10:02:52 bde Exp $ 381541Srgrimes */ 391541Srgrimes 401541Srgrimes#include <sys/param.h> 411541Srgrimes#include <sys/systm.h> 421541Srgrimes#include <sys/resourcevar.h> 433305Sphk#include <sys/signalvar.h> 441541Srgrimes#include <sys/proc.h> 451541Srgrimes#include <sys/buf.h> 461541Srgrimes#include <sys/vnode.h> 471541Srgrimes#include <sys/mount.h> 481541Srgrimes#include <sys/kernel.h> 491541Srgrimes 501541Srgrimes#include <vm/vm.h> 5112662Sdg#include <vm/vm_param.h> 5212662Sdg#include <vm/vm_extern.h> 531541Srgrimes 541541Srgrimes#include <nfs/rpcv2.h> 559336Sdfr#include <nfs/nfsproto.h> 561541Srgrimes#include <nfs/nfs.h> 571541Srgrimes#include <nfs/nfsmount.h> 581541Srgrimes#include <nfs/nqnfs.h> 599336Sdfr#include <nfs/nfsnode.h> 601541Srgrimes 6112588Sbdeextern struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 6212588Sbde struct proc *p)); 6312588Sbde 641541Srgrimesextern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 651541Srgrimesextern int nfs_numasync; 669336Sdfrextern struct nfsstats nfsstats; 671541Srgrimes 681541Srgrimes/* 699336Sdfr * Ifdefs for FreeBSD-current's merged VM/buffer cache. It is unfortunate 709336Sdfr * that this isn't done inside getblk() and brelse() so these calls 719336Sdfr * wouldn't need to be here. 729336Sdfr */ 739336Sdfr#ifdef B_VMIO 749336Sdfr#define vnode_pager_uncache(vp) 759336Sdfr#else 769336Sdfr#define vfs_busy_pages(bp, f) 779336Sdfr#define vfs_unbusy_pages(bp) 789336Sdfr#define vfs_dirty_pages(bp) 799336Sdfr#endif 809336Sdfr 819336Sdfr/* 821541Srgrimes * Vnode op for read using bio 831541Srgrimes * Any similarity to readip() is purely coincidental 841541Srgrimes */ 851549Srgrimesint 861541Srgrimesnfs_bioread(vp, uio, ioflag, cred) 871541Srgrimes register struct vnode *vp; 881541Srgrimes register struct uio *uio; 891541Srgrimes int ioflag; 901541Srgrimes struct ucred *cred; 911541Srgrimes{ 921541Srgrimes register struct nfsnode *np = VTONFS(vp); 939336Sdfr register int biosize, diff, i; 941549Srgrimes struct buf *bp = 0, *rabp; 951541Srgrimes struct vattr vattr; 961541Srgrimes struct proc *p; 979336Sdfr struct nfsmount *nmp = VFSTONFS(vp->v_mount); 985455Sdg daddr_t lbn, rabn; 998692Sdg int bufsize; 1007871Sdg int nra, error = 0, n = 0, on = 0, not_readin; 1011541Srgrimes 1021541Srgrimes#ifdef DIAGNOSTIC 1031541Srgrimes if (uio->uio_rw != UIO_READ) 1041541Srgrimes panic("nfs_read mode"); 1051541Srgrimes#endif 1061541Srgrimes if (uio->uio_resid == 0) 1071541Srgrimes return (0); 1089336Sdfr if (uio->uio_offset < 0) 1091541Srgrimes return (EINVAL); 1101541Srgrimes p = uio->uio_procp; 1119336Sdfr if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 1129336Sdfr (void)nfs_fsinfo(nmp, vp, cred, p); 1139428Sdfr biosize = vp->v_mount->mnt_stat.f_iosize; 1141541Srgrimes /* 1151541Srgrimes * For nfs, cache consistency can only be maintained approximately. 1161541Srgrimes * Although RFC1094 does not specify the criteria, the following is 1171541Srgrimes * believed to be compatible with the reference port. 1181541Srgrimes * For nqnfs, full cache consistency is maintained within the loop. 1191541Srgrimes * For nfs: 1201541Srgrimes * If the file's modify time on the server has changed since the 1211541Srgrimes * last read rpc or you have written to the file, 1221541Srgrimes * you may have lost data cache consistency with the 1231541Srgrimes * server, so flush all of the file's data out of the cache. 1241541Srgrimes * Then force a getattr rpc to ensure that you have up to date 1251541Srgrimes * attributes. 1261541Srgrimes * NB: This implies that cache data can be read when up to 1271541Srgrimes * NFS_ATTRTIMEO seconds out of date. If you find that you need current 1281541Srgrimes * attributes this could be forced by setting n_attrstamp to 0 before 1291541Srgrimes * the VOP_GETATTR() call. 1301541Srgrimes */ 13110219Sdfr if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { 1321541Srgrimes if (np->n_flag & NMODIFIED) { 1339336Sdfr if (vp->v_type != VREG) { 1349336Sdfr if (vp->v_type != VDIR) 1359336Sdfr panic("nfs: bioread, not dir"); 1369336Sdfr nfs_invaldir(vp); 1373305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1383305Sphk if (error) 1391541Srgrimes return (error); 1401541Srgrimes } 1411541Srgrimes np->n_attrstamp = 0; 1423305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 1433305Sphk if (error) 1441541Srgrimes return (error); 1451541Srgrimes np->n_mtime = vattr.va_mtime.ts_sec; 1461541Srgrimes } else { 1473305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 1483305Sphk if (error) 1491541Srgrimes return (error); 1501541Srgrimes if (np->n_mtime != vattr.va_mtime.ts_sec) { 1519336Sdfr if (vp->v_type == VDIR) 1529336Sdfr nfs_invaldir(vp); 1533305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1543305Sphk if (error) 1551541Srgrimes return (error); 1561541Srgrimes np->n_mtime = vattr.va_mtime.ts_sec; 1571541Srgrimes } 1581541Srgrimes } 1591541Srgrimes } 1601541Srgrimes do { 1611541Srgrimes 1621541Srgrimes /* 1631541Srgrimes * Get a valid lease. If cached data is stale, flush it. 1641541Srgrimes */ 1651541Srgrimes if (nmp->nm_flag & NFSMNT_NQNFS) { 1669336Sdfr if (NQNFS_CKINVALID(vp, np, ND_READ)) { 1671541Srgrimes do { 1689336Sdfr error = nqnfs_getlease(vp, ND_READ, cred, p); 1691541Srgrimes } while (error == NQNFS_EXPIRED); 1701541Srgrimes if (error) 1711541Srgrimes return (error); 1721541Srgrimes if (np->n_lrev != np->n_brev || 1731541Srgrimes (np->n_flag & NQNFSNONCACHE) || 1741541Srgrimes ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 1759336Sdfr if (vp->v_type == VDIR) 1769336Sdfr nfs_invaldir(vp); 1773305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1783305Sphk if (error) 1791541Srgrimes return (error); 1801541Srgrimes np->n_brev = np->n_lrev; 1811541Srgrimes } 1821541Srgrimes } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 1839336Sdfr nfs_invaldir(vp); 1843305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1853305Sphk if (error) 1861541Srgrimes return (error); 1871541Srgrimes } 1881541Srgrimes } 1891541Srgrimes if (np->n_flag & NQNFSNONCACHE) { 1901541Srgrimes switch (vp->v_type) { 1911541Srgrimes case VREG: 1929336Sdfr return (nfs_readrpc(vp, uio, cred)); 1931541Srgrimes case VLNK: 1949336Sdfr return (nfs_readlinkrpc(vp, uio, cred)); 1951541Srgrimes case VDIR: 1961541Srgrimes break; 1973305Sphk default: 1988876Srgrimes printf(" NQNFSNONCACHE: type %x unexpected\n", 1993305Sphk vp->v_type); 2001541Srgrimes }; 2011541Srgrimes } 2021541Srgrimes switch (vp->v_type) { 2031541Srgrimes case VREG: 2041541Srgrimes nfsstats.biocache_reads++; 2051541Srgrimes lbn = uio->uio_offset / biosize; 2069336Sdfr on = uio->uio_offset & (biosize - 1); 2071541Srgrimes not_readin = 1; 2081541Srgrimes 2091541Srgrimes /* 2101541Srgrimes * Start the read ahead(s), as required. 2111541Srgrimes */ 2129336Sdfr if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 2131541Srgrimes for (nra = 0; nra < nmp->nm_readahead && 2141541Srgrimes (lbn + 1 + nra) * biosize < np->n_size; nra++) { 2155455Sdg rabn = lbn + 1 + nra; 2161541Srgrimes if (!incore(vp, rabn)) { 2171541Srgrimes rabp = nfs_getcacheblk(vp, rabn, biosize, p); 2181541Srgrimes if (!rabp) 2191541Srgrimes return (EINTR); 2208692Sdg if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 2211541Srgrimes rabp->b_flags |= (B_READ | B_ASYNC); 2225455Sdg vfs_busy_pages(rabp, 0); 2231541Srgrimes if (nfs_asyncio(rabp, cred)) { 2245455Sdg rabp->b_flags |= B_INVAL|B_ERROR; 2255455Sdg vfs_unbusy_pages(rabp); 2261541Srgrimes brelse(rabp); 2271541Srgrimes } 2285471Sdg } else { 2295471Sdg brelse(rabp); 2301541Srgrimes } 2311541Srgrimes } 2321541Srgrimes } 2331541Srgrimes } 2341541Srgrimes 2351541Srgrimes /* 2361541Srgrimes * If the block is in the cache and has the required data 2371541Srgrimes * in a valid region, just copy it out. 2381541Srgrimes * Otherwise, get the block and write back/read in, 2391541Srgrimes * as required. 2401541Srgrimes */ 2411541Srgrimesagain: 2428692Sdg bufsize = biosize; 2438692Sdg if ((lbn + 1) * biosize > np->n_size) { 2448692Sdg bufsize = np->n_size - lbn * biosize; 2458692Sdg bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 2468692Sdg } 2478692Sdg bp = nfs_getcacheblk(vp, lbn, bufsize, p); 2487871Sdg if (!bp) 2497871Sdg return (EINTR); 2507871Sdg if ((bp->b_flags & B_CACHE) == 0) { 2517871Sdg bp->b_flags |= B_READ; 2527871Sdg not_readin = 0; 2537871Sdg vfs_busy_pages(bp, 0); 2547871Sdg error = nfs_doio(bp, cred, p); 2557871Sdg if (error) { 2567871Sdg brelse(bp); 2577871Sdg return (error); 2581541Srgrimes } 2591541Srgrimes } 2608692Sdg if (bufsize > on) { 2618692Sdg n = min((unsigned)(bufsize - on), uio->uio_resid); 2628692Sdg } else { 2638692Sdg n = 0; 2648692Sdg } 2651541Srgrimes diff = np->n_size - uio->uio_offset; 2661541Srgrimes if (diff < n) 2671541Srgrimes n = diff; 2681541Srgrimes if (not_readin && n > 0) { 2691541Srgrimes if (on < bp->b_validoff || (on + n) > bp->b_validend) { 2706148Sdg bp->b_flags |= B_NOCACHE; 2711541Srgrimes if (bp->b_dirtyend > 0) { 2721541Srgrimes if ((bp->b_flags & B_DELWRI) == 0) 2731541Srgrimes panic("nfsbioread"); 2741541Srgrimes if (VOP_BWRITE(bp) == EINTR) 2751541Srgrimes return (EINTR); 2761541Srgrimes } else 2771541Srgrimes brelse(bp); 2781541Srgrimes goto again; 2791541Srgrimes } 2801541Srgrimes } 2811541Srgrimes vp->v_lastr = lbn; 2821541Srgrimes diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 2831541Srgrimes if (diff < n) 2841541Srgrimes n = diff; 2851541Srgrimes break; 2861541Srgrimes case VLNK: 2871541Srgrimes nfsstats.biocache_readlinks++; 2881541Srgrimes bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 2891541Srgrimes if (!bp) 2901541Srgrimes return (EINTR); 2917871Sdg if ((bp->b_flags & B_CACHE) == 0) { 2921541Srgrimes bp->b_flags |= B_READ; 2935455Sdg vfs_busy_pages(bp, 0); 2943305Sphk error = nfs_doio(bp, cred, p); 2953305Sphk if (error) { 2965455Sdg bp->b_flags |= B_ERROR; 2971541Srgrimes brelse(bp); 2981541Srgrimes return (error); 2991541Srgrimes } 3001541Srgrimes } 3011541Srgrimes n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 3021541Srgrimes on = 0; 3031541Srgrimes break; 3041541Srgrimes case VDIR: 3051541Srgrimes nfsstats.biocache_readdirs++; 3069336Sdfr lbn = uio->uio_offset / NFS_DIRBLKSIZ; 3079336Sdfr on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 3085455Sdg bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 3091541Srgrimes if (!bp) 3109336Sdfr return (EINTR); 3117871Sdg if ((bp->b_flags & B_CACHE) == 0) { 3129336Sdfr bp->b_flags |= B_READ; 3139336Sdfr vfs_busy_pages(bp, 0); 3149336Sdfr error = nfs_doio(bp, cred, p); 3159336Sdfr if (error) { 3169336Sdfr brelse(bp); 3179336Sdfr while (error == NFSERR_BAD_COOKIE) { 3189336Sdfr nfs_invaldir(vp); 3199336Sdfr error = nfs_vinvalbuf(vp, 0, cred, p, 1); 3209336Sdfr /* 3219336Sdfr * Yuck! The directory has been modified on the 3229336Sdfr * server. The only way to get the block is by 3239336Sdfr * reading from the beginning to get all the 3249336Sdfr * offset cookies. 3259336Sdfr */ 3269336Sdfr for (i = 0; i <= lbn && !error; i++) { 3279336Sdfr bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 3289336Sdfr if (!bp) 3299336Sdfr return (EINTR); 3309336Sdfr if ((bp->b_flags & B_DONE) == 0) { 3319336Sdfr bp->b_flags |= B_READ; 3329336Sdfr vfs_busy_pages(bp, 0); 3339336Sdfr error = nfs_doio(bp, cred, p); 3349336Sdfr if (error) 3359336Sdfr brelse(bp); 3369336Sdfr } 3379336Sdfr } 3381541Srgrimes } 3399336Sdfr if (error) 3409336Sdfr return (error); 3419336Sdfr } 3421541Srgrimes } 3431541Srgrimes 3441541Srgrimes /* 3451541Srgrimes * If not eof and read aheads are enabled, start one. 3461541Srgrimes * (You need the current block first, so that you have the 3479336Sdfr * directory offset cookie of the next block.) 3481541Srgrimes */ 3491541Srgrimes if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 3509336Sdfr (np->n_direofoffset == 0 || 3519336Sdfr (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 3529336Sdfr !(np->n_flag & NQNFSNONCACHE) && 3539336Sdfr !incore(vp, lbn + 1)) { 3549336Sdfr rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 3551541Srgrimes if (rabp) { 3568692Sdg if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 3571541Srgrimes rabp->b_flags |= (B_READ | B_ASYNC); 3585455Sdg vfs_busy_pages(rabp, 0); 3591541Srgrimes if (nfs_asyncio(rabp, cred)) { 3606148Sdg rabp->b_flags |= B_INVAL|B_ERROR; 3615455Sdg vfs_unbusy_pages(rabp); 3621541Srgrimes brelse(rabp); 3631541Srgrimes } 3645471Sdg } else { 3655471Sdg brelse(rabp); 3661541Srgrimes } 3671541Srgrimes } 3681541Srgrimes } 3699336Sdfr n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 3701541Srgrimes break; 3713305Sphk default: 3729336Sdfr printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 3733305Sphk break; 3741541Srgrimes }; 3751541Srgrimes 3761541Srgrimes if (n > 0) { 3777871Sdg error = uiomove(bp->b_data + on, (int)n, uio); 3781541Srgrimes } 3791541Srgrimes switch (vp->v_type) { 3801541Srgrimes case VREG: 3811541Srgrimes break; 3821541Srgrimes case VLNK: 3831541Srgrimes n = 0; 3841541Srgrimes break; 3851541Srgrimes case VDIR: 3869336Sdfr if (np->n_flag & NQNFSNONCACHE) 3879336Sdfr bp->b_flags |= B_INVAL; 3881541Srgrimes break; 3893305Sphk default: 3909336Sdfr printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 3913305Sphk } 3927871Sdg brelse(bp); 3931541Srgrimes } while (error == 0 && uio->uio_resid > 0 && n > 0); 3941541Srgrimes return (error); 3951541Srgrimes} 3961541Srgrimes 3971541Srgrimes/* 3981541Srgrimes * Vnode op for write using bio 3991541Srgrimes */ 4001549Srgrimesint 4011541Srgrimesnfs_write(ap) 4021541Srgrimes struct vop_write_args /* { 4031541Srgrimes struct vnode *a_vp; 4041541Srgrimes struct uio *a_uio; 4051541Srgrimes int a_ioflag; 4061541Srgrimes struct ucred *a_cred; 4071541Srgrimes } */ *ap; 4081541Srgrimes{ 4091541Srgrimes register int biosize; 4101541Srgrimes register struct uio *uio = ap->a_uio; 4111541Srgrimes struct proc *p = uio->uio_procp; 4121541Srgrimes register struct vnode *vp = ap->a_vp; 4131541Srgrimes struct nfsnode *np = VTONFS(vp); 4141541Srgrimes register struct ucred *cred = ap->a_cred; 4151541Srgrimes int ioflag = ap->a_ioflag; 4161541Srgrimes struct buf *bp; 4171541Srgrimes struct vattr vattr; 4189336Sdfr struct nfsmount *nmp = VFSTONFS(vp->v_mount); 41911921Sphk daddr_t lbn; 4208692Sdg int bufsize; 4219336Sdfr int n, on, error = 0, iomode, must_commit; 4221541Srgrimes 4231541Srgrimes#ifdef DIAGNOSTIC 4241541Srgrimes if (uio->uio_rw != UIO_WRITE) 4251541Srgrimes panic("nfs_write mode"); 4261541Srgrimes if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 4271541Srgrimes panic("nfs_write proc"); 4281541Srgrimes#endif 4291541Srgrimes if (vp->v_type != VREG) 4301541Srgrimes return (EIO); 4311541Srgrimes if (np->n_flag & NWRITEERR) { 4321541Srgrimes np->n_flag &= ~NWRITEERR; 4331541Srgrimes return (np->n_error); 4341541Srgrimes } 4359336Sdfr if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 4369336Sdfr (void)nfs_fsinfo(nmp, vp, cred, p); 4371541Srgrimes if (ioflag & (IO_APPEND | IO_SYNC)) { 4381541Srgrimes if (np->n_flag & NMODIFIED) { 4391541Srgrimes np->n_attrstamp = 0; 4403305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4413305Sphk if (error) 4421541Srgrimes return (error); 4431541Srgrimes } 4441541Srgrimes if (ioflag & IO_APPEND) { 4451541Srgrimes np->n_attrstamp = 0; 4463305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 4473305Sphk if (error) 4481541Srgrimes return (error); 4491541Srgrimes uio->uio_offset = np->n_size; 4501541Srgrimes } 4511541Srgrimes } 4521541Srgrimes if (uio->uio_offset < 0) 4531541Srgrimes return (EINVAL); 4541541Srgrimes if (uio->uio_resid == 0) 4551541Srgrimes return (0); 4561541Srgrimes /* 4571541Srgrimes * Maybe this should be above the vnode op call, but so long as 4581541Srgrimes * file servers have no limits, i don't think it matters 4591541Srgrimes */ 4601541Srgrimes if (p && uio->uio_offset + uio->uio_resid > 4611541Srgrimes p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 4621541Srgrimes psignal(p, SIGXFSZ); 4631541Srgrimes return (EFBIG); 4641541Srgrimes } 4651541Srgrimes /* 4661541Srgrimes * I use nm_rsize, not nm_wsize so that all buffer cache blocks 4671541Srgrimes * will be the same size within a filesystem. nfs_writerpc will 4681541Srgrimes * still use nm_wsize when sizing the rpc's. 4691541Srgrimes */ 4709428Sdfr biosize = vp->v_mount->mnt_stat.f_iosize; 4711541Srgrimes do { 4721541Srgrimes 4731541Srgrimes /* 4743664Sphk * XXX make sure we aren't cached in the VM page cache 4753664Sphk */ 4763664Sphk /* 4771541Srgrimes * Check for a valid write lease. 4781541Srgrimes */ 4791541Srgrimes if ((nmp->nm_flag & NFSMNT_NQNFS) && 4809336Sdfr NQNFS_CKINVALID(vp, np, ND_WRITE)) { 4811541Srgrimes do { 4829336Sdfr error = nqnfs_getlease(vp, ND_WRITE, cred, p); 4831541Srgrimes } while (error == NQNFS_EXPIRED); 4841541Srgrimes if (error) 4851541Srgrimes return (error); 4861541Srgrimes if (np->n_lrev != np->n_brev || 4871541Srgrimes (np->n_flag & NQNFSNONCACHE)) { 4883305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4893305Sphk if (error) 4901541Srgrimes return (error); 4911541Srgrimes np->n_brev = np->n_lrev; 4921541Srgrimes } 4931541Srgrimes } 4949336Sdfr if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 4959336Sdfr iomode = NFSV3WRITE_FILESYNC; 4969336Sdfr error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 4979336Sdfr if (must_commit) 4989336Sdfr nfs_clearcommit(vp->v_mount); 4999336Sdfr return (error); 5009336Sdfr } 5011541Srgrimes nfsstats.biocache_writes++; 5021541Srgrimes lbn = uio->uio_offset / biosize; 5031541Srgrimes on = uio->uio_offset & (biosize-1); 5041541Srgrimes n = min((unsigned)(biosize - on), uio->uio_resid); 5051541Srgrimesagain: 5068692Sdg if (uio->uio_offset + n > np->n_size) { 5078692Sdg np->n_size = uio->uio_offset + n; 5088692Sdg vnode_pager_setsize(vp, (u_long)np->n_size); 5098692Sdg } 5108692Sdg bufsize = biosize; 5118692Sdg if ((lbn + 1) * biosize > np->n_size) { 5128692Sdg bufsize = np->n_size - lbn * biosize; 5138692Sdg bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 5148692Sdg } 5158692Sdg bp = nfs_getcacheblk(vp, lbn, bufsize, p); 5161541Srgrimes if (!bp) 5171541Srgrimes return (EINTR); 5181541Srgrimes if (bp->b_wcred == NOCRED) { 5191541Srgrimes crhold(cred); 5201541Srgrimes bp->b_wcred = cred; 5211541Srgrimes } 5221541Srgrimes np->n_flag |= NMODIFIED; 5238692Sdg 5248692Sdg if ((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend > np->n_size) { 5258692Sdg bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 5261541Srgrimes } 5271541Srgrimes 5281541Srgrimes /* 5291541Srgrimes * If the new write will leave a contiguous dirty 5301541Srgrimes * area, just update the b_dirtyoff and b_dirtyend, 5311541Srgrimes * otherwise force a write rpc of the old dirty area. 5321541Srgrimes */ 5331541Srgrimes if (bp->b_dirtyend > 0 && 5341541Srgrimes (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 5351541Srgrimes bp->b_proc = p; 5361541Srgrimes if (VOP_BWRITE(bp) == EINTR) 5371541Srgrimes return (EINTR); 5381541Srgrimes goto again; 5391541Srgrimes } 5401541Srgrimes 5411541Srgrimes /* 5421541Srgrimes * Check for valid write lease and get one as required. 5431541Srgrimes * In case getblk() and/or bwrite() delayed us. 5441541Srgrimes */ 5451541Srgrimes if ((nmp->nm_flag & NFSMNT_NQNFS) && 5469336Sdfr NQNFS_CKINVALID(vp, np, ND_WRITE)) { 5471541Srgrimes do { 5489336Sdfr error = nqnfs_getlease(vp, ND_WRITE, cred, p); 5491541Srgrimes } while (error == NQNFS_EXPIRED); 5501541Srgrimes if (error) { 5511541Srgrimes brelse(bp); 5521541Srgrimes return (error); 5531541Srgrimes } 5541541Srgrimes if (np->n_lrev != np->n_brev || 5551541Srgrimes (np->n_flag & NQNFSNONCACHE)) { 5561541Srgrimes brelse(bp); 5573305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 5583305Sphk if (error) 5591541Srgrimes return (error); 5601541Srgrimes np->n_brev = np->n_lrev; 5611541Srgrimes goto again; 5621541Srgrimes } 5631541Srgrimes } 5643305Sphk error = uiomove((char *)bp->b_data + on, n, uio); 5653305Sphk if (error) { 5661541Srgrimes bp->b_flags |= B_ERROR; 5671541Srgrimes brelse(bp); 5681541Srgrimes return (error); 5691541Srgrimes } 5701541Srgrimes if (bp->b_dirtyend > 0) { 5711541Srgrimes bp->b_dirtyoff = min(on, bp->b_dirtyoff); 5721541Srgrimes bp->b_dirtyend = max((on + n), bp->b_dirtyend); 5731541Srgrimes } else { 5741541Srgrimes bp->b_dirtyoff = on; 5751541Srgrimes bp->b_dirtyend = on + n; 5761541Srgrimes } 5771541Srgrimes if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 5781541Srgrimes bp->b_validoff > bp->b_dirtyend) { 5791541Srgrimes bp->b_validoff = bp->b_dirtyoff; 5801541Srgrimes bp->b_validend = bp->b_dirtyend; 5811541Srgrimes } else { 5821541Srgrimes bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 5831541Srgrimes bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 5841541Srgrimes } 5851541Srgrimes /* 5861541Srgrimes * If the lease is non-cachable or IO_SYNC do bwrite(). 5871541Srgrimes */ 5881541Srgrimes if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 5891541Srgrimes bp->b_proc = p; 5903305Sphk error = VOP_BWRITE(bp); 5913305Sphk if (error) 5921541Srgrimes return (error); 5939336Sdfr if (np->n_flag & NQNFSNONCACHE) { 5949336Sdfr error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 5959336Sdfr if (error) 5969336Sdfr return (error); 5979336Sdfr } 5981541Srgrimes } else if ((n + on) == biosize && 5991541Srgrimes (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 6001541Srgrimes bp->b_proc = (struct proc *)0; 6019336Sdfr bp->b_flags |= B_ASYNC; 6029336Sdfr (void)nfs_writebp(bp, 0); 6031541Srgrimes } else 6041541Srgrimes bdwrite(bp); 6051541Srgrimes } while (uio->uio_resid > 0 && n > 0); 6061541Srgrimes return (0); 6071541Srgrimes} 6081541Srgrimes 6091541Srgrimes/* 6101541Srgrimes * Get an nfs cache block. 6111541Srgrimes * Allocate a new one if the block isn't currently in the cache 6121541Srgrimes * and return the block marked busy. If the calling process is 6131541Srgrimes * interrupted by a signal for an interruptible mount point, return 6141541Srgrimes * NULL. 6151541Srgrimes */ 6161541Srgrimesstruct buf * 6171541Srgrimesnfs_getcacheblk(vp, bn, size, p) 6181541Srgrimes struct vnode *vp; 6191541Srgrimes daddr_t bn; 6201541Srgrimes int size; 6211541Srgrimes struct proc *p; 6221541Srgrimes{ 6231541Srgrimes register struct buf *bp; 6241541Srgrimes struct nfsmount *nmp = VFSTONFS(vp->v_mount); 6259428Sdfr int biosize = vp->v_mount->mnt_stat.f_iosize; 6261541Srgrimes 6271541Srgrimes if (nmp->nm_flag & NFSMNT_INT) { 6281541Srgrimes bp = getblk(vp, bn, size, PCATCH, 0); 6291541Srgrimes while (bp == (struct buf *)0) { 6301541Srgrimes if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 6311541Srgrimes return ((struct buf *)0); 6321541Srgrimes bp = getblk(vp, bn, size, 0, 2 * hz); 6331541Srgrimes } 6341541Srgrimes } else 6351541Srgrimes bp = getblk(vp, bn, size, 0, 0); 6365455Sdg 6375455Sdg if( vp->v_type == VREG) 6389336Sdfr bp->b_blkno = (bn * biosize) / DEV_BSIZE; 6395455Sdg 6401541Srgrimes return (bp); 6411541Srgrimes} 6421541Srgrimes 6431541Srgrimes/* 6441541Srgrimes * Flush and invalidate all dirty buffers. If another process is already 6451541Srgrimes * doing the flush, just wait for completion. 6461541Srgrimes */ 6471549Srgrimesint 6481541Srgrimesnfs_vinvalbuf(vp, flags, cred, p, intrflg) 6491541Srgrimes struct vnode *vp; 6501541Srgrimes int flags; 6511541Srgrimes struct ucred *cred; 6521541Srgrimes struct proc *p; 6531541Srgrimes int intrflg; 6541541Srgrimes{ 6551541Srgrimes register struct nfsnode *np = VTONFS(vp); 6561541Srgrimes struct nfsmount *nmp = VFSTONFS(vp->v_mount); 6571541Srgrimes int error = 0, slpflag, slptimeo; 6581541Srgrimes 6591541Srgrimes if ((nmp->nm_flag & NFSMNT_INT) == 0) 6601541Srgrimes intrflg = 0; 6611541Srgrimes if (intrflg) { 6621541Srgrimes slpflag = PCATCH; 6631541Srgrimes slptimeo = 2 * hz; 6641541Srgrimes } else { 6651541Srgrimes slpflag = 0; 6661541Srgrimes slptimeo = 0; 6671541Srgrimes } 6681541Srgrimes /* 6691541Srgrimes * First wait for any other process doing a flush to complete. 6701541Srgrimes */ 6711541Srgrimes while (np->n_flag & NFLUSHINPROG) { 6721541Srgrimes np->n_flag |= NFLUSHWANT; 6731541Srgrimes error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 6741541Srgrimes slptimeo); 6751541Srgrimes if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 6761541Srgrimes return (EINTR); 6771541Srgrimes } 6781541Srgrimes 6791541Srgrimes /* 6801541Srgrimes * Now, flush as required. 6811541Srgrimes */ 6821541Srgrimes np->n_flag |= NFLUSHINPROG; 6831541Srgrimes error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 6841541Srgrimes while (error) { 6851541Srgrimes if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 6861541Srgrimes np->n_flag &= ~NFLUSHINPROG; 6871541Srgrimes if (np->n_flag & NFLUSHWANT) { 6881541Srgrimes np->n_flag &= ~NFLUSHWANT; 6891541Srgrimes wakeup((caddr_t)&np->n_flag); 6901541Srgrimes } 6911541Srgrimes return (EINTR); 6921541Srgrimes } 6931541Srgrimes error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 6941541Srgrimes } 6951541Srgrimes np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 6961541Srgrimes if (np->n_flag & NFLUSHWANT) { 6971541Srgrimes np->n_flag &= ~NFLUSHWANT; 6981541Srgrimes wakeup((caddr_t)&np->n_flag); 6991541Srgrimes } 7001541Srgrimes return (0); 7011541Srgrimes} 7021541Srgrimes 7031541Srgrimes/* 7041541Srgrimes * Initiate asynchronous I/O. Return an error if no nfsiods are available. 7051541Srgrimes * This is mainly to avoid queueing async I/O requests when the nfsiods 7061541Srgrimes * are all hung on a dead server. 7071541Srgrimes */ 7081549Srgrimesint 7091541Srgrimesnfs_asyncio(bp, cred) 7101541Srgrimes register struct buf *bp; 7111541Srgrimes struct ucred *cred; 7121541Srgrimes{ 7131541Srgrimes register int i; 7141541Srgrimes 7151541Srgrimes if (nfs_numasync == 0) 7161541Srgrimes return (EIO); 7171541Srgrimes for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 7181541Srgrimes if (nfs_iodwant[i]) { 7191541Srgrimes if (bp->b_flags & B_READ) { 7201541Srgrimes if (bp->b_rcred == NOCRED && cred != NOCRED) { 7211541Srgrimes crhold(cred); 7221541Srgrimes bp->b_rcred = cred; 7231541Srgrimes } 7241541Srgrimes } else { 7259336Sdfr bp->b_flags |= B_WRITEINPROG; 7261541Srgrimes if (bp->b_wcred == NOCRED && cred != NOCRED) { 7271541Srgrimes crhold(cred); 7281541Srgrimes bp->b_wcred = cred; 7291541Srgrimes } 7301541Srgrimes } 7318876Srgrimes 7321541Srgrimes TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist); 7331541Srgrimes nfs_iodwant[i] = (struct proc *)0; 7341541Srgrimes wakeup((caddr_t)&nfs_iodwant[i]); 7351541Srgrimes return (0); 7361541Srgrimes } 7379336Sdfr 7389336Sdfr /* 7399336Sdfr * If it is a read or a write already marked B_WRITEINPROG or B_NOCACHE 7409336Sdfr * return EIO so the process will call nfs_doio() and do it 7419336Sdfr * synchronously. 7429336Sdfr */ 7439336Sdfr if (bp->b_flags & (B_READ | B_WRITEINPROG | B_NOCACHE)) 7449336Sdfr return (EIO); 7459336Sdfr 7469336Sdfr /* 7479336Sdfr * Just turn the async write into a delayed write, instead of 7489336Sdfr * doing in synchronously. Hopefully, at least one of the nfsiods 7499336Sdfr * is currently doing a write for this file and will pick up the 7509336Sdfr * delayed writes before going back to sleep. 7519336Sdfr */ 7529336Sdfr bp->b_flags |= B_DELWRI; 7539336Sdfr reassignbuf(bp, bp->b_vp); 7549336Sdfr biodone(bp); 7559336Sdfr return (0); 7561541Srgrimes} 7571541Srgrimes 7581541Srgrimes/* 7591541Srgrimes * Do an I/O operation to/from a cache block. This may be called 7601541Srgrimes * synchronously or from an nfsiod. 7611541Srgrimes */ 7621541Srgrimesint 7631541Srgrimesnfs_doio(bp, cr, p) 7641541Srgrimes register struct buf *bp; 7653305Sphk struct ucred *cr; 7661541Srgrimes struct proc *p; 7671541Srgrimes{ 7681541Srgrimes register struct uio *uiop; 7691541Srgrimes register struct vnode *vp; 7701541Srgrimes struct nfsnode *np; 7711541Srgrimes struct nfsmount *nmp; 7729336Sdfr int error = 0, diff, len, iomode, must_commit = 0; 7731541Srgrimes struct uio uio; 7741541Srgrimes struct iovec io; 7751541Srgrimes 7761541Srgrimes vp = bp->b_vp; 7771541Srgrimes np = VTONFS(vp); 7781541Srgrimes nmp = VFSTONFS(vp->v_mount); 7791541Srgrimes uiop = &uio; 7801541Srgrimes uiop->uio_iov = &io; 7811541Srgrimes uiop->uio_iovcnt = 1; 7821541Srgrimes uiop->uio_segflg = UIO_SYSSPACE; 7831541Srgrimes uiop->uio_procp = p; 7841541Srgrimes 7851541Srgrimes /* 7861541Srgrimes * Historically, paging was done with physio, but no more. 7871541Srgrimes */ 7883664Sphk if (bp->b_flags & B_PHYS) { 7893664Sphk /* 7903664Sphk * ...though reading /dev/drum still gets us here. 7913664Sphk */ 7921541Srgrimes io.iov_len = uiop->uio_resid = bp->b_bcount; 7933664Sphk /* mapping was done by vmapbuf() */ 7941541Srgrimes io.iov_base = bp->b_data; 7959336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 7963664Sphk if (bp->b_flags & B_READ) { 7973664Sphk uiop->uio_rw = UIO_READ; 7983664Sphk nfsstats.read_physios++; 7993664Sphk error = nfs_readrpc(vp, uiop, cr); 8003664Sphk } else { 8019336Sdfr int com; 8029336Sdfr 8039336Sdfr iomode = NFSV3WRITE_DATASYNC; 8043664Sphk uiop->uio_rw = UIO_WRITE; 8053664Sphk nfsstats.write_physios++; 8069336Sdfr error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 8073664Sphk } 8083664Sphk if (error) { 8093664Sphk bp->b_flags |= B_ERROR; 8103664Sphk bp->b_error = error; 8113664Sphk } 8123664Sphk } else if (bp->b_flags & B_READ) { 8133664Sphk io.iov_len = uiop->uio_resid = bp->b_bcount; 8143664Sphk io.iov_base = bp->b_data; 8151541Srgrimes uiop->uio_rw = UIO_READ; 8161541Srgrimes switch (vp->v_type) { 8171541Srgrimes case VREG: 8189336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 8191541Srgrimes nfsstats.read_bios++; 8201541Srgrimes error = nfs_readrpc(vp, uiop, cr); 8211541Srgrimes if (!error) { 8221541Srgrimes bp->b_validoff = 0; 8231541Srgrimes if (uiop->uio_resid) { 8241541Srgrimes /* 8251541Srgrimes * If len > 0, there is a hole in the file and 8261541Srgrimes * no writes after the hole have been pushed to 8271541Srgrimes * the server yet. 8281541Srgrimes * Just zero fill the rest of the valid area. 8291541Srgrimes */ 8301541Srgrimes diff = bp->b_bcount - uiop->uio_resid; 8319336Sdfr len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 8321541Srgrimes + diff); 8331541Srgrimes if (len > 0) { 8341541Srgrimes len = min(len, uiop->uio_resid); 8351541Srgrimes bzero((char *)bp->b_data + diff, len); 8361541Srgrimes bp->b_validend = diff + len; 8371541Srgrimes } else 8381541Srgrimes bp->b_validend = diff; 8391541Srgrimes } else 8401541Srgrimes bp->b_validend = bp->b_bcount; 8411541Srgrimes } 8421541Srgrimes if (p && (vp->v_flag & VTEXT) && 8431541Srgrimes (((nmp->nm_flag & NFSMNT_NQNFS) && 8449336Sdfr NQNFS_CKINVALID(vp, np, ND_READ) && 8451541Srgrimes np->n_lrev != np->n_brev) || 8461541Srgrimes (!(nmp->nm_flag & NFSMNT_NQNFS) && 8471541Srgrimes np->n_mtime != np->n_vattr.va_mtime.ts_sec))) { 8481541Srgrimes uprintf("Process killed due to text file modification\n"); 8491541Srgrimes psignal(p, SIGKILL); 8509336Sdfr#ifdef __NetBSD__ 8519336Sdfr p->p_holdcnt++; 8529336Sdfr#else 8531541Srgrimes p->p_flag |= P_NOSWAP; 8549336Sdfr#endif 8551541Srgrimes } 8561541Srgrimes break; 8571541Srgrimes case VLNK: 8589336Sdfr uiop->uio_offset = (off_t)0; 8591541Srgrimes nfsstats.readlink_bios++; 8601541Srgrimes error = nfs_readlinkrpc(vp, uiop, cr); 8611541Srgrimes break; 8621541Srgrimes case VDIR: 8631541Srgrimes nfsstats.readdir_bios++; 8649336Sdfr uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 8659336Sdfr if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 8669336Sdfr error = nfs_readdirplusrpc(vp, uiop, cr); 8679336Sdfr if (error == NFSERR_NOTSUPP) 8689336Sdfr nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 8699336Sdfr } 8709336Sdfr if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 8719336Sdfr error = nfs_readdirrpc(vp, uiop, cr); 8721541Srgrimes break; 8733305Sphk default: 8743305Sphk printf("nfs_doio: type %x unexpected\n",vp->v_type); 8753305Sphk break; 8761541Srgrimes }; 8771541Srgrimes if (error) { 8781541Srgrimes bp->b_flags |= B_ERROR; 8791541Srgrimes bp->b_error = error; 8801541Srgrimes } 8811541Srgrimes } else { 8828692Sdg if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 8838692Sdg bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 8848692Sdg 8858692Sdg if (bp->b_dirtyend > bp->b_dirtyoff) { 8868692Sdg io.iov_len = uiop->uio_resid = bp->b_dirtyend 8879336Sdfr - bp->b_dirtyoff; 8889336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 8899336Sdfr + bp->b_dirtyoff; 8908692Sdg io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 8918692Sdg uiop->uio_rw = UIO_WRITE; 8928692Sdg nfsstats.write_bios++; 8939336Sdfr if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 8949336Sdfr iomode = NFSV3WRITE_UNSTABLE; 8958692Sdg else 8969336Sdfr iomode = NFSV3WRITE_FILESYNC; 8979336Sdfr bp->b_flags |= B_WRITEINPROG; 8989336Sdfr error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 8999336Sdfr if (!error && iomode == NFSV3WRITE_UNSTABLE) 9009336Sdfr bp->b_flags |= B_NEEDCOMMIT; 9019336Sdfr else 9029336Sdfr bp->b_flags &= ~B_NEEDCOMMIT; 9039336Sdfr bp->b_flags &= ~B_WRITEINPROG; 9048692Sdg 9059336Sdfr /* 9069336Sdfr * For an interrupted write, the buffer is still valid 9079336Sdfr * and the write hasn't been pushed to the server yet, 9089336Sdfr * so we can't set B_ERROR and report the interruption 9099336Sdfr * by setting B_EINTR. For the B_ASYNC case, B_EINTR 9109336Sdfr * is not relevant, so the rpc attempt is essentially 9119336Sdfr * a noop. For the case of a V3 write rpc not being 9129336Sdfr * committed to stable storage, the block is still 9139336Sdfr * dirty and requires either a commit rpc or another 9149336Sdfr * write rpc with iomode == NFSV3WRITE_FILESYNC before 9159336Sdfr * the block is reused. This is indicated by setting 9169336Sdfr * the B_DELWRI and B_NEEDCOMMIT flags. 9179336Sdfr */ 9189336Sdfr if (error == EINTR 9199336Sdfr || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 9208692Sdg bp->b_flags &= ~(B_INVAL|B_NOCACHE); 9218692Sdg bp->b_flags |= B_DELWRI; 9221541Srgrimes 9231541Srgrimes /* 9241541Srgrimes * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 9251541Srgrimes * buffer to the clean list, we have to reassign it back to the 9261541Srgrimes * dirty one. Ugh. 9271541Srgrimes */ 9288692Sdg if (bp->b_flags & B_ASYNC) 9298692Sdg reassignbuf(bp, vp); 9308692Sdg else 9318692Sdg bp->b_flags |= B_EINTR; 9328692Sdg } else { 9338692Sdg if (error) { 9348692Sdg bp->b_flags |= B_ERROR; 9358692Sdg bp->b_error = np->n_error = error; 9368692Sdg np->n_flag |= NWRITEERR; 9378692Sdg } 9388692Sdg bp->b_dirtyoff = bp->b_dirtyend = 0; 9398692Sdg } 9401541Srgrimes } else { 9418692Sdg bp->b_resid = 0; 9428692Sdg biodone(bp); 9438692Sdg return (0); 9441541Srgrimes } 9451541Srgrimes } 9461541Srgrimes bp->b_resid = uiop->uio_resid; 9479336Sdfr if (must_commit) 9489336Sdfr nfs_clearcommit(vp->v_mount); 9491541Srgrimes biodone(bp); 9501541Srgrimes return (error); 9511541Srgrimes} 952