nfs_bio.c revision 54480
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 3750477Speter * $FreeBSD: head/sys/nfsclient/nfs_bio.c 54480 1999-12-12 06:09:57Z dillon $ 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> 501541Srgrimes 511541Srgrimes#include <vm/vm.h> 5212662Sdg#include <vm/vm_extern.h> 5325930Sdfr#include <vm/vm_page.h> 5425930Sdfr#include <vm/vm_object.h> 5525930Sdfr#include <vm/vm_pager.h> 5625930Sdfr#include <vm/vnode_pager.h> 571541Srgrimes 581541Srgrimes#include <nfs/rpcv2.h> 599336Sdfr#include <nfs/nfsproto.h> 601541Srgrimes#include <nfs/nfs.h> 611541Srgrimes#include <nfs/nfsmount.h> 621541Srgrimes#include <nfs/nqnfs.h> 639336Sdfr#include <nfs/nfsnode.h> 641541Srgrimes 6512911Sphkstatic struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, 6612588Sbde struct proc *p)); 6712588Sbde 681541Srgrimesextern int nfs_numasync; 6942957Sdillonextern int nfs_pbuf_freecnt; 709336Sdfrextern struct nfsstats nfsstats; 711541Srgrimes 721541Srgrimes/* 7325930Sdfr * Vnode op for VM getpages. 7425930Sdfr */ 7525930Sdfrint 7625930Sdfrnfs_getpages(ap) 7736563Speter struct vop_getpages_args /* { 7836563Speter struct vnode *a_vp; 7936563Speter vm_page_t *a_m; 8036563Speter int a_count; 8136563Speter int a_reqpage; 8236563Speter vm_ooffset_t a_offset; 8336563Speter } */ *ap; 8425930Sdfr{ 8546349Salc int i, error, nextoff, size, toff, count, npages; 8632755Sdyson struct uio uio; 8732755Sdyson struct iovec iov; 8832755Sdyson vm_offset_t kva; 8934206Sdyson struct buf *bp; 9036563Speter struct vnode *vp; 9136563Speter struct proc *p; 9236563Speter struct ucred *cred; 9336563Speter struct nfsmount *nmp; 9436563Speter vm_page_t *pages; 9525930Sdfr 9636563Speter vp = ap->a_vp; 9736563Speter p = curproc; /* XXX */ 9836563Speter cred = curproc->p_ucred; /* XXX */ 9936563Speter nmp = VFSTONFS(vp->v_mount); 10036563Speter pages = ap->a_m; 10136563Speter count = ap->a_count; 10236563Speter 10336563Speter if (vp->v_object == NULL) { 10432286Sdyson printf("nfs_getpages: called with non-merged cache vnode??\n"); 10536563Speter return VM_PAGER_ERROR; 10625930Sdfr } 10725930Sdfr 10836563Speter if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && 10936563Speter (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 11036563Speter (void)nfs_fsinfo(nmp, vp, cred, p); 11146349Salc 11246349Salc npages = btoc(count); 11346349Salc 11434206Sdyson /* 11546349Salc * If the requested page is partially valid, just return it and 11646349Salc * allow the pager to zero-out the blanks. Partially valid pages 11746349Salc * can only occur at the file EOF. 11846349Salc */ 11946349Salc 12046349Salc { 12146349Salc vm_page_t m = pages[ap->a_reqpage]; 12246349Salc 12346349Salc if (m->valid != 0) { 12446349Salc /* handled by vm_fault now */ 12546349Salc /* vm_page_zero_invalid(m, TRUE); */ 12646349Salc for (i = 0; i < npages; ++i) { 12746349Salc if (i != ap->a_reqpage) 12846349Salc vnode_pager_freepage(pages[i]); 12946349Salc } 13046349Salc return(0); 13146349Salc } 13246349Salc } 13346349Salc 13446349Salc /* 13534206Sdyson * We use only the kva address for the buffer, but this is extremely 13634206Sdyson * convienient and fast. 13734206Sdyson */ 13842957Sdillon bp = getpbuf(&nfs_pbuf_freecnt); 13925930Sdfr 14034206Sdyson kva = (vm_offset_t) bp->b_data; 14136563Speter pmap_qenter(kva, pages, npages); 14234206Sdyson 14332755Sdyson iov.iov_base = (caddr_t) kva; 14436563Speter iov.iov_len = count; 14532755Sdyson uio.uio_iov = &iov; 14632755Sdyson uio.uio_iovcnt = 1; 14736563Speter uio.uio_offset = IDX_TO_OFF(pages[0]->pindex); 14836563Speter uio.uio_resid = count; 14932755Sdyson uio.uio_segflg = UIO_SYSSPACE; 15032755Sdyson uio.uio_rw = UIO_READ; 15136563Speter uio.uio_procp = p; 15225930Sdfr 15336563Speter error = nfs_readrpc(vp, &uio, cred); 15434206Sdyson pmap_qremove(kva, npages); 15532755Sdyson 15642957Sdillon relpbuf(bp, &nfs_pbuf_freecnt); 15734206Sdyson 15842957Sdillon if (error && (uio.uio_resid == count)) { 15942957Sdillon printf("nfs_getpages: error %d\n", error); 16042957Sdillon for (i = 0; i < npages; ++i) { 16142957Sdillon if (i != ap->a_reqpage) 16242957Sdillon vnode_pager_freepage(pages[i]); 16342957Sdillon } 16434206Sdyson return VM_PAGER_ERROR; 16542957Sdillon } 16634206Sdyson 16745347Sjulian /* 16845347Sjulian * Calculate the number of bytes read and validate only that number 16945347Sjulian * of bytes. Note that due to pending writes, size may be 0. This 17045347Sjulian * does not mean that the remaining data is invalid! 17145347Sjulian */ 17245347Sjulian 17336563Speter size = count - uio.uio_resid; 17434206Sdyson 17534206Sdyson for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { 17634206Sdyson vm_page_t m; 17734206Sdyson nextoff = toff + PAGE_SIZE; 17836563Speter m = pages[i]; 17934206Sdyson 18034206Sdyson m->flags &= ~PG_ZERO; 18134206Sdyson 18234206Sdyson if (nextoff <= size) { 18345347Sjulian /* 18445347Sjulian * Read operation filled an entire page 18545347Sjulian */ 18634206Sdyson m->valid = VM_PAGE_BITS_ALL; 18749945Salc vm_page_undirty(m); 18845347Sjulian } else if (size > toff) { 18945347Sjulian /* 19046349Salc * Read operation filled a partial page. 19145347Sjulian */ 19246349Salc m->valid = 0; 19345347Sjulian vm_page_set_validclean(m, 0, size - toff); 19446349Salc /* handled by vm_fault now */ 19546349Salc /* vm_page_zero_invalid(m, TRUE); */ 19634206Sdyson } 19734206Sdyson 19825930Sdfr if (i != ap->a_reqpage) { 19934206Sdyson /* 20034206Sdyson * Whether or not to leave the page activated is up in 20134206Sdyson * the air, but we should put the page on a page queue 20234206Sdyson * somewhere (it already is in the object). Result: 20334206Sdyson * It appears that emperical results show that 20434206Sdyson * deactivating pages is best. 20534206Sdyson */ 20634206Sdyson 20734206Sdyson /* 20834206Sdyson * Just in case someone was asking for this page we 20934206Sdyson * now tell them that it is ok to use. 21034206Sdyson */ 21134206Sdyson if (!error) { 21234206Sdyson if (m->flags & PG_WANTED) 21334206Sdyson vm_page_activate(m); 21434206Sdyson else 21534206Sdyson vm_page_deactivate(m); 21638799Sdfr vm_page_wakeup(m); 21734206Sdyson } else { 21834206Sdyson vnode_pager_freepage(m); 21934206Sdyson } 22025930Sdfr } 22125930Sdfr } 22225930Sdfr return 0; 22325930Sdfr} 22425930Sdfr 22525930Sdfr/* 22634206Sdyson * Vnode op for VM putpages. 22734096Smsmith */ 22834096Smsmithint 22934096Smsmithnfs_putpages(ap) 23036563Speter struct vop_putpages_args /* { 23136563Speter struct vnode *a_vp; 23236563Speter vm_page_t *a_m; 23336563Speter int a_count; 23436563Speter int a_sync; 23536563Speter int *a_rtvals; 23636563Speter vm_ooffset_t a_offset; 23736563Speter } */ *ap; 23834096Smsmith{ 23934206Sdyson struct uio uio; 24034206Sdyson struct iovec iov; 24134206Sdyson vm_offset_t kva; 24234206Sdyson struct buf *bp; 24336563Speter int iomode, must_commit, i, error, npages, count; 24446349Salc off_t offset; 24534206Sdyson int *rtvals; 24636563Speter struct vnode *vp; 24736563Speter struct proc *p; 24836563Speter struct ucred *cred; 24936563Speter struct nfsmount *nmp; 25046349Salc struct nfsnode *np; 25136563Speter vm_page_t *pages; 25234206Sdyson 25336563Speter vp = ap->a_vp; 25446349Salc np = VTONFS(vp); 25536563Speter p = curproc; /* XXX */ 25636563Speter cred = curproc->p_ucred; /* XXX */ 25736563Speter nmp = VFSTONFS(vp->v_mount); 25836563Speter pages = ap->a_m; 25936563Speter count = ap->a_count; 26034206Sdyson rtvals = ap->a_rtvals; 26136563Speter npages = btoc(count); 26246349Salc offset = IDX_TO_OFF(pages[0]->pindex); 26334206Sdyson 26436563Speter if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && 26536563Speter (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 26636563Speter (void)nfs_fsinfo(nmp, vp, cred, p); 26734206Sdyson 26834206Sdyson for (i = 0; i < npages; i++) { 26934206Sdyson rtvals[i] = VM_PAGER_AGAIN; 27034206Sdyson } 27134206Sdyson 27234206Sdyson /* 27346349Salc * When putting pages, do not extend file past EOF. 27446349Salc */ 27546349Salc 27646349Salc if (offset + count > np->n_size) { 27746349Salc count = np->n_size - offset; 27846349Salc if (count < 0) 27946349Salc count = 0; 28046349Salc } 28146349Salc 28246349Salc /* 28334206Sdyson * We use only the kva address for the buffer, but this is extremely 28434206Sdyson * convienient and fast. 28534206Sdyson */ 28642957Sdillon bp = getpbuf(&nfs_pbuf_freecnt); 28734206Sdyson 28834206Sdyson kva = (vm_offset_t) bp->b_data; 28936563Speter pmap_qenter(kva, pages, npages); 29034206Sdyson 29134206Sdyson iov.iov_base = (caddr_t) kva; 29236563Speter iov.iov_len = count; 29334206Sdyson uio.uio_iov = &iov; 29434206Sdyson uio.uio_iovcnt = 1; 29546349Salc uio.uio_offset = offset; 29636563Speter uio.uio_resid = count; 29734206Sdyson uio.uio_segflg = UIO_SYSSPACE; 29834206Sdyson uio.uio_rw = UIO_WRITE; 29936563Speter uio.uio_procp = p; 30034206Sdyson 30134206Sdyson if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0) 30234206Sdyson iomode = NFSV3WRITE_UNSTABLE; 30334206Sdyson else 30434206Sdyson iomode = NFSV3WRITE_FILESYNC; 30534206Sdyson 30636563Speter error = nfs_writerpc(vp, &uio, cred, &iomode, &must_commit); 30734206Sdyson 30834206Sdyson pmap_qremove(kva, npages); 30942957Sdillon relpbuf(bp, &nfs_pbuf_freecnt); 31034206Sdyson 31134206Sdyson if (!error) { 31236563Speter int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE; 31334206Sdyson for (i = 0; i < nwritten; i++) { 31434206Sdyson rtvals[i] = VM_PAGER_OK; 31549945Salc vm_page_undirty(pages[i]); 31634206Sdyson } 31734206Sdyson if (must_commit) 31836563Speter nfs_clearcommit(vp->v_mount); 31934206Sdyson } 32036563Speter return rtvals[0]; 32134096Smsmith} 32234096Smsmith 32334096Smsmith/* 3241541Srgrimes * Vnode op for read using bio 3251541Srgrimes */ 3261549Srgrimesint 32746349Salcnfs_bioread(vp, uio, ioflag, cred) 3281541Srgrimes register struct vnode *vp; 3291541Srgrimes register struct uio *uio; 3301541Srgrimes int ioflag; 3311541Srgrimes struct ucred *cred; 3321541Srgrimes{ 3331541Srgrimes register struct nfsnode *np = VTONFS(vp); 33441791Sdt register int biosize, i; 3351549Srgrimes struct buf *bp = 0, *rabp; 3361541Srgrimes struct vattr vattr; 3371541Srgrimes struct proc *p; 3389336Sdfr struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3395455Sdg daddr_t lbn, rabn; 34046349Salc int bcount; 34151344Sdillon int seqcount; 34246349Salc int nra, error = 0, n = 0, on = 0; 3431541Srgrimes 3441541Srgrimes#ifdef DIAGNOSTIC 3451541Srgrimes if (uio->uio_rw != UIO_READ) 3461541Srgrimes panic("nfs_read mode"); 3471541Srgrimes#endif 3481541Srgrimes if (uio->uio_resid == 0) 3491541Srgrimes return (0); 35036473Speter if (uio->uio_offset < 0) /* XXX VDIR cookies can be negative */ 3511541Srgrimes return (EINVAL); 3521541Srgrimes p = uio->uio_procp; 35351344Sdillon 35436176Speter if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && 35536176Speter (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 3569336Sdfr (void)nfs_fsinfo(nmp, vp, cred, p); 35736473Speter if (vp->v_type != VDIR && 35836473Speter (uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 35936473Speter return (EFBIG); 3609428Sdfr biosize = vp->v_mount->mnt_stat.f_iosize; 36151344Sdillon seqcount = (int)((off_t)(ioflag >> 16) * biosize / BKVASIZE); 3621541Srgrimes /* 3631541Srgrimes * For nfs, cache consistency can only be maintained approximately. 3641541Srgrimes * Although RFC1094 does not specify the criteria, the following is 3651541Srgrimes * believed to be compatible with the reference port. 3661541Srgrimes * For nqnfs, full cache consistency is maintained within the loop. 3671541Srgrimes * For nfs: 3681541Srgrimes * If the file's modify time on the server has changed since the 3691541Srgrimes * last read rpc or you have written to the file, 3701541Srgrimes * you may have lost data cache consistency with the 3711541Srgrimes * server, so flush all of the file's data out of the cache. 3721541Srgrimes * Then force a getattr rpc to ensure that you have up to date 3731541Srgrimes * attributes. 3741541Srgrimes * NB: This implies that cache data can be read when up to 3751541Srgrimes * NFS_ATTRTIMEO seconds out of date. If you find that you need current 3761541Srgrimes * attributes this could be forced by setting n_attrstamp to 0 before 3771541Srgrimes * the VOP_GETATTR() call. 3781541Srgrimes */ 37910219Sdfr if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { 3801541Srgrimes if (np->n_flag & NMODIFIED) { 3819336Sdfr if (vp->v_type != VREG) { 3829336Sdfr if (vp->v_type != VDIR) 3839336Sdfr panic("nfs: bioread, not dir"); 3849336Sdfr nfs_invaldir(vp); 3853305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 3863305Sphk if (error) 3871541Srgrimes return (error); 3881541Srgrimes } 3891541Srgrimes np->n_attrstamp = 0; 3903305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 3913305Sphk if (error) 3921541Srgrimes return (error); 39318397Snate np->n_mtime = vattr.va_mtime.tv_sec; 3941541Srgrimes } else { 3953305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 3963305Sphk if (error) 3971541Srgrimes return (error); 39818397Snate if (np->n_mtime != vattr.va_mtime.tv_sec) { 3999336Sdfr if (vp->v_type == VDIR) 4009336Sdfr nfs_invaldir(vp); 4013305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4023305Sphk if (error) 4031541Srgrimes return (error); 40418397Snate np->n_mtime = vattr.va_mtime.tv_sec; 4051541Srgrimes } 4061541Srgrimes } 4071541Srgrimes } 4081541Srgrimes do { 4091541Srgrimes 4101541Srgrimes /* 4111541Srgrimes * Get a valid lease. If cached data is stale, flush it. 4121541Srgrimes */ 4131541Srgrimes if (nmp->nm_flag & NFSMNT_NQNFS) { 4149336Sdfr if (NQNFS_CKINVALID(vp, np, ND_READ)) { 4151541Srgrimes do { 4169336Sdfr error = nqnfs_getlease(vp, ND_READ, cred, p); 4171541Srgrimes } while (error == NQNFS_EXPIRED); 4181541Srgrimes if (error) 4191541Srgrimes return (error); 4201541Srgrimes if (np->n_lrev != np->n_brev || 4211541Srgrimes (np->n_flag & NQNFSNONCACHE) || 4221541Srgrimes ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { 4239336Sdfr if (vp->v_type == VDIR) 4249336Sdfr nfs_invaldir(vp); 4253305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4263305Sphk if (error) 4271541Srgrimes return (error); 4281541Srgrimes np->n_brev = np->n_lrev; 4291541Srgrimes } 4301541Srgrimes } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { 4319336Sdfr nfs_invaldir(vp); 4323305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 4333305Sphk if (error) 4341541Srgrimes return (error); 4351541Srgrimes } 4361541Srgrimes } 4371541Srgrimes if (np->n_flag & NQNFSNONCACHE) { 4381541Srgrimes switch (vp->v_type) { 4391541Srgrimes case VREG: 4409336Sdfr return (nfs_readrpc(vp, uio, cred)); 4411541Srgrimes case VLNK: 4429336Sdfr return (nfs_readlinkrpc(vp, uio, cred)); 4431541Srgrimes case VDIR: 4441541Srgrimes break; 4453305Sphk default: 44622521Sdyson printf(" NQNFSNONCACHE: type %x unexpected\n", 4473305Sphk vp->v_type); 4481541Srgrimes }; 4491541Srgrimes } 4501541Srgrimes switch (vp->v_type) { 4511541Srgrimes case VREG: 4521541Srgrimes nfsstats.biocache_reads++; 4531541Srgrimes lbn = uio->uio_offset / biosize; 4549336Sdfr on = uio->uio_offset & (biosize - 1); 4551541Srgrimes 4561541Srgrimes /* 4571541Srgrimes * Start the read ahead(s), as required. 4581541Srgrimes */ 4599336Sdfr if (nfs_numasync > 0 && nmp->nm_readahead > 0) { 46051344Sdillon for (nra = 0; nra < nmp->nm_readahead && nra < seqcount && 46113612Smpp (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { 4625455Sdg rabn = lbn + 1 + nra; 4631541Srgrimes if (!incore(vp, rabn)) { 4641541Srgrimes rabp = nfs_getcacheblk(vp, rabn, biosize, p); 4651541Srgrimes if (!rabp) 4661541Srgrimes return (EINTR); 4678692Sdg if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 4681541Srgrimes rabp->b_flags |= (B_READ | B_ASYNC); 4695455Sdg vfs_busy_pages(rabp, 0); 47046580Sphk if (nfs_asyncio(rabp, cred, p)) { 4715455Sdg rabp->b_flags |= B_INVAL|B_ERROR; 4725455Sdg vfs_unbusy_pages(rabp); 4731541Srgrimes brelse(rabp); 4741541Srgrimes } 47522521Sdyson } else 4765471Sdg brelse(rabp); 4771541Srgrimes } 4781541Srgrimes } 4791541Srgrimes } 4801541Srgrimes 4811541Srgrimes /* 48246349Salc * Obtain the buffer cache block. Figure out the buffer size 48346349Salc * when we are at EOF. nfs_getcacheblk() will also force 48446349Salc * uncached delayed-writes to be flushed to the server. 48546349Salc * 48646349Salc * Note that bcount is *not* DEV_BSIZE aligned. 4871541Srgrimes */ 48846349Salc 48946349Salc bcount = biosize; 49046349Salc if ((off_t)lbn * biosize >= np->n_size) { 49146349Salc bcount = 0; 49246349Salc } else if ((off_t)(lbn + 1) * biosize > np->n_size) { 49346349Salc bcount = np->n_size - (off_t)lbn * biosize; 4948692Sdg } 49546349Salc 49646349Salc bp = nfs_getcacheblk(vp, lbn, bcount, p); 4977871Sdg if (!bp) 4987871Sdg return (EINTR); 49942957Sdillon 50025930Sdfr /* 50146349Salc * If B_CACHE is not set, we must issue the read. If this 50246349Salc * fails, we return an error. 50325930Sdfr */ 50446349Salc 5057871Sdg if ((bp->b_flags & B_CACHE) == 0) { 50632755Sdyson bp->b_flags |= B_READ; 50732755Sdyson vfs_busy_pages(bp, 0); 50832755Sdyson error = nfs_doio(bp, cred, p); 50932755Sdyson if (error) { 51032755Sdyson brelse(bp); 51132755Sdyson return (error); 51232755Sdyson } 5131541Srgrimes } 51446349Salc 51546349Salc /* 51646349Salc * on is the offset into the current bp. Figure out how many 51746349Salc * bytes we can copy out of the bp. Note that bcount is 51846349Salc * NOT DEV_BSIZE aligned. 51946349Salc * 52046349Salc * Then figure out how many bytes we can copy into the uio. 52146349Salc */ 52246349Salc 52346349Salc n = 0; 52446349Salc if (on < bcount) 52546349Salc n = min((unsigned)(bcount - on), uio->uio_resid); 5261541Srgrimes break; 5271541Srgrimes case VLNK: 5281541Srgrimes nfsstats.biocache_readlinks++; 5291541Srgrimes bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); 5301541Srgrimes if (!bp) 5311541Srgrimes return (EINTR); 5327871Sdg if ((bp->b_flags & B_CACHE) == 0) { 53332755Sdyson bp->b_flags |= B_READ; 53432755Sdyson vfs_busy_pages(bp, 0); 53532755Sdyson error = nfs_doio(bp, cred, p); 53632755Sdyson if (error) { 53732755Sdyson bp->b_flags |= B_ERROR; 53832755Sdyson brelse(bp); 53932755Sdyson return (error); 54032755Sdyson } 5411541Srgrimes } 5421541Srgrimes n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); 5431541Srgrimes on = 0; 5441541Srgrimes break; 5451541Srgrimes case VDIR: 5461541Srgrimes nfsstats.biocache_readdirs++; 54724577Sdfr if (np->n_direofoffset 54824577Sdfr && uio->uio_offset >= np->n_direofoffset) { 54924577Sdfr return (0); 55024577Sdfr } 55136979Sbde lbn = (uoff_t)uio->uio_offset / NFS_DIRBLKSIZ; 5529336Sdfr on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); 5535455Sdg bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); 5541541Srgrimes if (!bp) 5559336Sdfr return (EINTR); 5567871Sdg if ((bp->b_flags & B_CACHE) == 0) { 5579336Sdfr bp->b_flags |= B_READ; 5589336Sdfr vfs_busy_pages(bp, 0); 5599336Sdfr error = nfs_doio(bp, cred, p); 56032912Stegge if (error) { 56132912Stegge brelse(bp); 56232912Stegge } 56332755Sdyson while (error == NFSERR_BAD_COOKIE) { 56446349Salc printf("got bad cookie vp %p bp %p\n", vp, bp); 56532755Sdyson nfs_invaldir(vp); 56632755Sdyson error = nfs_vinvalbuf(vp, 0, cred, p, 1); 56732755Sdyson /* 56832755Sdyson * Yuck! The directory has been modified on the 56932755Sdyson * server. The only way to get the block is by 57032755Sdyson * reading from the beginning to get all the 57132755Sdyson * offset cookies. 57246349Salc * 57346349Salc * Leave the last bp intact unless there is an error. 57446349Salc * Loop back up to the while if the error is another 57546349Salc * NFSERR_BAD_COOKIE (double yuch!). 57632755Sdyson */ 57732755Sdyson for (i = 0; i <= lbn && !error; i++) { 57832755Sdyson if (np->n_direofoffset 57932755Sdyson && (i * NFS_DIRBLKSIZ) >= np->n_direofoffset) 58024577Sdfr return (0); 58132755Sdyson bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); 58232755Sdyson if (!bp) 58332755Sdyson return (EINTR); 58446349Salc if ((bp->b_flags & B_CACHE) == 0) { 58546349Salc bp->b_flags |= B_READ; 58646349Salc vfs_busy_pages(bp, 0); 58746349Salc error = nfs_doio(bp, cred, p); 58846349Salc /* 58946349Salc * no error + B_INVAL == directory EOF, 59046349Salc * use the block. 59146349Salc */ 59246349Salc if (error == 0 && (bp->b_flags & B_INVAL)) 59346349Salc break; 59446349Salc } 59546349Salc /* 59646349Salc * An error will throw away the block and the 59746349Salc * for loop will break out. If no error and this 59846349Salc * is not the block we want, we throw away the 59946349Salc * block and go for the next one via the for loop. 60046349Salc */ 60146349Salc if (error || i < lbn) 60232755Sdyson brelse(bp); 6031541Srgrimes } 60432912Stegge } 60546349Salc /* 60646349Salc * The above while is repeated if we hit another cookie 60746349Salc * error. If we hit an error and it wasn't a cookie error, 60846349Salc * we give up. 60946349Salc */ 61032912Stegge if (error) 6119336Sdfr return (error); 6121541Srgrimes } 6131541Srgrimes 6141541Srgrimes /* 6151541Srgrimes * If not eof and read aheads are enabled, start one. 6161541Srgrimes * (You need the current block first, so that you have the 6179336Sdfr * directory offset cookie of the next block.) 6181541Srgrimes */ 6191541Srgrimes if (nfs_numasync > 0 && nmp->nm_readahead > 0 && 62039782Smckusick (bp->b_flags & B_INVAL) == 0 && 6219336Sdfr (np->n_direofoffset == 0 || 6229336Sdfr (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && 6239336Sdfr !(np->n_flag & NQNFSNONCACHE) && 6249336Sdfr !incore(vp, lbn + 1)) { 6259336Sdfr rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); 6261541Srgrimes if (rabp) { 6278692Sdg if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { 6281541Srgrimes rabp->b_flags |= (B_READ | B_ASYNC); 6295455Sdg vfs_busy_pages(rabp, 0); 63046580Sphk if (nfs_asyncio(rabp, cred, p)) { 6316148Sdg rabp->b_flags |= B_INVAL|B_ERROR; 6325455Sdg vfs_unbusy_pages(rabp); 6331541Srgrimes brelse(rabp); 6341541Srgrimes } 6355471Sdg } else { 6365471Sdg brelse(rabp); 6371541Srgrimes } 6381541Srgrimes } 6391541Srgrimes } 64026469Sdfr /* 64146349Salc * Unlike VREG files, whos buffer size ( bp->b_bcount ) is 64246349Salc * chopped for the EOF condition, we cannot tell how large 64346349Salc * NFS directories are going to be until we hit EOF. So 64446349Salc * an NFS directory buffer is *not* chopped to its EOF. Now, 64546349Salc * it just so happens that b_resid will effectively chop it 64646349Salc * to EOF. *BUT* this information is lost if the buffer goes 64746349Salc * away and is reconstituted into a B_CACHE state ( due to 64846349Salc * being VMIO ) later. So we keep track of the directory eof 64946349Salc * in np->n_direofoffset and chop it off as an extra step 65046349Salc * right here. 65126469Sdfr */ 65226469Sdfr n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); 65346349Salc if (np->n_direofoffset && n > np->n_direofoffset - uio->uio_offset) 65446349Salc n = np->n_direofoffset - uio->uio_offset; 6551541Srgrimes break; 6563305Sphk default: 6579336Sdfr printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 6583305Sphk break; 6591541Srgrimes }; 6601541Srgrimes 6611541Srgrimes if (n > 0) { 66234206Sdyson error = uiomove(bp->b_data + on, (int)n, uio); 6631541Srgrimes } 6641541Srgrimes switch (vp->v_type) { 6651541Srgrimes case VREG: 6661541Srgrimes break; 6671541Srgrimes case VLNK: 6681541Srgrimes n = 0; 6691541Srgrimes break; 6701541Srgrimes case VDIR: 67146349Salc /* 67246349Salc * Invalidate buffer if caching is disabled, forcing a 67346349Salc * re-read from the remote later. 67446349Salc */ 6759336Sdfr if (np->n_flag & NQNFSNONCACHE) 6769336Sdfr bp->b_flags |= B_INVAL; 6771541Srgrimes break; 6783305Sphk default: 6799336Sdfr printf(" nfs_bioread: type %x unexpected\n",vp->v_type); 6803305Sphk } 68132755Sdyson brelse(bp); 6821541Srgrimes } while (error == 0 && uio->uio_resid > 0 && n > 0); 6831541Srgrimes return (error); 6841541Srgrimes} 6851541Srgrimes 6861541Srgrimes/* 6871541Srgrimes * Vnode op for write using bio 6881541Srgrimes */ 6891549Srgrimesint 6901541Srgrimesnfs_write(ap) 6911541Srgrimes struct vop_write_args /* { 6921541Srgrimes struct vnode *a_vp; 6931541Srgrimes struct uio *a_uio; 6941541Srgrimes int a_ioflag; 6951541Srgrimes struct ucred *a_cred; 6961541Srgrimes } */ *ap; 6971541Srgrimes{ 69846349Salc int biosize; 69946349Salc struct uio *uio = ap->a_uio; 7001541Srgrimes struct proc *p = uio->uio_procp; 70146349Salc struct vnode *vp = ap->a_vp; 7021541Srgrimes struct nfsnode *np = VTONFS(vp); 70346349Salc struct ucred *cred = ap->a_cred; 7041541Srgrimes int ioflag = ap->a_ioflag; 7051541Srgrimes struct buf *bp; 7061541Srgrimes struct vattr vattr; 7079336Sdfr struct nfsmount *nmp = VFSTONFS(vp->v_mount); 70811921Sphk daddr_t lbn; 70946349Salc int bcount; 7109336Sdfr int n, on, error = 0, iomode, must_commit; 7111541Srgrimes 7121541Srgrimes#ifdef DIAGNOSTIC 7131541Srgrimes if (uio->uio_rw != UIO_WRITE) 7141541Srgrimes panic("nfs_write mode"); 7151541Srgrimes if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) 7161541Srgrimes panic("nfs_write proc"); 7171541Srgrimes#endif 7181541Srgrimes if (vp->v_type != VREG) 7191541Srgrimes return (EIO); 7201541Srgrimes if (np->n_flag & NWRITEERR) { 7211541Srgrimes np->n_flag &= ~NWRITEERR; 7221541Srgrimes return (np->n_error); 7231541Srgrimes } 72436176Speter if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && 72536176Speter (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 7269336Sdfr (void)nfs_fsinfo(nmp, vp, cred, p); 7271541Srgrimes if (ioflag & (IO_APPEND | IO_SYNC)) { 7281541Srgrimes if (np->n_flag & NMODIFIED) { 7291541Srgrimes np->n_attrstamp = 0; 7303305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 7313305Sphk if (error) 7321541Srgrimes return (error); 7331541Srgrimes } 7341541Srgrimes if (ioflag & IO_APPEND) { 7351541Srgrimes np->n_attrstamp = 0; 7363305Sphk error = VOP_GETATTR(vp, &vattr, cred, p); 7373305Sphk if (error) 7381541Srgrimes return (error); 7391541Srgrimes uio->uio_offset = np->n_size; 7401541Srgrimes } 7411541Srgrimes } 7421541Srgrimes if (uio->uio_offset < 0) 7431541Srgrimes return (EINVAL); 74436473Speter if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) 74536473Speter return (EFBIG); 7461541Srgrimes if (uio->uio_resid == 0) 7471541Srgrimes return (0); 7481541Srgrimes /* 7491541Srgrimes * Maybe this should be above the vnode op call, but so long as 7501541Srgrimes * file servers have no limits, i don't think it matters 7511541Srgrimes */ 7521541Srgrimes if (p && uio->uio_offset + uio->uio_resid > 7531541Srgrimes p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 7541541Srgrimes psignal(p, SIGXFSZ); 7551541Srgrimes return (EFBIG); 7561541Srgrimes } 75746349Salc 7589428Sdfr biosize = vp->v_mount->mnt_stat.f_iosize; 75946349Salc 7601541Srgrimes do { 7611541Srgrimes /* 7621541Srgrimes * Check for a valid write lease. 7631541Srgrimes */ 7641541Srgrimes if ((nmp->nm_flag & NFSMNT_NQNFS) && 7659336Sdfr NQNFS_CKINVALID(vp, np, ND_WRITE)) { 7661541Srgrimes do { 7679336Sdfr error = nqnfs_getlease(vp, ND_WRITE, cred, p); 7681541Srgrimes } while (error == NQNFS_EXPIRED); 7691541Srgrimes if (error) 7701541Srgrimes return (error); 7711541Srgrimes if (np->n_lrev != np->n_brev || 7721541Srgrimes (np->n_flag & NQNFSNONCACHE)) { 7733305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 7743305Sphk if (error) 7751541Srgrimes return (error); 7761541Srgrimes np->n_brev = np->n_lrev; 7771541Srgrimes } 7781541Srgrimes } 7799336Sdfr if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { 7809336Sdfr iomode = NFSV3WRITE_FILESYNC; 7819336Sdfr error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); 7829336Sdfr if (must_commit) 7839336Sdfr nfs_clearcommit(vp->v_mount); 7849336Sdfr return (error); 7859336Sdfr } 7861541Srgrimes nfsstats.biocache_writes++; 7871541Srgrimes lbn = uio->uio_offset / biosize; 7881541Srgrimes on = uio->uio_offset & (biosize-1); 7891541Srgrimes n = min((unsigned)(biosize - on), uio->uio_resid); 7901541Srgrimesagain: 79146349Salc /* 79246349Salc * Handle direct append and file extension cases, calculate 79346349Salc * unaligned buffer size. 79446349Salc */ 79546349Salc 79646349Salc if (uio->uio_offset == np->n_size && n) { 79746349Salc /* 79846349Salc * special append case. Obtain buffer prior to 79946349Salc * resizing it to maintain B_CACHE. 80046349Salc */ 80146349Salc long save; 80246349Salc 80346349Salc bcount = on; 80446349Salc bp = nfs_getcacheblk(vp, lbn, bcount, p); 80549659Sdt if (!bp) 80649659Sdt return (EINTR); 80746349Salc save = bp->b_flags & B_CACHE; 80846349Salc 8098692Sdg np->n_size = uio->uio_offset + n; 81025023Sdfr np->n_flag |= NMODIFIED; 81141026Speter vnode_pager_setsize(vp, np->n_size); 81246349Salc 81346349Salc bcount += n; 81446349Salc allocbuf(bp, bcount); 81546349Salc bp->b_flags |= save; 81646349Salc } else { 81746349Salc if (uio->uio_offset + n > np->n_size) { 81846349Salc np->n_size = uio->uio_offset + n; 81946349Salc np->n_flag |= NMODIFIED; 82046349Salc vnode_pager_setsize(vp, np->n_size); 82146349Salc } 82246349Salc bcount = biosize; 82346349Salc if ((off_t)(lbn + 1) * biosize > np->n_size) 82446349Salc bcount = np->n_size - (off_t)lbn * biosize; 82546349Salc bp = nfs_getcacheblk(vp, lbn, bcount, p); 82649659Sdt if (!bp) 82749659Sdt return (EINTR); 8288692Sdg } 82946349Salc 83046349Salc /* 83146349Salc * Issue a READ if B_CACHE is not set. In special-append 83246349Salc * mode, B_CACHE is based on the buffer prior to the write 83346349Salc * op and is typically set, avoiding the read. If a read 83446349Salc * is required in special append mode, the server will 83546349Salc * probably send us a short-read since we extended the file 83646349Salc * on our end, resulting in b_resid == 0 and, thusly, 83746349Salc * B_CACHE getting set. 83846349Salc * 83946349Salc * We can also avoid issuing the read if the write covers 84046349Salc * the entire buffer. We have to make sure the buffer state 84146349Salc * is reasonable in this case since we will not be initiating 84246349Salc * I/O. See the comments in kern/vfs_bio.c's getblk() for 84346349Salc * more information. 84446349Salc * 84546349Salc * B_CACHE may also be set due to the buffer being cached 84646349Salc * normally. 84746349Salc */ 84846349Salc 84946349Salc if (on == 0 && n == bcount) { 85046349Salc bp->b_flags |= B_CACHE; 85146349Salc bp->b_flags &= ~(B_ERROR | B_INVAL); 8528692Sdg } 85346349Salc 85446349Salc if ((bp->b_flags & B_CACHE) == 0) { 85546349Salc bp->b_flags |= B_READ; 85646349Salc vfs_busy_pages(bp, 0); 85746349Salc error = nfs_doio(bp, cred, p); 85846349Salc if (error) { 85946349Salc brelse(bp); 86046349Salc return (error); 86146349Salc } 86246349Salc } 8631541Srgrimes if (!bp) 8641541Srgrimes return (EINTR); 8651541Srgrimes if (bp->b_wcred == NOCRED) { 8661541Srgrimes crhold(cred); 8671541Srgrimes bp->b_wcred = cred; 8681541Srgrimes } 8691541Srgrimes np->n_flag |= NMODIFIED; 8708692Sdg 87145347Sjulian /* 87245347Sjulian * If dirtyend exceeds file size, chop it down. If this 87345347Sjulian * creates a reverse-indexed or degenerate situation with 87445347Sjulian * dirtyoff/end, 0 them. 87545347Sjulian */ 87645347Sjulian 87741791Sdt if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size) 87841791Sdt bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE; 87945347Sjulian if (bp->b_dirtyoff >= bp->b_dirtyend) 88045347Sjulian bp->b_dirtyoff = bp->b_dirtyend = 0; 88131617Sdyson 8821541Srgrimes /* 88331617Sdyson * If the new write will leave a contiguous dirty 88431617Sdyson * area, just update the b_dirtyoff and b_dirtyend, 88531617Sdyson * otherwise force a write rpc of the old dirty area. 88646349Salc * 88746349Salc * While it is possible to merge discontiguous writes due to 88846349Salc * our having a B_CACHE buffer ( and thus valid read data 88946349Salc * for the hole), we don't because it could lead to 89046349Salc * significant cache coherency problems with multiple clients, 89146349Salc * especially if locking is implemented later on. 89246349Salc * 89346349Salc * as an optimization we could theoretically maintain 89446349Salc * a linked list of discontinuous areas, but we would still 89546349Salc * have to commit them separately so there isn't much 89646349Salc * advantage to it except perhaps a bit of asynchronization. 89731617Sdyson */ 89842957Sdillon 89931617Sdyson if (bp->b_dirtyend > 0 && 90031617Sdyson (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { 90147964Smckusick if (VOP_BWRITE(bp->b_vp, bp) == EINTR) 90231617Sdyson return (EINTR); 90331617Sdyson goto again; 90431617Sdyson } 90531617Sdyson 90631617Sdyson /* 9071541Srgrimes * Check for valid write lease and get one as required. 9081541Srgrimes * In case getblk() and/or bwrite() delayed us. 9091541Srgrimes */ 9101541Srgrimes if ((nmp->nm_flag & NFSMNT_NQNFS) && 9119336Sdfr NQNFS_CKINVALID(vp, np, ND_WRITE)) { 9121541Srgrimes do { 9139336Sdfr error = nqnfs_getlease(vp, ND_WRITE, cred, p); 9141541Srgrimes } while (error == NQNFS_EXPIRED); 9151541Srgrimes if (error) { 9161541Srgrimes brelse(bp); 9171541Srgrimes return (error); 9181541Srgrimes } 9191541Srgrimes if (np->n_lrev != np->n_brev || 9201541Srgrimes (np->n_flag & NQNFSNONCACHE)) { 9211541Srgrimes brelse(bp); 9223305Sphk error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 9233305Sphk if (error) 9241541Srgrimes return (error); 9251541Srgrimes np->n_brev = np->n_lrev; 9261541Srgrimes goto again; 9271541Srgrimes } 9281541Srgrimes } 92934206Sdyson 9303305Sphk error = uiomove((char *)bp->b_data + on, n, uio); 93154480Sdillon 93254480Sdillon /* 93354480Sdillon * Since this block is being modified, it must be written 93454480Sdillon * again and not just committed. Since write clustering does 93554480Sdillon * not work for the stage 1 data write, only the stage 2 93654480Sdillon * commit rpc, we have to clear B_CLUSTEROK as well. 93754480Sdillon */ 93854480Sdillon bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 93954480Sdillon 9403305Sphk if (error) { 9411541Srgrimes bp->b_flags |= B_ERROR; 9421541Srgrimes brelse(bp); 9431541Srgrimes return (error); 9441541Srgrimes } 94534206Sdyson 94634206Sdyson /* 94745347Sjulian * Only update dirtyoff/dirtyend if not a degenerate 94845347Sjulian * condition. 94945347Sjulian */ 95045347Sjulian if (n) { 95145347Sjulian if (bp->b_dirtyend > 0) { 95245347Sjulian bp->b_dirtyoff = min(on, bp->b_dirtyoff); 95345347Sjulian bp->b_dirtyend = max((on + n), bp->b_dirtyend); 95445347Sjulian } else { 95545347Sjulian bp->b_dirtyoff = on; 95645347Sjulian bp->b_dirtyend = on + n; 95745347Sjulian } 95846349Salc vfs_bio_set_validclean(bp, on, n); 9591541Srgrimes } 96045347Sjulian 96144679Sjulian /* 9621541Srgrimes * If the lease is non-cachable or IO_SYNC do bwrite(). 96346349Salc * 96446349Salc * IO_INVAL appears to be unused. The idea appears to be 96546349Salc * to turn off caching in this case. Very odd. XXX 9661541Srgrimes */ 9671541Srgrimes if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { 96834206Sdyson if (ioflag & IO_INVAL) 96946349Salc bp->b_flags |= B_NOCACHE; 97047964Smckusick error = VOP_BWRITE(bp->b_vp, bp); 9713305Sphk if (error) 9721541Srgrimes return (error); 9739336Sdfr if (np->n_flag & NQNFSNONCACHE) { 9749336Sdfr error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); 9759336Sdfr if (error) 9769336Sdfr return (error); 9779336Sdfr } 9781541Srgrimes } else if ((n + on) == biosize && 9791541Srgrimes (nmp->nm_flag & NFSMNT_NQNFS) == 0) { 9809336Sdfr bp->b_flags |= B_ASYNC; 98146580Sphk (void)nfs_writebp(bp, 0, 0); 98246349Salc } else { 9831541Srgrimes bdwrite(bp); 98446349Salc } 9851541Srgrimes } while (uio->uio_resid > 0 && n > 0); 9861541Srgrimes return (0); 9871541Srgrimes} 9881541Srgrimes 9891541Srgrimes/* 9901541Srgrimes * Get an nfs cache block. 99154480Sdillon * 9921541Srgrimes * Allocate a new one if the block isn't currently in the cache 9931541Srgrimes * and return the block marked busy. If the calling process is 9941541Srgrimes * interrupted by a signal for an interruptible mount point, return 9951541Srgrimes * NULL. 99654480Sdillon * 99754480Sdillon * The caller must carefully deal with the possible B_INVAL state of 99854480Sdillon * the buffer. nfs_doio() clears B_INVAL (and nfs_asyncio() clears it 99954480Sdillon * indirectly), so synchronous reads can be issued without worrying about 100054480Sdillon * the B_INVAL state. We have to be a little more careful when dealing 100154480Sdillon * with writes (see comments in nfs_write()) when extending a file past 100254480Sdillon * its EOF. 10031541Srgrimes */ 100412911Sphkstatic struct buf * 10051541Srgrimesnfs_getcacheblk(vp, bn, size, p) 10061541Srgrimes struct vnode *vp; 10071541Srgrimes daddr_t bn; 10081541Srgrimes int size; 10091541Srgrimes struct proc *p; 10101541Srgrimes{ 10111541Srgrimes register struct buf *bp; 101232755Sdyson struct mount *mp; 101332755Sdyson struct nfsmount *nmp; 10141541Srgrimes 101532755Sdyson mp = vp->v_mount; 101632755Sdyson nmp = VFSTONFS(mp); 101732755Sdyson 10181541Srgrimes if (nmp->nm_flag & NFSMNT_INT) { 10191541Srgrimes bp = getblk(vp, bn, size, PCATCH, 0); 10201541Srgrimes while (bp == (struct buf *)0) { 10211541Srgrimes if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) 10221541Srgrimes return ((struct buf *)0); 10231541Srgrimes bp = getblk(vp, bn, size, 0, 2 * hz); 10241541Srgrimes } 102546349Salc } else { 10261541Srgrimes bp = getblk(vp, bn, size, 0, 0); 102746349Salc } 10285455Sdg 102941791Sdt if (vp->v_type == VREG) { 103032755Sdyson int biosize; 103146349Salc 103232755Sdyson biosize = mp->mnt_stat.f_iosize; 103341791Sdt bp->b_blkno = bn * (biosize / DEV_BSIZE); 103432755Sdyson } 10351541Srgrimes return (bp); 10361541Srgrimes} 10371541Srgrimes 10381541Srgrimes/* 10391541Srgrimes * Flush and invalidate all dirty buffers. If another process is already 10401541Srgrimes * doing the flush, just wait for completion. 10411541Srgrimes */ 10421549Srgrimesint 10431541Srgrimesnfs_vinvalbuf(vp, flags, cred, p, intrflg) 10441541Srgrimes struct vnode *vp; 10451541Srgrimes int flags; 10461541Srgrimes struct ucred *cred; 10471541Srgrimes struct proc *p; 10481541Srgrimes int intrflg; 10491541Srgrimes{ 10501541Srgrimes register struct nfsnode *np = VTONFS(vp); 10511541Srgrimes struct nfsmount *nmp = VFSTONFS(vp->v_mount); 10521541Srgrimes int error = 0, slpflag, slptimeo; 10531541Srgrimes 105432755Sdyson if (vp->v_flag & VXLOCK) { 105532755Sdyson return (0); 105632755Sdyson } 105732755Sdyson 10581541Srgrimes if ((nmp->nm_flag & NFSMNT_INT) == 0) 10591541Srgrimes intrflg = 0; 10601541Srgrimes if (intrflg) { 10611541Srgrimes slpflag = PCATCH; 10621541Srgrimes slptimeo = 2 * hz; 10631541Srgrimes } else { 10641541Srgrimes slpflag = 0; 10651541Srgrimes slptimeo = 0; 10661541Srgrimes } 10671541Srgrimes /* 10681541Srgrimes * First wait for any other process doing a flush to complete. 10691541Srgrimes */ 10701541Srgrimes while (np->n_flag & NFLUSHINPROG) { 10711541Srgrimes np->n_flag |= NFLUSHWANT; 10721541Srgrimes error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", 10731541Srgrimes slptimeo); 10741541Srgrimes if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) 10751541Srgrimes return (EINTR); 10761541Srgrimes } 10771541Srgrimes 10781541Srgrimes /* 10791541Srgrimes * Now, flush as required. 10801541Srgrimes */ 10811541Srgrimes np->n_flag |= NFLUSHINPROG; 10821541Srgrimes error = vinvalbuf(vp, flags, cred, p, slpflag, 0); 10831541Srgrimes while (error) { 10841541Srgrimes if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { 10851541Srgrimes np->n_flag &= ~NFLUSHINPROG; 10861541Srgrimes if (np->n_flag & NFLUSHWANT) { 10871541Srgrimes np->n_flag &= ~NFLUSHWANT; 10881541Srgrimes wakeup((caddr_t)&np->n_flag); 10891541Srgrimes } 10901541Srgrimes return (EINTR); 10911541Srgrimes } 10921541Srgrimes error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); 10931541Srgrimes } 10941541Srgrimes np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); 10951541Srgrimes if (np->n_flag & NFLUSHWANT) { 10961541Srgrimes np->n_flag &= ~NFLUSHWANT; 10971541Srgrimes wakeup((caddr_t)&np->n_flag); 10981541Srgrimes } 10991541Srgrimes return (0); 11001541Srgrimes} 11011541Srgrimes 11021541Srgrimes/* 11031541Srgrimes * Initiate asynchronous I/O. Return an error if no nfsiods are available. 11041541Srgrimes * This is mainly to avoid queueing async I/O requests when the nfsiods 11051541Srgrimes * are all hung on a dead server. 110646349Salc * 110746349Salc * Note: nfs_asyncio() does not clear (B_ERROR|B_INVAL) but when the bp 110846349Salc * is eventually dequeued by the async daemon, nfs_doio() *will*. 11091541Srgrimes */ 11101549Srgrimesint 111146580Sphknfs_asyncio(bp, cred, procp) 11121541Srgrimes register struct buf *bp; 11131541Srgrimes struct ucred *cred; 111446580Sphk struct proc *procp; 11151541Srgrimes{ 111619449Sdfr struct nfsmount *nmp; 111719449Sdfr int i; 111819449Sdfr int gotiod; 111919449Sdfr int slpflag = 0; 112019449Sdfr int slptimeo = 0; 112119449Sdfr int error; 11221541Srgrimes 11231541Srgrimes if (nfs_numasync == 0) 11241541Srgrimes return (EIO); 112544679Sjulian 112619449Sdfr nmp = VFSTONFS(bp->b_vp->v_mount); 112719449Sdfragain: 112819449Sdfr if (nmp->nm_flag & NFSMNT_INT) 112919449Sdfr slpflag = PCATCH; 113019449Sdfr gotiod = FALSE; 113119449Sdfr 113219449Sdfr /* 113319449Sdfr * Find a free iod to process this request. 113419449Sdfr */ 11351541Srgrimes for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 113619449Sdfr if (nfs_iodwant[i]) { 113719449Sdfr /* 113819449Sdfr * Found one, so wake it up and tell it which 113919449Sdfr * mount to process. 114019449Sdfr */ 114119449Sdfr NFS_DPF(ASYNCIO, 114219449Sdfr ("nfs_asyncio: waking iod %d for mount %p\n", 114319449Sdfr i, nmp)); 114419449Sdfr nfs_iodwant[i] = (struct proc *)0; 114519449Sdfr nfs_iodmount[i] = nmp; 114619449Sdfr nmp->nm_bufqiods++; 114719449Sdfr wakeup((caddr_t)&nfs_iodwant[i]); 114819449Sdfr gotiod = TRUE; 114925023Sdfr break; 115019449Sdfr } 115119449Sdfr 115219449Sdfr /* 115319449Sdfr * If none are free, we may already have an iod working on this mount 115419449Sdfr * point. If so, it will process our request. 115519449Sdfr */ 115619449Sdfr if (!gotiod) { 115719449Sdfr if (nmp->nm_bufqiods > 0) { 115819449Sdfr NFS_DPF(ASYNCIO, 115919449Sdfr ("nfs_asyncio: %d iods are already processing mount %p\n", 116019449Sdfr nmp->nm_bufqiods, nmp)); 116119449Sdfr gotiod = TRUE; 116219449Sdfr } 116319449Sdfr } 116419449Sdfr 116519449Sdfr /* 116619449Sdfr * If we have an iod which can process the request, then queue 116719449Sdfr * the buffer. 116819449Sdfr */ 116919449Sdfr if (gotiod) { 117019449Sdfr /* 117119449Sdfr * Ensure that the queue never grows too large. 117219449Sdfr */ 117319449Sdfr while (nmp->nm_bufqlen >= 2*nfs_numasync) { 117419449Sdfr NFS_DPF(ASYNCIO, 117519449Sdfr ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); 117619449Sdfr nmp->nm_bufqwant = TRUE; 117719449Sdfr error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, 117819449Sdfr "nfsaio", slptimeo); 117919449Sdfr if (error) { 118046580Sphk if (nfs_sigintr(nmp, NULL, procp)) 118119449Sdfr return (EINTR); 118219449Sdfr if (slpflag == PCATCH) { 118319449Sdfr slpflag = 0; 118419449Sdfr slptimeo = 2 * hz; 118519449Sdfr } 118619449Sdfr } 118719449Sdfr /* 118819449Sdfr * We might have lost our iod while sleeping, 118919449Sdfr * so check and loop if nescessary. 119019449Sdfr */ 119119449Sdfr if (nmp->nm_bufqiods == 0) { 119219449Sdfr NFS_DPF(ASYNCIO, 119319449Sdfr ("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); 119419449Sdfr goto again; 119519449Sdfr } 119619449Sdfr } 119719449Sdfr 11981541Srgrimes if (bp->b_flags & B_READ) { 11991541Srgrimes if (bp->b_rcred == NOCRED && cred != NOCRED) { 12001541Srgrimes crhold(cred); 12011541Srgrimes bp->b_rcred = cred; 12021541Srgrimes } 12031541Srgrimes } else { 12049336Sdfr bp->b_flags |= B_WRITEINPROG; 12051541Srgrimes if (bp->b_wcred == NOCRED && cred != NOCRED) { 12061541Srgrimes crhold(cred); 12071541Srgrimes bp->b_wcred = cred; 12081541Srgrimes } 12091541Srgrimes } 12108876Srgrimes 121148225Smckusick BUF_KERNPROC(bp); 121219449Sdfr TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); 121319449Sdfr nmp->nm_bufqlen++; 12141541Srgrimes return (0); 121519449Sdfr } 12169336Sdfr 12179336Sdfr /* 121819449Sdfr * All the iods are busy on other mounts, so return EIO to 121919449Sdfr * force the caller to process the i/o synchronously. 12209336Sdfr */ 122119449Sdfr NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); 122219449Sdfr return (EIO); 12231541Srgrimes} 12241541Srgrimes 12251541Srgrimes/* 12261541Srgrimes * Do an I/O operation to/from a cache block. This may be called 12271541Srgrimes * synchronously or from an nfsiod. 12281541Srgrimes */ 12291541Srgrimesint 12301541Srgrimesnfs_doio(bp, cr, p) 123144679Sjulian struct buf *bp; 12323305Sphk struct ucred *cr; 12331541Srgrimes struct proc *p; 12341541Srgrimes{ 123544679Sjulian struct uio *uiop; 123644679Sjulian struct vnode *vp; 12371541Srgrimes struct nfsnode *np; 12381541Srgrimes struct nfsmount *nmp; 123946349Salc int error = 0, iomode, must_commit = 0; 12401541Srgrimes struct uio uio; 12411541Srgrimes struct iovec io; 12421541Srgrimes 12431541Srgrimes vp = bp->b_vp; 12441541Srgrimes np = VTONFS(vp); 12451541Srgrimes nmp = VFSTONFS(vp->v_mount); 12461541Srgrimes uiop = &uio; 12471541Srgrimes uiop->uio_iov = &io; 12481541Srgrimes uiop->uio_iovcnt = 1; 12491541Srgrimes uiop->uio_segflg = UIO_SYSSPACE; 12501541Srgrimes uiop->uio_procp = p; 12511541Srgrimes 125246349Salc /* 125346349Salc * clear B_ERROR and B_INVAL state prior to initiating the I/O. We 125446349Salc * do this here so we do not have to do it in all the code that 125546349Salc * calls us. 125646349Salc */ 125746349Salc bp->b_flags &= ~(B_ERROR | B_INVAL); 125846349Salc 125944679Sjulian KASSERT(!(bp->b_flags & B_DONE), ("nfs_doio: bp %p already marked done", bp)); 126044679Sjulian 12611541Srgrimes /* 12621541Srgrimes * Historically, paging was done with physio, but no more. 12631541Srgrimes */ 12643664Sphk if (bp->b_flags & B_PHYS) { 12653664Sphk /* 12663664Sphk * ...though reading /dev/drum still gets us here. 12673664Sphk */ 12681541Srgrimes io.iov_len = uiop->uio_resid = bp->b_bcount; 12693664Sphk /* mapping was done by vmapbuf() */ 12701541Srgrimes io.iov_base = bp->b_data; 12719336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 12723664Sphk if (bp->b_flags & B_READ) { 12733664Sphk uiop->uio_rw = UIO_READ; 12743664Sphk nfsstats.read_physios++; 12753664Sphk error = nfs_readrpc(vp, uiop, cr); 12763664Sphk } else { 12779336Sdfr int com; 12789336Sdfr 12799336Sdfr iomode = NFSV3WRITE_DATASYNC; 12803664Sphk uiop->uio_rw = UIO_WRITE; 12813664Sphk nfsstats.write_physios++; 12829336Sdfr error = nfs_writerpc(vp, uiop, cr, &iomode, &com); 12833664Sphk } 12843664Sphk if (error) { 12853664Sphk bp->b_flags |= B_ERROR; 12863664Sphk bp->b_error = error; 12873664Sphk } 12883664Sphk } else if (bp->b_flags & B_READ) { 12893664Sphk io.iov_len = uiop->uio_resid = bp->b_bcount; 12903664Sphk io.iov_base = bp->b_data; 12911541Srgrimes uiop->uio_rw = UIO_READ; 12921541Srgrimes switch (vp->v_type) { 12931541Srgrimes case VREG: 12949336Sdfr uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; 12951541Srgrimes nfsstats.read_bios++; 12961541Srgrimes error = nfs_readrpc(vp, uiop, cr); 12971541Srgrimes if (!error) { 12981541Srgrimes if (uiop->uio_resid) { 12991541Srgrimes /* 130046349Salc * If we had a short read with no error, we must have 130146349Salc * hit a file hole. We should zero-fill the remainder. 130246349Salc * This can also occur if the server hits the file EOF. 130346349Salc * 130446349Salc * Holes used to be able to occur due to pending 130546349Salc * writes, but that is not possible any longer. 13061541Srgrimes */ 130746349Salc int nread = bp->b_bcount - uiop->uio_resid; 130846349Salc int left = bp->b_bcount - nread; 130946349Salc 131046349Salc if (left > 0) 131146349Salc bzero((char *)bp->b_data + nread, left); 131246349Salc uiop->uio_resid = 0; 131346349Salc } 13141541Srgrimes } 13151541Srgrimes if (p && (vp->v_flag & VTEXT) && 13161541Srgrimes (((nmp->nm_flag & NFSMNT_NQNFS) && 13179336Sdfr NQNFS_CKINVALID(vp, np, ND_READ) && 13181541Srgrimes np->n_lrev != np->n_brev) || 13191541Srgrimes (!(nmp->nm_flag & NFSMNT_NQNFS) && 132018397Snate np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { 13211541Srgrimes uprintf("Process killed due to text file modification\n"); 13221541Srgrimes psignal(p, SIGKILL); 132345361Speter PHOLD(p); 13241541Srgrimes } 13251541Srgrimes break; 13261541Srgrimes case VLNK: 13279336Sdfr uiop->uio_offset = (off_t)0; 13281541Srgrimes nfsstats.readlink_bios++; 13291541Srgrimes error = nfs_readlinkrpc(vp, uiop, cr); 13301541Srgrimes break; 13311541Srgrimes case VDIR: 13321541Srgrimes nfsstats.readdir_bios++; 13339336Sdfr uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; 13349336Sdfr if (nmp->nm_flag & NFSMNT_RDIRPLUS) { 13359336Sdfr error = nfs_readdirplusrpc(vp, uiop, cr); 13369336Sdfr if (error == NFSERR_NOTSUPP) 13379336Sdfr nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 13389336Sdfr } 13399336Sdfr if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) 13409336Sdfr error = nfs_readdirrpc(vp, uiop, cr); 134146349Salc /* 134246349Salc * end-of-directory sets B_INVAL but does not generate an 134346349Salc * error. 134446349Salc */ 134539782Smckusick if (error == 0 && uiop->uio_resid == bp->b_bcount) 134639782Smckusick bp->b_flags |= B_INVAL; 13471541Srgrimes break; 13483305Sphk default: 13493305Sphk printf("nfs_doio: type %x unexpected\n",vp->v_type); 13503305Sphk break; 13511541Srgrimes }; 13521541Srgrimes if (error) { 13531541Srgrimes bp->b_flags |= B_ERROR; 13541541Srgrimes bp->b_error = error; 13551541Srgrimes } 13561541Srgrimes } else { 135751344Sdillon /* 135851344Sdillon * If we only need to commit, try to commit 135951344Sdillon */ 136051344Sdillon if (bp->b_flags & B_NEEDCOMMIT) { 136151344Sdillon int retv; 136251344Sdillon off_t off; 136351344Sdillon 136451344Sdillon off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; 136551344Sdillon bp->b_flags |= B_WRITEINPROG; 136651344Sdillon retv = nfs_commit( 136751344Sdillon bp->b_vp, off, bp->b_dirtyend-bp->b_dirtyoff, 136851344Sdillon bp->b_wcred, p); 136951344Sdillon bp->b_flags &= ~B_WRITEINPROG; 137051344Sdillon if (retv == 0) { 137151344Sdillon bp->b_dirtyoff = bp->b_dirtyend = 0; 137254480Sdillon bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 137351344Sdillon bp->b_resid = 0; 137451344Sdillon biodone(bp); 137551344Sdillon return (0); 137651344Sdillon } 137751344Sdillon if (retv == NFSERR_STALEWRITEVERF) { 137851344Sdillon nfs_clearcommit(bp->b_vp->v_mount); 137951344Sdillon } 138051344Sdillon } 138151344Sdillon 138251344Sdillon /* 138351344Sdillon * Setup for actual write 138451344Sdillon */ 138551344Sdillon 138641791Sdt if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size) 138741791Sdt bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE; 13888692Sdg 13898692Sdg if (bp->b_dirtyend > bp->b_dirtyoff) { 13908692Sdg io.iov_len = uiop->uio_resid = bp->b_dirtyend 13919336Sdfr - bp->b_dirtyoff; 139241791Sdt uiop->uio_offset = (off_t)bp->b_blkno * DEV_BSIZE 13939336Sdfr + bp->b_dirtyoff; 13948692Sdg io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; 13958692Sdg uiop->uio_rw = UIO_WRITE; 13968692Sdg nfsstats.write_bios++; 139744679Sjulian 139825785Sdfr if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC) 13999336Sdfr iomode = NFSV3WRITE_UNSTABLE; 14008692Sdg else 14019336Sdfr iomode = NFSV3WRITE_FILESYNC; 140244679Sjulian 14039336Sdfr bp->b_flags |= B_WRITEINPROG; 14049336Sdfr error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); 140551475Sdillon 140651475Sdillon /* 140751475Sdillon * When setting B_NEEDCOMMIT also set B_CLUSTEROK to try 140851475Sdillon * to cluster the buffers needing commit. This will allow 140951475Sdillon * the system to submit a single commit rpc for the whole 141054480Sdillon * cluster. We can do this even if the buffer is not 100% 141154480Sdillon * dirty (relative to the NFS blocksize), so we optimize the 141254480Sdillon * append-to-file-case. 141354480Sdillon * 141454480Sdillon * (when clearing B_NEEDCOMMIT, B_CLUSTEROK must also be 141554480Sdillon * cleared because write clustering only works for commit 141654480Sdillon * rpc's, not for the data portion of the write). 141751475Sdillon */ 141851475Sdillon 141925003Sdfr if (!error && iomode == NFSV3WRITE_UNSTABLE) { 142025003Sdfr bp->b_flags |= B_NEEDCOMMIT; 142125003Sdfr if (bp->b_dirtyoff == 0 142246349Salc && bp->b_dirtyend == bp->b_bcount) 142325003Sdfr bp->b_flags |= B_CLUSTEROK; 142444679Sjulian } else { 142554480Sdillon bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 142644679Sjulian } 14279336Sdfr bp->b_flags &= ~B_WRITEINPROG; 14288692Sdg 14299336Sdfr /* 14309336Sdfr * For an interrupted write, the buffer is still valid 14319336Sdfr * and the write hasn't been pushed to the server yet, 14329336Sdfr * so we can't set B_ERROR and report the interruption 14339336Sdfr * by setting B_EINTR. For the B_ASYNC case, B_EINTR 14349336Sdfr * is not relevant, so the rpc attempt is essentially 14359336Sdfr * a noop. For the case of a V3 write rpc not being 14369336Sdfr * committed to stable storage, the block is still 14379336Sdfr * dirty and requires either a commit rpc or another 14389336Sdfr * write rpc with iomode == NFSV3WRITE_FILESYNC before 14399336Sdfr * the block is reused. This is indicated by setting 14409336Sdfr * the B_DELWRI and B_NEEDCOMMIT flags. 144142957Sdillon * 144242957Sdillon * If the buffer is marked B_PAGING, it does not reside on 144344679Sjulian * the vp's paging queues so we cannot call bdirty(). The 144444679Sjulian * bp in this case is not an NFS cache block so we should 144544679Sjulian * be safe. XXX 14469336Sdfr */ 14479336Sdfr if (error == EINTR 14489336Sdfr || (!error && (bp->b_flags & B_NEEDCOMMIT))) { 144934266Sjulian int s; 145034266Sjulian 145144679Sjulian s = splbio(); 14528692Sdg bp->b_flags &= ~(B_INVAL|B_NOCACHE); 145342957Sdillon if ((bp->b_flags & B_PAGING) == 0) { 145444679Sjulian bdirty(bp); 145544679Sjulian bp->b_flags &= ~B_DONE; 145642957Sdillon } 145747749Speter if (error && (bp->b_flags & B_ASYNC) == 0) 145832755Sdyson bp->b_flags |= B_EINTR; 145944679Sjulian splx(s); 14608692Sdg } else { 146144679Sjulian if (error) { 146244679Sjulian bp->b_flags |= B_ERROR; 146344679Sjulian bp->b_error = np->n_error = error; 146444679Sjulian np->n_flag |= NWRITEERR; 146544679Sjulian } 146644679Sjulian bp->b_dirtyoff = bp->b_dirtyend = 0; 14678692Sdg } 14681541Srgrimes } else { 14698692Sdg bp->b_resid = 0; 14708692Sdg biodone(bp); 14718692Sdg return (0); 14721541Srgrimes } 14731541Srgrimes } 14741541Srgrimes bp->b_resid = uiop->uio_resid; 14759336Sdfr if (must_commit) 147644679Sjulian nfs_clearcommit(vp->v_mount); 14771541Srgrimes biodone(bp); 14781541Srgrimes return (error); 14791541Srgrimes} 1480