nfs_bio.c revision 25023
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 * 3622521Sdyson * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 3725023Sdfr * $Id: nfs_bio.c,v 1.35 1997/04/18 14:11:59 dfr Exp $ 381541Srgrimes */ 391541Srgrimes 4022521Sdyson 411541Srgrimes#include <sys/param.h> 421541Srgrimes#include <sys/systm.h> 431541Srgrimes#include <sys/resourcevar.h> 443305Sphk#include <sys/signalvar.h> 451541Srgrimes#include <sys/proc.h> 461541Srgrimes#include <sys/buf.h> 471541Srgrimes#include <sys/vnode.h> 481541Srgrimes#include <sys/mount.h> 491541Srgrimes#include <sys/kernel.h> 5018866Sdfr#include <sys/sysctl.h> 511541Srgrimes 521541Srgrimes#include <vm/vm.h> 5312662Sdg#include <vm/vm_param.h> 5412662Sdg#include <vm/vm_extern.h> 551541Srgrimes 561541Srgrimes#include <nfs/rpcv2.h> 579336Sdfr#include <nfs/nfsproto.h> 581541Srgrimes#include <nfs/nfs.h> 591541Srgrimes#include <nfs/nfsmount.h> 601541Srgrimes#include <nfs/nqnfs.h> 619336Sdfr#include <nfs/nfsnode.h> 621541Srgrimes 6312911Sphkstatic struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 6412588Sbde struct proc *p)); 6512588Sbde 661541Srgrimesextern int nfs_numasync; 679336Sdfrextern struct nfsstats nfsstats; 681541Srgrimes 691541Srgrimes/* 701541Srgrimes * Vnode op for read using bio 711541Srgrimes * Any similarity to readip() is purely coincidental 721541Srgrimes */ 731549Srgrimesint 741541Srgrimesnfs_bioread(vp, uio, ioflag, cred) 751541Srgrimes register struct vnode *vp; 761541Srgrimes register struct uio *uio; 771541Srgrimes int ioflag; 781541Srgrimes struct ucred *cred; 791541Srgrimes{ 801541Srgrimes register struct nfsnode *np = VTONFS(vp); 819336Sdfr register int biosize, diff, i; 821549Srgrimes struct buf *bp = 0, *rabp; 831541Srgrimes struct vattr vattr; 841541Srgrimes struct proc *p; 859336Sdfr struct nfsmount *nmp = VFSTONFS(vp->v_mount); 865455Sdg daddr_t lbn, rabn; 878692Sdg int bufsize; 887871Sdg int nra, error = 0, n = 0, on = 0, not_readin; 891541Srgrimes 901541Srgrimes#ifdef DIAGNOSTIC 911541Srgrimes if (uio->uio_rw != UIO_READ) 921541Srgrimes panic("nfs_read mode"); 931541Srgrimes#endif 941541Srgrimes if (uio->uio_resid == 0) 951541Srgrimes return (0); 969336Sdfr if (uio->uio_offset < 0) 971541Srgrimes return (EINVAL); 981541Srgrimes p = uio->uio_procp; 999336Sdfr if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 1009336Sdfr (void)nfs_fsinfo(nmp, vp, cred, p); 1019428Sdfr biosize = vp->v_mount->mnt_stat.f_iosize; 1021541Srgrimes /* 1031541Srgrimes * For nfs, cache consistency can only be maintained approximately. 1041541Srgrimes * Although RFC1094 does not specify the criteria, the following is 1051541Srgrimes * believed to be compatible with the reference port. 1061541Srgrimes * For nqnfs, full cache consistency is maintained within the loop. 1071541Srgrimes * For nfs: 1081541Srgrimes * If the file's modify time on the server has changed since the 1091541Srgrimes * last read rpc or you have written to the file, 1101541Srgrimes * you may have lost data cache consistency with the 1111541Srgrimes * server, so flush all of the file's data out of the cache. 1121541Srgrimes * Then force a getattr rpc to ensure that you have up to date 1131541Srgrimes * attributes. 1141541Srgrimes * NB: This implies that cache data can be read when up to 1151541Srgrimes * NFS_ATTRTIMEO seconds out of date. If you find that you need current 1161541Srgrimes * attributes this could be forced by setting n_attrstamp to 0 before 1171541Srgrimes * the VOP_GETATTR() call. 1181541Srgrimes */ 11910219Sdfr if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { 1201541Srgrimes if (np->n_flag & NMODIFIED) { 1219336Sdfr if (vp->v_type != VREG) { 1229336Sdfr if (vp->v_type != VDIR) 1239336Sdfr panic("nfs: bioread, not dir"); 1249336Sdfr nfs_invaldir(vp); 1253305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1263305Sphk if (error) 1271541Srgrimes return (error); 1281541Srgrimes } 1291541Srgrimes np->n_attrstamp = 0; 1303305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 1313305Sphk if (error) 1321541Srgrimes return (error); 13318397Snate np->n_mtime = vattr.va_mtime.tv_sec; 1341541Srgrimes } else { 1353305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 1363305Sphk if (error) 1371541Srgrimes return (error); 13818397Snate if (np->n_mtime != vattr.va_mtime.tv_sec) { 1399336Sdfr if (vp->v_type == VDIR) 1409336Sdfr nfs_invaldir(vp); 1413305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1423305Sphk if (error) 1431541Srgrimes return (error); 14418397Snate np->n_mtime = vattr.va_mtime.tv_sec; 1451541Srgrimes } 1461541Srgrimes } 1471541Srgrimes } 1481541Srgrimes do { 1491541Srgrimes 1501541Srgrimes /* 1511541Srgrimes * Get a valid lease. If cached data is stale, flush it. 1521541Srgrimes */ 1531541Srgrimes if (nmp->nm_flag & NFSMNT_NQNFS) { 1549336Sdfr if (NQNFS_CKINVALID(vp, np, ND_READ)) { 1551541Srgrimes do { 1569336Sdfr error = nqnfs_getlease(vp, ND_READ, cred, p); 1571541Srgrimes } while (error == NQNFS_EXPIRED); 1581541Srgrimes if (error) 1591541Srgrimes return (error); 1601541Srgrimes if (np->n_lrev != np->n_brev || 1611541Srgrimes (np->n_flag & NQNFSNONCACHE) || 1621541Srgrimes ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 1639336Sdfr if (vp->v_type == VDIR) 1649336Sdfr nfs_invaldir(vp); 1653305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1663305Sphk if (error) 1671541Srgrimes return (error); 1681541Srgrimes np->n_brev = np->n_lrev; 1691541Srgrimes } 1701541Srgrimes } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 1719336Sdfr nfs_invaldir(vp); 1723305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 1733305Sphk if (error) 1741541Srgrimes return (error); 1751541Srgrimes } 1761541Srgrimes } 1771541Srgrimes if (np->n_flag & NQNFSNONCACHE) { 1781541Srgrimes switch (vp->v_type) { 1791541Srgrimes case VREG: 1809336Sdfr return (nfs_readrpc(vp, uio, cred)); 1811541Srgrimes case VLNK: 1829336Sdfr return (nfs_readlinkrpc(vp, uio, cred)); 1831541Srgrimes case VDIR: 1841541Srgrimes break; 1853305Sphk default: 18622521Sdyson printf(" NQNFSNONCACHE: type %x unexpected\n", 1873305Sphk vp->v_type); 1881541Srgrimes }; 1891541Srgrimes } 1901541Srgrimes switch (vp->v_type) { 1911541Srgrimes case VREG: 1921541Srgrimes nfsstats.biocache_reads++; 1931541Srgrimes lbn = uio->uio_offset / biosize; 1949336Sdfr on = uio->uio_offset & (biosize - 1); 1951541Srgrimes not_readin = 1; 1961541Srgrimes 1971541Srgrimes /* 1981541Srgrimes * Start the read ahead(s), as required. 1991541Srgrimes */ 2009336Sdfr if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 2011541Srgrimes for (nra = 0; nra < nmp->nm_readahead && 20213612Smpp (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { 2035455Sdg rabn = lbn + 1 + nra; 2041541Srgrimes if (!incore(vp, rabn)) { 2051541Srgrimes rabp = nfs_getcacheblk(vp, rabn, biosize, p); 2061541Srgrimes if (!rabp) 2071541Srgrimes return (EINTR); 2088692Sdg if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 2091541Srgrimes rabp->b_flags |= (B_READ | B_ASYNC); 2105455Sdg vfs_busy_pages(rabp, 0); 2111541Srgrimes if (nfs_asyncio(rabp, cred)) { 2125455Sdg rabp->b_flags |= B_INVAL|B_ERROR; 2135455Sdg vfs_unbusy_pages(rabp); 2141541Srgrimes brelse(rabp); 2151541Srgrimes } 21622521Sdyson } else 2175471Sdg brelse(rabp); 2181541Srgrimes } 2191541Srgrimes } 2201541Srgrimes } 2211541Srgrimes 2221541Srgrimes /* 2231541Srgrimes * If the block is in the cache and has the required data 2241541Srgrimes * in a valid region, just copy it out. 2251541Srgrimes * Otherwise, get the block and write back/read in, 2261541Srgrimes * as required. 2271541Srgrimes */ 2281541Srgrimesagain: 2298692Sdg bufsize = biosize; 23013612Smpp if ((off_t)(lbn + 1) * biosize > np->n_size && 23113612Smpp (off_t)(lbn + 1) * biosize - np->n_size < biosize) { 2328692Sdg bufsize = np->n_size - lbn * biosize; 2338692Sdg bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 2348692Sdg } 2358692Sdg bp = nfs_getcacheblk(vp, lbn, bufsize, p); 2367871Sdg if (!bp) 2377871Sdg return (EINTR); 2387871Sdg if ((bp->b_flags & B_CACHE) == 0) { 2397871Sdg bp->b_flags |= B_READ; 24016192Spst bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL); 2417871Sdg not_readin = 0; 2427871Sdg vfs_busy_pages(bp, 0); 2437871Sdg error = nfs_doio(bp, cred, p); 2447871Sdg if (error) { 2457871Sdg brelse(bp); 2467871Sdg return (error); 2471541Srgrimes } 2481541Srgrimes } 2498692Sdg if (bufsize > on) { 2508692Sdg n = min((unsigned)(bufsize - on), uio->uio_resid); 2518692Sdg } else { 2528692Sdg n = 0; 2538692Sdg } 2541541Srgrimes diff = np->n_size - uio->uio_offset; 2551541Srgrimes if (diff < n) 2561541Srgrimes n = diff; 2571541Srgrimes if (not_readin && n > 0) { 2581541Srgrimes if (on < bp->b_validoff || (on + n) > bp->b_validend) { 2596148Sdg bp->b_flags |= B_NOCACHE; 26022521Sdyson bp->b_flags |= B_INVAFTERWRITE; 2611541Srgrimes if (bp->b_dirtyend > 0) { 2621541Srgrimes if ((bp->b_flags & B_DELWRI) == 0) 2631541Srgrimes panic("nfsbioread"); 2641541Srgrimes if (VOP_BWRITE(bp) == EINTR) 2651541Srgrimes return (EINTR); 2661541Srgrimes } else 2671541Srgrimes brelse(bp); 2681541Srgrimes goto again; 2691541Srgrimes } 2701541Srgrimes } 2711541Srgrimes vp->v_lastr = lbn; 2721541Srgrimes diff = (on >= bp->b_validend) ? 0 : (bp->b_validend - on); 2731541Srgrimes if (diff < n) 2741541Srgrimes n = diff; 2751541Srgrimes break; 2761541Srgrimes case VLNK: 2771541Srgrimes nfsstats.biocache_readlinks++; 2781541Srgrimes bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 2791541Srgrimes if (!bp) 2801541Srgrimes return (EINTR); 2817871Sdg if ((bp->b_flags & B_CACHE) == 0) { 2821541Srgrimes bp->b_flags |= B_READ; 2835455Sdg vfs_busy_pages(bp, 0); 2843305Sphk error = nfs_doio(bp, cred, p); 2853305Sphk if (error) { 2865455Sdg bp->b_flags |= B_ERROR; 2871541Srgrimes brelse(bp); 2881541Srgrimes return (error); 2891541Srgrimes } 2901541Srgrimes } 2911541Srgrimes n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 2921541Srgrimes on = 0; 2931541Srgrimes break; 2941541Srgrimes case VDIR: 2951541Srgrimes nfsstats.biocache_readdirs++; 29624577Sdfr if (np->n_direofoffset 29724577Sdfr && uio->uio_offset >= np->n_direofoffset) { 29824577Sdfr return (0); 29924577Sdfr } 3009336Sdfr lbn = uio->uio_offset / NFS_DIRBLKSIZ; 3019336Sdfr on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 3025455Sdg bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 3031541Srgrimes if (!bp) 3049336Sdfr return (EINTR); 3057871Sdg if ((bp->b_flags & B_CACHE) == 0) { 3069336Sdfr bp->b_flags |= B_READ; 3079336Sdfr vfs_busy_pages(bp, 0); 3089336Sdfr error = nfs_doio(bp, cred, p); 3099336Sdfr if (error) { 31019070Sdfr vfs_unbusy_pages(bp); 3119336Sdfr brelse(bp); 3129336Sdfr while (error == NFSERR_BAD_COOKIE) { 3139336Sdfr nfs_invaldir(vp); 3149336Sdfr error = nfs_vinvalbuf(vp, 0, cred, p, 1); 3159336Sdfr /* 3169336Sdfr * Yuck! The directory has been modified on the 3179336Sdfr * server. The only way to get the block is by 3189336Sdfr * reading from the beginning to get all the 3199336Sdfr * offset cookies. 3209336Sdfr */ 3219336Sdfr for (i = 0; i <= lbn && !error; i++) { 32224577Sdfr if (np->n_direofoffset 32324577Sdfr && (i * NFS_DIRBLKSIZ) >= np->n_direofoffset) 32424577Sdfr return (0); 3259336Sdfr bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 3269336Sdfr if (!bp) 3279336Sdfr return (EINTR); 3289336Sdfr if ((bp->b_flags & B_DONE) == 0) { 3299336Sdfr bp->b_flags |= B_READ; 3309336Sdfr vfs_busy_pages(bp, 0); 3319336Sdfr error = nfs_doio(bp, cred, p); 33219070Sdfr if (error) { 33319070Sdfr vfs_unbusy_pages(bp); 3349336Sdfr brelse(bp); 33519070Sdfr } else if (i < lbn) 33619070Sdfr brelse(bp); 3379336Sdfr } 3389336Sdfr } 3391541Srgrimes } 3409336Sdfr if (error) 3419336Sdfr return (error); 3429336Sdfr } 3431541Srgrimes } 3441541Srgrimes 3451541Srgrimes /* 3461541Srgrimes * If not eof and read aheads are enabled, start one. 3471541Srgrimes * (You need the current block first, so that you have the 3489336Sdfr * directory offset cookie of the next block.) 3491541Srgrimes */ 3501541Srgrimes if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 3519336Sdfr (np->n_direofoffset == 0 || 3529336Sdfr (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 3539336Sdfr !(np->n_flag & NQNFSNONCACHE) && 3549336Sdfr !incore(vp, lbn + 1)) { 3559336Sdfr rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 3561541Srgrimes if (rabp) { 3578692Sdg if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 3581541Srgrimes rabp->b_flags |= (B_READ | B_ASYNC); 3595455Sdg vfs_busy_pages(rabp, 0); 3601541Srgrimes if (nfs_asyncio(rabp, cred)) { 3616148Sdg rabp->b_flags |= B_INVAL|B_ERROR; 3625455Sdg vfs_unbusy_pages(rabp); 3631541Srgrimes brelse(rabp); 3641541Srgrimes } 3655471Sdg } else { 3665471Sdg brelse(rabp); 3671541Srgrimes } 3681541Srgrimes } 3691541Srgrimes } 3709336Sdfr n = min(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 3711541Srgrimes break; 3723305Sphk default: 3739336Sdfr printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 3743305Sphk break; 3751541Srgrimes }; 3761541Srgrimes 3771541Srgrimes if (n > 0) { 3787871Sdg error = uiomove(bp->b_data + on, (int)n, uio); 3791541Srgrimes } 3801541Srgrimes switch (vp->v_type) { 3811541Srgrimes case VREG: 3821541Srgrimes break; 3831541Srgrimes case VLNK: 3841541Srgrimes n = 0; 3851541Srgrimes break; 3861541Srgrimes case VDIR: 3879336Sdfr if (np->n_flag & NQNFSNONCACHE) 3889336Sdfr bp->b_flags |= B_INVAL; 3891541Srgrimes break; 3903305Sphk default: 3919336Sdfr printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 3923305Sphk } 3937871Sdg brelse(bp); 3941541Srgrimes } while (error == 0 && uio->uio_resid > 0 && n > 0); 3951541Srgrimes return (error); 3961541Srgrimes} 3971541Srgrimes 3981541Srgrimes/* 3991541Srgrimes * Vnode op for write using bio 4001541Srgrimes */ 4011549Srgrimesint 4021541Srgrimesnfs_write(ap) 4031541Srgrimes struct vop_write_args /* { 4041541Srgrimes struct vnode *a_vp; 4051541Srgrimes struct uio *a_uio; 4061541Srgrimes int a_ioflag; 4071541Srgrimes struct ucred *a_cred; 4081541Srgrimes } */ *ap; 4091541Srgrimes{ 4101541Srgrimes register int biosize; 4111541Srgrimes register struct uio *uio = ap->a_uio; 4121541Srgrimes struct proc *p = uio->uio_procp; 4131541Srgrimes register struct vnode *vp = ap->a_vp; 4141541Srgrimes struct nfsnode *np = VTONFS(vp); 4151541Srgrimes register struct ucred *cred = ap->a_cred; 4161541Srgrimes int ioflag = ap->a_ioflag; 4171541Srgrimes struct buf *bp; 4181541Srgrimes struct vattr vattr; 4199336Sdfr struct nfsmount *nmp = VFSTONFS(vp->v_mount); 42011921Sphk daddr_t lbn; 4218692Sdg int bufsize; 4229336Sdfr int n, on, error = 0, iomode, must_commit; 4231541Srgrimes 4241541Srgrimes#ifdef DIAGNOSTIC 4251541Srgrimes if (uio->uio_rw != UIO_WRITE) 4261541Srgrimes panic("nfs_write mode"); 4271541Srgrimes if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 4281541Srgrimes panic("nfs_write proc"); 4291541Srgrimes#endif 4301541Srgrimes if (vp->v_type != VREG) 4311541Srgrimes return (EIO); 4321541Srgrimes if (np->n_flag & NWRITEERR) { 4331541Srgrimes np->n_flag &= ~NWRITEERR; 4341541Srgrimes return (np->n_error); 4351541Srgrimes } 4369336Sdfr if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3) 4379336Sdfr (void)nfs_fsinfo(nmp, vp, cred, p); 4381541Srgrimes if (ioflag & (IO_APPEND | IO_SYNC)) { 4391541Srgrimes if (np->n_flag & NMODIFIED) { 4401541Srgrimes np->n_attrstamp = 0; 4413305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4423305Sphk if (error) 4431541Srgrimes return (error); 4441541Srgrimes } 4451541Srgrimes if (ioflag & IO_APPEND) { 4461541Srgrimes np->n_attrstamp = 0; 4473305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 4483305Sphk if (error) 4491541Srgrimes return (error); 4501541Srgrimes uio->uio_offset = np->n_size; 4511541Srgrimes } 4521541Srgrimes } 4531541Srgrimes if (uio->uio_offset < 0) 4541541Srgrimes return (EINVAL); 4551541Srgrimes if (uio->uio_resid == 0) 4561541Srgrimes return (0); 4571541Srgrimes /* 4581541Srgrimes * Maybe this should be above the vnode op call, but so long as 4591541Srgrimes * file servers have no limits, i don't think it matters 4601541Srgrimes */ 4611541Srgrimes if (p && uio->uio_offset + uio->uio_resid > 4621541Srgrimes p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 4631541Srgrimes psignal(p, SIGXFSZ); 4641541Srgrimes return (EFBIG); 4651541Srgrimes } 4661541Srgrimes /* 4671541Srgrimes * I use nm_rsize, not nm_wsize so that all buffer cache blocks 4681541Srgrimes * will be the same size within a filesystem. nfs_writerpc will 4691541Srgrimes * still use nm_wsize when sizing the rpc's. 4701541Srgrimes */ 4719428Sdfr biosize = vp->v_mount->mnt_stat.f_iosize; 4721541Srgrimes do { 4731541Srgrimes /* 4741541Srgrimes * Check for a valid write lease. 4751541Srgrimes */ 4761541Srgrimes if ((nmp->nm_flag & NFSMNT_NQNFS) && 4779336Sdfr NQNFS_CKINVALID(vp, np, ND_WRITE)) { 4781541Srgrimes do { 4799336Sdfr error = nqnfs_getlease(vp, ND_WRITE, cred, p); 4801541Srgrimes } while (error == NQNFS_EXPIRED); 4811541Srgrimes if (error) 4821541Srgrimes return (error); 4831541Srgrimes if (np->n_lrev != np->n_brev || 4841541Srgrimes (np->n_flag & NQNFSNONCACHE)) { 4853305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4863305Sphk if (error) 4871541Srgrimes return (error); 4881541Srgrimes np->n_brev = np->n_lrev; 4891541Srgrimes } 4901541Srgrimes } 4919336Sdfr if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 4929336Sdfr iomode = NFSV3WRITE_FILESYNC; 4939336Sdfr error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 4949336Sdfr if (must_commit) 4959336Sdfr nfs_clearcommit(vp->v_mount); 4969336Sdfr return (error); 4979336Sdfr } 4981541Srgrimes nfsstats.biocache_writes++; 4991541Srgrimes lbn = uio->uio_offset / biosize; 5001541Srgrimes on = uio->uio_offset & (biosize-1); 5011541Srgrimes n = min((unsigned)(biosize - on), uio->uio_resid); 5021541Srgrimesagain: 5038692Sdg if (uio->uio_offset + n > np->n_size) { 5048692Sdg np->n_size = uio->uio_offset + n; 50525023Sdfr np->n_flag |= NMODIFIED; 5068692Sdg vnode_pager_setsize(vp, (u_long)np->n_size); 5078692Sdg } 5088692Sdg bufsize = biosize; 5098692Sdg if ((lbn + 1) * biosize > np->n_size) { 5108692Sdg bufsize = np->n_size - lbn * biosize; 5118692Sdg bufsize = (bufsize + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 5128692Sdg } 5138692Sdg bp = nfs_getcacheblk(vp, lbn, bufsize, p); 5141541Srgrimes if (!bp) 5151541Srgrimes return (EINTR); 5161541Srgrimes if (bp->b_wcred == NOCRED) { 5171541Srgrimes crhold(cred); 5181541Srgrimes bp->b_wcred = cred; 5191541Srgrimes } 5201541Srgrimes np->n_flag |= NMODIFIED; 5218692Sdg 5228692Sdg if ((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend > np->n_size) { 5238692Sdg bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 5241541Srgrimes } 5251541Srgrimes 5261541Srgrimes /* 5271541Srgrimes * If the new write will leave a contiguous dirty 5281541Srgrimes * area, just update the b_dirtyoff and b_dirtyend, 5291541Srgrimes * otherwise force a write rpc of the old dirty area. 5301541Srgrimes */ 5311541Srgrimes if (bp->b_dirtyend > 0 && 5321541Srgrimes (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 5331541Srgrimes bp->b_proc = p; 5341541Srgrimes if (VOP_BWRITE(bp) == EINTR) 5351541Srgrimes return (EINTR); 5361541Srgrimes goto again; 5371541Srgrimes } 5381541Srgrimes 5391541Srgrimes /* 5401541Srgrimes * Check for valid write lease and get one as required. 5411541Srgrimes * In case getblk() and/or bwrite() delayed us. 5421541Srgrimes */ 5431541Srgrimes if ((nmp->nm_flag & NFSMNT_NQNFS) && 5449336Sdfr NQNFS_CKINVALID(vp, np, ND_WRITE)) { 5451541Srgrimes do { 5469336Sdfr error = nqnfs_getlease(vp, ND_WRITE, cred, p); 5471541Srgrimes } while (error == NQNFS_EXPIRED); 5481541Srgrimes if (error) { 5491541Srgrimes brelse(bp); 5501541Srgrimes return (error); 5511541Srgrimes } 5521541Srgrimes if (np->n_lrev != np->n_brev || 5531541Srgrimes (np->n_flag & NQNFSNONCACHE)) { 5541541Srgrimes brelse(bp); 5553305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 5563305Sphk if (error) 5571541Srgrimes return (error); 5581541Srgrimes np->n_brev = np->n_lrev; 5591541Srgrimes goto again; 5601541Srgrimes } 5611541Srgrimes } 5623305Sphk error = uiomove((char *)bp->b_data + on, n, uio); 5633305Sphk if (error) { 5641541Srgrimes bp->b_flags |= B_ERROR; 5651541Srgrimes brelse(bp); 5661541Srgrimes return (error); 5671541Srgrimes } 5681541Srgrimes if (bp->b_dirtyend > 0) { 5691541Srgrimes bp->b_dirtyoff = min(on, bp->b_dirtyoff); 5701541Srgrimes bp->b_dirtyend = max((on + n), bp->b_dirtyend); 5711541Srgrimes } else { 5721541Srgrimes bp->b_dirtyoff = on; 5731541Srgrimes bp->b_dirtyend = on + n; 5741541Srgrimes } 5751541Srgrimes if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff || 5761541Srgrimes bp->b_validoff > bp->b_dirtyend) { 5771541Srgrimes bp->b_validoff = bp->b_dirtyoff; 5781541Srgrimes bp->b_validend = bp->b_dirtyend; 5791541Srgrimes } else { 5801541Srgrimes bp->b_validoff = min(bp->b_validoff, bp->b_dirtyoff); 5811541Srgrimes bp->b_validend = max(bp->b_validend, bp->b_dirtyend); 5821541Srgrimes } 58317186Sdfr 5841541Srgrimes /* 58517186Sdfr * Since this block is being modified, it must be written 58617186Sdfr * again and not just committed. 58717186Sdfr */ 58817186Sdfr bp->b_flags &= ~B_NEEDCOMMIT; 58917186Sdfr 59017186Sdfr /* 5911541Srgrimes * If the lease is non-cachable or IO_SYNC do bwrite(). 5921541Srgrimes */ 5931541Srgrimes if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 5941541Srgrimes bp->b_proc = p; 5953305Sphk error = VOP_BWRITE(bp); 5963305Sphk if (error) 5971541Srgrimes return (error); 5989336Sdfr if (np->n_flag & NQNFSNONCACHE) { 5999336Sdfr error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 6009336Sdfr if (error) 6019336Sdfr return (error); 6029336Sdfr } 6031541Srgrimes } else if ((n + on) == biosize && 6041541Srgrimes (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 6051541Srgrimes bp->b_proc = (struct proc *)0; 6069336Sdfr bp->b_flags |= B_ASYNC; 6079336Sdfr (void)nfs_writebp(bp, 0); 6081541Srgrimes } else 6091541Srgrimes bdwrite(bp); 6101541Srgrimes } while (uio->uio_resid > 0 && n > 0); 6111541Srgrimes return (0); 6121541Srgrimes} 6131541Srgrimes 6141541Srgrimes/* 6151541Srgrimes * Get an nfs cache block. 6161541Srgrimes * Allocate a new one if the block isn't currently in the cache 6171541Srgrimes * and return the block marked busy. If the calling process is 6181541Srgrimes * interrupted by a signal for an interruptible mount point, return 6191541Srgrimes * NULL. 6201541Srgrimes */ 62112911Sphkstatic struct buf * 6221541Srgrimesnfs_getcacheblk(vp, bn, size, p) 6231541Srgrimes struct vnode *vp; 6241541Srgrimes daddr_t bn; 6251541Srgrimes int size; 6261541Srgrimes struct proc *p; 6271541Srgrimes{ 6281541Srgrimes register struct buf *bp; 6291541Srgrimes struct nfsmount *nmp = VFSTONFS(vp->v_mount); 6309428Sdfr int biosize = vp->v_mount->mnt_stat.f_iosize; 6311541Srgrimes 6321541Srgrimes if (nmp->nm_flag & NFSMNT_INT) { 6331541Srgrimes bp = getblk(vp, bn, size, PCATCH, 0); 6341541Srgrimes while (bp == (struct buf *)0) { 6351541Srgrimes if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 6361541Srgrimes return ((struct buf *)0); 6371541Srgrimes bp = getblk(vp, bn, size, 0, 2 * hz); 6381541Srgrimes } 6391541Srgrimes } else 6401541Srgrimes bp = getblk(vp, bn, size, 0, 0); 6415455Sdg 6425455Sdg if( vp->v_type == VREG) 6439336Sdfr bp->b_blkno = (bn * biosize) / DEV_BSIZE; 6445455Sdg 6451541Srgrimes return (bp); 6461541Srgrimes} 6471541Srgrimes 6481541Srgrimes/* 6491541Srgrimes * Flush and invalidate all dirty buffers. If another process is already 6501541Srgrimes * doing the flush, just wait for completion. 6511541Srgrimes */ 6521549Srgrimesint 6531541Srgrimesnfs_vinvalbuf(vp, flags, cred, p, intrflg) 6541541Srgrimes struct vnode *vp; 6551541Srgrimes int flags; 6561541Srgrimes struct ucred *cred; 6571541Srgrimes struct proc *p; 6581541Srgrimes int intrflg; 6591541Srgrimes{ 6601541Srgrimes register struct nfsnode *np = VTONFS(vp); 6611541Srgrimes struct nfsmount *nmp = VFSTONFS(vp->v_mount); 6621541Srgrimes int error = 0, slpflag, slptimeo; 6631541Srgrimes 6641541Srgrimes if ((nmp->nm_flag & NFSMNT_INT) == 0) 6651541Srgrimes intrflg = 0; 6661541Srgrimes if (intrflg) { 6671541Srgrimes slpflag = PCATCH; 6681541Srgrimes slptimeo = 2 * hz; 6691541Srgrimes } else { 6701541Srgrimes slpflag = 0; 6711541Srgrimes slptimeo = 0; 6721541Srgrimes } 6731541Srgrimes /* 6741541Srgrimes * First wait for any other process doing a flush to complete. 6751541Srgrimes */ 6761541Srgrimes while (np->n_flag & NFLUSHINPROG) { 6771541Srgrimes np->n_flag |= NFLUSHWANT; 6781541Srgrimes error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 6791541Srgrimes slptimeo); 6801541Srgrimes if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 6811541Srgrimes return (EINTR); 6821541Srgrimes } 6831541Srgrimes 6841541Srgrimes /* 6851541Srgrimes * Now, flush as required. 6861541Srgrimes */ 6871541Srgrimes np->n_flag |= NFLUSHINPROG; 6881541Srgrimes error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 6891541Srgrimes while (error) { 6901541Srgrimes if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 6911541Srgrimes np->n_flag &= ~NFLUSHINPROG; 6921541Srgrimes if (np->n_flag & NFLUSHWANT) { 6931541Srgrimes np->n_flag &= ~NFLUSHWANT; 6941541Srgrimes wakeup((caddr_t)&np->n_flag); 6951541Srgrimes } 6961541Srgrimes return (EINTR); 6971541Srgrimes } 6981541Srgrimes error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 6991541Srgrimes } 7001541Srgrimes np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 7011541Srgrimes if (np->n_flag & NFLUSHWANT) { 7021541Srgrimes np->n_flag &= ~NFLUSHWANT; 7031541Srgrimes wakeup((caddr_t)&np->n_flag); 7041541Srgrimes } 7051541Srgrimes return (0); 7061541Srgrimes} 7071541Srgrimes 7081541Srgrimes/* 7091541Srgrimes * Initiate asynchronous I/O. Return an error if no nfsiods are available. 7101541Srgrimes * This is mainly to avoid queueing async I/O requests when the nfsiods 7111541Srgrimes * are all hung on a dead server. 7121541Srgrimes */ 7131549Srgrimesint 7141541Srgrimesnfs_asyncio(bp, cred) 7151541Srgrimes register struct buf *bp; 7161541Srgrimes struct ucred *cred; 7171541Srgrimes{ 71819449Sdfr struct nfsmount *nmp; 71919449Sdfr int i; 72019449Sdfr int gotiod; 72119449Sdfr int slpflag = 0; 72219449Sdfr int slptimeo = 0; 72319449Sdfr int error; 7241541Srgrimes 7251541Srgrimes if (nfs_numasync == 0) 7261541Srgrimes return (EIO); 72719449Sdfr 72819449Sdfr nmp = VFSTONFS(bp->b_vp->v_mount); 72919449Sdfragain: 73019449Sdfr if (nmp->nm_flag & NFSMNT_INT) 73119449Sdfr slpflag = PCATCH; 73219449Sdfr gotiod = FALSE; 73319449Sdfr 73419449Sdfr /* 73519449Sdfr * Find a free iod to process this request. 73619449Sdfr */ 7371541Srgrimes for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 73819449Sdfr if (nfs_iodwant[i]) { 73919449Sdfr /* 74019449Sdfr * Found one, so wake it up and tell it which 74119449Sdfr * mount to process. 74219449Sdfr */ 74319449Sdfr NFS_DPF(ASYNCIO, 74419449Sdfr ("nfs_asyncio: waking iod %d for mount %p\n", 74519449Sdfr i, nmp)); 74619449Sdfr nfs_iodwant[i] = (struct proc *)0; 74719449Sdfr nfs_iodmount[i] = nmp; 74819449Sdfr nmp->nm_bufqiods++; 74919449Sdfr wakeup((caddr_t)&nfs_iodwant[i]); 75019449Sdfr gotiod = TRUE; 75125023Sdfr break; 75219449Sdfr } 75319449Sdfr 75419449Sdfr /* 75519449Sdfr * If none are free, we may already have an iod working on this mount 75619449Sdfr * point. If so, it will process our request. 75719449Sdfr */ 75819449Sdfr if (!gotiod) { 75919449Sdfr if (nmp->nm_bufqiods > 0) { 76019449Sdfr NFS_DPF(ASYNCIO, 76119449Sdfr ("nfs_asyncio: %d iods are already processing mount %p\n", 76219449Sdfr nmp->nm_bufqiods, nmp)); 76319449Sdfr gotiod = TRUE; 76419449Sdfr } 76519449Sdfr } 76619449Sdfr 76719449Sdfr /* 76819449Sdfr * If we have an iod which can process the request, then queue 76919449Sdfr * the buffer. 77019449Sdfr */ 77119449Sdfr if (gotiod) { 77219449Sdfr /* 77319449Sdfr * Ensure that the queue never grows too large. 77419449Sdfr */ 77519449Sdfr while (nmp->nm_bufqlen >= 2*nfs_numasync) { 77619449Sdfr NFS_DPF(ASYNCIO, 77719449Sdfr ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); 77819449Sdfr nmp->nm_bufqwant = TRUE; 77919449Sdfr error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 78019449Sdfr "nfsaio", slptimeo); 78119449Sdfr if (error) { 78219449Sdfr if (nfs_sigintr(nmp, NULL, bp->b_proc)) 78319449Sdfr return (EINTR); 78419449Sdfr if (slpflag == PCATCH) { 78519449Sdfr slpflag = 0; 78619449Sdfr slptimeo = 2 * hz; 78719449Sdfr } 78819449Sdfr } 78919449Sdfr /* 79019449Sdfr * We might have lost our iod while sleeping, 79119449Sdfr * so check and loop if nescessary. 79219449Sdfr */ 79319449Sdfr if (nmp->nm_bufqiods == 0) { 79419449Sdfr NFS_DPF(ASYNCIO, 79519449Sdfr ("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); 79619449Sdfr goto again; 79719449Sdfr } 79819449Sdfr } 79919449Sdfr 8001541Srgrimes if (bp->b_flags & B_READ) { 8011541Srgrimes if (bp->b_rcred == NOCRED && cred != NOCRED) { 8021541Srgrimes crhold(cred); 8031541Srgrimes bp->b_rcred = cred; 8041541Srgrimes } 8051541Srgrimes } else { 8069336Sdfr bp->b_flags |= B_WRITEINPROG; 8071541Srgrimes if (bp->b_wcred == NOCRED && cred != NOCRED) { 8081541Srgrimes crhold(cred); 8091541Srgrimes bp->b_wcred = cred; 8101541Srgrimes } 8111541Srgrimes } 8128876Srgrimes 81319449Sdfr TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 81419449Sdfr nmp->nm_bufqlen++; 8151541Srgrimes return (0); 81619449Sdfr } 8179336Sdfr 8189336Sdfr /* 81919449Sdfr * All the iods are busy on other mounts, so return EIO to 82019449Sdfr * force the caller to process the i/o synchronously. 8219336Sdfr */ 82219449Sdfr NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); 82319449Sdfr return (EIO); 8241541Srgrimes} 8251541Srgrimes 8261541Srgrimes/* 8271541Srgrimes * Do an I/O operation to/from a cache block. This may be called 8281541Srgrimes * synchronously or from an nfsiod. 8291541Srgrimes */ 8301541Srgrimesint 8311541Srgrimesnfs_doio(bp, cr, p) 8321541Srgrimes register struct buf *bp; 8333305Sphk struct ucred *cr; 8341541Srgrimes struct proc *p; 8351541Srgrimes{ 8361541Srgrimes register struct uio *uiop; 8371541Srgrimes register struct vnode *vp; 8381541Srgrimes struct nfsnode *np; 8391541Srgrimes struct nfsmount *nmp; 8409336Sdfr int error = 0, diff, len, iomode, must_commit = 0; 8411541Srgrimes struct uio uio; 8421541Srgrimes struct iovec io; 8431541Srgrimes 8441541Srgrimes vp = bp->b_vp; 8451541Srgrimes np = VTONFS(vp); 8461541Srgrimes nmp = VFSTONFS(vp->v_mount); 8471541Srgrimes uiop = &uio; 8481541Srgrimes uiop->uio_iov = &io; 8491541Srgrimes uiop->uio_iovcnt = 1; 8501541Srgrimes uiop->uio_segflg = UIO_SYSSPACE; 8511541Srgrimes uiop->uio_procp = p; 8521541Srgrimes 8531541Srgrimes /* 8541541Srgrimes * Historically, paging was done with physio, but no more. 8551541Srgrimes */ 8563664Sphk if (bp->b_flags & B_PHYS) { 8573664Sphk /* 8583664Sphk * ...though reading /dev/drum still gets us here. 8593664Sphk */ 8601541Srgrimes io.iov_len = uiop->uio_resid = bp->b_bcount; 8613664Sphk /* mapping was done by vmapbuf() */ 8621541Srgrimes io.iov_base = bp->b_data; 8639336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 8643664Sphk if (bp->b_flags & B_READ) { 8653664Sphk uiop->uio_rw = UIO_READ; 8663664Sphk nfsstats.read_physios++; 8673664Sphk error = nfs_readrpc(vp, uiop, cr); 8683664Sphk } else { 8699336Sdfr int com; 8709336Sdfr 8719336Sdfr iomode = NFSV3WRITE_DATASYNC; 8723664Sphk uiop->uio_rw = UIO_WRITE; 8733664Sphk nfsstats.write_physios++; 8749336Sdfr error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 8753664Sphk } 8763664Sphk if (error) { 8773664Sphk bp->b_flags |= B_ERROR; 8783664Sphk bp->b_error = error; 8793664Sphk } 8803664Sphk } else if (bp->b_flags & B_READ) { 8813664Sphk io.iov_len = uiop->uio_resid = bp->b_bcount; 8823664Sphk io.iov_base = bp->b_data; 8831541Srgrimes uiop->uio_rw = UIO_READ; 8841541Srgrimes switch (vp->v_type) { 8851541Srgrimes case VREG: 8869336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 8871541Srgrimes nfsstats.read_bios++; 8881541Srgrimes error = nfs_readrpc(vp, uiop, cr); 8891541Srgrimes if (!error) { 8901541Srgrimes bp->b_validoff = 0; 8911541Srgrimes if (uiop->uio_resid) { 8921541Srgrimes /* 8931541Srgrimes * If len > 0, there is a hole in the file and 8941541Srgrimes * no writes after the hole have been pushed to 8951541Srgrimes * the server yet. 8961541Srgrimes * Just zero fill the rest of the valid area. 8971541Srgrimes */ 8981541Srgrimes diff = bp->b_bcount - uiop->uio_resid; 8999336Sdfr len = np->n_size - (((u_quad_t)bp->b_blkno) * DEV_BSIZE 9001541Srgrimes + diff); 9011541Srgrimes if (len > 0) { 9021541Srgrimes len = min(len, uiop->uio_resid); 9031541Srgrimes bzero((char *)bp->b_data + diff, len); 9041541Srgrimes bp->b_validend = diff + len; 9051541Srgrimes } else 9061541Srgrimes bp->b_validend = diff; 9071541Srgrimes } else 9081541Srgrimes bp->b_validend = bp->b_bcount; 9091541Srgrimes } 9101541Srgrimes if (p && (vp->v_flag & VTEXT) && 9111541Srgrimes (((nmp->nm_flag & NFSMNT_NQNFS) && 9129336Sdfr NQNFS_CKINVALID(vp, np, ND_READ) && 9131541Srgrimes np->n_lrev != np->n_brev) || 9141541Srgrimes (!(nmp->nm_flag & NFSMNT_NQNFS) && 91518397Snate np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 9161541Srgrimes uprintf("Process killed due to text file modification\n"); 9171541Srgrimes psignal(p, SIGKILL); 9189336Sdfr#ifdef __NetBSD__ 9199336Sdfr p->p_holdcnt++; 9209336Sdfr#else 9211541Srgrimes p->p_flag |= P_NOSWAP; 9229336Sdfr#endif 9231541Srgrimes } 9241541Srgrimes break; 9251541Srgrimes case VLNK: 9269336Sdfr uiop->uio_offset = (off_t)0; 9271541Srgrimes nfsstats.readlink_bios++; 9281541Srgrimes error = nfs_readlinkrpc(vp, uiop, cr); 9291541Srgrimes break; 9301541Srgrimes case VDIR: 9311541Srgrimes nfsstats.readdir_bios++; 9329336Sdfr uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 9339336Sdfr if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 9349336Sdfr error = nfs_readdirplusrpc(vp, uiop, cr); 9359336Sdfr if (error == NFSERR_NOTSUPP) 9369336Sdfr nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 9379336Sdfr } 9389336Sdfr if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 9399336Sdfr error = nfs_readdirrpc(vp, uiop, cr); 9401541Srgrimes break; 9413305Sphk default: 9423305Sphk printf("nfs_doio: type %x unexpected\n",vp->v_type); 9433305Sphk break; 9441541Srgrimes }; 9451541Srgrimes if (error) { 9461541Srgrimes bp->b_flags |= B_ERROR; 9471541Srgrimes bp->b_error = error; 9481541Srgrimes } 9491541Srgrimes } else { 9508692Sdg if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size) 9518692Sdg bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE); 9528692Sdg 9538692Sdg if (bp->b_dirtyend > bp->b_dirtyoff) { 9548692Sdg io.iov_len = uiop->uio_resid = bp->b_dirtyend 9559336Sdfr - bp->b_dirtyoff; 9569336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE 9579336Sdfr + bp->b_dirtyoff; 9588692Sdg io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 9598692Sdg uiop->uio_rw = UIO_WRITE; 9608692Sdg nfsstats.write_bios++; 9619336Sdfr if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC) 9629336Sdfr iomode = NFSV3WRITE_UNSTABLE; 9638692Sdg else 9649336Sdfr iomode = NFSV3WRITE_FILESYNC; 9659336Sdfr bp->b_flags |= B_WRITEINPROG; 9669336Sdfr error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 96725003Sdfr if (!error && iomode == NFSV3WRITE_UNSTABLE) { 96825003Sdfr bp->b_flags |= B_NEEDCOMMIT; 96925003Sdfr if (bp->b_dirtyoff == 0 97025003Sdfr && bp->b_dirtyend == bp->b_bufsize) 97125003Sdfr bp->b_flags |= B_CLUSTEROK; 97225003Sdfr } else 9739336Sdfr bp->b_flags &= ~B_NEEDCOMMIT; 9749336Sdfr bp->b_flags &= ~B_WRITEINPROG; 9758692Sdg 9769336Sdfr /* 9779336Sdfr * For an interrupted write, the buffer is still valid 9789336Sdfr * and the write hasn't been pushed to the server yet, 9799336Sdfr * so we can't set B_ERROR and report the interruption 9809336Sdfr * by setting B_EINTR. For the B_ASYNC case, B_EINTR 9819336Sdfr * is not relevant, so the rpc attempt is essentially 9829336Sdfr * a noop. For the case of a V3 write rpc not being 9839336Sdfr * committed to stable storage, the block is still 9849336Sdfr * dirty and requires either a commit rpc or another 9859336Sdfr * write rpc with iomode == NFSV3WRITE_FILESYNC before 9869336Sdfr * the block is reused. This is indicated by setting 9879336Sdfr * the B_DELWRI and B_NEEDCOMMIT flags. 9889336Sdfr */ 9899336Sdfr if (error == EINTR 9909336Sdfr || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 9918692Sdg bp->b_flags &= ~(B_INVAL|B_NOCACHE); 9928692Sdg bp->b_flags |= B_DELWRI; 9931541Srgrimes 9941541Srgrimes /* 9951541Srgrimes * Since for the B_ASYNC case, nfs_bwrite() has reassigned the 9961541Srgrimes * buffer to the clean list, we have to reassign it back to the 9971541Srgrimes * dirty one. Ugh. 9981541Srgrimes */ 9998692Sdg if (bp->b_flags & B_ASYNC) 10008692Sdg reassignbuf(bp, vp); 10018692Sdg else 10028692Sdg bp->b_flags |= B_EINTR; 10038692Sdg } else { 10048692Sdg if (error) { 10058692Sdg bp->b_flags |= B_ERROR; 10068692Sdg bp->b_error = np->n_error = error; 10078692Sdg np->n_flag |= NWRITEERR; 10088692Sdg } 10098692Sdg bp->b_dirtyoff = bp->b_dirtyend = 0; 10108692Sdg } 10111541Srgrimes } else { 10128692Sdg bp->b_resid = 0; 10138692Sdg biodone(bp); 10148692Sdg return (0); 10151541Srgrimes } 10161541Srgrimes } 10171541Srgrimes bp->b_resid = uiop->uio_resid; 10189336Sdfr if (must_commit) 10199336Sdfr nfs_clearcommit(vp->v_mount); 10201541Srgrimes biodone(bp); 10211541Srgrimes return (error); 10221541Srgrimes} 1023