Index: stable/4/sys/nfs/nfs_bio.c =================================================================== --- stable/4/sys/nfs/nfs_bio.c (revision 131097) +++ stable/4/sys/nfs/nfs_bio.c (revision 131098) @@ -1,1652 +1,1654 @@ /* * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, struct proc *p)); extern int nfs_numasync; extern int nfs_pbuf_freecnt; extern struct nfsstats nfsstats; /* * Vnode op for VM getpages. */ int nfs_getpages(ap) struct vop_getpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_reqpage; vm_ooffset_t a_offset; } */ *ap; { int i, error, nextoff, size, toff, count, npages; struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; struct vnode *vp; struct proc *p; struct ucred *cred; struct nfsmount *nmp; vm_page_t *pages; vp = ap->a_vp; p = curproc; /* XXX */ cred = curproc->p_ucred; /* XXX */ nmp = VFSTONFS(vp->v_mount); pages = ap->a_m; count = ap->a_count; if (vp->v_object == NULL) { printf("nfs_getpages: called with non-merged cache vnode??\n"); return VM_PAGER_ERROR; } if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) (void)nfs_fsinfo(nmp, vp, cred, p); npages = btoc(count); /* * If the requested page is partially valid, just return it and * allow the pager to zero-out the blanks. Partially valid pages * can only occur at the file EOF. */ { vm_page_t m = pages[ap->a_reqpage]; if (m->valid != 0) { /* handled by vm_fault now */ /* vm_page_zero_invalid(m, TRUE); */ for (i = 0; i < npages; ++i) { if (i != ap->a_reqpage) vnode_pager_freepage(pages[i]); } return(0); } } /* * We use only the kva address for the buffer, but this is extremely * convienient and fast. */ bp = getpbuf(&nfs_pbuf_freecnt); kva = (vm_offset_t) bp->b_data; pmap_qenter(kva, pages, npages); iov.iov_base = (caddr_t) kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = IDX_TO_OFF(pages[0]->pindex); uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_procp = p; error = nfs_readrpc(vp, &uio, cred); pmap_qremove(kva, npages); relpbuf(bp, &nfs_pbuf_freecnt); if (error && (uio.uio_resid == count)) { printf("nfs_getpages: error %d\n", error); for (i = 0; i < npages; ++i) { if (i != ap->a_reqpage) vnode_pager_freepage(pages[i]); } return VM_PAGER_ERROR; } /* * Calculate the number of bytes read and validate only that number * of bytes. Note that due to pending writes, size may be 0. This * does not mean that the remaining data is invalid! */ size = count - uio.uio_resid; for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { vm_page_t m; nextoff = toff + PAGE_SIZE; m = pages[i]; m->flags &= ~PG_ZERO; if (nextoff <= size) { /* * Read operation filled an entire page */ m->valid = VM_PAGE_BITS_ALL; vm_page_undirty(m); } else if (size > toff) { /* * Read operation filled a partial page. */ m->valid = 0; vm_page_set_validclean(m, 0, size - toff); /* handled by vm_fault now */ /* vm_page_zero_invalid(m, TRUE); */ } else { /* * Read operation was short. If no error occured * we may have hit a zero-fill section. We simply * leave valid set to 0. */ ; } if (i != ap->a_reqpage) { /* * Whether or not to leave the page activated is up in * the air, but we should put the page on a page queue * somewhere (it already is in the object). Result: * It appears that emperical results show that * deactivating pages is best. */ /* * Just in case someone was asking for this page we * now tell them that it is ok to use. */ if (!error) { if (m->flags & PG_WANTED) vm_page_activate(m); else vm_page_deactivate(m); vm_page_wakeup(m); } else { vnode_pager_freepage(m); } } } return 0; } /* * Vnode op for VM putpages. */ int nfs_putpages(ap) struct vop_putpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; vm_ooffset_t a_offset; } */ *ap; { struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; int iomode, must_commit, i, error, npages, count; off_t offset; int *rtvals; struct vnode *vp; struct proc *p; struct ucred *cred; struct nfsmount *nmp; struct nfsnode *np; vm_page_t *pages; vp = ap->a_vp; np = VTONFS(vp); p = curproc; /* XXX */ cred = curproc->p_ucred; /* XXX */ nmp = VFSTONFS(vp->v_mount); pages = ap->a_m; count = ap->a_count; rtvals = ap->a_rtvals; npages = btoc(count); offset = IDX_TO_OFF(pages[0]->pindex); if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) (void)nfs_fsinfo(nmp, vp, cred, p); for (i = 0; i < npages; i++) { rtvals[i] = VM_PAGER_AGAIN; } /* * When putting pages, do not extend file past EOF. */ if (offset + count > np->n_size) { count = np->n_size - offset; if (count < 0) count = 0; } /* * We use only the kva address for the buffer, but this is extremely * convienient and fast. */ bp = getpbuf(&nfs_pbuf_freecnt); kva = (vm_offset_t) bp->b_data; pmap_qenter(kva, pages, npages); iov.iov_base = (caddr_t) kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = offset; uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_WRITE; uio.uio_procp = p; if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0) iomode = NFSV3WRITE_UNSTABLE; else iomode = NFSV3WRITE_FILESYNC; error = nfs_writerpc(vp, &uio, cred, &iomode, &must_commit); pmap_qremove(kva, npages); relpbuf(bp, &nfs_pbuf_freecnt); if (!error) { int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE; for (i = 0; i < nwritten; i++) { rtvals[i] = VM_PAGER_OK; vm_page_undirty(pages[i]); } if (must_commit) nfs_clearcommit(vp->v_mount); } return rtvals[0]; } /* * Vnode op for read using bio */ int nfs_bioread(vp, uio, ioflag, cred) register struct vnode *vp; register struct uio *uio; int ioflag; struct ucred *cred; { register struct nfsnode *np = VTONFS(vp); register int biosize, i; struct buf *bp = 0, *rabp; struct vattr vattr; struct proc *p; struct nfsmount *nmp = VFSTONFS(vp->v_mount); daddr_t lbn, rabn; int bcount; int seqcount; int nra, error = 0, n = 0, on = 0; #ifdef DIAGNOSTIC if (uio->uio_rw != UIO_READ) panic("nfs_read mode"); #endif if (uio->uio_resid == 0) return (0); if (uio->uio_offset < 0) /* XXX VDIR cookies can be negative */ return (EINVAL); p = uio->uio_procp; if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) (void)nfs_fsinfo(nmp, vp, cred, p); if (vp->v_type != VDIR && (uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) return (EFBIG); biosize = vp->v_mount->mnt_stat.f_iosize; seqcount = (int)((off_t)(ioflag >> IO_SEQSHIFT) * biosize / BKVASIZE); /* * For nfs, cache consistency can only be maintained approximately. * Although RFC1094 does not specify the criteria, the following is * believed to be compatible with the reference port. * For nqnfs, full cache consistency is maintained within the loop. * For nfs: * If the file's modify time on the server has changed since the * last read rpc or you have written to the file, * you may have lost data cache consistency with the * server, so flush all of the file's data out of the cache. * Then force a getattr rpc to ensure that you have up to date * attributes. * NB: This implies that cache data can be read when up to * NFS_ATTRTIMEO seconds out of date. If you find that you need current * attributes this could be forced by setting n_attrstamp to 0 before * the VOP_GETATTR() call. */ if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) { if (np->n_flag & NMODIFIED) { if (vp->v_type != VREG) { if (vp->v_type != VDIR) panic("nfs: bioread, not dir"); nfs_invaldir(vp); error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) return (error); } np->n_attrstamp = 0; error = VOP_GETATTR(vp, &vattr, cred, p); if (error) return (error); np->n_mtime = vattr.va_mtime.tv_sec; } else { error = VOP_GETATTR(vp, &vattr, cred, p); if (error) return (error); - if (np->n_mtime != vattr.va_mtime.tv_sec) { + if ((np->n_flag & NSIZECHANGED) + || np->n_mtime != vattr.va_mtime.tv_sec) { if (vp->v_type == VDIR) nfs_invaldir(vp); error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) return (error); np->n_mtime = vattr.va_mtime.tv_sec; + np->n_flag &= ~NSIZECHANGED; } } } do { /* * Get a valid lease. If cached data is stale, flush it. */ if (nmp->nm_flag & NFSMNT_NQNFS) { if (NQNFS_CKINVALID(vp, np, ND_READ)) { do { error = nqnfs_getlease(vp, ND_READ, cred, p); } while (error == NQNFS_EXPIRED); if (error) return (error); if (np->n_lrev != np->n_brev || (np->n_flag & NQNFSNONCACHE) || ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) { if (vp->v_type == VDIR) nfs_invaldir(vp); error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) return (error); np->n_brev = np->n_lrev; } } else if (vp->v_type == VDIR && (np->n_flag & NMODIFIED)) { nfs_invaldir(vp); error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) return (error); } } if (np->n_flag & NQNFSNONCACHE) { switch (vp->v_type) { case VREG: return (nfs_readrpc(vp, uio, cred)); case VLNK: return (nfs_readlinkrpc(vp, uio, cred)); case VDIR: break; default: printf(" NQNFSNONCACHE: type %x unexpected\n", vp->v_type); }; } switch (vp->v_type) { case VREG: nfsstats.biocache_reads++; lbn = uio->uio_offset / biosize; on = uio->uio_offset & (biosize - 1); /* * Start the read ahead(s), as required. */ if (nfs_numasync > 0 && nmp->nm_readahead > 0) { for (nra = 0; nra < nmp->nm_readahead && nra < seqcount && (off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) { rabn = lbn + 1 + nra; if (!incore(vp, rabn)) { rabp = nfs_getcacheblk(vp, rabn, biosize, p); if (!rabp) return (EINTR); if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { rabp->b_flags |= (B_READ | B_ASYNC); vfs_busy_pages(rabp, 0); if (nfs_asyncio(rabp, cred, p)) { rabp->b_flags |= B_INVAL|B_ERROR; vfs_unbusy_pages(rabp); brelse(rabp); break; } } else { brelse(rabp); } } } } /* * Obtain the buffer cache block. Figure out the buffer size * when we are at EOF. If we are modifying the size of the * buffer based on an EOF condition we need to hold * nfs_rslock() through obtaining the buffer to prevent * a potential writer-appender from messing with n_size. * Otherwise we may accidently truncate the buffer and * lose dirty data. * * Note that bcount is *not* DEV_BSIZE aligned. */ again: bcount = biosize; if ((off_t)lbn * biosize >= np->n_size) { bcount = 0; } else if ((off_t)(lbn + 1) * biosize > np->n_size) { bcount = np->n_size - (off_t)lbn * biosize; } if (bcount != biosize) { switch(nfs_rslock(np, p)) { case ENOLCK: goto again; /* not reached */ case EINTR: case ERESTART: return(EINTR); /* not reached */ default: break; } } bp = nfs_getcacheblk(vp, lbn, bcount, p); if (bcount != biosize) nfs_rsunlock(np, p); if (!bp) return (EINTR); /* * If B_CACHE is not set, we must issue the read. If this * fails, we return an error. */ if ((bp->b_flags & B_CACHE) == 0) { bp->b_flags |= B_READ; vfs_busy_pages(bp, 0); error = nfs_doio(bp, cred, p); if (error) { brelse(bp); return (error); } } /* * on is the offset into the current bp. Figure out how many * bytes we can copy out of the bp. Note that bcount is * NOT DEV_BSIZE aligned. * * Then figure out how many bytes we can copy into the uio. */ n = 0; if (on < bcount) n = min((unsigned)(bcount - on), uio->uio_resid); break; case VLNK: nfsstats.biocache_readlinks++; bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, p); if (!bp) return (EINTR); if ((bp->b_flags & B_CACHE) == 0) { bp->b_flags |= B_READ; vfs_busy_pages(bp, 0); error = nfs_doio(bp, cred, p); if (error) { bp->b_flags |= B_ERROR; brelse(bp); return (error); } } n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); on = 0; break; case VDIR: nfsstats.biocache_readdirs++; if (np->n_direofoffset && uio->uio_offset >= np->n_direofoffset) { return (0); } lbn = (uoff_t)uio->uio_offset / NFS_DIRBLKSIZ; on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, p); if (!bp) return (EINTR); if ((bp->b_flags & B_CACHE) == 0) { bp->b_flags |= B_READ; vfs_busy_pages(bp, 0); error = nfs_doio(bp, cred, p); if (error) { brelse(bp); } while (error == NFSERR_BAD_COOKIE) { printf("got bad cookie vp %p bp %p\n", vp, bp); nfs_invaldir(vp); error = nfs_vinvalbuf(vp, 0, cred, p, 1); /* * Yuck! The directory has been modified on the * server. The only way to get the block is by * reading from the beginning to get all the * offset cookies. * * Leave the last bp intact unless there is an error. * Loop back up to the while if the error is another * NFSERR_BAD_COOKIE (double yuch!). */ for (i = 0; i <= lbn && !error; i++) { if (np->n_direofoffset && (i * NFS_DIRBLKSIZ) >= np->n_direofoffset) return (0); bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, p); if (!bp) return (EINTR); if ((bp->b_flags & B_CACHE) == 0) { bp->b_flags |= B_READ; vfs_busy_pages(bp, 0); error = nfs_doio(bp, cred, p); /* * no error + B_INVAL == directory EOF, * use the block. */ if (error == 0 && (bp->b_flags & B_INVAL)) break; } /* * An error will throw away the block and the * for loop will break out. If no error and this * is not the block we want, we throw away the * block and go for the next one via the for loop. */ if (error || i < lbn) brelse(bp); } } /* * The above while is repeated if we hit another cookie * error. If we hit an error and it wasn't a cookie error, * we give up. */ if (error) return (error); } /* * If not eof and read aheads are enabled, start one. * (You need the current block first, so that you have the * directory offset cookie of the next block.) */ if (nfs_numasync > 0 && nmp->nm_readahead > 0 && (bp->b_flags & B_INVAL) == 0 && (np->n_direofoffset == 0 || (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && !(np->n_flag & NQNFSNONCACHE) && !incore(vp, lbn + 1)) { rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, p); if (rabp) { if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { rabp->b_flags |= (B_READ | B_ASYNC); vfs_busy_pages(rabp, 0); if (nfs_asyncio(rabp, cred, p)) { rabp->b_flags |= B_INVAL|B_ERROR; vfs_unbusy_pages(rabp); brelse(rabp); } } else { brelse(rabp); } } } /* * Unlike VREG files, whos buffer size ( bp->b_bcount ) is * chopped for the EOF condition, we cannot tell how large * NFS directories are going to be until we hit EOF. So * an NFS directory buffer is *not* chopped to its EOF. Now, * it just so happens that b_resid will effectively chop it * to EOF. *BUT* this information is lost if the buffer goes * away and is reconstituted into a B_CACHE state ( due to * being VMIO ) later. So we keep track of the directory eof * in np->n_direofoffset and chop it off as an extra step * right here. */ n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); if (np->n_direofoffset && n > np->n_direofoffset - uio->uio_offset) n = np->n_direofoffset - uio->uio_offset; break; default: printf(" nfs_bioread: type %x unexpected\n",vp->v_type); break; }; if (n > 0) { error = uiomove(bp->b_data + on, (int)n, uio); } switch (vp->v_type) { case VREG: break; case VLNK: n = 0; break; case VDIR: /* * Invalidate buffer if caching is disabled, forcing a * re-read from the remote later. */ if (np->n_flag & NQNFSNONCACHE) bp->b_flags |= B_INVAL; break; default: printf(" nfs_bioread: type %x unexpected\n",vp->v_type); } brelse(bp); } while (error == 0 && uio->uio_resid > 0 && n > 0); return (error); } /* * Vnode op for write using bio */ int nfs_write(ap) struct vop_write_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { int biosize; struct uio *uio = ap->a_uio; struct proc *p = uio->uio_procp; struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct ucred *cred = ap->a_cred; int ioflag = ap->a_ioflag; struct buf *bp; struct vattr vattr; struct nfsmount *nmp = VFSTONFS(vp->v_mount); daddr_t lbn; int bcount; int n, on, error = 0, iomode, must_commit; int haverslock = 0; #ifdef DIAGNOSTIC if (uio->uio_rw != UIO_WRITE) panic("nfs_write mode"); if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) panic("nfs_write proc"); #endif if (vp->v_type != VREG) return (EIO); if (np->n_flag & NWRITEERR) { np->n_flag &= ~NWRITEERR; return (np->n_error); } if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) (void)nfs_fsinfo(nmp, vp, cred, p); /* * Synchronously flush pending buffers if we are in synchronous * mode or if we are appending. */ if (ioflag & (IO_APPEND | IO_SYNC)) { if (np->n_flag & NMODIFIED) { np->n_attrstamp = 0; error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) return (error); } } /* * If IO_APPEND then load uio_offset. We restart here if we cannot * get the append lock. */ restart: if (ioflag & IO_APPEND) { np->n_attrstamp = 0; error = VOP_GETATTR(vp, &vattr, cred, p); if (error) return (error); uio->uio_offset = np->n_size; } if (uio->uio_offset < 0) return (EINVAL); if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize) return (EFBIG); if (uio->uio_resid == 0) return (0); /* * We need to obtain the rslock if we intend to modify np->n_size * in order to guarentee the append point with multiple contending * writers, to guarentee that no other appenders modify n_size * while we are trying to obtain a truncated buffer (i.e. to avoid * accidently truncating data written by another appender due to * the race), and to ensure that the buffer is populated prior to * our extending of the file. We hold rslock through the entire * operation. * * Note that we do not synchronize the case where someone truncates * the file while we are appending to it because attempting to lock * this case may deadlock other parts of the system unexpectedly. */ if ((ioflag & IO_APPEND) || uio->uio_offset + uio->uio_resid > np->n_size) { switch(nfs_rslock(np, p)) { case ENOLCK: goto restart; /* not reached */ case EINTR: case ERESTART: return(EINTR); /* not reached */ default: break; } haverslock = 1; } /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, i don't think it matters */ if (p && uio->uio_offset + uio->uio_resid > p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { psignal(p, SIGXFSZ); if (haverslock) nfs_rsunlock(np, p); return (EFBIG); } biosize = vp->v_mount->mnt_stat.f_iosize; do { /* * Check for a valid write lease. */ if ((nmp->nm_flag & NFSMNT_NQNFS) && NQNFS_CKINVALID(vp, np, ND_WRITE)) { do { error = nqnfs_getlease(vp, ND_WRITE, cred, p); } while (error == NQNFS_EXPIRED); if (error) break; if (np->n_lrev != np->n_brev || (np->n_flag & NQNFSNONCACHE)) { error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) break; np->n_brev = np->n_lrev; } } if ((np->n_flag & NQNFSNONCACHE) && uio->uio_iovcnt == 1) { iomode = NFSV3WRITE_FILESYNC; error = nfs_writerpc(vp, uio, cred, &iomode, &must_commit); if (must_commit) nfs_clearcommit(vp->v_mount); break; } nfsstats.biocache_writes++; lbn = uio->uio_offset / biosize; on = uio->uio_offset & (biosize-1); n = min((unsigned)(biosize - on), uio->uio_resid); again: /* * Handle direct append and file extension cases, calculate * unaligned buffer size. */ if (uio->uio_offset == np->n_size && n) { /* * Get the buffer (in its pre-append state to maintain * B_CACHE if it was previously set). Resize the * nfsnode after we have locked the buffer to prevent * readers from reading garbage. */ bcount = on; bp = nfs_getcacheblk(vp, lbn, bcount, p); if (bp != NULL) { long save; np->n_size = uio->uio_offset + n; np->n_flag |= NMODIFIED; vnode_pager_setsize(vp, np->n_size); save = bp->b_flags & B_CACHE; bcount += n; allocbuf(bp, bcount); bp->b_flags |= save; } } else { /* * Obtain the locked cache block first, and then * adjust the file's size as appropriate. */ bcount = on + n; if ((off_t)lbn * biosize + bcount < np->n_size) { if ((off_t)(lbn + 1) * biosize < np->n_size) bcount = biosize; else bcount = np->n_size - (off_t)lbn * biosize; } bp = nfs_getcacheblk(vp, lbn, bcount, p); if (uio->uio_offset + n > np->n_size) { np->n_size = uio->uio_offset + n; np->n_flag |= NMODIFIED; vnode_pager_setsize(vp, np->n_size); } } if (!bp) { error = EINTR; break; } /* * Issue a READ if B_CACHE is not set. In special-append * mode, B_CACHE is based on the buffer prior to the write * op and is typically set, avoiding the read. If a read * is required in special append mode, the server will * probably send us a short-read since we extended the file * on our end, resulting in b_resid == 0 and, thusly, * B_CACHE getting set. * * We can also avoid issuing the read if the write covers * the entire buffer. We have to make sure the buffer state * is reasonable in this case since we will not be initiating * I/O. See the comments in kern/vfs_bio.c's getblk() for * more information. * * B_CACHE may also be set due to the buffer being cached * normally. */ if (on == 0 && n == bcount) { bp->b_flags |= B_CACHE; bp->b_flags &= ~(B_ERROR | B_INVAL); } if ((bp->b_flags & B_CACHE) == 0) { bp->b_flags |= B_READ; vfs_busy_pages(bp, 0); error = nfs_doio(bp, cred, p); if (error) { brelse(bp); break; } } if (!bp) { error = EINTR; break; } if (bp->b_wcred == NOCRED) { crhold(cred); bp->b_wcred = cred; } np->n_flag |= NMODIFIED; /* * If dirtyend exceeds file size, chop it down. This should * not normally occur but there is an append race where it * might occur XXX, so we log it. * * If the chopping creates a reverse-indexed or degenerate * situation with dirtyoff/end, we 0 both of them. */ if (bp->b_dirtyend > bcount) { printf("NFS append race @%lx:%d\n", (long)bp->b_blkno * DEV_BSIZE, bp->b_dirtyend - bcount); bp->b_dirtyend = bcount; } if (bp->b_dirtyoff >= bp->b_dirtyend) bp->b_dirtyoff = bp->b_dirtyend = 0; /* * If the new write will leave a contiguous dirty * area, just update the b_dirtyoff and b_dirtyend, * otherwise force a write rpc of the old dirty area. * * While it is possible to merge discontiguous writes due to * our having a B_CACHE buffer ( and thus valid read data * for the hole), we don't because it could lead to * significant cache coherency problems with multiple clients, * especially if locking is implemented later on. * * as an optimization we could theoretically maintain * a linked list of discontinuous areas, but we would still * have to commit them separately so there isn't much * advantage to it except perhaps a bit of asynchronization. */ if (bp->b_dirtyend > 0 && (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { if (VOP_BWRITE(bp->b_vp, bp) == EINTR) { error = EINTR; break; } goto again; } /* * Check for valid write lease and get one as required. * In case getblk() and/or bwrite() delayed us. */ if ((nmp->nm_flag & NFSMNT_NQNFS) && NQNFS_CKINVALID(vp, np, ND_WRITE)) { do { error = nqnfs_getlease(vp, ND_WRITE, cred, p); } while (error == NQNFS_EXPIRED); if (error) { brelse(bp); break; } if (np->n_lrev != np->n_brev || (np->n_flag & NQNFSNONCACHE)) { brelse(bp); error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) break; np->n_brev = np->n_lrev; goto again; } } error = uiomove((char *)bp->b_data + on, n, uio); /* * Since this block is being modified, it must be written * again and not just committed. Since write clustering does * not work for the stage 1 data write, only the stage 2 * commit rpc, we have to clear B_CLUSTEROK as well. */ bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); if (error) { bp->b_flags |= B_ERROR; brelse(bp); break; } /* * Only update dirtyoff/dirtyend if not a degenerate * condition. */ if (n) { if (bp->b_dirtyend > 0) { bp->b_dirtyoff = min(on, bp->b_dirtyoff); bp->b_dirtyend = max((on + n), bp->b_dirtyend); } else { bp->b_dirtyoff = on; bp->b_dirtyend = on + n; } vfs_bio_set_validclean(bp, on, n); } /* * If IO_NOWDRAIN then set B_NOWDRAIN (e.g. nfs-backed VN * filesystem). XXX also use for loopback NFS mounts. */ if (ioflag & IO_NOWDRAIN) bp->b_flags |= B_NOWDRAIN; /* * If the lease is non-cachable or IO_SYNC do bwrite(). * * IO_INVAL appears to be unused. The idea appears to be * to turn off caching in this case. Very odd. XXX */ if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { if (ioflag & IO_INVAL) bp->b_flags |= B_NOCACHE; error = VOP_BWRITE(bp->b_vp, bp); if (error) break; if (np->n_flag & NQNFSNONCACHE) { error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1); if (error) break; } } else if ((n + on) == biosize && (nmp->nm_flag & NFSMNT_NQNFS) == 0) { bp->b_flags |= B_ASYNC; (void)nfs_writebp(bp, 0, 0); } else { bdwrite(bp); } } while (uio->uio_resid > 0 && n > 0); if (haverslock) nfs_rsunlock(np, p); return (error); } /* * Get an nfs cache block. * * Allocate a new one if the block isn't currently in the cache * and return the block marked busy. If the calling process is * interrupted by a signal for an interruptible mount point, return * NULL. * * The caller must carefully deal with the possible B_INVAL state of * the buffer. nfs_doio() clears B_INVAL (and nfs_asyncio() clears it * indirectly), so synchronous reads can be issued without worrying about * the B_INVAL state. We have to be a little more careful when dealing * with writes (see comments in nfs_write()) when extending a file past * its EOF. */ static struct buf * nfs_getcacheblk(vp, bn, size, p) struct vnode *vp; daddr_t bn; int size; struct proc *p; { register struct buf *bp; struct mount *mp; struct nfsmount *nmp; mp = vp->v_mount; nmp = VFSTONFS(mp); if (nmp->nm_flag & NFSMNT_INT) { bp = getblk(vp, bn, size, PCATCH, 0); while (bp == (struct buf *)0) { if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) return ((struct buf *)0); bp = getblk(vp, bn, size, 0, 2 * hz); } } else { bp = getblk(vp, bn, size, 0, 0); } if (vp->v_type == VREG) { int biosize; biosize = mp->mnt_stat.f_iosize; bp->b_blkno = bn * (biosize / DEV_BSIZE); } return (bp); } /* * Flush and invalidate all dirty buffers. If another process is already * doing the flush, just wait for completion. */ int nfs_vinvalbuf(vp, flags, cred, p, intrflg) struct vnode *vp; int flags; struct ucred *cred; struct proc *p; int intrflg; { register struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, slpflag, slptimeo; if (vp->v_flag & VXLOCK) { return (0); } if ((nmp->nm_flag & NFSMNT_INT) == 0) intrflg = 0; if (intrflg) { slpflag = PCATCH; slptimeo = 2 * hz; } else { slpflag = 0; slptimeo = 0; } /* * First wait for any other process doing a flush to complete. */ while (np->n_flag & NFLUSHINPROG) { np->n_flag |= NFLUSHWANT; error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval", slptimeo); if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) return (EINTR); } /* * Now, flush as required. */ np->n_flag |= NFLUSHINPROG; error = vinvalbuf(vp, flags, cred, p, slpflag, 0); while (error) { if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) { np->n_flag &= ~NFLUSHINPROG; if (np->n_flag & NFLUSHWANT) { np->n_flag &= ~NFLUSHWANT; wakeup((caddr_t)&np->n_flag); } return (EINTR); } error = vinvalbuf(vp, flags, cred, p, 0, slptimeo); } np->n_flag &= ~(NMODIFIED | NFLUSHINPROG); if (np->n_flag & NFLUSHWANT) { np->n_flag &= ~NFLUSHWANT; wakeup((caddr_t)&np->n_flag); } return (0); } /* * Initiate asynchronous I/O. Return an error if no nfsiods are available. * This is mainly to avoid queueing async I/O requests when the nfsiods * are all hung on a dead server. * * Note: nfs_asyncio() does not clear (B_ERROR|B_INVAL) but when the bp * is eventually dequeued by the async daemon, nfs_doio() *will*. */ int nfs_asyncio(bp, cred, procp) register struct buf *bp; struct ucred *cred; struct proc *procp; { struct nfsmount *nmp; int i; int gotiod; int slpflag = 0; int slptimeo = 0; int error; /* * If no async daemons then return EIO to force caller to run the rpc * synchronously. */ if (nfs_numasync == 0) return (EIO); nmp = VFSTONFS(bp->b_vp->v_mount); /* * Commits are usually short and sweet so lets save some cpu and * leave the async daemons for more important rpc's (such as reads * and writes). */ if ((bp->b_flags & (B_READ|B_NEEDCOMMIT)) == B_NEEDCOMMIT && (nmp->nm_bufqiods > nfs_numasync / 2)) { return(EIO); } again: if (nmp->nm_flag & NFSMNT_INT) slpflag = PCATCH; gotiod = FALSE; /* * Find a free iod to process this request. */ for (i = 0; i < NFS_MAXASYNCDAEMON; i++) if (nfs_iodwant[i]) { /* * Found one, so wake it up and tell it which * mount to process. */ NFS_DPF(ASYNCIO, ("nfs_asyncio: waking iod %d for mount %p\n", i, nmp)); nfs_iodwant[i] = (struct proc *)0; nfs_iodmount[i] = nmp; nmp->nm_bufqiods++; wakeup((caddr_t)&nfs_iodwant[i]); gotiod = TRUE; break; } /* * If none are free, we may already have an iod working on this mount * point. If so, it will process our request. */ if (!gotiod) { if (nmp->nm_bufqiods > 0) { NFS_DPF(ASYNCIO, ("nfs_asyncio: %d iods are already processing mount %p\n", nmp->nm_bufqiods, nmp)); gotiod = TRUE; } } /* * If we have an iod which can process the request, then queue * the buffer. */ if (gotiod) { /* * Ensure that the queue never grows too large. We still want * to asynchronize so we block rather then return EIO. */ while (nmp->nm_bufqlen >= 2*nfs_numasync) { NFS_DPF(ASYNCIO, ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); nmp->nm_bufqwant = TRUE; error = tsleep(&nmp->nm_bufq, slpflag | PRIBIO, "nfsaio", slptimeo); if (error) { if (nfs_sigintr(nmp, NULL, procp)) return (EINTR); if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } } /* * We might have lost our iod while sleeping, * so check and loop if nescessary. */ if (nmp->nm_bufqiods == 0) { NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); goto again; } } if (bp->b_flags & B_READ) { if (bp->b_rcred == NOCRED && cred != NOCRED) { crhold(cred); bp->b_rcred = cred; } } else { bp->b_flags |= B_WRITEINPROG; if (bp->b_wcred == NOCRED && cred != NOCRED) { crhold(cred); bp->b_wcred = cred; } } BUF_KERNPROC(bp); TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); nmp->nm_bufqlen++; return (0); } /* * All the iods are busy on other mounts, so return EIO to * force the caller to process the i/o synchronously. */ NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); return (EIO); } /* * Do an I/O operation to/from a cache block. This may be called * synchronously or from an nfsiod. */ int nfs_doio(bp, cr, p) struct buf *bp; struct ucred *cr; struct proc *p; { struct uio *uiop; struct vnode *vp; struct nfsnode *np; struct nfsmount *nmp; int error = 0, iomode, must_commit = 0; struct uio uio; struct iovec io; vp = bp->b_vp; np = VTONFS(vp); nmp = VFSTONFS(vp->v_mount); uiop = &uio; uiop->uio_iov = &io; uiop->uio_iovcnt = 1; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_procp = p; /* * clear B_ERROR and B_INVAL state prior to initiating the I/O. We * do this here so we do not have to do it in all the code that * calls us. */ bp->b_flags &= ~(B_ERROR | B_INVAL); KASSERT(!(bp->b_flags & B_DONE), ("nfs_doio: bp %p already marked done", bp)); /* * Historically, paging was done with physio, but no more. */ if (bp->b_flags & B_PHYS) { /* * ...though reading /dev/drum still gets us here. */ io.iov_len = uiop->uio_resid = bp->b_bcount; /* mapping was done by vmapbuf() */ io.iov_base = bp->b_data; uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; if (bp->b_flags & B_READ) { uiop->uio_rw = UIO_READ; nfsstats.read_physios++; error = nfs_readrpc(vp, uiop, cr); } else { int com; iomode = NFSV3WRITE_DATASYNC; uiop->uio_rw = UIO_WRITE; nfsstats.write_physios++; error = nfs_writerpc(vp, uiop, cr, &iomode, &com); } if (error) { bp->b_flags |= B_ERROR; bp->b_error = error; } } else if (bp->b_flags & B_READ) { io.iov_len = uiop->uio_resid = bp->b_bcount; io.iov_base = bp->b_data; uiop->uio_rw = UIO_READ; switch (vp->v_type) { case VREG: uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; nfsstats.read_bios++; error = nfs_readrpc(vp, uiop, cr); if (!error) { if (uiop->uio_resid) { /* * If we had a short read with no error, we must have * hit a file hole. We should zero-fill the remainder. * This can also occur if the server hits the file EOF. * * Holes used to be able to occur due to pending * writes, but that is not possible any longer. */ int nread = bp->b_bcount - uiop->uio_resid; int left = uiop->uio_resid; if (left > 0) bzero((char *)bp->b_data + nread, left); uiop->uio_resid = 0; } } if (p && (vp->v_flag & VTEXT) && (((nmp->nm_flag & NFSMNT_NQNFS) && NQNFS_CKINVALID(vp, np, ND_READ) && np->n_lrev != np->n_brev) || (!(nmp->nm_flag & NFSMNT_NQNFS) && np->n_mtime != np->n_vattr.va_mtime.tv_sec))) { uprintf("Process killed due to text file modification\n"); psignal(p, SIGKILL); PHOLD(p); } break; case VLNK: uiop->uio_offset = (off_t)0; nfsstats.readlink_bios++; error = nfs_readlinkrpc(vp, uiop, cr); break; case VDIR: nfsstats.readdir_bios++; uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; if (nmp->nm_flag & NFSMNT_RDIRPLUS) { error = nfs_readdirplusrpc(vp, uiop, cr); if (error == NFSERR_NOTSUPP) nmp->nm_flag &= ~NFSMNT_RDIRPLUS; } if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) error = nfs_readdirrpc(vp, uiop, cr); /* * end-of-directory sets B_INVAL but does not generate an * error. */ if (error == 0 && uiop->uio_resid == bp->b_bcount) bp->b_flags |= B_INVAL; break; default: printf("nfs_doio: type %x unexpected\n",vp->v_type); break; }; if (error) { bp->b_flags |= B_ERROR; bp->b_error = error; } } else { /* * If we only need to commit, try to commit */ if (bp->b_flags & B_NEEDCOMMIT) { int retv; off_t off; off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; bp->b_flags |= B_WRITEINPROG; retv = nfs_commit( bp->b_vp, off, bp->b_dirtyend-bp->b_dirtyoff, bp->b_wcred, p); bp->b_flags &= ~B_WRITEINPROG; if (retv == 0) { bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); bp->b_resid = 0; biodone(bp); return (0); } if (retv == NFSERR_STALEWRITEVERF) { nfs_clearcommit(bp->b_vp->v_mount); } } /* * Setup for actual write */ if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size) bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE; if (bp->b_dirtyend > bp->b_dirtyoff) { io.iov_len = uiop->uio_resid = bp->b_dirtyend - bp->b_dirtyoff; uiop->uio_offset = (off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyoff; io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; uiop->uio_rw = UIO_WRITE; nfsstats.write_bios++; if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC) iomode = NFSV3WRITE_UNSTABLE; else iomode = NFSV3WRITE_FILESYNC; bp->b_flags |= B_WRITEINPROG; error = nfs_writerpc(vp, uiop, cr, &iomode, &must_commit); /* * When setting B_NEEDCOMMIT also set B_CLUSTEROK to try * to cluster the buffers needing commit. This will allow * the system to submit a single commit rpc for the whole * cluster. We can do this even if the buffer is not 100% * dirty (relative to the NFS blocksize), so we optimize the * append-to-file-case. * * (when clearing B_NEEDCOMMIT, B_CLUSTEROK must also be * cleared because write clustering only works for commit * rpc's, not for the data portion of the write). */ if (!error && iomode == NFSV3WRITE_UNSTABLE) { bp->b_flags |= B_NEEDCOMMIT; if (bp->b_dirtyoff == 0 && bp->b_dirtyend == bp->b_bcount) bp->b_flags |= B_CLUSTEROK; } else { bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); } bp->b_flags &= ~B_WRITEINPROG; /* * For an interrupted write, the buffer is still valid * and the write hasn't been pushed to the server yet, * so we can't set B_ERROR and report the interruption * by setting B_EINTR. For the B_ASYNC case, B_EINTR * is not relevant, so the rpc attempt is essentially * a noop. For the case of a V3 write rpc not being * committed to stable storage, the block is still * dirty and requires either a commit rpc or another * write rpc with iomode == NFSV3WRITE_FILESYNC before * the block is reused. This is indicated by setting * the B_DELWRI and B_NEEDCOMMIT flags. * * If the buffer is marked B_PAGING, it does not reside on * the vp's paging queues so we cannot call bdirty(). The * bp in this case is not an NFS cache block so we should * be safe. XXX */ if (error == EINTR || (!error && (bp->b_flags & B_NEEDCOMMIT))) { int s; s = splbio(); bp->b_flags &= ~(B_INVAL|B_NOCACHE); if ((bp->b_flags & B_PAGING) == 0) { bdirty(bp); bp->b_flags &= ~B_DONE; } if (error && (bp->b_flags & B_ASYNC) == 0) bp->b_flags |= B_EINTR; splx(s); } else { if (error) { bp->b_flags |= B_ERROR; bp->b_error = np->n_error = error; np->n_flag |= NWRITEERR; } bp->b_dirtyoff = bp->b_dirtyend = 0; } } else { bp->b_resid = 0; biodone(bp); return (0); } } bp->b_resid = uiop->uio_resid; if (must_commit) nfs_clearcommit(vp->v_mount); biodone(bp); return (error); } /* * Used to aid in handling ftruncate() operations on the NFS client side. * Truncation creates a number of special problems for NFS. We have to * throw away VM pages and buffer cache buffers that are beyond EOF, and * we have to properly handle VM pages or (potentially dirty) buffers * that straddle the truncation point. */ int nfs_meta_setsize(struct vnode *vp, struct ucred *cred, struct proc *p, u_quad_t nsize) { struct nfsnode *np = VTONFS(vp); u_quad_t tsize = np->n_size; int biosize = vp->v_mount->mnt_stat.f_iosize; int error = 0; np->n_size = nsize; if (np->n_size < tsize) { struct buf *bp; daddr_t lbn; int bufsize; /* * vtruncbuf() doesn't get the buffer overlapping the * truncation point. We may have a B_DELWRI and/or B_CACHE * buffer that now needs to be truncated. */ error = vtruncbuf(vp, cred, p, nsize, biosize); lbn = nsize / biosize; bufsize = nsize & (biosize - 1); bp = nfs_getcacheblk(vp, lbn, bufsize, p); if (bp->b_dirtyoff > bp->b_bcount) bp->b_dirtyoff = bp->b_bcount; if (bp->b_dirtyend > bp->b_bcount) bp->b_dirtyend = bp->b_bcount; bp->b_flags |= B_RELBUF; /* don't leave garbage around */ brelse(bp); } else { vnode_pager_setsize(vp, nsize); } return(error); } Index: stable/4/sys/nfs/nfs_subs.c =================================================================== --- stable/4/sys/nfs/nfs_subs.c (revision 131097) +++ stable/4/sys/nfs/nfs_subs.c (revision 131098) @@ -1,2238 +1,2245 @@ /* * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95 * $FreeBSD$ */ /* * These functions support the macros and help fiddle mbuf chains for * the nfs op functions. They do things like create the rpc header and * copy data between mbuf chains and uio lists. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Data items converted to xdr at startup, since they are constant * This is kinda hokey, but may save a little time doing byte swaps */ u_int32_t nfs_xdrneg1; u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr, rpc_mismatch, rpc_auth_unix, rpc_msgaccepted, rpc_auth_kerb; u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false; /* And other global data */ static u_int32_t nfs_xid = 0; static enum vtype nv2tov_type[8]= { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON }; enum vtype nv3tov_type[8]= { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO }; int nfs_ticks; int nfs_pbuf_freecnt = -1; /* start out unlimited */ struct nfs_reqq nfs_reqq; struct nfssvc_sockhead nfssvc_sockhead; int nfssvc_sockhead_flag; struct nfsd_head nfsd_head; int nfsd_head_flag; struct nfs_bufq nfs_bufq; struct nqtimerhead nqtimerhead; struct nqfhhashhead *nqfhhashtbl; u_long nqfhhash; static void (*nfs_prev_lease_updatetime) __P((int)); static int nfs_prev_nfssvc_sy_narg; static sy_call_t *nfs_prev_nfssvc_sy_call; #ifndef NFS_NOSERVER static vop_t *nfs_prev_vop_lease_check; /* * Mapping of old NFS Version 2 RPC numbers to generic numbers. */ int nfsv3_procid[NFS_NPROCS] = { NFSPROC_NULL, NFSPROC_GETATTR, NFSPROC_SETATTR, NFSPROC_NOOP, NFSPROC_LOOKUP, NFSPROC_READLINK, NFSPROC_READ, NFSPROC_NOOP, NFSPROC_WRITE, NFSPROC_CREATE, NFSPROC_REMOVE, NFSPROC_RENAME, NFSPROC_LINK, NFSPROC_SYMLINK, NFSPROC_MKDIR, NFSPROC_RMDIR, NFSPROC_READDIR, NFSPROC_FSSTAT, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP }; #endif /* NFS_NOSERVER */ /* * and the reverse mapping from generic to Version 2 procedure numbers */ int nfsv2_procid[NFS_NPROCS] = { NFSV2PROC_NULL, NFSV2PROC_GETATTR, NFSV2PROC_SETATTR, NFSV2PROC_LOOKUP, NFSV2PROC_NOOP, NFSV2PROC_READLINK, NFSV2PROC_READ, NFSV2PROC_WRITE, NFSV2PROC_CREATE, NFSV2PROC_MKDIR, NFSV2PROC_SYMLINK, NFSV2PROC_CREATE, NFSV2PROC_REMOVE, NFSV2PROC_RMDIR, NFSV2PROC_RENAME, NFSV2PROC_LINK, NFSV2PROC_READDIR, NFSV2PROC_NOOP, NFSV2PROC_STATFS, NFSV2PROC_NOOP, NFSV2PROC_NOOP, NFSV2PROC_NOOP, NFSV2PROC_NOOP, NFSV2PROC_NOOP, NFSV2PROC_NOOP, NFSV2PROC_NOOP, }; #ifndef NFS_NOSERVER /* * Maps errno values to nfs error numbers. * Use NFSERR_IO as the catch all for ones not specifically defined in * RFC 1094. */ static u_char nfsrv_v2errmap[ELAST] = { NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO /* << Last is 86 */ }; /* * Maps errno values to nfs error numbers. * Although it is not obvious whether or not NFS clients really care if * a returned error value is in the specified list for the procedure, the * safest thing to do is filter them appropriately. For Version 2, the * X/Open XNFS document is the only specification that defines error values * for each RPC (The RFC simply lists all possible error values for all RPCs), * so I have decided to not do this for Version 2. * The first entry is the default error return and the rest are the valid * errors for that RPC in increasing numeric order. */ static short nfsv3err_null[] = { 0, 0, }; static short nfsv3err_getattr[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_setattr[] = { NFSERR_IO, NFSERR_PERM, NFSERR_IO, NFSERR_ACCES, NFSERR_INVAL, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOT_SYNC, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_lookup[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_IO, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_NAMETOL, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_access[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_readlink[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_INVAL, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_read[] = { NFSERR_IO, NFSERR_IO, NFSERR_NXIO, NFSERR_ACCES, NFSERR_INVAL, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_write[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_INVAL, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_create[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NOTDIR, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_NAMETOL, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_mkdir[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NOTDIR, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_NAMETOL, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_symlink[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NOTDIR, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_NAMETOL, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_mknod[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NOTDIR, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_NAMETOL, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_BADTYPE, 0, }; static short nfsv3err_remove[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_IO, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_ROFS, NFSERR_NAMETOL, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_rmdir[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NOTDIR, NFSERR_INVAL, NFSERR_ROFS, NFSERR_NAMETOL, NFSERR_NOTEMPTY, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_rename[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_XDEV, NFSERR_NOTDIR, NFSERR_ISDIR, NFSERR_INVAL, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_MLINK, NFSERR_NAMETOL, NFSERR_NOTEMPTY, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_link[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_XDEV, NFSERR_NOTDIR, NFSERR_INVAL, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_MLINK, NFSERR_NAMETOL, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_readdir[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_BAD_COOKIE, NFSERR_TOOSMALL, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_readdirplus[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_BAD_COOKIE, NFSERR_NOTSUPP, NFSERR_TOOSMALL, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_fsstat[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_fsinfo[] = { NFSERR_STALE, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_pathconf[] = { NFSERR_STALE, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short nfsv3err_commit[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, 0, }; static short *nfsrv_v3errmap[] = { nfsv3err_null, nfsv3err_getattr, nfsv3err_setattr, nfsv3err_lookup, nfsv3err_access, nfsv3err_readlink, nfsv3err_read, nfsv3err_write, nfsv3err_create, nfsv3err_mkdir, nfsv3err_symlink, nfsv3err_mknod, nfsv3err_remove, nfsv3err_rmdir, nfsv3err_rename, nfsv3err_link, nfsv3err_readdir, nfsv3err_readdirplus, nfsv3err_fsstat, nfsv3err_fsinfo, nfsv3err_pathconf, nfsv3err_commit, }; #endif /* NFS_NOSERVER */ extern struct nfsrtt nfsrtt; extern time_t nqnfsstarttime; extern int nqsrv_clockskew; extern int nqsrv_writeslack; extern int nqsrv_maxlease; extern struct nfsstats nfsstats; extern int nqnfs_piggy[NFS_NPROCS]; extern nfstype nfsv2_type[9]; extern nfstype nfsv3_type[9]; extern struct nfsnodehashhead *nfsnodehashtbl; extern u_long nfsnodehash; struct nfssvc_args; extern int nfssvc(struct proc *, struct nfssvc_args *, int *); LIST_HEAD(nfsnodehashhead, nfsnode); int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); u_quad_t nfs_curusec() { struct timeval tv; getmicrotime(&tv); return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec); } /* * Create the header for an rpc request packet * The hsiz is the size of the rest of the nfs request header. * (just used to decide if a cluster is a good idea) */ struct mbuf * nfsm_reqh(vp, procid, hsiz, bposp) struct vnode *vp; u_long procid; int hsiz; caddr_t *bposp; { register struct mbuf *mb; register u_int32_t *tl; register caddr_t bpos; struct mbuf *mb2; struct nfsmount *nmp; int nqflag; MGET(mb, M_WAIT, MT_DATA); if (hsiz >= MINCLSIZE) MCLGET(mb, M_WAIT); mb->m_len = 0; bpos = mtod(mb, caddr_t); /* * For NQNFS, add lease request. */ if (vp) { nmp = VFSTONFS(vp->v_mount); if (nmp->nm_flag & NFSMNT_NQNFS) { nqflag = NQNFS_NEEDLEASE(vp, procid); if (nqflag) { nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED); *tl++ = txdr_unsigned(nqflag); *tl = txdr_unsigned(nmp->nm_leaseterm); } else { nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); *tl = 0; } } } /* Finally, return values */ *bposp = bpos; return (mb); } /* * Build the RPC header and fill in the authorization info. * The authorization string argument is only used when the credentials * come from outside of the kernel. * Returns the head of the mbuf list. */ struct mbuf * nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, verf_str, mrest, mrest_len, mbp, xidp) register struct ucred *cr; int nmflag; int procid; int auth_type; int auth_len; char *auth_str; int verf_len; char *verf_str; struct mbuf *mrest; int mrest_len; struct mbuf **mbp; u_int32_t *xidp; { register struct mbuf *mb; register u_int32_t *tl; register caddr_t bpos; register int i; struct mbuf *mreq, *mb2; int siz, grpsiz, authsiz; authsiz = nfsm_rndup(auth_len); MGETHDR(mb, M_WAIT, MT_DATA); if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { MCLGET(mb, M_WAIT); } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); } else { MH_ALIGN(mb, 8 * NFSX_UNSIGNED); } mb->m_len = 0; mreq = mb; bpos = mtod(mb, caddr_t); /* * First the RPC header. */ nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); /* Get a pretty random xid to start with */ if (!nfs_xid) nfs_xid = random(); /* * Skip zero xid if it should ever happen. */ if (++nfs_xid == 0) nfs_xid++; *tl++ = *xidp = txdr_unsigned(nfs_xid); *tl++ = rpc_call; *tl++ = rpc_vers; if (nmflag & NFSMNT_NQNFS) { *tl++ = txdr_unsigned(NQNFS_PROG); *tl++ = txdr_unsigned(NQNFS_VER3); } else { *tl++ = txdr_unsigned(NFS_PROG); if (nmflag & NFSMNT_NFSV3) *tl++ = txdr_unsigned(NFS_VER3); else *tl++ = txdr_unsigned(NFS_VER2); } if (nmflag & NFSMNT_NFSV3) *tl++ = txdr_unsigned(procid); else *tl++ = txdr_unsigned(nfsv2_procid[procid]); /* * And then the authorization cred. */ *tl++ = txdr_unsigned(auth_type); *tl = txdr_unsigned(authsiz); switch (auth_type) { case RPCAUTH_UNIX: nfsm_build(tl, u_int32_t *, auth_len); *tl++ = 0; /* stamp ?? */ *tl++ = 0; /* NULL hostname */ *tl++ = txdr_unsigned(cr->cr_uid); *tl++ = txdr_unsigned(cr->cr_groups[0]); grpsiz = (auth_len >> 2) - 5; *tl++ = txdr_unsigned(grpsiz); for (i = 1; i <= grpsiz; i++) *tl++ = txdr_unsigned(cr->cr_groups[i]); break; case RPCAUTH_KERB4: siz = auth_len; while (siz > 0) { if (M_TRAILINGSPACE(mb) == 0) { MGET(mb2, M_WAIT, MT_DATA); if (siz >= MINCLSIZE) MCLGET(mb2, M_WAIT); mb->m_next = mb2; mb = mb2; mb->m_len = 0; bpos = mtod(mb, caddr_t); } i = min(siz, M_TRAILINGSPACE(mb)); bcopy(auth_str, bpos, i); mb->m_len += i; auth_str += i; bpos += i; siz -= i; } if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { for (i = 0; i < siz; i++) *bpos++ = '\0'; mb->m_len += siz; } break; }; /* * And the verifier... */ nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (verf_str) { *tl++ = txdr_unsigned(RPCAUTH_KERB4); *tl = txdr_unsigned(verf_len); siz = verf_len; while (siz > 0) { if (M_TRAILINGSPACE(mb) == 0) { MGET(mb2, M_WAIT, MT_DATA); if (siz >= MINCLSIZE) MCLGET(mb2, M_WAIT); mb->m_next = mb2; mb = mb2; mb->m_len = 0; bpos = mtod(mb, caddr_t); } i = min(siz, M_TRAILINGSPACE(mb)); bcopy(verf_str, bpos, i); mb->m_len += i; verf_str += i; bpos += i; siz -= i; } if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { for (i = 0; i < siz; i++) *bpos++ = '\0'; mb->m_len += siz; } } else { *tl++ = txdr_unsigned(RPCAUTH_NULL); *tl = 0; } mb->m_next = mrest; mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; mreq->m_pkthdr.rcvif = (struct ifnet *)0; *mbp = mb; return (mreq); } /* * copies mbuf chain to the uio scatter/gather list */ int nfsm_mbuftouio(mrep, uiop, siz, dpos) struct mbuf **mrep; register struct uio *uiop; int siz; caddr_t *dpos; { register char *mbufcp, *uiocp; register int xfer, left, len; register struct mbuf *mp; long uiosiz, rem; int error = 0; mp = *mrep; mbufcp = *dpos; len = mtod(mp, caddr_t)+mp->m_len-mbufcp; rem = nfsm_rndup(siz)-siz; while (siz > 0) { if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) return (EFBIG); left = uiop->uio_iov->iov_len; uiocp = uiop->uio_iov->iov_base; if (left > siz) left = siz; uiosiz = left; while (left > 0) { while (len == 0) { mp = mp->m_next; if (mp == NULL) return (EBADRPC); mbufcp = mtod(mp, caddr_t); len = mp->m_len; } xfer = (left > len) ? len : left; #ifdef notdef /* Not Yet.. */ if (uiop->uio_iov->iov_op != NULL) (*(uiop->uio_iov->iov_op)) (mbufcp, uiocp, xfer); else #endif if (uiop->uio_segflg == UIO_SYSSPACE) bcopy(mbufcp, uiocp, xfer); else copyout(mbufcp, uiocp, xfer); left -= xfer; len -= xfer; mbufcp += xfer; uiocp += xfer; uiop->uio_offset += xfer; uiop->uio_resid -= xfer; } if (uiop->uio_iov->iov_len <= siz) { uiop->uio_iovcnt--; uiop->uio_iov++; } else { uiop->uio_iov->iov_base += uiosiz; uiop->uio_iov->iov_len -= uiosiz; } siz -= uiosiz; } *dpos = mbufcp; *mrep = mp; if (rem > 0) { if (len < rem) error = nfs_adv(mrep, dpos, rem, len); else *dpos += rem; } return (error); } /* * copies a uio scatter/gather list to an mbuf chain. * NOTE: can ony handle iovcnt == 1 */ int nfsm_uiotombuf(uiop, mq, siz, bpos) register struct uio *uiop; struct mbuf **mq; int siz; caddr_t *bpos; { register char *uiocp; register struct mbuf *mp, *mp2; register int xfer, left, mlen; int uiosiz, clflg, rem; char *cp; #ifdef DIAGNOSTIC if (uiop->uio_iovcnt != 1) panic("nfsm_uiotombuf: iovcnt != 1"); #endif if (siz > MLEN) /* or should it >= MCLBYTES ?? */ clflg = 1; else clflg = 0; rem = nfsm_rndup(siz)-siz; mp = mp2 = *mq; while (siz > 0) { left = uiop->uio_iov->iov_len; uiocp = uiop->uio_iov->iov_base; if (left > siz) left = siz; uiosiz = left; while (left > 0) { mlen = M_TRAILINGSPACE(mp); if (mlen == 0) { MGET(mp, M_WAIT, MT_DATA); if (clflg) MCLGET(mp, M_WAIT); mp->m_len = 0; mp2->m_next = mp; mp2 = mp; mlen = M_TRAILINGSPACE(mp); } xfer = (left > mlen) ? mlen : left; #ifdef notdef /* Not Yet.. */ if (uiop->uio_iov->iov_op != NULL) (*(uiop->uio_iov->iov_op)) (uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); else #endif if (uiop->uio_segflg == UIO_SYSSPACE) bcopy(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); else copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); mp->m_len += xfer; left -= xfer; uiocp += xfer; uiop->uio_offset += xfer; uiop->uio_resid -= xfer; } uiop->uio_iov->iov_base += uiosiz; uiop->uio_iov->iov_len -= uiosiz; siz -= uiosiz; } if (rem > 0) { if (rem > M_TRAILINGSPACE(mp)) { MGET(mp, M_WAIT, MT_DATA); mp->m_len = 0; mp2->m_next = mp; } cp = mtod(mp, caddr_t)+mp->m_len; for (left = 0; left < rem; left++) *cp++ = '\0'; mp->m_len += rem; *bpos = cp; } else *bpos = mtod(mp, caddr_t)+mp->m_len; *mq = mp; return (0); } /* * Help break down an mbuf chain by setting the first siz bytes contiguous * pointed to by returned val. * This is used by the macros nfsm_dissect and nfsm_dissecton for tough * cases. (The macros use the vars. dpos and dpos2) */ int nfsm_disct(mdp, dposp, siz, left, cp2) struct mbuf **mdp; caddr_t *dposp; int siz; int left; caddr_t *cp2; { register struct mbuf *mp, *mp2; register int siz2, xfer; register caddr_t p; mp = *mdp; while (left == 0) { *mdp = mp = mp->m_next; if (mp == NULL) return (EBADRPC); left = mp->m_len; *dposp = mtod(mp, caddr_t); } if (left >= siz) { *cp2 = *dposp; *dposp += siz; } else if (mp->m_next == NULL) { return (EBADRPC); } else if (siz > MHLEN) { panic("nfs S too big"); } else { MGET(mp2, M_WAIT, MT_DATA); mp2->m_next = mp->m_next; mp->m_next = mp2; mp->m_len -= left; mp = mp2; *cp2 = p = mtod(mp, caddr_t); bcopy(*dposp, p, left); /* Copy what was left */ siz2 = siz-left; p += left; mp2 = mp->m_next; /* Loop around copying up the siz2 bytes */ while (siz2 > 0) { if (mp2 == NULL) return (EBADRPC); xfer = (siz2 > mp2->m_len) ? mp2->m_len : siz2; if (xfer > 0) { bcopy(mtod(mp2, caddr_t), p, xfer); NFSMADV(mp2, xfer); mp2->m_len -= xfer; p += xfer; siz2 -= xfer; } if (siz2 > 0) mp2 = mp2->m_next; } mp->m_len = siz; *mdp = mp2; *dposp = mtod(mp2, caddr_t); } return (0); } /* * Advance the position in the mbuf chain. */ int nfs_adv(mdp, dposp, offs, left) struct mbuf **mdp; caddr_t *dposp; int offs; int left; { register struct mbuf *m; register int s; m = *mdp; s = left; while (s < offs) { offs -= s; m = m->m_next; if (m == NULL) return (EBADRPC); s = m->m_len; } *mdp = m; *dposp = mtod(m, caddr_t)+offs; return (0); } /* * Copy a string into mbufs for the hard cases... */ int nfsm_strtmbuf(mb, bpos, cp, siz) struct mbuf **mb; char **bpos; const char *cp; long siz; { register struct mbuf *m1 = NULL, *m2; long left, xfer, len, tlen; u_int32_t *tl; int putsize; putsize = 1; m2 = *mb; left = M_TRAILINGSPACE(m2); if (left > 0) { tl = ((u_int32_t *)(*bpos)); *tl++ = txdr_unsigned(siz); putsize = 0; left -= NFSX_UNSIGNED; m2->m_len += NFSX_UNSIGNED; if (left > 0) { bcopy(cp, (caddr_t) tl, left); siz -= left; cp += left; m2->m_len += left; left = 0; } } /* Loop around adding mbufs */ while (siz > 0) { MGET(m1, M_WAIT, MT_DATA); if (siz > MLEN) MCLGET(m1, M_WAIT); m1->m_len = NFSMSIZ(m1); m2->m_next = m1; m2 = m1; tl = mtod(m1, u_int32_t *); tlen = 0; if (putsize) { *tl++ = txdr_unsigned(siz); m1->m_len -= NFSX_UNSIGNED; tlen = NFSX_UNSIGNED; putsize = 0; } if (siz < m1->m_len) { len = nfsm_rndup(siz); xfer = siz; if (xfer < len) *(tl+(xfer>>2)) = 0; } else { xfer = len = m1->m_len; } bcopy(cp, (caddr_t) tl, xfer); m1->m_len = len+tlen; siz -= xfer; cp += xfer; } *mb = m1; *bpos = mtod(m1, caddr_t)+m1->m_len; return (0); } /* * Called once to initialize data structures... */ int nfs_init(vfsp) struct vfsconf *vfsp; { register int i; nfsmount_zone = zinit("NFSMOUNT", sizeof(struct nfsmount), 0, 0, 1); nfs_mount_type = vfsp->vfc_typenum; nfsrtt.pos = 0; rpc_vers = txdr_unsigned(RPC_VER2); rpc_call = txdr_unsigned(RPC_CALL); rpc_reply = txdr_unsigned(RPC_REPLY); rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); rpc_mismatch = txdr_unsigned(RPC_MISMATCH); rpc_autherr = txdr_unsigned(RPC_AUTHERR); rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); nfs_prog = txdr_unsigned(NFS_PROG); nqnfs_prog = txdr_unsigned(NQNFS_PROG); nfs_true = txdr_unsigned(TRUE); nfs_false = txdr_unsigned(FALSE); nfs_xdrneg1 = txdr_unsigned(-1); nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; if (nfs_ticks < 1) nfs_ticks = 1; /* Ensure async daemons disabled */ for (i = 0; i < NFS_MAXASYNCDAEMON; i++) { nfs_iodwant[i] = (struct proc *)0; nfs_iodmount[i] = (struct nfsmount *)0; } nfs_nhinit(); /* Init the nfsnode table */ #ifndef NFS_NOSERVER nfsrv_init(0); /* Init server data structures */ nfsrv_initcache(); /* Init the server request cache */ #endif /* * Initialize the nqnfs server stuff. */ if (nqnfsstarttime == 0) { nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease + nqsrv_clockskew + nqsrv_writeslack; NQLOADNOVRAM(nqnfsstarttime); CIRCLEQ_INIT(&nqtimerhead); nqfhhashtbl = hashinit(NQLCHSZ, M_NQLEASE, &nqfhhash); } /* * Initialize reply list and start timer */ TAILQ_INIT(&nfs_reqq); nfs_timer(0); /* * Set up lease_check and lease_updatetime so that other parts * of the system can call us, if we are loadable. */ #ifndef NFS_NOSERVER nfs_prev_vop_lease_check = default_vnodeop_p[VOFFSET(vop_lease)]; default_vnodeop_p[VOFFSET(vop_lease)] = (vop_t *)nqnfs_vop_lease_check; #endif nfs_prev_lease_updatetime = lease_updatetime; lease_updatetime = nfs_lease_updatetime; nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg; sysent[SYS_nfssvc].sy_narg = 2; nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call; sysent[SYS_nfssvc].sy_call = (sy_call_t *)nfssvc; nfs_pbuf_freecnt = nswbuf / 2 + 1; return (0); } int nfs_uninit(vfsp) struct vfsconf *vfsp; { untimeout(nfs_timer, (void *)NULL, nfs_timer_handle); nfs_mount_type = -1; #ifndef NFS_NOSERVER default_vnodeop_p[VOFFSET(vop_lease)] = nfs_prev_vop_lease_check; #endif lease_updatetime = nfs_prev_lease_updatetime; sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg; sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call; return (0); } /* * Attribute cache routines. * nfs_loadattrcache() - loads or updates the cache contents from attributes * that are on the mbuf list * nfs_getattrcache() - returns valid attributes if found in cache, returns * error otherwise */ /* * Load the attribute cache (that lives in the nfsnode entry) with * the values on the mbuf list and * Iff vap not NULL * copy the attributes to *vaper */ int nfs_loadattrcache(vpp, mdp, dposp, vaper, dontshrink) struct vnode **vpp; struct mbuf **mdp; caddr_t *dposp; struct vattr *vaper; int dontshrink; { register struct vnode *vp = *vpp; register struct vattr *vap; register struct nfs_fattr *fp; register struct nfsnode *np; register int32_t t1; caddr_t cp2; int error = 0, rdev; struct mbuf *md; enum vtype vtyp; u_short vmode; struct timespec mtime; int v3 = NFS_ISV3(vp); md = *mdp; t1 = (mtod(md, caddr_t) + md->m_len) - *dposp; if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0) return (error); fp = (struct nfs_fattr *)cp2; if (v3) { vtyp = nfsv3tov_type(fp->fa_type); vmode = fxdr_unsigned(u_short, fp->fa_mode); rdev = makeudev(fxdr_unsigned(int, fp->fa3_rdev.specdata1), fxdr_unsigned(int, fp->fa3_rdev.specdata2)); fxdr_nfsv3time(&fp->fa3_mtime, &mtime); } else { vtyp = nfsv2tov_type(fp->fa_type); vmode = fxdr_unsigned(u_short, fp->fa_mode); /* * XXX * * The duplicate information returned in fa_type and fa_mode * is an ambiguity in the NFS version 2 protocol. * * VREG should be taken literally as a regular file. If a * server intents to return some type information differently * in the upper bits of the mode field (e.g. for sockets, or * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we * leave the examination of the mode bits even in the VREG * case to avoid breakage for bogus servers, but we make sure * that there are actually type bits set in the upper part of * fa_mode (and failing that, trust the va_type field). * * NFSv3 cleared the issue, and requires fa_mode to not * contain any type information (while also introduing sockets * and FIFOs for fa_type). */ if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0)) vtyp = IFTOVT(vmode); rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); fxdr_nfsv2time(&fp->fa2_mtime, &mtime); /* * Really ugly NFSv2 kludge. */ if (vtyp == VCHR && rdev == 0xffffffff) vtyp = VFIFO; } /* * If v_type == VNON it is a new node, so fill in the v_type, * n_mtime fields. Check to see if it represents a special * device, and if so, check for a possible alias. Once the * correct vnode has been obtained, fill in the rest of the * information. */ np = VTONFS(vp); if (vp->v_type != vtyp) { vp->v_type = vtyp; if (vp->v_type == VFIFO) { vp->v_op = fifo_nfsv2nodeop_p; } if (vp->v_type == VCHR || vp->v_type == VBLK) { vp->v_op = spec_nfsv2nodeop_p; addaliasu(vp, rdev); } np->n_mtime = mtime.tv_sec; } vap = &np->n_vattr; vap->va_type = vtyp; vap->va_mode = (vmode & 07777); vap->va_rdev = rdev; vap->va_mtime = mtime; vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; if (v3) { vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid); vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid); vap->va_size = fxdr_hyper(&fp->fa3_size); vap->va_blocksize = NFS_FABLKSIZE; vap->va_bytes = fxdr_hyper(&fp->fa3_used); vap->va_fileid = fxdr_unsigned(int32_t, fp->fa3_fileid.nfsuquad[1]); fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime); vap->va_flags = 0; vap->va_filerev = 0; } else { vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid); vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid); vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize); vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks) * NFS_FABLKSIZE; vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); vap->va_flags = 0; vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t, fp->fa2_ctime.nfsv2_sec); vap->va_ctime.tv_nsec = 0; vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); vap->va_filerev = 0; } np->n_attrstamp = time_second; if (vap->va_size != np->n_size) { if (vap->va_type == VREG) { if (dontshrink && vap->va_size < np->n_size) { /* * We've been told not to shrink the file; * zero np->n_attrstamp to indicate that * the attributes are stale. */ vap->va_size = np->n_size; np->n_attrstamp = 0; } else if (np->n_flag & NMODIFIED) { - if (vap->va_size < np->n_size) + /* + * We've modified the file: Use the larger + * of our size, and the server's size. + */ + if (vap->va_size < np->n_size) { vap->va_size = np->n_size; - else + } else { np->n_size = vap->va_size; + np->n_flag |= NSIZECHANGED; + } } else { np->n_size = vap->va_size; + np->n_flag |= NSIZECHANGED; } vnode_pager_setsize(vp, np->n_size); } else { np->n_size = vap->va_size; } } if (vaper != NULL) { bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); if (np->n_flag & NCHG) { if (np->n_flag & NACC) vaper->va_atime = np->n_atim; if (np->n_flag & NUPD) vaper->va_mtime = np->n_mtim; } } return (0); } #ifdef NFS_ACDEBUG #include SYSCTL_DECL(_vfs_nfs); static int nfs_acdebug; SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, ""); #endif /* * Check the time stamp * If the cache is valid, copy contents to *vap and return 0 * otherwise return an error */ int nfs_getattrcache(vp, vaper) register struct vnode *vp; struct vattr *vaper; { register struct nfsnode *np; register struct vattr *vap; struct nfsmount *nmp; int timeo; np = VTONFS(vp); vap = &np->n_vattr; nmp = VFSTONFS(vp->v_mount); /* XXX n_mtime doesn't seem to be updated on a miss-and-reload */ timeo = (time_second - np->n_mtime) / 10; #ifdef NFS_ACDEBUG if (nfs_acdebug>1) printf("nfs_getattrcache: initial timeo = %d\n", timeo); #endif if (vap->va_type == VDIR) { if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin) timeo = nmp->nm_acdirmin; else if (timeo > nmp->nm_acdirmax) timeo = nmp->nm_acdirmax; } else { if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin) timeo = nmp->nm_acregmin; else if (timeo > nmp->nm_acregmax) timeo = nmp->nm_acregmax; } #ifdef NFS_ACDEBUG if (nfs_acdebug > 2) printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n", nmp->nm_acregmin, nmp->nm_acregmax, nmp->nm_acdirmin, nmp->nm_acdirmax); if (nfs_acdebug) printf("nfs_getattrcache: age = %d; final timeo = %d\n", (time_second - np->n_attrstamp), timeo); #endif if ((time_second - np->n_attrstamp) >= timeo) { nfsstats.attrcache_misses++; return (ENOENT); } nfsstats.attrcache_hits++; if (vap->va_size != np->n_size) { if (vap->va_type == VREG) { if (np->n_flag & NMODIFIED) { if (vap->va_size < np->n_size) vap->va_size = np->n_size; else np->n_size = vap->va_size; } else { np->n_size = vap->va_size; } vnode_pager_setsize(vp, np->n_size); } else { np->n_size = vap->va_size; } } bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr)); if (np->n_flag & NCHG) { if (np->n_flag & NACC) vaper->va_atime = np->n_atim; if (np->n_flag & NUPD) vaper->va_mtime = np->n_mtim; } return (0); } #ifndef NFS_NOSERVER /* * Set up nameidata for a lookup() call and do it. * * If pubflag is set, this call is done for a lookup operation on the * public filehandle. In that case we allow crossing mountpoints and * absolute pathnames. However, the caller is expected to check that * the lookup result is within the public fs, and deny access if * it is not. * * nfs_namei() clears out garbage fields that namei() might leave garbage. * This is mainly ni_vp and ni_dvp when an error occurs, and ni_dvp when no * error occurs but the parent was not requested. * * dirp may be set whether an error is returned or not, and must be * released by the caller. */ int nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag) register struct nameidata *ndp; fhandle_t *fhp; int len; struct nfssvc_sock *slp; struct sockaddr *nam; struct mbuf **mdp; caddr_t *dposp; struct vnode **retdirp; struct proc *p; int kerbflag, pubflag; { register int i, rem; register struct mbuf *md; register char *fromcp, *tocp, *cp; struct iovec aiov; struct uio auio; struct vnode *dp; int error, rdonly, linklen; struct componentname *cnp = &ndp->ni_cnd; *retdirp = (struct vnode *)0; cnp->cn_pnbuf = zalloc(namei_zone); /* * Copy the name from the mbuf list to ndp->ni_pnbuf * and set the various ndp fields appropriately. */ fromcp = *dposp; tocp = cnp->cn_pnbuf; md = *mdp; rem = mtod(md, caddr_t) + md->m_len - fromcp; for (i = 0; i < len; i++) { while (rem == 0) { md = md->m_next; if (md == NULL) { error = EBADRPC; goto out; } fromcp = mtod(md, caddr_t); rem = md->m_len; } if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { error = EACCES; goto out; } *tocp++ = *fromcp++; rem--; } *tocp = '\0'; *mdp = md; *dposp = fromcp; len = nfsm_rndup(len)-len; if (len > 0) { if (rem >= len) *dposp += len; else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) goto out; } /* * Extract and set starting directory. */ error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp, nam, &rdonly, kerbflag, pubflag); if (error) goto out; if (dp->v_type != VDIR) { vrele(dp); error = ENOTDIR; goto out; } if (rdonly) cnp->cn_flags |= RDONLY; /* * Set return directory. Reference to dp is implicitly transfered * to the returned pointer */ *retdirp = dp; if (pubflag) { /* * Oh joy. For WebNFS, handle those pesky '%' escapes, * and the 'native path' indicator. */ cp = zalloc(namei_zone); fromcp = cnp->cn_pnbuf; tocp = cp; if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { switch ((unsigned char)*fromcp) { case WEBNFS_NATIVE_CHAR: /* * 'Native' path for us is the same * as a path according to the NFS spec, * just skip the escape char. */ fromcp++; break; /* * More may be added in the future, range 0x80-0xff */ default: error = EIO; zfree(namei_zone, cp); goto out; } } /* * Translate the '%' escapes, URL-style. */ while (*fromcp != '\0') { if (*fromcp == WEBNFS_ESC_CHAR) { if (fromcp[1] != '\0' && fromcp[2] != '\0') { fromcp++; *tocp++ = HEXSTRTOI(fromcp); fromcp += 2; continue; } else { error = ENOENT; zfree(namei_zone, cp); goto out; } } else *tocp++ = *fromcp++; } *tocp = '\0'; zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; ndp->ni_segflg = UIO_SYSSPACE; if (pubflag) { ndp->ni_rootdir = rootvnode; ndp->ni_loopcnt = 0; if (cnp->cn_pnbuf[0] == '/') dp = rootvnode; } else { cnp->cn_flags |= NOCROSSMOUNT; } /* * Initialize for scan, set ni_startdir and bump ref on dp again * becuase lookup() will dereference ni_startdir. */ cnp->cn_proc = p; VREF(dp); ndp->ni_startdir = dp; for (;;) { cnp->cn_nameptr = cnp->cn_pnbuf; /* * Call lookup() to do the real work. If an error occurs, * ndp->ni_vp and ni_dvp are left uninitialized or NULL and * we do not have to dereference anything before returning. * In either case ni_startdir will be dereferenced and NULLed * out. */ error = lookup(ndp); if (error) break; /* * Check for encountering a symbolic link. Trivial * termination occurs if no symlink encountered. * Note: zfree is safe because error is 0, so we will * not zfree it again when we break. */ if ((cnp->cn_flags & ISSYMLINK) == 0) { nfsrv_object_create(ndp->ni_vp); if (cnp->cn_flags & (SAVENAME | SAVESTART)) cnp->cn_flags |= HASBUF; else zfree(namei_zone, cnp->cn_pnbuf); break; } /* * Validate symlink */ if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) VOP_UNLOCK(ndp->ni_dvp, 0, p); if (!pubflag) { error = EINVAL; goto badlink2; } if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { error = ELOOP; goto badlink2; } if (ndp->ni_pathlen > 1) cp = zalloc(namei_zone); else cp = cnp->cn_pnbuf; aiov.iov_base = cp; aiov.iov_len = MAXPATHLEN; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_procp = (struct proc *)0; auio.uio_resid = MAXPATHLEN; error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); if (error) { badlink1: if (ndp->ni_pathlen > 1) zfree(namei_zone, cp); badlink2: vrele(ndp->ni_dvp); vput(ndp->ni_vp); break; } linklen = MAXPATHLEN - auio.uio_resid; if (linklen == 0) { error = ENOENT; goto badlink1; } if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { error = ENAMETOOLONG; goto badlink1; } /* * Adjust or replace path */ if (ndp->ni_pathlen > 1) { bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } else cnp->cn_pnbuf[linklen] = '\0'; ndp->ni_pathlen += linklen; /* * Cleanup refs for next loop and check if root directory * should replace current directory. Normally ni_dvp * becomes the new base directory and is cleaned up when * we loop. Explicitly null pointers after invalidation * to clarify operation. */ vput(ndp->ni_vp); ndp->ni_vp = NULL; if (cnp->cn_pnbuf[0] == '/') { vrele(ndp->ni_dvp); ndp->ni_dvp = ndp->ni_rootdir; VREF(ndp->ni_dvp); } ndp->ni_startdir = ndp->ni_dvp; ndp->ni_dvp = NULL; } /* * nfs_namei() guarentees that fields will not contain garbage * whether an error occurs or not. This allows the caller to track * cleanup state trivially. */ out: if (error) { zfree(namei_zone, cnp->cn_pnbuf); ndp->ni_vp = NULL; ndp->ni_dvp = NULL; ndp->ni_startdir = NULL; cnp->cn_flags &= ~HASBUF; } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { ndp->ni_dvp = NULL; } return (error); } /* * A fiddled version of m_adj() that ensures null fill to a long * boundary and only trims off the back end */ void nfsm_adj(mp, len, nul) struct mbuf *mp; register int len; int nul; { register struct mbuf *m; register int count, i; register char *cp; /* * Trim from tail. Scan the mbuf chain, * calculating its length and finding the last mbuf. * If the adjustment only affects this mbuf, then just * adjust and return. Otherwise, rescan and truncate * after the remaining size. */ count = 0; m = mp; for (;;) { count += m->m_len; if (m->m_next == (struct mbuf *)0) break; m = m->m_next; } if (m->m_len > len) { m->m_len -= len; if (nul > 0) { cp = mtod(m, caddr_t)+m->m_len-nul; for (i = 0; i < nul; i++) *cp++ = '\0'; } return; } count -= len; if (count < 0) count = 0; /* * Correct length for chain is "count". * Find the mbuf with last data, adjust its length, * and toss data from remaining mbufs on chain. */ for (m = mp; m; m = m->m_next) { if (m->m_len >= count) { m->m_len = count; if (nul > 0) { cp = mtod(m, caddr_t)+m->m_len-nul; for (i = 0; i < nul; i++) *cp++ = '\0'; } break; } count -= m->m_len; } for (m = m->m_next;m;m = m->m_next) m->m_len = 0; } /* * Make these functions instead of macros, so that the kernel text size * doesn't get too big... */ void nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) struct nfsrv_descript *nfsd; int before_ret; register struct vattr *before_vap; int after_ret; struct vattr *after_vap; struct mbuf **mbp; char **bposp; { register struct mbuf *mb = *mbp, *mb2; register char *bpos = *bposp; register u_int32_t *tl; if (before_ret) { nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); *tl = nfs_false; } else { nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); *tl++ = nfs_true; txdr_hyper(before_vap->va_size, tl); tl += 2; txdr_nfsv3time(&(before_vap->va_mtime), tl); tl += 2; txdr_nfsv3time(&(before_vap->va_ctime), tl); } *bposp = bpos; *mbp = mb; nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); } void nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) struct nfsrv_descript *nfsd; int after_ret; struct vattr *after_vap; struct mbuf **mbp; char **bposp; { register struct mbuf *mb = *mbp, *mb2; register char *bpos = *bposp; register u_int32_t *tl; register struct nfs_fattr *fp; if (after_ret) { nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); *tl = nfs_false; } else { nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); *tl++ = nfs_true; fp = (struct nfs_fattr *)tl; nfsm_srvfattr(nfsd, after_vap, fp); } *mbp = mb; *bposp = bpos; } void nfsm_srvfattr(nfsd, vap, fp) register struct nfsrv_descript *nfsd; register struct vattr *vap; register struct nfs_fattr *fp; { fp->fa_nlink = txdr_unsigned(vap->va_nlink); fp->fa_uid = txdr_unsigned(vap->va_uid); fp->fa_gid = txdr_unsigned(vap->va_gid); if (nfsd->nd_flag & ND_NFSV3) { fp->fa_type = vtonfsv3_type(vap->va_type); fp->fa_mode = vtonfsv3_mode(vap->va_mode); txdr_hyper(vap->va_size, &fp->fa3_size); txdr_hyper(vap->va_bytes, &fp->fa3_used); fp->fa3_rdev.specdata1 = txdr_unsigned(umajor(vap->va_rdev)); fp->fa3_rdev.specdata2 = txdr_unsigned(uminor(vap->va_rdev)); fp->fa3_fsid.nfsuquad[0] = 0; fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); fp->fa3_fileid.nfsuquad[0] = 0; fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid); txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); } else { fp->fa_type = vtonfsv2_type(vap->va_type); fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); fp->fa2_size = txdr_unsigned(vap->va_size); fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); if (vap->va_type == VFIFO) fp->fa2_rdev = 0xffffffff; else fp->fa2_rdev = txdr_unsigned(vap->va_rdev); fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); fp->fa2_fsid = txdr_unsigned(vap->va_fsid); fp->fa2_fileid = txdr_unsigned(vap->va_fileid); txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); } } /* * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) * - look up fsid in mount list (if not found ret error) * - get vp and export rights by calling VFS_FHTOVP() * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon * - if not lockflag unlock it with VOP_UNLOCK() */ int nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag) fhandle_t *fhp; int lockflag; struct vnode **vpp; struct ucred *cred; struct nfssvc_sock *slp; struct sockaddr *nam; int *rdonlyp; int kerbflag; int pubflag; { struct proc *p = curproc; /* XXX */ register struct mount *mp; register int i; struct ucred *credanon; int error, exflags; #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */ struct sockaddr_int *saddr; #endif *vpp = (struct vnode *)0; if (nfs_ispublicfh(fhp)) { if (!pubflag || !nfs_pub.np_valid) return (ESTALE); fhp = &nfs_pub.np_handle; } mp = vfs_getvfs(&fhp->fh_fsid); if (!mp) return (ESTALE); error = VFS_CHECKEXP(mp, nam, &exflags, &credanon); if (error) return (error); error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); if (error) return (error); #ifdef MNT_EXNORESPORT if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { saddr = (struct sockaddr_in *)nam; if (saddr->sin_family == AF_INET && ntohs(saddr->sin_port) >= IPPORT_RESERVED) { vput(*vpp); *vpp = NULL; return (NFSERR_AUTHERR | AUTH_TOOWEAK); } } #endif /* * Check/setup credentials. */ if (exflags & MNT_EXKERB) { if (!kerbflag) { vput(*vpp); *vpp = NULL; return (NFSERR_AUTHERR | AUTH_TOOWEAK); } } else if (kerbflag) { vput(*vpp); *vpp = NULL; return (NFSERR_AUTHERR | AUTH_TOOWEAK); } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) { cred->cr_uid = credanon->cr_uid; for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++) cred->cr_groups[i] = credanon->cr_groups[i]; cred->cr_ngroups = i; } if (exflags & MNT_EXRDONLY) *rdonlyp = 1; else *rdonlyp = 0; nfsrv_object_create(*vpp); if (!lockflag) VOP_UNLOCK(*vpp, 0, p); return (0); } /* * WebNFS: check if a filehandle is a public filehandle. For v3, this * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has * transformed this to all zeroes in both cases, so check for it. */ int nfs_ispublicfh(fhp) fhandle_t *fhp; { char *cp = (char *)fhp; int i; for (i = 0; i < NFSX_V3FH; i++) if (*cp++ != 0) return (FALSE); return (TRUE); } #endif /* NFS_NOSERVER */ /* * This function compares two net addresses by family and returns TRUE * if they are the same host. * If there is any doubt, return FALSE. * The AF_INET family is handled as a special case so that address mbufs * don't need to be saved to store "struct in_addr", which is only 4 bytes. */ int netaddr_match(family, haddr, nam) int family; union nethostaddr *haddr; struct sockaddr *nam; { register struct sockaddr_in *inetaddr; switch (family) { case AF_INET: inetaddr = (struct sockaddr_in *)nam; if (inetaddr->sin_family == AF_INET && inetaddr->sin_addr.s_addr == haddr->had_inetaddr) return (1); break; default: break; }; return (0); } static nfsuint64 nfs_nullcookie = { { 0, 0 } }; /* * This function finds the directory cookie that corresponds to the * logical byte offset given. */ nfsuint64 * nfs_getcookie(np, off, add) register struct nfsnode *np; off_t off; int add; { register struct nfsdmap *dp, *dp2; register int pos; pos = (uoff_t)off / NFS_DIRBLKSIZ; if (pos == 0 || off < 0) { #ifdef DIAGNOSTIC if (add) panic("nfs getcookie add at <= 0"); #endif return (&nfs_nullcookie); } pos--; dp = np->n_cookies.lh_first; if (!dp) { if (add) { MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap), M_NFSDIROFF, M_WAITOK); dp->ndm_eocookie = 0; LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list); } else return ((nfsuint64 *)0); } while (pos >= NFSNUMCOOKIES) { pos -= NFSNUMCOOKIES; if (dp->ndm_list.le_next) { if (!add && dp->ndm_eocookie < NFSNUMCOOKIES && pos >= dp->ndm_eocookie) return ((nfsuint64 *)0); dp = dp->ndm_list.le_next; } else if (add) { MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap), M_NFSDIROFF, M_WAITOK); dp2->ndm_eocookie = 0; LIST_INSERT_AFTER(dp, dp2, ndm_list); dp = dp2; } else return ((nfsuint64 *)0); } if (pos >= dp->ndm_eocookie) { if (add) dp->ndm_eocookie = pos + 1; else return ((nfsuint64 *)0); } return (&dp->ndm_cookies[pos]); } /* * Invalidate cached directory information, except for the actual directory * blocks (which are invalidated separately). * Done mainly to avoid the use of stale offset cookies. */ void nfs_invaldir(vp) register struct vnode *vp; { register struct nfsnode *np = VTONFS(vp); #ifdef DIAGNOSTIC if (vp->v_type != VDIR) panic("nfs: invaldir not dir"); #endif np->n_direofoffset = 0; np->n_cookieverf.nfsuquad[0] = 0; np->n_cookieverf.nfsuquad[1] = 0; if (np->n_cookies.lh_first) np->n_cookies.lh_first->ndm_eocookie = 0; } /* * The write verifier has changed (probably due to a server reboot), so all * B_NEEDCOMMIT blocks will have to be written again. Since they are on the * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT * and B_CLUSTEROK flags. Once done the new write verifier can be set for the * mount point. * * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data * writes are not clusterable. */ void nfs_clearcommit(mp) struct mount *mp; { register struct vnode *vp, *nvp; register struct buf *bp, *nbp; int s; s = splbio(); loop: for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) { if (vp->v_mount != mp) /* Paranoia */ goto loop; nvp = TAILQ_NEXT(vp, v_nmntvnodes); for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { nbp = TAILQ_NEXT(bp, b_vnbufs); if (BUF_REFCNT(bp) == 0 && (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == (B_DELWRI | B_NEEDCOMMIT)) bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); } } splx(s); } #ifndef NFS_NOSERVER /* * Map errnos to NFS error numbers. For Version 3 also filter out error * numbers not specified for the associated procedure. */ int nfsrv_errmap(nd, err) struct nfsrv_descript *nd; register int err; { register short *defaulterrp, *errp; if (nd->nd_flag & ND_NFSV3) { if (nd->nd_procnum <= NFSPROC_COMMIT) { errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; while (*++errp) { if (*errp == err) return (err); else if (*errp > err) break; } return ((int)*defaulterrp); } else return (err & 0xffff); } if (err <= ELAST) return ((int)nfsrv_v2errmap[err - 1]); return (NFSERR_IO); } int nfsrv_object_create(vp) struct vnode *vp; { if (vp == NULL || vp->v_type != VREG) return (1); return (vfs_object_create(vp, curproc, curproc ? curproc->p_ucred : NULL)); } /* * Sort the group list in increasing numerical order. * (Insertion sort by Chris Torek, who was grossed out by the bubble sort * that used to be here.) */ void nfsrvw_sort(list, num) register gid_t *list; register int num; { register int i, j; gid_t v; /* Insertion sort. */ for (i = 1; i < num; i++) { v = list[i]; /* find correct slot for value v, moving others up */ for (j = i; --j >= 0 && v < list[j];) list[j + 1] = list[j]; list[j + 1] = v; } } /* * copy credentials making sure that the result can be compared with bcmp(). */ void nfsrv_setcred(incred, outcred) register struct ucred *incred, *outcred; { register int i; bzero((caddr_t)outcred, sizeof (struct ucred)); outcred->cr_ref = 1; outcred->cr_uid = incred->cr_uid; outcred->cr_ngroups = incred->cr_ngroups; for (i = 0; i < incred->cr_ngroups; i++) outcred->cr_groups[i] = incred->cr_groups[i]; nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups); } #endif /* NFS_NOSERVER */ Index: stable/4/sys/nfs/nfsnode.h =================================================================== --- stable/4/sys/nfs/nfsnode.h (revision 131097) +++ stable/4/sys/nfs/nfsnode.h (revision 131098) @@ -1,211 +1,212 @@ /* * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)nfsnode.h 8.9 (Berkeley) 5/14/95 * $FreeBSD$ */ #ifndef _NFS_NFSNODE_H_ #define _NFS_NFSNODE_H_ #if !defined(_NFS_NFS_H_) && !defined(_KERNEL) #include #endif /* * Silly rename structure that hangs off the nfsnode until the name * can be removed by nfs_inactive() */ struct sillyrename { struct ucred *s_cred; struct vnode *s_dvp; long s_namlen; char s_name[20]; }; /* * This structure is used to save the logical directory offset to * NFS cookie mappings. * The mappings are stored in a list headed * by n_cookies, as required. * There is one mapping for each NFS_DIRBLKSIZ bytes of directory information * stored in increasing logical offset byte order. */ #define NFSNUMCOOKIES 31 struct nfsdmap { LIST_ENTRY(nfsdmap) ndm_list; int ndm_eocookie; nfsuint64 ndm_cookies[NFSNUMCOOKIES]; }; /* * The nfsnode is the nfs equivalent to ufs's inode. Any similarity * is purely coincidental. * There is a unique nfsnode allocated for each active file, * each current directory, each mounted-on file, text file, and the root. * An nfsnode is 'named' by its file handle. (nget/nfs_node.c) * If this structure exceeds 256 bytes (it is currently 256 using 4.4BSD-Lite * type definitions), file handles of > 32 bytes should probably be split out * into a separate MALLOC()'d data structure. (Reduce the size of nfsfh_t by * changing the definition in nfsproto.h of NFS_SMALLFH.) * NB: Hopefully the current order of the fields is such that everything will * be well aligned and, therefore, tightly packed. */ struct nfsnode { struct lock n_lock; LIST_ENTRY(nfsnode) n_hash; /* Hash chain */ CIRCLEQ_ENTRY(nfsnode) n_timer; /* Nqnfs timer chain */ u_quad_t n_size; /* Current size of file */ u_quad_t n_brev; /* Modify rev when cached */ u_quad_t n_lrev; /* Modify rev for lease */ struct vattr n_vattr; /* Vnode attribute cache */ time_t n_attrstamp; /* Attr. cache timestamp */ u_int32_t n_mode; /* ACCESS mode cache */ uid_t n_modeuid; /* credentials having mode */ time_t n_modestamp; /* mode cache timestamp */ time_t n_mtime; /* Prev modify time. */ time_t n_ctime; /* Prev create time. */ time_t n_expiry; /* Lease expiry time */ nfsfh_t *n_fhp; /* NFS File Handle */ struct vnode *n_vnode; /* associated vnode */ struct lockf *n_lockf; /* Locking record of file */ int n_error; /* Save write error value */ union { struct timespec nf_atim; /* Special file times */ nfsuint64 nd_cookieverf; /* Cookie verifier (dir only) */ } n_un1; union { struct timespec nf_mtim; off_t nd_direof; /* Dir. EOF offset cache */ } n_un2; union { struct sillyrename *nf_silly; /* Ptr to silly rename struct */ LIST_HEAD(, nfsdmap) nd_cook; /* cookies */ } n_un3; short n_fhsize; /* size in bytes, of fh */ short n_flag; /* Flag for locking.. */ nfsfh_t n_fh; /* Small File Handle */ struct lock n_rslock; }; #define n_atim n_un1.nf_atim #define n_mtim n_un2.nf_mtim #define n_sillyrename n_un3.nf_silly #define n_cookieverf n_un1.nd_cookieverf #define n_direofoffset n_un2.nd_direof #define n_cookies n_un3.nd_cook /* * Flags for n_flag */ #define NFLUSHWANT 0x0001 /* Want wakeup from a flush in prog. */ #define NFLUSHINPROG 0x0002 /* Avoid multiple calls to vinvalbuf() */ #define NMODIFIED 0x0004 /* Might have a modified buffer in bio */ #define NWRITEERR 0x0008 /* Flag write errors so close will know */ #define NQNFSNONCACHE 0x0020 /* Non-cachable lease */ #define NQNFSWRITE 0x0040 /* Write lease */ #define NQNFSEVICTED 0x0080 /* Has been evicted */ #define NACC 0x0100 /* Special file accessed */ #define NUPD 0x0200 /* Special file updated */ #define NCHG 0x0400 /* Special file times changed */ #define NLOCKED 0x0800 /* node is locked */ #define NWANTED 0x0100 /* someone wants to lock */ +#define NSIZECHANGED 0x2000 /* File size has changed: need cache inval */ /* * Convert between nfsnode pointers and vnode pointers */ #define VTONFS(vp) ((struct nfsnode *)(vp)->v_data) #define NFSTOV(np) ((struct vnode *)(np)->n_vnode) /* * Queue head for nfsiod's */ extern TAILQ_HEAD(nfs_bufq, buf) nfs_bufq; extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; extern struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; #if defined(_KERNEL) /* * nfs_rslock - Attempt to obtain lock on nfsnode * * Attempt to obtain a lock on the passed nfsnode, returning ENOLCK * if the lock could not be obtained due to our having to sleep. This * function is generally used to lock around code that modifies an * NFS file's size. In order to avoid deadlocks the lock * should not be obtained while other locks are being held. */ static __inline int nfs_rslock(struct nfsnode *np, struct proc *p) { return(lockmgr(&np->n_rslock, LK_EXCLUSIVE | LK_CANRECURSE | LK_SLEEPFAIL, NULL, p)); } static __inline void nfs_rsunlock(struct nfsnode *np, struct proc *p) { (void)lockmgr(&np->n_rslock, LK_RELEASE, NULL, p); } extern vop_t **fifo_nfsv2nodeop_p; extern vop_t **nfsv2_vnodeop_p; extern vop_t **spec_nfsv2nodeop_p; /* * Prototypes for NFS vnode operations */ int nfs_getpages __P((struct vop_getpages_args *)); int nfs_putpages __P((struct vop_putpages_args *)); int nfs_write __P((struct vop_write_args *)); int nqnfs_vop_lease_check __P((struct vop_lease_args *)); int nfs_inactive __P((struct vop_inactive_args *)); int nfs_reclaim __P((struct vop_reclaim_args *)); /* other stuff */ int nfs_removeit __P((struct sillyrename *)); int nfs_nget __P((struct mount *,nfsfh_t *,int,struct nfsnode **)); nfsuint64 *nfs_getcookie __P((struct nfsnode *, off_t, int)); void nfs_invaldir __P((struct vnode *)); #define nqnfs_lease_updatetime nfs_lease_updatetime #endif /* _KERNEL */ #endif