17c478bd9Sstevel@tonic-gate /* 27c478bd9Sstevel@tonic-gate * CDDL HEADER START 37c478bd9Sstevel@tonic-gate * 47c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5a5652762Spraks * Common Development and Distribution License (the "License"). 6a5652762Spraks * You may not use this file except in compliance with the License. 77c478bd9Sstevel@tonic-gate * 87c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 97c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 107c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 117c478bd9Sstevel@tonic-gate * and limitations under the License. 127c478bd9Sstevel@tonic-gate * 137c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 147c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 157c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 167c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 177c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 187c478bd9Sstevel@tonic-gate * 197c478bd9Sstevel@tonic-gate * CDDL HEADER END 207c478bd9Sstevel@tonic-gate */ 217c478bd9Sstevel@tonic-gate /* 2260946fe0Smec * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate * 257c478bd9Sstevel@tonic-gate * Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T. 267c478bd9Sstevel@tonic-gate * All rights reserved. 277c478bd9Sstevel@tonic-gate */ 287c478bd9Sstevel@tonic-gate 297c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 307c478bd9Sstevel@tonic-gate 317c478bd9Sstevel@tonic-gate #include <sys/param.h> 327c478bd9Sstevel@tonic-gate #include <sys/types.h> 337c478bd9Sstevel@tonic-gate #include <sys/systm.h> 347c478bd9Sstevel@tonic-gate #include <sys/cred.h> 357c478bd9Sstevel@tonic-gate #include <sys/time.h> 367c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 377c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 38aa59c4cbSrsb #include <sys/vfs_opreg.h> 397c478bd9Sstevel@tonic-gate #include <sys/file.h> 407c478bd9Sstevel@tonic-gate #include <sys/filio.h> 417c478bd9Sstevel@tonic-gate #include <sys/uio.h> 427c478bd9Sstevel@tonic-gate #include <sys/buf.h> 437c478bd9Sstevel@tonic-gate #include <sys/mman.h> 447c478bd9Sstevel@tonic-gate #include <sys/pathname.h> 457c478bd9Sstevel@tonic-gate #include <sys/dirent.h> 467c478bd9Sstevel@tonic-gate #include <sys/debug.h> 477c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 487c478bd9Sstevel@tonic-gate #include <sys/fcntl.h> 497c478bd9Sstevel@tonic-gate #include <sys/flock.h> 507c478bd9Sstevel@tonic-gate #include <sys/swap.h> 517c478bd9Sstevel@tonic-gate #include <sys/errno.h> 527c478bd9Sstevel@tonic-gate #include <sys/strsubr.h> 537c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 547c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 557c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 567c478bd9Sstevel@tonic-gate #include <sys/pathconf.h> 577c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 587c478bd9Sstevel@tonic-gate #include <sys/dnlc.h> 597c478bd9Sstevel@tonic-gate #include <sys/acl.h> 607c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 617c478bd9Sstevel@tonic-gate #include <sys/policy.h> 627c478bd9Sstevel@tonic-gate #include <sys/sdt.h> 637c478bd9Sstevel@tonic-gate 647c478bd9Sstevel@tonic-gate #include <rpc/types.h> 657c478bd9Sstevel@tonic-gate #include <rpc/auth.h> 667c478bd9Sstevel@tonic-gate #include <rpc/clnt.h> 677c478bd9Sstevel@tonic-gate 687c478bd9Sstevel@tonic-gate #include <nfs/nfs.h> 697c478bd9Sstevel@tonic-gate #include <nfs/nfs_clnt.h> 707c478bd9Sstevel@tonic-gate #include <nfs/rnode.h> 717c478bd9Sstevel@tonic-gate #include <nfs/nfs_acl.h> 727c478bd9Sstevel@tonic-gate #include <nfs/lm.h> 737c478bd9Sstevel@tonic-gate 747c478bd9Sstevel@tonic-gate #include <vm/hat.h> 757c478bd9Sstevel@tonic-gate #include <vm/as.h> 767c478bd9Sstevel@tonic-gate #include <vm/page.h> 777c478bd9Sstevel@tonic-gate #include <vm/pvn.h> 787c478bd9Sstevel@tonic-gate #include <vm/seg.h> 797c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 807c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 817c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 827c478bd9Sstevel@tonic-gate 837c478bd9Sstevel@tonic-gate #include <fs/fs_subr.h> 847c478bd9Sstevel@tonic-gate 857c478bd9Sstevel@tonic-gate #include <sys/ddi.h> 867c478bd9Sstevel@tonic-gate 877c478bd9Sstevel@tonic-gate static int nfs_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int, 887c478bd9Sstevel@tonic-gate cred_t *); 897c478bd9Sstevel@tonic-gate static int nfswrite(vnode_t *, caddr_t, uint_t, int, cred_t *); 907c478bd9Sstevel@tonic-gate static int nfsread(vnode_t *, caddr_t, uint_t, int, size_t *, cred_t *); 917c478bd9Sstevel@tonic-gate static int nfssetattr(vnode_t *, struct vattr *, int, cred_t *); 927c478bd9Sstevel@tonic-gate static int nfslookup_dnlc(vnode_t *, char *, vnode_t **, cred_t *); 937c478bd9Sstevel@tonic-gate static int nfslookup_otw(vnode_t *, char *, vnode_t **, cred_t *, int); 94da6c28aaSamw static int nfsrename(vnode_t *, char *, vnode_t *, char *, cred_t *, 95da6c28aaSamw caller_context_t *); 967c478bd9Sstevel@tonic-gate static int nfsreaddir(vnode_t *, rddir_cache *, cred_t *); 977c478bd9Sstevel@tonic-gate static int nfs_bio(struct buf *, cred_t *); 987c478bd9Sstevel@tonic-gate static int nfs_getapage(vnode_t *, u_offset_t, size_t, uint_t *, 997c478bd9Sstevel@tonic-gate page_t *[], size_t, struct seg *, caddr_t, 1007c478bd9Sstevel@tonic-gate enum seg_rw, cred_t *); 1017c478bd9Sstevel@tonic-gate static void nfs_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *, 1027c478bd9Sstevel@tonic-gate cred_t *); 1037c478bd9Sstevel@tonic-gate static int nfs_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t, 1047c478bd9Sstevel@tonic-gate int, cred_t *); 1057c478bd9Sstevel@tonic-gate static int nfs_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t, 1067c478bd9Sstevel@tonic-gate int, cred_t *); 1077c478bd9Sstevel@tonic-gate static void nfs_delmap_callback(struct as *, void *, uint_t); 1087c478bd9Sstevel@tonic-gate 1097c478bd9Sstevel@tonic-gate /* 1107c478bd9Sstevel@tonic-gate * Error flags used to pass information about certain special errors 1117c478bd9Sstevel@tonic-gate * which need to be handled specially. 1127c478bd9Sstevel@tonic-gate */ 1137c478bd9Sstevel@tonic-gate #define NFS_EOF -98 1147c478bd9Sstevel@tonic-gate 1157c478bd9Sstevel@tonic-gate /* 1167c478bd9Sstevel@tonic-gate * These are the vnode ops routines which implement the vnode interface to 1177c478bd9Sstevel@tonic-gate * the networked file system. These routines just take their parameters, 1187c478bd9Sstevel@tonic-gate * make them look networkish by putting the right info into interface structs, 1197c478bd9Sstevel@tonic-gate * and then calling the appropriate remote routine(s) to do the work. 1207c478bd9Sstevel@tonic-gate * 1217c478bd9Sstevel@tonic-gate * Note on directory name lookup cacheing: If we detect a stale fhandle, 1227c478bd9Sstevel@tonic-gate * we purge the directory cache relative to that vnode. This way, the 1237c478bd9Sstevel@tonic-gate * user won't get burned by the cache repeatedly. See <nfs/rnode.h> for 1247c478bd9Sstevel@tonic-gate * more details on rnode locking. 1257c478bd9Sstevel@tonic-gate */ 1267c478bd9Sstevel@tonic-gate 127da6c28aaSamw static int nfs_open(vnode_t **, int, cred_t *, caller_context_t *); 128da6c28aaSamw static int nfs_close(vnode_t *, int, int, offset_t, cred_t *, 129da6c28aaSamw caller_context_t *); 1307c478bd9Sstevel@tonic-gate static int nfs_read(vnode_t *, struct uio *, int, cred_t *, 1317c478bd9Sstevel@tonic-gate caller_context_t *); 1327c478bd9Sstevel@tonic-gate static int nfs_write(vnode_t *, struct uio *, int, cred_t *, 1337c478bd9Sstevel@tonic-gate caller_context_t *); 134da6c28aaSamw static int nfs_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *, 135da6c28aaSamw caller_context_t *); 136da6c28aaSamw static int nfs_getattr(vnode_t *, struct vattr *, int, cred_t *, 137da6c28aaSamw caller_context_t *); 1387c478bd9Sstevel@tonic-gate static int nfs_setattr(vnode_t *, struct vattr *, int, cred_t *, 1397c478bd9Sstevel@tonic-gate caller_context_t *); 140da6c28aaSamw static int nfs_access(vnode_t *, int, int, cred_t *, caller_context_t *); 1417c478bd9Sstevel@tonic-gate static int nfs_accessx(void *, int, cred_t *); 142da6c28aaSamw static int nfs_readlink(vnode_t *, struct uio *, cred_t *, 143da6c28aaSamw caller_context_t *); 144da6c28aaSamw static int nfs_fsync(vnode_t *, int, cred_t *, caller_context_t *); 145da6c28aaSamw static void nfs_inactive(vnode_t *, cred_t *, caller_context_t *); 1467c478bd9Sstevel@tonic-gate static int nfs_lookup(vnode_t *, char *, vnode_t **, struct pathname *, 147da6c28aaSamw int, vnode_t *, cred_t *, caller_context_t *, 148da6c28aaSamw int *, pathname_t *); 1497c478bd9Sstevel@tonic-gate static int nfs_create(vnode_t *, char *, struct vattr *, enum vcexcl, 150da6c28aaSamw int, vnode_t **, cred_t *, int, caller_context_t *, 151da6c28aaSamw vsecattr_t *); 152da6c28aaSamw static int nfs_remove(vnode_t *, char *, cred_t *, caller_context_t *, 153da6c28aaSamw int); 154da6c28aaSamw static int nfs_link(vnode_t *, vnode_t *, char *, cred_t *, 155da6c28aaSamw caller_context_t *, int); 156da6c28aaSamw static int nfs_rename(vnode_t *, char *, vnode_t *, char *, cred_t *, 157da6c28aaSamw caller_context_t *, int); 158da6c28aaSamw static int nfs_mkdir(vnode_t *, char *, struct vattr *, vnode_t **, 159da6c28aaSamw cred_t *, caller_context_t *, int, vsecattr_t *); 160da6c28aaSamw static int nfs_rmdir(vnode_t *, char *, vnode_t *, cred_t *, 161da6c28aaSamw caller_context_t *, int); 1627c478bd9Sstevel@tonic-gate static int nfs_symlink(vnode_t *, char *, struct vattr *, char *, 163da6c28aaSamw cred_t *, caller_context_t *, int); 164da6c28aaSamw static int nfs_readdir(vnode_t *, struct uio *, cred_t *, int *, 165da6c28aaSamw caller_context_t *, int); 166da6c28aaSamw static int nfs_fid(vnode_t *, fid_t *, caller_context_t *); 1677c478bd9Sstevel@tonic-gate static int nfs_rwlock(vnode_t *, int, caller_context_t *); 1687c478bd9Sstevel@tonic-gate static void nfs_rwunlock(vnode_t *, int, caller_context_t *); 169da6c28aaSamw static int nfs_seek(vnode_t *, offset_t, offset_t *, caller_context_t *); 1707c478bd9Sstevel@tonic-gate static int nfs_getpage(vnode_t *, offset_t, size_t, uint_t *, 1717c478bd9Sstevel@tonic-gate page_t *[], size_t, struct seg *, caddr_t, 172da6c28aaSamw enum seg_rw, cred_t *, caller_context_t *); 173da6c28aaSamw static int nfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *, 174da6c28aaSamw caller_context_t *); 175da6c28aaSamw static int nfs_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t, 176da6c28aaSamw uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *); 177da6c28aaSamw static int nfs_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t, 178da6c28aaSamw uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *); 1797c478bd9Sstevel@tonic-gate static int nfs_frlock(vnode_t *, int, struct flock64 *, int, offset_t, 180da6c28aaSamw struct flk_callback *, cred_t *, caller_context_t *); 1817c478bd9Sstevel@tonic-gate static int nfs_space(vnode_t *, int, struct flock64 *, int, offset_t, 1827c478bd9Sstevel@tonic-gate cred_t *, caller_context_t *); 183da6c28aaSamw static int nfs_realvp(vnode_t *, vnode_t **, caller_context_t *); 184da6c28aaSamw static int nfs_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t, 185da6c28aaSamw uint_t, uint_t, uint_t, cred_t *, caller_context_t *); 186da6c28aaSamw static int nfs_pathconf(vnode_t *, int, ulong_t *, cred_t *, 187da6c28aaSamw caller_context_t *); 1887c478bd9Sstevel@tonic-gate static int nfs_pageio(vnode_t *, page_t *, u_offset_t, size_t, int, 189da6c28aaSamw cred_t *, caller_context_t *); 190da6c28aaSamw static int nfs_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *, 191da6c28aaSamw caller_context_t *); 192da6c28aaSamw static int nfs_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *, 193da6c28aaSamw caller_context_t *); 194da6c28aaSamw static int nfs_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *, 195da6c28aaSamw caller_context_t *); 1967c478bd9Sstevel@tonic-gate 1977c478bd9Sstevel@tonic-gate struct vnodeops *nfs_vnodeops; 1987c478bd9Sstevel@tonic-gate 1997c478bd9Sstevel@tonic-gate const fs_operation_def_t nfs_vnodeops_template[] = { 200aa59c4cbSrsb VOPNAME_OPEN, { .vop_open = nfs_open }, 201aa59c4cbSrsb VOPNAME_CLOSE, { .vop_close = nfs_close }, 202aa59c4cbSrsb VOPNAME_READ, { .vop_read = nfs_read }, 203aa59c4cbSrsb VOPNAME_WRITE, { .vop_write = nfs_write }, 204aa59c4cbSrsb VOPNAME_IOCTL, { .vop_ioctl = nfs_ioctl }, 205aa59c4cbSrsb VOPNAME_GETATTR, { .vop_getattr = nfs_getattr }, 206aa59c4cbSrsb VOPNAME_SETATTR, { .vop_setattr = nfs_setattr }, 207aa59c4cbSrsb VOPNAME_ACCESS, { .vop_access = nfs_access }, 208aa59c4cbSrsb VOPNAME_LOOKUP, { .vop_lookup = nfs_lookup }, 209aa59c4cbSrsb VOPNAME_CREATE, { .vop_create = nfs_create }, 210aa59c4cbSrsb VOPNAME_REMOVE, { .vop_remove = nfs_remove }, 211aa59c4cbSrsb VOPNAME_LINK, { .vop_link = nfs_link }, 212aa59c4cbSrsb VOPNAME_RENAME, { .vop_rename = nfs_rename }, 213aa59c4cbSrsb VOPNAME_MKDIR, { .vop_mkdir = nfs_mkdir }, 214aa59c4cbSrsb VOPNAME_RMDIR, { .vop_rmdir = nfs_rmdir }, 215aa59c4cbSrsb VOPNAME_READDIR, { .vop_readdir = nfs_readdir }, 216aa59c4cbSrsb VOPNAME_SYMLINK, { .vop_symlink = nfs_symlink }, 217aa59c4cbSrsb VOPNAME_READLINK, { .vop_readlink = nfs_readlink }, 218aa59c4cbSrsb VOPNAME_FSYNC, { .vop_fsync = nfs_fsync }, 219aa59c4cbSrsb VOPNAME_INACTIVE, { .vop_inactive = nfs_inactive }, 220aa59c4cbSrsb VOPNAME_FID, { .vop_fid = nfs_fid }, 221aa59c4cbSrsb VOPNAME_RWLOCK, { .vop_rwlock = nfs_rwlock }, 222aa59c4cbSrsb VOPNAME_RWUNLOCK, { .vop_rwunlock = nfs_rwunlock }, 223aa59c4cbSrsb VOPNAME_SEEK, { .vop_seek = nfs_seek }, 224aa59c4cbSrsb VOPNAME_FRLOCK, { .vop_frlock = nfs_frlock }, 225aa59c4cbSrsb VOPNAME_SPACE, { .vop_space = nfs_space }, 226aa59c4cbSrsb VOPNAME_REALVP, { .vop_realvp = nfs_realvp }, 227aa59c4cbSrsb VOPNAME_GETPAGE, { .vop_getpage = nfs_getpage }, 228aa59c4cbSrsb VOPNAME_PUTPAGE, { .vop_putpage = nfs_putpage }, 229aa59c4cbSrsb VOPNAME_MAP, { .vop_map = nfs_map }, 230aa59c4cbSrsb VOPNAME_ADDMAP, { .vop_addmap = nfs_addmap }, 231aa59c4cbSrsb VOPNAME_DELMAP, { .vop_delmap = nfs_delmap }, 232aa59c4cbSrsb VOPNAME_DUMP, { .vop_dump = nfs_dump }, 233aa59c4cbSrsb VOPNAME_PATHCONF, { .vop_pathconf = nfs_pathconf }, 234aa59c4cbSrsb VOPNAME_PAGEIO, { .vop_pageio = nfs_pageio }, 235aa59c4cbSrsb VOPNAME_SETSECATTR, { .vop_setsecattr = nfs_setsecattr }, 236aa59c4cbSrsb VOPNAME_GETSECATTR, { .vop_getsecattr = nfs_getsecattr }, 237aa59c4cbSrsb VOPNAME_SHRLOCK, { .vop_shrlock = nfs_shrlock }, 238df2381bfSpraks VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 2397c478bd9Sstevel@tonic-gate NULL, NULL 2407c478bd9Sstevel@tonic-gate }; 2417c478bd9Sstevel@tonic-gate 2427c478bd9Sstevel@tonic-gate /* 2437c478bd9Sstevel@tonic-gate * XXX: This is referenced in modstubs.s 2447c478bd9Sstevel@tonic-gate */ 2457c478bd9Sstevel@tonic-gate struct vnodeops * 2467c478bd9Sstevel@tonic-gate nfs_getvnodeops(void) 2477c478bd9Sstevel@tonic-gate { 2487c478bd9Sstevel@tonic-gate return (nfs_vnodeops); 2497c478bd9Sstevel@tonic-gate } 2507c478bd9Sstevel@tonic-gate 2517c478bd9Sstevel@tonic-gate /* ARGSUSED */ 2527c478bd9Sstevel@tonic-gate static int 253da6c28aaSamw nfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 2547c478bd9Sstevel@tonic-gate { 2557c478bd9Sstevel@tonic-gate int error; 2567c478bd9Sstevel@tonic-gate struct vattr va; 2577c478bd9Sstevel@tonic-gate rnode_t *rp; 2587c478bd9Sstevel@tonic-gate vnode_t *vp; 2597c478bd9Sstevel@tonic-gate 2607c478bd9Sstevel@tonic-gate vp = *vpp; 2617c478bd9Sstevel@tonic-gate rp = VTOR(vp); 262108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 2637c478bd9Sstevel@tonic-gate return (EIO); 2647c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 2657c478bd9Sstevel@tonic-gate if (rp->r_cred == NULL) { 2667c478bd9Sstevel@tonic-gate crhold(cr); 2677c478bd9Sstevel@tonic-gate rp->r_cred = cr; 2687c478bd9Sstevel@tonic-gate } 2697c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 2707c478bd9Sstevel@tonic-gate 2717c478bd9Sstevel@tonic-gate /* 2727c478bd9Sstevel@tonic-gate * If there is no cached data or if close-to-open 2737c478bd9Sstevel@tonic-gate * consistency checking is turned off, we can avoid 2747c478bd9Sstevel@tonic-gate * the over the wire getattr. Otherwise, if the 2757c478bd9Sstevel@tonic-gate * file system is mounted readonly, then just verify 2767c478bd9Sstevel@tonic-gate * the caches are up to date using the normal mechanism. 2777c478bd9Sstevel@tonic-gate * Else, if the file is not mmap'd, then just mark 2787c478bd9Sstevel@tonic-gate * the attributes as timed out. They will be refreshed 2797c478bd9Sstevel@tonic-gate * and the caches validated prior to being used. 2807c478bd9Sstevel@tonic-gate * Else, the file system is mounted writeable so 2817c478bd9Sstevel@tonic-gate * force an over the wire GETATTR in order to ensure 2827c478bd9Sstevel@tonic-gate * that all cached data is valid. 2837c478bd9Sstevel@tonic-gate */ 2847c478bd9Sstevel@tonic-gate if (vp->v_count > 1 || 2857c478bd9Sstevel@tonic-gate ((vn_has_cached_data(vp) || HAVE_RDDIR_CACHE(rp)) && 2867c478bd9Sstevel@tonic-gate !(VTOMI(vp)->mi_flags & MI_NOCTO))) { 2877c478bd9Sstevel@tonic-gate if (vn_is_readonly(vp)) 2887c478bd9Sstevel@tonic-gate error = nfs_validate_caches(vp, cr); 2897c478bd9Sstevel@tonic-gate else if (rp->r_mapcnt == 0 && vp->v_count == 1) { 2907c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); 2917c478bd9Sstevel@tonic-gate error = 0; 2927c478bd9Sstevel@tonic-gate } else { 2937c478bd9Sstevel@tonic-gate va.va_mask = AT_ALL; 2947c478bd9Sstevel@tonic-gate error = nfs_getattr_otw(vp, &va, cr); 2957c478bd9Sstevel@tonic-gate } 2967c478bd9Sstevel@tonic-gate } else 2977c478bd9Sstevel@tonic-gate error = 0; 2987c478bd9Sstevel@tonic-gate 2997c478bd9Sstevel@tonic-gate return (error); 3007c478bd9Sstevel@tonic-gate } 3017c478bd9Sstevel@tonic-gate 302da6c28aaSamw /* ARGSUSED */ 3037c478bd9Sstevel@tonic-gate static int 304da6c28aaSamw nfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 305da6c28aaSamw caller_context_t *ct) 3067c478bd9Sstevel@tonic-gate { 3077c478bd9Sstevel@tonic-gate rnode_t *rp; 3087c478bd9Sstevel@tonic-gate int error; 3097c478bd9Sstevel@tonic-gate struct vattr va; 3107c478bd9Sstevel@tonic-gate 3117c478bd9Sstevel@tonic-gate /* 3127c478bd9Sstevel@tonic-gate * zone_enter(2) prevents processes from changing zones with NFS files 3137c478bd9Sstevel@tonic-gate * open; if we happen to get here from the wrong zone we can't do 3147c478bd9Sstevel@tonic-gate * anything over the wire. 3157c478bd9Sstevel@tonic-gate */ 316108322fbScarlsonj if (VTOMI(vp)->mi_zone != nfs_zone()) { 3177c478bd9Sstevel@tonic-gate /* 3187c478bd9Sstevel@tonic-gate * We could attempt to clean up locks, except we're sure 3197c478bd9Sstevel@tonic-gate * that the current process didn't acquire any locks on 3207c478bd9Sstevel@tonic-gate * the file: any attempt to lock a file belong to another zone 3217c478bd9Sstevel@tonic-gate * will fail, and one can't lock an NFS file and then change 3227c478bd9Sstevel@tonic-gate * zones, as that fails too. 3237c478bd9Sstevel@tonic-gate * 3247c478bd9Sstevel@tonic-gate * Returning an error here is the sane thing to do. A 3257c478bd9Sstevel@tonic-gate * subsequent call to VN_RELE() which translates to a 3267c478bd9Sstevel@tonic-gate * nfs_inactive() will clean up state: if the zone of the 3277c478bd9Sstevel@tonic-gate * vnode's origin is still alive and kicking, an async worker 3287c478bd9Sstevel@tonic-gate * thread will handle the request (from the correct zone), and 3297c478bd9Sstevel@tonic-gate * everything (minus the final nfs_getattr_otw() call) should 3307c478bd9Sstevel@tonic-gate * be OK. If the zone is going away nfs_async_inactive() will 3317c478bd9Sstevel@tonic-gate * throw away cached pages inline. 3327c478bd9Sstevel@tonic-gate */ 3337c478bd9Sstevel@tonic-gate return (EIO); 3347c478bd9Sstevel@tonic-gate } 3357c478bd9Sstevel@tonic-gate 3367c478bd9Sstevel@tonic-gate /* 3377c478bd9Sstevel@tonic-gate * If we are using local locking for this filesystem, then 3387c478bd9Sstevel@tonic-gate * release all of the SYSV style record locks. Otherwise, 3397c478bd9Sstevel@tonic-gate * we are doing network locking and we need to release all 3407c478bd9Sstevel@tonic-gate * of the network locks. All of the locks held by this 3417c478bd9Sstevel@tonic-gate * process on this file are released no matter what the 3427c478bd9Sstevel@tonic-gate * incoming reference count is. 3437c478bd9Sstevel@tonic-gate */ 3447c478bd9Sstevel@tonic-gate if (VTOMI(vp)->mi_flags & MI_LLOCK) { 3457c478bd9Sstevel@tonic-gate cleanlocks(vp, ttoproc(curthread)->p_pid, 0); 3467c478bd9Sstevel@tonic-gate cleanshares(vp, ttoproc(curthread)->p_pid); 3477c478bd9Sstevel@tonic-gate } else 3487c478bd9Sstevel@tonic-gate nfs_lockrelease(vp, flag, offset, cr); 3497c478bd9Sstevel@tonic-gate 3507c478bd9Sstevel@tonic-gate if (count > 1) 3517c478bd9Sstevel@tonic-gate return (0); 3527c478bd9Sstevel@tonic-gate 3537c478bd9Sstevel@tonic-gate /* 3547c478bd9Sstevel@tonic-gate * If the file has been `unlinked', then purge the 3557c478bd9Sstevel@tonic-gate * DNLC so that this vnode will get reycled quicker 3567c478bd9Sstevel@tonic-gate * and the .nfs* file on the server will get removed. 3577c478bd9Sstevel@tonic-gate */ 3587c478bd9Sstevel@tonic-gate rp = VTOR(vp); 3597c478bd9Sstevel@tonic-gate if (rp->r_unldvp != NULL) 3607c478bd9Sstevel@tonic-gate dnlc_purge_vp(vp); 3617c478bd9Sstevel@tonic-gate 3627c478bd9Sstevel@tonic-gate /* 3637c478bd9Sstevel@tonic-gate * If the file was open for write and there are pages, 3647c478bd9Sstevel@tonic-gate * then if the file system was mounted using the "no-close- 3657c478bd9Sstevel@tonic-gate * to-open" semantics, then start an asynchronous flush 3667c478bd9Sstevel@tonic-gate * of the all of the pages in the file. 3677c478bd9Sstevel@tonic-gate * else the file system was not mounted using the "no-close- 3687c478bd9Sstevel@tonic-gate * to-open" semantics, then do a synchronous flush and 3697c478bd9Sstevel@tonic-gate * commit of all of the dirty and uncommitted pages. 3707c478bd9Sstevel@tonic-gate * 3717c478bd9Sstevel@tonic-gate * The asynchronous flush of the pages in the "nocto" path 3727c478bd9Sstevel@tonic-gate * mostly just associates a cred pointer with the rnode so 3737c478bd9Sstevel@tonic-gate * writes which happen later will have a better chance of 3747c478bd9Sstevel@tonic-gate * working. It also starts the data being written to the 3757c478bd9Sstevel@tonic-gate * server, but without unnecessarily delaying the application. 3767c478bd9Sstevel@tonic-gate */ 3777c478bd9Sstevel@tonic-gate if ((flag & FWRITE) && vn_has_cached_data(vp)) { 3787c478bd9Sstevel@tonic-gate if ((VTOMI(vp)->mi_flags & MI_NOCTO)) { 379da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, B_ASYNC, 380da6c28aaSamw cr, ct); 3817c478bd9Sstevel@tonic-gate if (error == EAGAIN) 3827c478bd9Sstevel@tonic-gate error = 0; 3837c478bd9Sstevel@tonic-gate } else 384da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct); 3857c478bd9Sstevel@tonic-gate if (!error) { 3867c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 3877c478bd9Sstevel@tonic-gate error = rp->r_error; 3887c478bd9Sstevel@tonic-gate rp->r_error = 0; 3897c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3907c478bd9Sstevel@tonic-gate } 3917c478bd9Sstevel@tonic-gate } else { 3927c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 3937c478bd9Sstevel@tonic-gate error = rp->r_error; 3947c478bd9Sstevel@tonic-gate rp->r_error = 0; 3957c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3967c478bd9Sstevel@tonic-gate } 3977c478bd9Sstevel@tonic-gate 3987c478bd9Sstevel@tonic-gate /* 3997c478bd9Sstevel@tonic-gate * If RWRITEATTR is set, then issue an over the wire GETATTR to 4007c478bd9Sstevel@tonic-gate * refresh the attribute cache with a set of attributes which 4017c478bd9Sstevel@tonic-gate * weren't returned from a WRITE. This will enable the close- 4027c478bd9Sstevel@tonic-gate * to-open processing to work. 4037c478bd9Sstevel@tonic-gate */ 4047c478bd9Sstevel@tonic-gate if (rp->r_flags & RWRITEATTR) 4057c478bd9Sstevel@tonic-gate (void) nfs_getattr_otw(vp, &va, cr); 4067c478bd9Sstevel@tonic-gate 4077c478bd9Sstevel@tonic-gate return (error); 4087c478bd9Sstevel@tonic-gate } 4097c478bd9Sstevel@tonic-gate 4107c478bd9Sstevel@tonic-gate /* ARGSUSED */ 4117c478bd9Sstevel@tonic-gate static int 4127c478bd9Sstevel@tonic-gate nfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr, 4137c478bd9Sstevel@tonic-gate caller_context_t *ct) 4147c478bd9Sstevel@tonic-gate { 4157c478bd9Sstevel@tonic-gate rnode_t *rp; 4167c478bd9Sstevel@tonic-gate u_offset_t off; 4177c478bd9Sstevel@tonic-gate offset_t diff; 4187c478bd9Sstevel@tonic-gate int on; 4197c478bd9Sstevel@tonic-gate size_t n; 4207c478bd9Sstevel@tonic-gate caddr_t base; 4217c478bd9Sstevel@tonic-gate uint_t flags; 4227c478bd9Sstevel@tonic-gate int error; 4237c478bd9Sstevel@tonic-gate mntinfo_t *mi; 4247c478bd9Sstevel@tonic-gate 4257c478bd9Sstevel@tonic-gate rp = VTOR(vp); 4267c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 4277c478bd9Sstevel@tonic-gate 428108322fbScarlsonj if (nfs_zone() != mi->mi_zone) 4297c478bd9Sstevel@tonic-gate return (EIO); 4307c478bd9Sstevel@tonic-gate 4317c478bd9Sstevel@tonic-gate ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER)); 4327c478bd9Sstevel@tonic-gate 4337c478bd9Sstevel@tonic-gate if (vp->v_type != VREG) 4347c478bd9Sstevel@tonic-gate return (EISDIR); 4357c478bd9Sstevel@tonic-gate 4367c478bd9Sstevel@tonic-gate if (uiop->uio_resid == 0) 4377c478bd9Sstevel@tonic-gate return (0); 4387c478bd9Sstevel@tonic-gate 4397c478bd9Sstevel@tonic-gate if (uiop->uio_loffset > MAXOFF32_T) 4407c478bd9Sstevel@tonic-gate return (EFBIG); 4417c478bd9Sstevel@tonic-gate 4427c478bd9Sstevel@tonic-gate if (uiop->uio_loffset < 0 || 4437c478bd9Sstevel@tonic-gate uiop->uio_loffset + uiop->uio_resid > MAXOFF32_T) 4447c478bd9Sstevel@tonic-gate return (EINVAL); 4457c478bd9Sstevel@tonic-gate 4467c478bd9Sstevel@tonic-gate /* 4477c478bd9Sstevel@tonic-gate * Bypass VM if caching has been disabled (e.g., locking) or if 4487c478bd9Sstevel@tonic-gate * using client-side direct I/O and the file is not mmap'd and 4497c478bd9Sstevel@tonic-gate * there are no cached pages. 4507c478bd9Sstevel@tonic-gate */ 4517c478bd9Sstevel@tonic-gate if ((vp->v_flag & VNOCACHE) || 4527c478bd9Sstevel@tonic-gate (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) && 4537c478bd9Sstevel@tonic-gate rp->r_mapcnt == 0 && !vn_has_cached_data(vp))) { 4547c478bd9Sstevel@tonic-gate size_t bufsize; 4557c478bd9Sstevel@tonic-gate size_t resid = 0; 4567c478bd9Sstevel@tonic-gate 4577c478bd9Sstevel@tonic-gate /* 4587c478bd9Sstevel@tonic-gate * Let's try to do read in as large a chunk as we can 4597c478bd9Sstevel@tonic-gate * (Filesystem (NFS client) bsize if possible/needed). 4607c478bd9Sstevel@tonic-gate * For V3, this is 32K and for V2, this is 8K. 4617c478bd9Sstevel@tonic-gate */ 4627c478bd9Sstevel@tonic-gate bufsize = MIN(uiop->uio_resid, VTOMI(vp)->mi_curread); 4637c478bd9Sstevel@tonic-gate base = kmem_alloc(bufsize, KM_SLEEP); 4647c478bd9Sstevel@tonic-gate do { 4657c478bd9Sstevel@tonic-gate n = MIN(uiop->uio_resid, bufsize); 4667c478bd9Sstevel@tonic-gate error = nfsread(vp, base, uiop->uio_offset, n, 4677c478bd9Sstevel@tonic-gate &resid, cr); 4687c478bd9Sstevel@tonic-gate if (!error) { 4697c478bd9Sstevel@tonic-gate n -= resid; 4707c478bd9Sstevel@tonic-gate error = uiomove(base, n, UIO_READ, uiop); 4717c478bd9Sstevel@tonic-gate } 4727c478bd9Sstevel@tonic-gate } while (!error && uiop->uio_resid > 0 && n > 0); 4737c478bd9Sstevel@tonic-gate kmem_free(base, bufsize); 4747c478bd9Sstevel@tonic-gate return (error); 4757c478bd9Sstevel@tonic-gate } 4767c478bd9Sstevel@tonic-gate 4777c478bd9Sstevel@tonic-gate error = 0; 4787c478bd9Sstevel@tonic-gate 4797c478bd9Sstevel@tonic-gate do { 4807c478bd9Sstevel@tonic-gate off = uiop->uio_loffset & MAXBMASK; /* mapping offset */ 4817c478bd9Sstevel@tonic-gate on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */ 4827c478bd9Sstevel@tonic-gate n = MIN(MAXBSIZE - on, uiop->uio_resid); 4837c478bd9Sstevel@tonic-gate 4847c478bd9Sstevel@tonic-gate error = nfs_validate_caches(vp, cr); 4857c478bd9Sstevel@tonic-gate if (error) 4867c478bd9Sstevel@tonic-gate break; 4877c478bd9Sstevel@tonic-gate 4887c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 4895e4df02aSvv149972 while (rp->r_flags & RINCACHEPURGE) { 4905e4df02aSvv149972 if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) { 4915e4df02aSvv149972 mutex_exit(&rp->r_statelock); 4925e4df02aSvv149972 return (EINTR); 4935e4df02aSvv149972 } 4945e4df02aSvv149972 } 4957c478bd9Sstevel@tonic-gate diff = rp->r_size - uiop->uio_loffset; 4967c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 4977c478bd9Sstevel@tonic-gate if (diff <= 0) 4987c478bd9Sstevel@tonic-gate break; 4997c478bd9Sstevel@tonic-gate if (diff < n) 5007c478bd9Sstevel@tonic-gate n = (size_t)diff; 5017c478bd9Sstevel@tonic-gate 502a5652762Spraks if (vpm_enable) { 503a5652762Spraks /* 504a5652762Spraks * Copy data. 505a5652762Spraks */ 506a5652762Spraks error = vpm_data_copy(vp, off + on, n, uiop, 507a5652762Spraks 1, NULL, 0, S_READ); 508a5652762Spraks } else { 509a5652762Spraks base = segmap_getmapflt(segkmap, vp, off + on, n, 510a5652762Spraks 1, S_READ); 5117c478bd9Sstevel@tonic-gate error = uiomove(base + on, n, UIO_READ, uiop); 512a5652762Spraks } 5137c478bd9Sstevel@tonic-gate 5147c478bd9Sstevel@tonic-gate if (!error) { 5157c478bd9Sstevel@tonic-gate /* 5167c478bd9Sstevel@tonic-gate * If read a whole block or read to eof, 5177c478bd9Sstevel@tonic-gate * won't need this buffer again soon. 5187c478bd9Sstevel@tonic-gate */ 5197c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 5207c478bd9Sstevel@tonic-gate if (n + on == MAXBSIZE || 5217c478bd9Sstevel@tonic-gate uiop->uio_loffset == rp->r_size) 5227c478bd9Sstevel@tonic-gate flags = SM_DONTNEED; 5237c478bd9Sstevel@tonic-gate else 5247c478bd9Sstevel@tonic-gate flags = 0; 5257c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 526a5652762Spraks if (vpm_enable) { 527a5652762Spraks error = vpm_sync_pages(vp, off, n, flags); 528a5652762Spraks } else { 5297c478bd9Sstevel@tonic-gate error = segmap_release(segkmap, base, flags); 530a5652762Spraks } 531a5652762Spraks } else { 532a5652762Spraks if (vpm_enable) { 533a5652762Spraks (void) vpm_sync_pages(vp, off, n, 0); 534a5652762Spraks } else { 5357c478bd9Sstevel@tonic-gate (void) segmap_release(segkmap, base, 0); 536a5652762Spraks } 537a5652762Spraks } 5387c478bd9Sstevel@tonic-gate } while (!error && uiop->uio_resid > 0); 5397c478bd9Sstevel@tonic-gate 5407c478bd9Sstevel@tonic-gate return (error); 5417c478bd9Sstevel@tonic-gate } 5427c478bd9Sstevel@tonic-gate 5437c478bd9Sstevel@tonic-gate /* ARGSUSED */ 5447c478bd9Sstevel@tonic-gate static int 5457c478bd9Sstevel@tonic-gate nfs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr, 5467c478bd9Sstevel@tonic-gate caller_context_t *ct) 5477c478bd9Sstevel@tonic-gate { 5487c478bd9Sstevel@tonic-gate rnode_t *rp; 5497c478bd9Sstevel@tonic-gate u_offset_t off; 5507c478bd9Sstevel@tonic-gate caddr_t base; 5517c478bd9Sstevel@tonic-gate uint_t flags; 5527c478bd9Sstevel@tonic-gate int remainder; 5537c478bd9Sstevel@tonic-gate size_t n; 5547c478bd9Sstevel@tonic-gate int on; 5557c478bd9Sstevel@tonic-gate int error; 5567c478bd9Sstevel@tonic-gate int resid; 5577c478bd9Sstevel@tonic-gate offset_t offset; 5587c478bd9Sstevel@tonic-gate rlim_t limit; 5597c478bd9Sstevel@tonic-gate mntinfo_t *mi; 5607c478bd9Sstevel@tonic-gate 5617c478bd9Sstevel@tonic-gate rp = VTOR(vp); 5627c478bd9Sstevel@tonic-gate 5637c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 564108322fbScarlsonj if (nfs_zone() != mi->mi_zone) 5657c478bd9Sstevel@tonic-gate return (EIO); 5667c478bd9Sstevel@tonic-gate if (vp->v_type != VREG) 5677c478bd9Sstevel@tonic-gate return (EISDIR); 5687c478bd9Sstevel@tonic-gate 5697c478bd9Sstevel@tonic-gate if (uiop->uio_resid == 0) 5707c478bd9Sstevel@tonic-gate return (0); 5717c478bd9Sstevel@tonic-gate 5727c478bd9Sstevel@tonic-gate if (ioflag & FAPPEND) { 5737c478bd9Sstevel@tonic-gate struct vattr va; 5747c478bd9Sstevel@tonic-gate 5757c478bd9Sstevel@tonic-gate /* 5767c478bd9Sstevel@tonic-gate * Must serialize if appending. 5777c478bd9Sstevel@tonic-gate */ 5787c478bd9Sstevel@tonic-gate if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) { 5797c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_rwlock); 5807c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, 5817c478bd9Sstevel@tonic-gate INTR(vp))) 5827c478bd9Sstevel@tonic-gate return (EINTR); 5837c478bd9Sstevel@tonic-gate } 5847c478bd9Sstevel@tonic-gate 5857c478bd9Sstevel@tonic-gate va.va_mask = AT_SIZE; 5867c478bd9Sstevel@tonic-gate error = nfsgetattr(vp, &va, cr); 5877c478bd9Sstevel@tonic-gate if (error) 5887c478bd9Sstevel@tonic-gate return (error); 5897c478bd9Sstevel@tonic-gate uiop->uio_loffset = va.va_size; 5907c478bd9Sstevel@tonic-gate } 5917c478bd9Sstevel@tonic-gate 5927c478bd9Sstevel@tonic-gate if (uiop->uio_loffset > MAXOFF32_T) 5937c478bd9Sstevel@tonic-gate return (EFBIG); 5947c478bd9Sstevel@tonic-gate 5957c478bd9Sstevel@tonic-gate offset = uiop->uio_loffset + uiop->uio_resid; 5967c478bd9Sstevel@tonic-gate 5977c478bd9Sstevel@tonic-gate if (uiop->uio_loffset < 0 || offset > MAXOFF32_T) 5987c478bd9Sstevel@tonic-gate return (EINVAL); 5997c478bd9Sstevel@tonic-gate 6007c478bd9Sstevel@tonic-gate if (uiop->uio_llimit > (rlim64_t)MAXOFF32_T) { 6017c478bd9Sstevel@tonic-gate limit = MAXOFF32_T; 6027c478bd9Sstevel@tonic-gate } else { 6037c478bd9Sstevel@tonic-gate limit = (rlim_t)uiop->uio_llimit; 6047c478bd9Sstevel@tonic-gate } 6057c478bd9Sstevel@tonic-gate 6067c478bd9Sstevel@tonic-gate /* 6077c478bd9Sstevel@tonic-gate * Check to make sure that the process will not exceed 6087c478bd9Sstevel@tonic-gate * its limit on file size. It is okay to write up to 6097c478bd9Sstevel@tonic-gate * the limit, but not beyond. Thus, the write which 6107c478bd9Sstevel@tonic-gate * reaches the limit will be short and the next write 6117c478bd9Sstevel@tonic-gate * will return an error. 6127c478bd9Sstevel@tonic-gate */ 6137c478bd9Sstevel@tonic-gate remainder = 0; 6147c478bd9Sstevel@tonic-gate if (offset > limit) { 6157c478bd9Sstevel@tonic-gate remainder = offset - limit; 6167c478bd9Sstevel@tonic-gate uiop->uio_resid = limit - uiop->uio_offset; 6177c478bd9Sstevel@tonic-gate if (uiop->uio_resid <= 0) { 6187c478bd9Sstevel@tonic-gate proc_t *p = ttoproc(curthread); 6197c478bd9Sstevel@tonic-gate 6207c478bd9Sstevel@tonic-gate uiop->uio_resid += remainder; 6217c478bd9Sstevel@tonic-gate mutex_enter(&p->p_lock); 6227c478bd9Sstevel@tonic-gate (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], 6237c478bd9Sstevel@tonic-gate p->p_rctls, p, RCA_UNSAFE_SIGINFO); 6247c478bd9Sstevel@tonic-gate mutex_exit(&p->p_lock); 6257c478bd9Sstevel@tonic-gate return (EFBIG); 6267c478bd9Sstevel@tonic-gate } 6277c478bd9Sstevel@tonic-gate } 6287c478bd9Sstevel@tonic-gate 6297c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) 6307c478bd9Sstevel@tonic-gate return (EINTR); 6317c478bd9Sstevel@tonic-gate 6327c478bd9Sstevel@tonic-gate /* 6337c478bd9Sstevel@tonic-gate * Bypass VM if caching has been disabled (e.g., locking) or if 6347c478bd9Sstevel@tonic-gate * using client-side direct I/O and the file is not mmap'd and 6357c478bd9Sstevel@tonic-gate * there are no cached pages. 6367c478bd9Sstevel@tonic-gate */ 6377c478bd9Sstevel@tonic-gate if ((vp->v_flag & VNOCACHE) || 6387c478bd9Sstevel@tonic-gate (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) && 6397c478bd9Sstevel@tonic-gate rp->r_mapcnt == 0 && !vn_has_cached_data(vp))) { 6407c478bd9Sstevel@tonic-gate size_t bufsize; 6417c478bd9Sstevel@tonic-gate int count; 6427c478bd9Sstevel@tonic-gate uint_t org_offset; 6437c478bd9Sstevel@tonic-gate 6447c478bd9Sstevel@tonic-gate nfs_fwrite: 6457c478bd9Sstevel@tonic-gate if (rp->r_flags & RSTALE) { 6467c478bd9Sstevel@tonic-gate resid = uiop->uio_resid; 6477c478bd9Sstevel@tonic-gate offset = uiop->uio_loffset; 6487c478bd9Sstevel@tonic-gate error = rp->r_error; 6497c478bd9Sstevel@tonic-gate goto bottom; 6507c478bd9Sstevel@tonic-gate } 6517c478bd9Sstevel@tonic-gate bufsize = MIN(uiop->uio_resid, mi->mi_curwrite); 6527c478bd9Sstevel@tonic-gate base = kmem_alloc(bufsize, KM_SLEEP); 6537c478bd9Sstevel@tonic-gate do { 6547c478bd9Sstevel@tonic-gate resid = uiop->uio_resid; 6557c478bd9Sstevel@tonic-gate offset = uiop->uio_loffset; 6567c478bd9Sstevel@tonic-gate count = MIN(uiop->uio_resid, bufsize); 6577c478bd9Sstevel@tonic-gate org_offset = uiop->uio_offset; 6587c478bd9Sstevel@tonic-gate error = uiomove(base, count, UIO_WRITE, uiop); 6597c478bd9Sstevel@tonic-gate if (!error) { 6607c478bd9Sstevel@tonic-gate error = nfswrite(vp, base, org_offset, 6617c478bd9Sstevel@tonic-gate count, cr); 6627c478bd9Sstevel@tonic-gate } 6637c478bd9Sstevel@tonic-gate } while (!error && uiop->uio_resid > 0); 6647c478bd9Sstevel@tonic-gate kmem_free(base, bufsize); 6657c478bd9Sstevel@tonic-gate goto bottom; 6667c478bd9Sstevel@tonic-gate } 6677c478bd9Sstevel@tonic-gate 6687c478bd9Sstevel@tonic-gate do { 6697c478bd9Sstevel@tonic-gate off = uiop->uio_loffset & MAXBMASK; /* mapping offset */ 6707c478bd9Sstevel@tonic-gate on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */ 6717c478bd9Sstevel@tonic-gate n = MIN(MAXBSIZE - on, uiop->uio_resid); 6727c478bd9Sstevel@tonic-gate 6737c478bd9Sstevel@tonic-gate resid = uiop->uio_resid; 6747c478bd9Sstevel@tonic-gate offset = uiop->uio_loffset; 6757c478bd9Sstevel@tonic-gate 6767c478bd9Sstevel@tonic-gate if (rp->r_flags & RSTALE) { 6777c478bd9Sstevel@tonic-gate error = rp->r_error; 6787c478bd9Sstevel@tonic-gate break; 6797c478bd9Sstevel@tonic-gate } 6807c478bd9Sstevel@tonic-gate 6817c478bd9Sstevel@tonic-gate /* 6827c478bd9Sstevel@tonic-gate * Don't create dirty pages faster than they 6837c478bd9Sstevel@tonic-gate * can be cleaned so that the system doesn't 6847c478bd9Sstevel@tonic-gate * get imbalanced. If the async queue is 6857c478bd9Sstevel@tonic-gate * maxed out, then wait for it to drain before 6867c478bd9Sstevel@tonic-gate * creating more dirty pages. Also, wait for 6877c478bd9Sstevel@tonic-gate * any threads doing pagewalks in the vop_getattr 6887c478bd9Sstevel@tonic-gate * entry points so that they don't block for 6897c478bd9Sstevel@tonic-gate * long periods. 6907c478bd9Sstevel@tonic-gate */ 6917c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 6927c478bd9Sstevel@tonic-gate while ((mi->mi_max_threads != 0 && 6937c478bd9Sstevel@tonic-gate rp->r_awcount > 2 * mi->mi_max_threads) || 6947c478bd9Sstevel@tonic-gate rp->r_gcount > 0) 6957c478bd9Sstevel@tonic-gate cv_wait(&rp->r_cv, &rp->r_statelock); 6967c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 6977c478bd9Sstevel@tonic-gate 698a5652762Spraks if (vpm_enable) { 699a5652762Spraks /* 700a5652762Spraks * It will use kpm mappings, so no need to 701a5652762Spraks * pass an address. 702a5652762Spraks */ 703a5652762Spraks error = writerp(rp, NULL, n, uiop, 0); 704a5652762Spraks } else { 7057c478bd9Sstevel@tonic-gate if (segmap_kpm) { 7067c478bd9Sstevel@tonic-gate int pon = uiop->uio_loffset & PAGEOFFSET; 707a5652762Spraks size_t pn = MIN(PAGESIZE - pon, 708a5652762Spraks uiop->uio_resid); 7097c478bd9Sstevel@tonic-gate int pagecreate; 7107c478bd9Sstevel@tonic-gate 7117c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 7127c478bd9Sstevel@tonic-gate pagecreate = (pon == 0) && (pn == PAGESIZE || 7137c478bd9Sstevel@tonic-gate uiop->uio_loffset + pn >= rp->r_size); 7147c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 7157c478bd9Sstevel@tonic-gate 7167c478bd9Sstevel@tonic-gate base = segmap_getmapflt(segkmap, vp, off + on, 7177c478bd9Sstevel@tonic-gate pn, !pagecreate, S_WRITE); 7187c478bd9Sstevel@tonic-gate 719a5652762Spraks error = writerp(rp, base + pon, n, uiop, 720a5652762Spraks pagecreate); 7217c478bd9Sstevel@tonic-gate 7227c478bd9Sstevel@tonic-gate } else { 7237c478bd9Sstevel@tonic-gate base = segmap_getmapflt(segkmap, vp, off + on, 7247c478bd9Sstevel@tonic-gate n, 0, S_READ); 7257c478bd9Sstevel@tonic-gate error = writerp(rp, base + on, n, uiop, 0); 7267c478bd9Sstevel@tonic-gate } 727a5652762Spraks } 7287c478bd9Sstevel@tonic-gate 7297c478bd9Sstevel@tonic-gate if (!error) { 7307c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_NOAC) 7317c478bd9Sstevel@tonic-gate flags = SM_WRITE; 7327c478bd9Sstevel@tonic-gate else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) { 7337c478bd9Sstevel@tonic-gate /* 7347c478bd9Sstevel@tonic-gate * Have written a whole block. 7357c478bd9Sstevel@tonic-gate * Start an asynchronous write 7367c478bd9Sstevel@tonic-gate * and mark the buffer to 7377c478bd9Sstevel@tonic-gate * indicate that it won't be 7387c478bd9Sstevel@tonic-gate * needed again soon. 7397c478bd9Sstevel@tonic-gate */ 7407c478bd9Sstevel@tonic-gate flags = SM_WRITE | SM_ASYNC | SM_DONTNEED; 7417c478bd9Sstevel@tonic-gate } else 7427c478bd9Sstevel@tonic-gate flags = 0; 7437c478bd9Sstevel@tonic-gate if ((ioflag & (FSYNC|FDSYNC)) || 7447c478bd9Sstevel@tonic-gate (rp->r_flags & ROUTOFSPACE)) { 7457c478bd9Sstevel@tonic-gate flags &= ~SM_ASYNC; 7467c478bd9Sstevel@tonic-gate flags |= SM_WRITE; 7477c478bd9Sstevel@tonic-gate } 748a5652762Spraks if (vpm_enable) { 749a5652762Spraks error = vpm_sync_pages(vp, off, n, flags); 750a5652762Spraks } else { 7517c478bd9Sstevel@tonic-gate error = segmap_release(segkmap, base, flags); 752a5652762Spraks } 753a5652762Spraks } else { 754a5652762Spraks if (vpm_enable) { 755a5652762Spraks (void) vpm_sync_pages(vp, off, n, 0); 7567c478bd9Sstevel@tonic-gate } else { 7577c478bd9Sstevel@tonic-gate (void) segmap_release(segkmap, base, 0); 758a5652762Spraks } 7597c478bd9Sstevel@tonic-gate /* 7607c478bd9Sstevel@tonic-gate * In the event that we got an access error while 7617c478bd9Sstevel@tonic-gate * faulting in a page for a write-only file just 7627c478bd9Sstevel@tonic-gate * force a write. 7637c478bd9Sstevel@tonic-gate */ 7647c478bd9Sstevel@tonic-gate if (error == EACCES) 7657c478bd9Sstevel@tonic-gate goto nfs_fwrite; 7667c478bd9Sstevel@tonic-gate } 7677c478bd9Sstevel@tonic-gate } while (!error && uiop->uio_resid > 0); 7687c478bd9Sstevel@tonic-gate 7697c478bd9Sstevel@tonic-gate bottom: 7707c478bd9Sstevel@tonic-gate if (error) { 7717c478bd9Sstevel@tonic-gate uiop->uio_resid = resid + remainder; 7727c478bd9Sstevel@tonic-gate uiop->uio_loffset = offset; 7737c478bd9Sstevel@tonic-gate } else 7747c478bd9Sstevel@tonic-gate uiop->uio_resid += remainder; 7757c478bd9Sstevel@tonic-gate 7767c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_lkserlock); 7777c478bd9Sstevel@tonic-gate 7787c478bd9Sstevel@tonic-gate return (error); 7797c478bd9Sstevel@tonic-gate } 7807c478bd9Sstevel@tonic-gate 7817c478bd9Sstevel@tonic-gate /* 7827c478bd9Sstevel@tonic-gate * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED} 7837c478bd9Sstevel@tonic-gate */ 7847c478bd9Sstevel@tonic-gate static int 7857c478bd9Sstevel@tonic-gate nfs_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len, 7867c478bd9Sstevel@tonic-gate int flags, cred_t *cr) 7877c478bd9Sstevel@tonic-gate { 7887c478bd9Sstevel@tonic-gate struct buf *bp; 7897c478bd9Sstevel@tonic-gate int error; 7907c478bd9Sstevel@tonic-gate 791108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 7927c478bd9Sstevel@tonic-gate bp = pageio_setup(pp, len, vp, flags); 7937c478bd9Sstevel@tonic-gate ASSERT(bp != NULL); 7947c478bd9Sstevel@tonic-gate 7957c478bd9Sstevel@tonic-gate /* 7967c478bd9Sstevel@tonic-gate * pageio_setup should have set b_addr to 0. This 7977c478bd9Sstevel@tonic-gate * is correct since we want to do I/O on a page 7987c478bd9Sstevel@tonic-gate * boundary. bp_mapin will use this addr to calculate 7997c478bd9Sstevel@tonic-gate * an offset, and then set b_addr to the kernel virtual 8007c478bd9Sstevel@tonic-gate * address it allocated for us. 8017c478bd9Sstevel@tonic-gate */ 8027c478bd9Sstevel@tonic-gate ASSERT(bp->b_un.b_addr == 0); 8037c478bd9Sstevel@tonic-gate 8047c478bd9Sstevel@tonic-gate bp->b_edev = 0; 8057c478bd9Sstevel@tonic-gate bp->b_dev = 0; 8067c478bd9Sstevel@tonic-gate bp->b_lblkno = lbtodb(off); 8077c478bd9Sstevel@tonic-gate bp->b_file = vp; 8087c478bd9Sstevel@tonic-gate bp->b_offset = (offset_t)off; 8097c478bd9Sstevel@tonic-gate bp_mapin(bp); 8107c478bd9Sstevel@tonic-gate 8117c478bd9Sstevel@tonic-gate error = nfs_bio(bp, cr); 8127c478bd9Sstevel@tonic-gate 8137c478bd9Sstevel@tonic-gate bp_mapout(bp); 8147c478bd9Sstevel@tonic-gate pageio_done(bp); 8157c478bd9Sstevel@tonic-gate 8167c478bd9Sstevel@tonic-gate return (error); 8177c478bd9Sstevel@tonic-gate } 8187c478bd9Sstevel@tonic-gate 8197c478bd9Sstevel@tonic-gate /* 8207c478bd9Sstevel@tonic-gate * Write to file. Writes to remote server in largest size 8217c478bd9Sstevel@tonic-gate * chunks that the server can handle. Write is synchronous. 8227c478bd9Sstevel@tonic-gate */ 8237c478bd9Sstevel@tonic-gate static int 8247c478bd9Sstevel@tonic-gate nfswrite(vnode_t *vp, caddr_t base, uint_t offset, int count, cred_t *cr) 8257c478bd9Sstevel@tonic-gate { 8267c478bd9Sstevel@tonic-gate rnode_t *rp; 8277c478bd9Sstevel@tonic-gate mntinfo_t *mi; 8287c478bd9Sstevel@tonic-gate struct nfswriteargs wa; 8297c478bd9Sstevel@tonic-gate struct nfsattrstat ns; 8307c478bd9Sstevel@tonic-gate int error; 8317c478bd9Sstevel@tonic-gate int tsize; 8327c478bd9Sstevel@tonic-gate int douprintf; 8337c478bd9Sstevel@tonic-gate 8347c478bd9Sstevel@tonic-gate douprintf = 1; 8357c478bd9Sstevel@tonic-gate 8367c478bd9Sstevel@tonic-gate rp = VTOR(vp); 8377c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 8387c478bd9Sstevel@tonic-gate 839108322fbScarlsonj ASSERT(nfs_zone() == mi->mi_zone); 8407c478bd9Sstevel@tonic-gate 8417c478bd9Sstevel@tonic-gate wa.wa_args = &wa.wa_args_buf; 8427c478bd9Sstevel@tonic-gate wa.wa_fhandle = *VTOFH(vp); 8437c478bd9Sstevel@tonic-gate 8447c478bd9Sstevel@tonic-gate do { 8457c478bd9Sstevel@tonic-gate tsize = MIN(mi->mi_curwrite, count); 8467c478bd9Sstevel@tonic-gate wa.wa_data = base; 8477c478bd9Sstevel@tonic-gate wa.wa_begoff = offset; 8487c478bd9Sstevel@tonic-gate wa.wa_totcount = tsize; 8497c478bd9Sstevel@tonic-gate wa.wa_count = tsize; 8507c478bd9Sstevel@tonic-gate wa.wa_offset = offset; 8517c478bd9Sstevel@tonic-gate 8527c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 8537c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 8547c478bd9Sstevel@tonic-gate kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 8557c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 8567c478bd9Sstevel@tonic-gate } 8577c478bd9Sstevel@tonic-gate wa.wa_mblk = NULL; 8587c478bd9Sstevel@tonic-gate do { 8597c478bd9Sstevel@tonic-gate error = rfs2call(mi, RFS_WRITE, 8607c478bd9Sstevel@tonic-gate xdr_writeargs, (caddr_t)&wa, 8617c478bd9Sstevel@tonic-gate xdr_attrstat, (caddr_t)&ns, cr, 8627c478bd9Sstevel@tonic-gate &douprintf, &ns.ns_status, 0, NULL); 8637c478bd9Sstevel@tonic-gate } while (error == ENFS_TRYAGAIN); 8647c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 8657c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 8667c478bd9Sstevel@tonic-gate kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 8677c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 8687c478bd9Sstevel@tonic-gate } 8697c478bd9Sstevel@tonic-gate 8707c478bd9Sstevel@tonic-gate if (!error) { 8717c478bd9Sstevel@tonic-gate error = geterrno(ns.ns_status); 8727c478bd9Sstevel@tonic-gate /* 8737c478bd9Sstevel@tonic-gate * Can't check for stale fhandle and purge caches 8747c478bd9Sstevel@tonic-gate * here because pages are held by nfs_getpage. 8757c478bd9Sstevel@tonic-gate * Just mark the attribute cache as timed out 8767c478bd9Sstevel@tonic-gate * and set RWRITEATTR to indicate that the file 8777c478bd9Sstevel@tonic-gate * was modified with a WRITE operation. 8787c478bd9Sstevel@tonic-gate */ 8797c478bd9Sstevel@tonic-gate if (!error) { 8807c478bd9Sstevel@tonic-gate count -= tsize; 8817c478bd9Sstevel@tonic-gate base += tsize; 8827c478bd9Sstevel@tonic-gate offset += tsize; 8837c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 8847c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 885da6c28aaSamw KSTAT_IO_PTR(mi->mi_io_kstats)-> 886da6c28aaSamw writes++; 887da6c28aaSamw KSTAT_IO_PTR(mi->mi_io_kstats)-> 888da6c28aaSamw nwritten += tsize; 8897c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 8907c478bd9Sstevel@tonic-gate } 8917c478bd9Sstevel@tonic-gate lwp_stat_update(LWP_STAT_OUBLK, 1); 8927c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 8937c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE_LOCKED(rp); 8947c478bd9Sstevel@tonic-gate rp->r_flags |= RWRITEATTR; 8957c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 8967c478bd9Sstevel@tonic-gate } 8977c478bd9Sstevel@tonic-gate } 8987c478bd9Sstevel@tonic-gate } while (!error && count); 8997c478bd9Sstevel@tonic-gate 9007c478bd9Sstevel@tonic-gate return (error); 9017c478bd9Sstevel@tonic-gate } 9027c478bd9Sstevel@tonic-gate 9037c478bd9Sstevel@tonic-gate /* 9047c478bd9Sstevel@tonic-gate * Read from a file. Reads data in largest chunks our interface can handle. 9057c478bd9Sstevel@tonic-gate */ 9067c478bd9Sstevel@tonic-gate static int 9077c478bd9Sstevel@tonic-gate nfsread(vnode_t *vp, caddr_t base, uint_t offset, int count, size_t *residp, 9087c478bd9Sstevel@tonic-gate cred_t *cr) 9097c478bd9Sstevel@tonic-gate { 9107c478bd9Sstevel@tonic-gate mntinfo_t *mi; 9117c478bd9Sstevel@tonic-gate struct nfsreadargs ra; 9127c478bd9Sstevel@tonic-gate struct nfsrdresult rr; 9137c478bd9Sstevel@tonic-gate int tsize; 9147c478bd9Sstevel@tonic-gate int error; 9157c478bd9Sstevel@tonic-gate int douprintf; 9167c478bd9Sstevel@tonic-gate failinfo_t fi; 9177c478bd9Sstevel@tonic-gate rnode_t *rp; 9187c478bd9Sstevel@tonic-gate struct vattr va; 9197c478bd9Sstevel@tonic-gate hrtime_t t; 9207c478bd9Sstevel@tonic-gate 9217c478bd9Sstevel@tonic-gate rp = VTOR(vp); 9227c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 9237c478bd9Sstevel@tonic-gate 924108322fbScarlsonj ASSERT(nfs_zone() == mi->mi_zone); 9257c478bd9Sstevel@tonic-gate 9267c478bd9Sstevel@tonic-gate douprintf = 1; 9277c478bd9Sstevel@tonic-gate 9287c478bd9Sstevel@tonic-gate ra.ra_fhandle = *VTOFH(vp); 9297c478bd9Sstevel@tonic-gate 9307c478bd9Sstevel@tonic-gate fi.vp = vp; 9317c478bd9Sstevel@tonic-gate fi.fhp = (caddr_t)&ra.ra_fhandle; 9327c478bd9Sstevel@tonic-gate fi.copyproc = nfscopyfh; 9337c478bd9Sstevel@tonic-gate fi.lookupproc = nfslookup; 9347c478bd9Sstevel@tonic-gate fi.xattrdirproc = acl_getxattrdir2; 9357c478bd9Sstevel@tonic-gate 9367c478bd9Sstevel@tonic-gate do { 9377c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 9387c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 9397c478bd9Sstevel@tonic-gate kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 9407c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 9417c478bd9Sstevel@tonic-gate } 9427c478bd9Sstevel@tonic-gate 9437c478bd9Sstevel@tonic-gate do { 9447c478bd9Sstevel@tonic-gate tsize = MIN(mi->mi_curread, count); 9457c478bd9Sstevel@tonic-gate rr.rr_data = base; 9467c478bd9Sstevel@tonic-gate ra.ra_offset = offset; 9477c478bd9Sstevel@tonic-gate ra.ra_totcount = tsize; 9487c478bd9Sstevel@tonic-gate ra.ra_count = tsize; 9497c478bd9Sstevel@tonic-gate t = gethrtime(); 9507c478bd9Sstevel@tonic-gate error = rfs2call(mi, RFS_READ, 9517c478bd9Sstevel@tonic-gate xdr_readargs, (caddr_t)&ra, 9527c478bd9Sstevel@tonic-gate xdr_rdresult, (caddr_t)&rr, cr, 9537c478bd9Sstevel@tonic-gate &douprintf, &rr.rr_status, 0, &fi); 9547c478bd9Sstevel@tonic-gate } while (error == ENFS_TRYAGAIN); 9557c478bd9Sstevel@tonic-gate 9567c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 9577c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 9587c478bd9Sstevel@tonic-gate kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 9597c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 9607c478bd9Sstevel@tonic-gate } 9617c478bd9Sstevel@tonic-gate 9627c478bd9Sstevel@tonic-gate if (!error) { 9637c478bd9Sstevel@tonic-gate error = geterrno(rr.rr_status); 9647c478bd9Sstevel@tonic-gate if (!error) { 9657c478bd9Sstevel@tonic-gate count -= rr.rr_count; 9667c478bd9Sstevel@tonic-gate base += rr.rr_count; 9677c478bd9Sstevel@tonic-gate offset += rr.rr_count; 9687c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 9697c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 9707c478bd9Sstevel@tonic-gate KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 9717c478bd9Sstevel@tonic-gate KSTAT_IO_PTR(mi->mi_io_kstats)->nread += 9727c478bd9Sstevel@tonic-gate rr.rr_count; 9737c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 9747c478bd9Sstevel@tonic-gate } 9757c478bd9Sstevel@tonic-gate lwp_stat_update(LWP_STAT_INBLK, 1); 9767c478bd9Sstevel@tonic-gate } 9777c478bd9Sstevel@tonic-gate } 9787c478bd9Sstevel@tonic-gate } while (!error && count && rr.rr_count == tsize); 9797c478bd9Sstevel@tonic-gate 9807c478bd9Sstevel@tonic-gate *residp = count; 9817c478bd9Sstevel@tonic-gate 9827c478bd9Sstevel@tonic-gate if (!error) { 9837c478bd9Sstevel@tonic-gate /* 9847c478bd9Sstevel@tonic-gate * Since no error occurred, we have the current 9857c478bd9Sstevel@tonic-gate * attributes and we need to do a cache check and then 9867c478bd9Sstevel@tonic-gate * potentially update the cached attributes. We can't 9877c478bd9Sstevel@tonic-gate * use the normal attribute check and cache mechanisms 9887c478bd9Sstevel@tonic-gate * because they might cause a cache flush which would 9897c478bd9Sstevel@tonic-gate * deadlock. Instead, we just check the cache to see 9907c478bd9Sstevel@tonic-gate * if the attributes have changed. If it is, then we 9917c478bd9Sstevel@tonic-gate * just mark the attributes as out of date. The next 9927c478bd9Sstevel@tonic-gate * time that the attributes are checked, they will be 9937c478bd9Sstevel@tonic-gate * out of date, new attributes will be fetched, and 9947c478bd9Sstevel@tonic-gate * the page cache will be flushed. If the attributes 9957c478bd9Sstevel@tonic-gate * weren't changed, then we just update the cached 9967c478bd9Sstevel@tonic-gate * attributes with these attributes. 9977c478bd9Sstevel@tonic-gate */ 9987c478bd9Sstevel@tonic-gate /* 9997c478bd9Sstevel@tonic-gate * If NFS_ACL is supported on the server, then the 10007c478bd9Sstevel@tonic-gate * attributes returned by server may have minimal 10017c478bd9Sstevel@tonic-gate * permissions sometimes denying access to users having 10027c478bd9Sstevel@tonic-gate * proper access. To get the proper attributes, mark 10037c478bd9Sstevel@tonic-gate * the attributes as expired so that they will be 10047c478bd9Sstevel@tonic-gate * regotten via the NFS_ACL GETATTR2 procedure. 10057c478bd9Sstevel@tonic-gate */ 10067c478bd9Sstevel@tonic-gate error = nattr_to_vattr(vp, &rr.rr_attr, &va); 10077c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 10087c478bd9Sstevel@tonic-gate if (error || !CACHE_VALID(rp, va.va_mtime, va.va_size) || 10097c478bd9Sstevel@tonic-gate (mi->mi_flags & MI_ACL)) { 10107c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 10117c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); 10127c478bd9Sstevel@tonic-gate } else { 10137c478bd9Sstevel@tonic-gate if (rp->r_mtime <= t) { 10147c478bd9Sstevel@tonic-gate nfs_attrcache_va(vp, &va); 10157c478bd9Sstevel@tonic-gate } 10167c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 10177c478bd9Sstevel@tonic-gate } 10187c478bd9Sstevel@tonic-gate } 10197c478bd9Sstevel@tonic-gate 10207c478bd9Sstevel@tonic-gate return (error); 10217c478bd9Sstevel@tonic-gate } 10227c478bd9Sstevel@tonic-gate 10237c478bd9Sstevel@tonic-gate /* ARGSUSED */ 10247c478bd9Sstevel@tonic-gate static int 1025da6c28aaSamw nfs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp, 1026da6c28aaSamw caller_context_t *ct) 10277c478bd9Sstevel@tonic-gate { 10287c478bd9Sstevel@tonic-gate 1029108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 10307c478bd9Sstevel@tonic-gate return (EIO); 10317c478bd9Sstevel@tonic-gate switch (cmd) { 10327c478bd9Sstevel@tonic-gate case _FIODIRECTIO: 10337c478bd9Sstevel@tonic-gate return (nfs_directio(vp, (int)arg, cr)); 10347c478bd9Sstevel@tonic-gate default: 10357c478bd9Sstevel@tonic-gate return (ENOTTY); 10367c478bd9Sstevel@tonic-gate } 10377c478bd9Sstevel@tonic-gate } 10387c478bd9Sstevel@tonic-gate 1039da6c28aaSamw /* ARGSUSED */ 10407c478bd9Sstevel@tonic-gate static int 1041da6c28aaSamw nfs_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 1042da6c28aaSamw caller_context_t *ct) 10437c478bd9Sstevel@tonic-gate { 10447c478bd9Sstevel@tonic-gate int error; 10457c478bd9Sstevel@tonic-gate rnode_t *rp; 10467c478bd9Sstevel@tonic-gate 1047108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 10487c478bd9Sstevel@tonic-gate return (EIO); 10497c478bd9Sstevel@tonic-gate /* 10507c478bd9Sstevel@tonic-gate * If it has been specified that the return value will 10517c478bd9Sstevel@tonic-gate * just be used as a hint, and we are only being asked 10527c478bd9Sstevel@tonic-gate * for size, fsid or rdevid, then return the client's 10537c478bd9Sstevel@tonic-gate * notion of these values without checking to make sure 10547c478bd9Sstevel@tonic-gate * that the attribute cache is up to date. 10557c478bd9Sstevel@tonic-gate * The whole point is to avoid an over the wire GETATTR 10567c478bd9Sstevel@tonic-gate * call. 10577c478bd9Sstevel@tonic-gate */ 10587c478bd9Sstevel@tonic-gate rp = VTOR(vp); 10597c478bd9Sstevel@tonic-gate if (flags & ATTR_HINT) { 10607c478bd9Sstevel@tonic-gate if (vap->va_mask == 10617c478bd9Sstevel@tonic-gate (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) { 10627c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 10637c478bd9Sstevel@tonic-gate if (vap->va_mask | AT_SIZE) 10647c478bd9Sstevel@tonic-gate vap->va_size = rp->r_size; 10657c478bd9Sstevel@tonic-gate if (vap->va_mask | AT_FSID) 10667c478bd9Sstevel@tonic-gate vap->va_fsid = rp->r_attr.va_fsid; 10677c478bd9Sstevel@tonic-gate if (vap->va_mask | AT_RDEV) 10687c478bd9Sstevel@tonic-gate vap->va_rdev = rp->r_attr.va_rdev; 10697c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 10707c478bd9Sstevel@tonic-gate return (0); 10717c478bd9Sstevel@tonic-gate } 10727c478bd9Sstevel@tonic-gate } 10737c478bd9Sstevel@tonic-gate 10747c478bd9Sstevel@tonic-gate /* 10757c478bd9Sstevel@tonic-gate * Only need to flush pages if asking for the mtime 10767c478bd9Sstevel@tonic-gate * and if there any dirty pages or any outstanding 10777c478bd9Sstevel@tonic-gate * asynchronous (write) requests for this file. 10787c478bd9Sstevel@tonic-gate */ 10797c478bd9Sstevel@tonic-gate if (vap->va_mask & AT_MTIME) { 10807c478bd9Sstevel@tonic-gate if (vn_has_cached_data(vp) && 10817c478bd9Sstevel@tonic-gate ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) { 10827c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 10837c478bd9Sstevel@tonic-gate rp->r_gcount++; 10847c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 1085da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct); 10867c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 10877c478bd9Sstevel@tonic-gate if (error && (error == ENOSPC || error == EDQUOT)) { 10887c478bd9Sstevel@tonic-gate if (!rp->r_error) 10897c478bd9Sstevel@tonic-gate rp->r_error = error; 10907c478bd9Sstevel@tonic-gate } 10917c478bd9Sstevel@tonic-gate if (--rp->r_gcount == 0) 10927c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 10937c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 10947c478bd9Sstevel@tonic-gate } 10957c478bd9Sstevel@tonic-gate } 10967c478bd9Sstevel@tonic-gate 10977c478bd9Sstevel@tonic-gate return (nfsgetattr(vp, vap, cr)); 10987c478bd9Sstevel@tonic-gate } 10997c478bd9Sstevel@tonic-gate 11007c478bd9Sstevel@tonic-gate /*ARGSUSED4*/ 11017c478bd9Sstevel@tonic-gate static int 11027c478bd9Sstevel@tonic-gate nfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 11037c478bd9Sstevel@tonic-gate caller_context_t *ct) 11047c478bd9Sstevel@tonic-gate { 11057c478bd9Sstevel@tonic-gate int error; 11067c478bd9Sstevel@tonic-gate uint_t mask; 11077c478bd9Sstevel@tonic-gate struct vattr va; 11087c478bd9Sstevel@tonic-gate 11097c478bd9Sstevel@tonic-gate mask = vap->va_mask; 11107c478bd9Sstevel@tonic-gate 11117c478bd9Sstevel@tonic-gate if (mask & AT_NOSET) 11127c478bd9Sstevel@tonic-gate return (EINVAL); 11137c478bd9Sstevel@tonic-gate 11147c478bd9Sstevel@tonic-gate if ((mask & AT_SIZE) && 11157c478bd9Sstevel@tonic-gate vap->va_type == VREG && 11167c478bd9Sstevel@tonic-gate vap->va_size > MAXOFF32_T) 11177c478bd9Sstevel@tonic-gate return (EFBIG); 11187c478bd9Sstevel@tonic-gate 1119108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 11207c478bd9Sstevel@tonic-gate return (EIO); 11217c478bd9Sstevel@tonic-gate 11227c478bd9Sstevel@tonic-gate va.va_mask = AT_UID | AT_MODE; 11237c478bd9Sstevel@tonic-gate 11247c478bd9Sstevel@tonic-gate error = nfsgetattr(vp, &va, cr); 11257c478bd9Sstevel@tonic-gate if (error) 11267c478bd9Sstevel@tonic-gate return (error); 11277c478bd9Sstevel@tonic-gate 11287c478bd9Sstevel@tonic-gate error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs_accessx, 11297c478bd9Sstevel@tonic-gate vp); 11307c478bd9Sstevel@tonic-gate 11317c478bd9Sstevel@tonic-gate if (error) 11327c478bd9Sstevel@tonic-gate return (error); 11337c478bd9Sstevel@tonic-gate 11347c478bd9Sstevel@tonic-gate return (nfssetattr(vp, vap, flags, cr)); 11357c478bd9Sstevel@tonic-gate } 11367c478bd9Sstevel@tonic-gate 11377c478bd9Sstevel@tonic-gate static int 11387c478bd9Sstevel@tonic-gate nfssetattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr) 11397c478bd9Sstevel@tonic-gate { 11407c478bd9Sstevel@tonic-gate int error; 11417c478bd9Sstevel@tonic-gate uint_t mask; 11427c478bd9Sstevel@tonic-gate struct nfssaargs args; 11437c478bd9Sstevel@tonic-gate struct nfsattrstat ns; 11447c478bd9Sstevel@tonic-gate int douprintf; 11457c478bd9Sstevel@tonic-gate rnode_t *rp; 11467c478bd9Sstevel@tonic-gate struct vattr va; 11477c478bd9Sstevel@tonic-gate mode_t omode; 11487c478bd9Sstevel@tonic-gate mntinfo_t *mi; 11497c478bd9Sstevel@tonic-gate vsecattr_t *vsp; 11507c478bd9Sstevel@tonic-gate hrtime_t t; 11517c478bd9Sstevel@tonic-gate 11527c478bd9Sstevel@tonic-gate mask = vap->va_mask; 11537c478bd9Sstevel@tonic-gate 1154108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 11557c478bd9Sstevel@tonic-gate 11567c478bd9Sstevel@tonic-gate rp = VTOR(vp); 11577c478bd9Sstevel@tonic-gate 11587c478bd9Sstevel@tonic-gate /* 11597c478bd9Sstevel@tonic-gate * Only need to flush pages if there are any pages and 11607c478bd9Sstevel@tonic-gate * if the file is marked as dirty in some fashion. The 11617c478bd9Sstevel@tonic-gate * file must be flushed so that we can accurately 11627c478bd9Sstevel@tonic-gate * determine the size of the file and the cached data 11637c478bd9Sstevel@tonic-gate * after the SETATTR returns. A file is considered to 11647c478bd9Sstevel@tonic-gate * be dirty if it is either marked with RDIRTY, has 11657c478bd9Sstevel@tonic-gate * outstanding i/o's active, or is mmap'd. In this 11667c478bd9Sstevel@tonic-gate * last case, we can't tell whether there are dirty 11677c478bd9Sstevel@tonic-gate * pages, so we flush just to be sure. 11687c478bd9Sstevel@tonic-gate */ 11697c478bd9Sstevel@tonic-gate if (vn_has_cached_data(vp) && 11707c478bd9Sstevel@tonic-gate ((rp->r_flags & RDIRTY) || 11717c478bd9Sstevel@tonic-gate rp->r_count > 0 || 11727c478bd9Sstevel@tonic-gate rp->r_mapcnt > 0)) { 11737c478bd9Sstevel@tonic-gate ASSERT(vp->v_type != VCHR); 1174da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, NULL); 11757c478bd9Sstevel@tonic-gate if (error && (error == ENOSPC || error == EDQUOT)) { 11767c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 11777c478bd9Sstevel@tonic-gate if (!rp->r_error) 11787c478bd9Sstevel@tonic-gate rp->r_error = error; 11797c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 11807c478bd9Sstevel@tonic-gate } 11817c478bd9Sstevel@tonic-gate } 11827c478bd9Sstevel@tonic-gate 11837c478bd9Sstevel@tonic-gate /* 11847c478bd9Sstevel@tonic-gate * If the system call was utime(2) or utimes(2) and the 11857c478bd9Sstevel@tonic-gate * application did not specify the times, then set the 11867c478bd9Sstevel@tonic-gate * mtime nanosecond field to 1 billion. This will get 11877c478bd9Sstevel@tonic-gate * translated from 1 billion nanoseconds to 1 million 11887c478bd9Sstevel@tonic-gate * microseconds in the over the wire request. The 11897c478bd9Sstevel@tonic-gate * server will use 1 million in the microsecond field 11907c478bd9Sstevel@tonic-gate * to tell whether both the mtime and atime should be 11917c478bd9Sstevel@tonic-gate * set to the server's current time. 11927c478bd9Sstevel@tonic-gate * 11937c478bd9Sstevel@tonic-gate * This is an overload of the protocol and should be 11947c478bd9Sstevel@tonic-gate * documented in the NFS Version 2 protocol specification. 11957c478bd9Sstevel@tonic-gate */ 11967c478bd9Sstevel@tonic-gate if ((mask & AT_MTIME) && !(flags & ATTR_UTIME)) { 11977c478bd9Sstevel@tonic-gate vap->va_mtime.tv_nsec = 1000000000; 11987c478bd9Sstevel@tonic-gate if (NFS_TIME_T_OK(vap->va_mtime.tv_sec) && 11997c478bd9Sstevel@tonic-gate NFS_TIME_T_OK(vap->va_atime.tv_sec)) { 12007c478bd9Sstevel@tonic-gate error = vattr_to_sattr(vap, &args.saa_sa); 12017c478bd9Sstevel@tonic-gate } else { 12027c478bd9Sstevel@tonic-gate /* 12037c478bd9Sstevel@tonic-gate * Use server times. vap time values will not be used. 12047c478bd9Sstevel@tonic-gate * To ensure no time overflow, make sure vap has 12057c478bd9Sstevel@tonic-gate * valid values, but retain the original values. 12067c478bd9Sstevel@tonic-gate */ 12077c478bd9Sstevel@tonic-gate timestruc_t mtime = vap->va_mtime; 12087c478bd9Sstevel@tonic-gate timestruc_t atime = vap->va_atime; 12097c478bd9Sstevel@tonic-gate time_t now; 12107c478bd9Sstevel@tonic-gate 12117c478bd9Sstevel@tonic-gate now = gethrestime_sec(); 12127c478bd9Sstevel@tonic-gate if (NFS_TIME_T_OK(now)) { 12137c478bd9Sstevel@tonic-gate /* Just in case server does not know of this */ 12147c478bd9Sstevel@tonic-gate vap->va_mtime.tv_sec = now; 12157c478bd9Sstevel@tonic-gate vap->va_atime.tv_sec = now; 12167c478bd9Sstevel@tonic-gate } else { 12177c478bd9Sstevel@tonic-gate vap->va_mtime.tv_sec = 0; 12187c478bd9Sstevel@tonic-gate vap->va_atime.tv_sec = 0; 12197c478bd9Sstevel@tonic-gate } 12207c478bd9Sstevel@tonic-gate error = vattr_to_sattr(vap, &args.saa_sa); 12217c478bd9Sstevel@tonic-gate /* set vap times back on */ 12227c478bd9Sstevel@tonic-gate vap->va_mtime = mtime; 12237c478bd9Sstevel@tonic-gate vap->va_atime = atime; 12247c478bd9Sstevel@tonic-gate } 12257c478bd9Sstevel@tonic-gate } else { 12267c478bd9Sstevel@tonic-gate /* Either do not set times or use the client specified times */ 12277c478bd9Sstevel@tonic-gate error = vattr_to_sattr(vap, &args.saa_sa); 12287c478bd9Sstevel@tonic-gate } 12297c478bd9Sstevel@tonic-gate if (error) { 12307c478bd9Sstevel@tonic-gate /* req time field(s) overflow - return immediately */ 12317c478bd9Sstevel@tonic-gate return (error); 12327c478bd9Sstevel@tonic-gate } 12337c478bd9Sstevel@tonic-gate args.saa_fh = *VTOFH(vp); 12347c478bd9Sstevel@tonic-gate 12357c478bd9Sstevel@tonic-gate va.va_mask = AT_MODE; 12367c478bd9Sstevel@tonic-gate error = nfsgetattr(vp, &va, cr); 12377c478bd9Sstevel@tonic-gate if (error) 12387c478bd9Sstevel@tonic-gate return (error); 12397c478bd9Sstevel@tonic-gate omode = va.va_mode; 12407c478bd9Sstevel@tonic-gate 12417c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 12427c478bd9Sstevel@tonic-gate 12437c478bd9Sstevel@tonic-gate douprintf = 1; 12447c478bd9Sstevel@tonic-gate 12457c478bd9Sstevel@tonic-gate t = gethrtime(); 12467c478bd9Sstevel@tonic-gate 12477c478bd9Sstevel@tonic-gate error = rfs2call(mi, RFS_SETATTR, 12487c478bd9Sstevel@tonic-gate xdr_saargs, (caddr_t)&args, 12497c478bd9Sstevel@tonic-gate xdr_attrstat, (caddr_t)&ns, cr, 12507c478bd9Sstevel@tonic-gate &douprintf, &ns.ns_status, 0, NULL); 12517c478bd9Sstevel@tonic-gate 12527c478bd9Sstevel@tonic-gate /* 12537c478bd9Sstevel@tonic-gate * Purge the access cache and ACL cache if changing either the 12547c478bd9Sstevel@tonic-gate * owner of the file, the group owner, or the mode. These may 12557c478bd9Sstevel@tonic-gate * change the access permissions of the file, so purge old 12567c478bd9Sstevel@tonic-gate * information and start over again. 12577c478bd9Sstevel@tonic-gate */ 12587c478bd9Sstevel@tonic-gate if ((mask & (AT_UID | AT_GID | AT_MODE)) && (mi->mi_flags & MI_ACL)) { 12597c478bd9Sstevel@tonic-gate (void) nfs_access_purge_rp(rp); 12607c478bd9Sstevel@tonic-gate if (rp->r_secattr != NULL) { 12617c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 12627c478bd9Sstevel@tonic-gate vsp = rp->r_secattr; 12637c478bd9Sstevel@tonic-gate rp->r_secattr = NULL; 12647c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 12657c478bd9Sstevel@tonic-gate if (vsp != NULL) 12667c478bd9Sstevel@tonic-gate nfs_acl_free(vsp); 12677c478bd9Sstevel@tonic-gate } 12687c478bd9Sstevel@tonic-gate } 12697c478bd9Sstevel@tonic-gate 12707c478bd9Sstevel@tonic-gate if (!error) { 12717c478bd9Sstevel@tonic-gate error = geterrno(ns.ns_status); 12727c478bd9Sstevel@tonic-gate if (!error) { 12737c478bd9Sstevel@tonic-gate /* 12747c478bd9Sstevel@tonic-gate * If changing the size of the file, invalidate 12757c478bd9Sstevel@tonic-gate * any local cached data which is no longer part 12767c478bd9Sstevel@tonic-gate * of the file. We also possibly invalidate the 12777c478bd9Sstevel@tonic-gate * last page in the file. We could use 12787c478bd9Sstevel@tonic-gate * pvn_vpzero(), but this would mark the page as 12797c478bd9Sstevel@tonic-gate * modified and require it to be written back to 12807c478bd9Sstevel@tonic-gate * the server for no particularly good reason. 12817c478bd9Sstevel@tonic-gate * This way, if we access it, then we bring it 12827c478bd9Sstevel@tonic-gate * back in. A read should be cheaper than a 12837c478bd9Sstevel@tonic-gate * write. 12847c478bd9Sstevel@tonic-gate */ 12857c478bd9Sstevel@tonic-gate if (mask & AT_SIZE) { 12867c478bd9Sstevel@tonic-gate nfs_invalidate_pages(vp, 12877c478bd9Sstevel@tonic-gate (vap->va_size & PAGEMASK), cr); 12887c478bd9Sstevel@tonic-gate } 12897c478bd9Sstevel@tonic-gate (void) nfs_cache_fattr(vp, &ns.ns_attr, &va, t, cr); 12907c478bd9Sstevel@tonic-gate /* 12917c478bd9Sstevel@tonic-gate * If NFS_ACL is supported on the server, then the 12927c478bd9Sstevel@tonic-gate * attributes returned by server may have minimal 12937c478bd9Sstevel@tonic-gate * permissions sometimes denying access to users having 12947c478bd9Sstevel@tonic-gate * proper access. To get the proper attributes, mark 12957c478bd9Sstevel@tonic-gate * the attributes as expired so that they will be 12967c478bd9Sstevel@tonic-gate * regotten via the NFS_ACL GETATTR2 procedure. 12977c478bd9Sstevel@tonic-gate */ 12987c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) { 12997c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); 13007c478bd9Sstevel@tonic-gate } 13017c478bd9Sstevel@tonic-gate /* 13027c478bd9Sstevel@tonic-gate * This next check attempts to deal with NFS 13037c478bd9Sstevel@tonic-gate * servers which can not handle increasing 13047c478bd9Sstevel@tonic-gate * the size of the file via setattr. Most 13057c478bd9Sstevel@tonic-gate * of these servers do not return an error, 13067c478bd9Sstevel@tonic-gate * but do not change the size of the file. 13077c478bd9Sstevel@tonic-gate * Hence, this check and then attempt to set 13087c478bd9Sstevel@tonic-gate * the file size by writing 1 byte at the 13097c478bd9Sstevel@tonic-gate * offset of the end of the file that we need. 13107c478bd9Sstevel@tonic-gate */ 13117c478bd9Sstevel@tonic-gate if ((mask & AT_SIZE) && 13127c478bd9Sstevel@tonic-gate ns.ns_attr.na_size < (uint32_t)vap->va_size) { 13137c478bd9Sstevel@tonic-gate char zb = '\0'; 13147c478bd9Sstevel@tonic-gate 13157c478bd9Sstevel@tonic-gate error = nfswrite(vp, &zb, 13167c478bd9Sstevel@tonic-gate vap->va_size - sizeof (zb), 13177c478bd9Sstevel@tonic-gate sizeof (zb), cr); 13187c478bd9Sstevel@tonic-gate } 13197c478bd9Sstevel@tonic-gate /* 13207c478bd9Sstevel@tonic-gate * Some servers will change the mode to clear the setuid 13217c478bd9Sstevel@tonic-gate * and setgid bits when changing the uid or gid. The 13227c478bd9Sstevel@tonic-gate * client needs to compensate appropriately. 13237c478bd9Sstevel@tonic-gate */ 13247c478bd9Sstevel@tonic-gate if (mask & (AT_UID | AT_GID)) { 13257c478bd9Sstevel@tonic-gate int terror; 13267c478bd9Sstevel@tonic-gate 13277c478bd9Sstevel@tonic-gate va.va_mask = AT_MODE; 13287c478bd9Sstevel@tonic-gate terror = nfsgetattr(vp, &va, cr); 13297c478bd9Sstevel@tonic-gate if (!terror && 13307c478bd9Sstevel@tonic-gate (((mask & AT_MODE) && 13317c478bd9Sstevel@tonic-gate va.va_mode != vap->va_mode) || 13327c478bd9Sstevel@tonic-gate (!(mask & AT_MODE) && 13337c478bd9Sstevel@tonic-gate va.va_mode != omode))) { 13347c478bd9Sstevel@tonic-gate va.va_mask = AT_MODE; 13357c478bd9Sstevel@tonic-gate if (mask & AT_MODE) 13367c478bd9Sstevel@tonic-gate va.va_mode = vap->va_mode; 13377c478bd9Sstevel@tonic-gate else 13387c478bd9Sstevel@tonic-gate va.va_mode = omode; 13397c478bd9Sstevel@tonic-gate (void) nfssetattr(vp, &va, 0, cr); 13407c478bd9Sstevel@tonic-gate } 13417c478bd9Sstevel@tonic-gate } 13427c478bd9Sstevel@tonic-gate } else { 13437c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); 13447c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, vp, cr); 13457c478bd9Sstevel@tonic-gate } 13467c478bd9Sstevel@tonic-gate } else { 13477c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); 13487c478bd9Sstevel@tonic-gate } 13497c478bd9Sstevel@tonic-gate 13507c478bd9Sstevel@tonic-gate return (error); 13517c478bd9Sstevel@tonic-gate } 13527c478bd9Sstevel@tonic-gate 13537c478bd9Sstevel@tonic-gate static int 13547c478bd9Sstevel@tonic-gate nfs_accessx(void *vp, int mode, cred_t *cr) 13557c478bd9Sstevel@tonic-gate { 1356108322fbScarlsonj ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone); 1357da6c28aaSamw return (nfs_access(vp, mode, 0, cr, NULL)); 13587c478bd9Sstevel@tonic-gate } 13597c478bd9Sstevel@tonic-gate 1360da6c28aaSamw /* ARGSUSED */ 13617c478bd9Sstevel@tonic-gate static int 1362da6c28aaSamw nfs_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct) 13637c478bd9Sstevel@tonic-gate { 13647c478bd9Sstevel@tonic-gate struct vattr va; 13657c478bd9Sstevel@tonic-gate int error; 13667c478bd9Sstevel@tonic-gate mntinfo_t *mi; 13677c478bd9Sstevel@tonic-gate int shift = 0; 13687c478bd9Sstevel@tonic-gate 13697c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 13707c478bd9Sstevel@tonic-gate 1371108322fbScarlsonj if (nfs_zone() != mi->mi_zone) 13727c478bd9Sstevel@tonic-gate return (EIO); 13737c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) { 13747c478bd9Sstevel@tonic-gate error = acl_access2(vp, mode, flags, cr); 13757c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) 13767c478bd9Sstevel@tonic-gate return (error); 13777c478bd9Sstevel@tonic-gate } 13787c478bd9Sstevel@tonic-gate 13797c478bd9Sstevel@tonic-gate va.va_mask = AT_MODE | AT_UID | AT_GID; 13807c478bd9Sstevel@tonic-gate error = nfsgetattr(vp, &va, cr); 13817c478bd9Sstevel@tonic-gate if (error) 13827c478bd9Sstevel@tonic-gate return (error); 13837c478bd9Sstevel@tonic-gate 13847c478bd9Sstevel@tonic-gate /* 13857c478bd9Sstevel@tonic-gate * Disallow write attempts on read-only 13867c478bd9Sstevel@tonic-gate * file systems, unless the file is a 13877c478bd9Sstevel@tonic-gate * device node. 13887c478bd9Sstevel@tonic-gate */ 13897c478bd9Sstevel@tonic-gate if ((mode & VWRITE) && vn_is_readonly(vp) && !IS_DEVVP(vp)) 13907c478bd9Sstevel@tonic-gate return (EROFS); 13917c478bd9Sstevel@tonic-gate 13927c478bd9Sstevel@tonic-gate /* 13937c478bd9Sstevel@tonic-gate * Disallow attempts to access mandatory lock files. 13947c478bd9Sstevel@tonic-gate */ 13957c478bd9Sstevel@tonic-gate if ((mode & (VWRITE | VREAD | VEXEC)) && 13967c478bd9Sstevel@tonic-gate MANDLOCK(vp, va.va_mode)) 13977c478bd9Sstevel@tonic-gate return (EACCES); 13987c478bd9Sstevel@tonic-gate 13997c478bd9Sstevel@tonic-gate /* 14007c478bd9Sstevel@tonic-gate * Access check is based on only 14017c478bd9Sstevel@tonic-gate * one of owner, group, public. 14027c478bd9Sstevel@tonic-gate * If not owner, then check group. 14037c478bd9Sstevel@tonic-gate * If not a member of the group, 14047c478bd9Sstevel@tonic-gate * then check public access. 14057c478bd9Sstevel@tonic-gate */ 14067c478bd9Sstevel@tonic-gate if (crgetuid(cr) != va.va_uid) { 14077c478bd9Sstevel@tonic-gate shift += 3; 14087c478bd9Sstevel@tonic-gate if (!groupmember(va.va_gid, cr)) 14097c478bd9Sstevel@tonic-gate shift += 3; 14107c478bd9Sstevel@tonic-gate } 14117c478bd9Sstevel@tonic-gate found: 14127c478bd9Sstevel@tonic-gate mode &= ~(va.va_mode << shift); 14137c478bd9Sstevel@tonic-gate if (mode == 0) 14147c478bd9Sstevel@tonic-gate return (0); 14157c478bd9Sstevel@tonic-gate 14167c478bd9Sstevel@tonic-gate return (secpolicy_vnode_access(cr, vp, va.va_uid, mode)); 14177c478bd9Sstevel@tonic-gate } 14187c478bd9Sstevel@tonic-gate 14197c478bd9Sstevel@tonic-gate static int nfs_do_symlink_cache = 1; 14207c478bd9Sstevel@tonic-gate 1421da6c28aaSamw /* ARGSUSED */ 14227c478bd9Sstevel@tonic-gate static int 1423da6c28aaSamw nfs_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct) 14247c478bd9Sstevel@tonic-gate { 14257c478bd9Sstevel@tonic-gate int error; 14267c478bd9Sstevel@tonic-gate struct nfsrdlnres rl; 14277c478bd9Sstevel@tonic-gate rnode_t *rp; 14287c478bd9Sstevel@tonic-gate int douprintf; 14297c478bd9Sstevel@tonic-gate failinfo_t fi; 14307c478bd9Sstevel@tonic-gate 14317c478bd9Sstevel@tonic-gate /* 14327c478bd9Sstevel@tonic-gate * We want to be consistent with UFS semantics so we will return 14337c478bd9Sstevel@tonic-gate * EINVAL instead of ENXIO. This violates the XNFS spec and 14347c478bd9Sstevel@tonic-gate * the RFC 1094, which are wrong any way. BUGID 1138002. 14357c478bd9Sstevel@tonic-gate */ 14367c478bd9Sstevel@tonic-gate if (vp->v_type != VLNK) 14377c478bd9Sstevel@tonic-gate return (EINVAL); 14387c478bd9Sstevel@tonic-gate 1439108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 14407c478bd9Sstevel@tonic-gate return (EIO); 14417c478bd9Sstevel@tonic-gate 14427c478bd9Sstevel@tonic-gate rp = VTOR(vp); 14437c478bd9Sstevel@tonic-gate if (nfs_do_symlink_cache && rp->r_symlink.contents != NULL) { 14447c478bd9Sstevel@tonic-gate error = nfs_validate_caches(vp, cr); 14457c478bd9Sstevel@tonic-gate if (error) 14467c478bd9Sstevel@tonic-gate return (error); 14477c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 14487c478bd9Sstevel@tonic-gate if (rp->r_symlink.contents != NULL) { 14497c478bd9Sstevel@tonic-gate error = uiomove(rp->r_symlink.contents, 14507c478bd9Sstevel@tonic-gate rp->r_symlink.len, UIO_READ, uiop); 14517c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 14527c478bd9Sstevel@tonic-gate return (error); 14537c478bd9Sstevel@tonic-gate } 14547c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 14557c478bd9Sstevel@tonic-gate } 14567c478bd9Sstevel@tonic-gate 14577c478bd9Sstevel@tonic-gate 14587c478bd9Sstevel@tonic-gate rl.rl_data = kmem_alloc(NFS_MAXPATHLEN, KM_SLEEP); 14597c478bd9Sstevel@tonic-gate 14607c478bd9Sstevel@tonic-gate fi.vp = vp; 14617c478bd9Sstevel@tonic-gate fi.fhp = NULL; /* no need to update, filehandle not copied */ 14627c478bd9Sstevel@tonic-gate fi.copyproc = nfscopyfh; 14637c478bd9Sstevel@tonic-gate fi.lookupproc = nfslookup; 14647c478bd9Sstevel@tonic-gate fi.xattrdirproc = acl_getxattrdir2; 14657c478bd9Sstevel@tonic-gate 14667c478bd9Sstevel@tonic-gate douprintf = 1; 14677c478bd9Sstevel@tonic-gate 14687c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(vp), RFS_READLINK, 14697c478bd9Sstevel@tonic-gate xdr_fhandle, (caddr_t)VTOFH(vp), 14707c478bd9Sstevel@tonic-gate xdr_rdlnres, (caddr_t)&rl, cr, 14717c478bd9Sstevel@tonic-gate &douprintf, &rl.rl_status, 0, &fi); 14727c478bd9Sstevel@tonic-gate 14737c478bd9Sstevel@tonic-gate if (error) { 14747c478bd9Sstevel@tonic-gate 14757c478bd9Sstevel@tonic-gate kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN); 14767c478bd9Sstevel@tonic-gate return (error); 14777c478bd9Sstevel@tonic-gate } 14787c478bd9Sstevel@tonic-gate 14797c478bd9Sstevel@tonic-gate error = geterrno(rl.rl_status); 14807c478bd9Sstevel@tonic-gate if (!error) { 14817c478bd9Sstevel@tonic-gate error = uiomove(rl.rl_data, (int)rl.rl_count, UIO_READ, uiop); 14827c478bd9Sstevel@tonic-gate if (nfs_do_symlink_cache && rp->r_symlink.contents == NULL) { 14837c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 14847c478bd9Sstevel@tonic-gate if (rp->r_symlink.contents == NULL) { 14857c478bd9Sstevel@tonic-gate rp->r_symlink.contents = rl.rl_data; 14867c478bd9Sstevel@tonic-gate rp->r_symlink.len = (int)rl.rl_count; 14877c478bd9Sstevel@tonic-gate rp->r_symlink.size = NFS_MAXPATHLEN; 14887c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 14897c478bd9Sstevel@tonic-gate } else { 14907c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 14917c478bd9Sstevel@tonic-gate 14927c478bd9Sstevel@tonic-gate kmem_free((void *)rl.rl_data, 14937c478bd9Sstevel@tonic-gate NFS_MAXPATHLEN); 14947c478bd9Sstevel@tonic-gate } 14957c478bd9Sstevel@tonic-gate } else { 14967c478bd9Sstevel@tonic-gate 14977c478bd9Sstevel@tonic-gate kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN); 14987c478bd9Sstevel@tonic-gate } 14997c478bd9Sstevel@tonic-gate } else { 15007c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, vp, cr); 15017c478bd9Sstevel@tonic-gate 15027c478bd9Sstevel@tonic-gate kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN); 15037c478bd9Sstevel@tonic-gate } 15047c478bd9Sstevel@tonic-gate 15057c478bd9Sstevel@tonic-gate /* 15067c478bd9Sstevel@tonic-gate * Conform to UFS semantics (see comment above) 15077c478bd9Sstevel@tonic-gate */ 15087c478bd9Sstevel@tonic-gate return (error == ENXIO ? EINVAL : error); 15097c478bd9Sstevel@tonic-gate } 15107c478bd9Sstevel@tonic-gate 15117c478bd9Sstevel@tonic-gate /* 15127c478bd9Sstevel@tonic-gate * Flush local dirty pages to stable storage on the server. 15137c478bd9Sstevel@tonic-gate * 15147c478bd9Sstevel@tonic-gate * If FNODSYNC is specified, then there is nothing to do because 15157c478bd9Sstevel@tonic-gate * metadata changes are not cached on the client before being 15167c478bd9Sstevel@tonic-gate * sent to the server. 15177c478bd9Sstevel@tonic-gate */ 1518da6c28aaSamw /* ARGSUSED */ 15197c478bd9Sstevel@tonic-gate static int 1520da6c28aaSamw nfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 15217c478bd9Sstevel@tonic-gate { 15227c478bd9Sstevel@tonic-gate int error; 15237c478bd9Sstevel@tonic-gate 15247c478bd9Sstevel@tonic-gate if ((syncflag & FNODSYNC) || IS_SWAPVP(vp)) 15257c478bd9Sstevel@tonic-gate return (0); 15267c478bd9Sstevel@tonic-gate 1527108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 15287c478bd9Sstevel@tonic-gate return (EIO); 15297c478bd9Sstevel@tonic-gate 1530da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct); 15317c478bd9Sstevel@tonic-gate if (!error) 15327c478bd9Sstevel@tonic-gate error = VTOR(vp)->r_error; 15337c478bd9Sstevel@tonic-gate return (error); 15347c478bd9Sstevel@tonic-gate } 15357c478bd9Sstevel@tonic-gate 15367c478bd9Sstevel@tonic-gate 15377c478bd9Sstevel@tonic-gate /* 15387c478bd9Sstevel@tonic-gate * Weirdness: if the file was removed or the target of a rename 15397c478bd9Sstevel@tonic-gate * operation while it was open, it got renamed instead. Here we 15407c478bd9Sstevel@tonic-gate * remove the renamed file. 15417c478bd9Sstevel@tonic-gate */ 1542da6c28aaSamw /* ARGSUSED */ 15437c478bd9Sstevel@tonic-gate static void 1544da6c28aaSamw nfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 15457c478bd9Sstevel@tonic-gate { 15467c478bd9Sstevel@tonic-gate rnode_t *rp; 15477c478bd9Sstevel@tonic-gate 15487c478bd9Sstevel@tonic-gate ASSERT(vp != DNLC_NO_VNODE); 15497c478bd9Sstevel@tonic-gate 15507c478bd9Sstevel@tonic-gate /* 15517c478bd9Sstevel@tonic-gate * If this is coming from the wrong zone, we let someone in the right 15527c478bd9Sstevel@tonic-gate * zone take care of it asynchronously. We can get here due to 15537c478bd9Sstevel@tonic-gate * VN_RELE() being called from pageout() or fsflush(). This call may 15547c478bd9Sstevel@tonic-gate * potentially turn into an expensive no-op if, for instance, v_count 15557c478bd9Sstevel@tonic-gate * gets incremented in the meantime, but it's still correct. 15567c478bd9Sstevel@tonic-gate */ 1557108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) { 15587c478bd9Sstevel@tonic-gate nfs_async_inactive(vp, cr, nfs_inactive); 15597c478bd9Sstevel@tonic-gate return; 15607c478bd9Sstevel@tonic-gate } 15617c478bd9Sstevel@tonic-gate 15627c478bd9Sstevel@tonic-gate rp = VTOR(vp); 15637c478bd9Sstevel@tonic-gate redo: 15647c478bd9Sstevel@tonic-gate if (rp->r_unldvp != NULL) { 15657c478bd9Sstevel@tonic-gate /* 15667c478bd9Sstevel@tonic-gate * Save the vnode pointer for the directory where the 15677c478bd9Sstevel@tonic-gate * unlinked-open file got renamed, then set it to NULL 15687c478bd9Sstevel@tonic-gate * to prevent another thread from getting here before 15697c478bd9Sstevel@tonic-gate * we're done with the remove. While we have the 15707c478bd9Sstevel@tonic-gate * statelock, make local copies of the pertinent rnode 15717c478bd9Sstevel@tonic-gate * fields. If we weren't to do this in an atomic way, the 15727c478bd9Sstevel@tonic-gate * the unl* fields could become inconsistent with respect 15737c478bd9Sstevel@tonic-gate * to each other due to a race condition between this 15747c478bd9Sstevel@tonic-gate * code and nfs_remove(). See bug report 1034328. 15757c478bd9Sstevel@tonic-gate */ 15767c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 15777c478bd9Sstevel@tonic-gate if (rp->r_unldvp != NULL) { 15787c478bd9Sstevel@tonic-gate vnode_t *unldvp; 15797c478bd9Sstevel@tonic-gate char *unlname; 15807c478bd9Sstevel@tonic-gate cred_t *unlcred; 15817c478bd9Sstevel@tonic-gate struct nfsdiropargs da; 15827c478bd9Sstevel@tonic-gate enum nfsstat status; 15837c478bd9Sstevel@tonic-gate int douprintf; 15847c478bd9Sstevel@tonic-gate int error; 15857c478bd9Sstevel@tonic-gate 15867c478bd9Sstevel@tonic-gate unldvp = rp->r_unldvp; 15877c478bd9Sstevel@tonic-gate rp->r_unldvp = NULL; 15887c478bd9Sstevel@tonic-gate unlname = rp->r_unlname; 15897c478bd9Sstevel@tonic-gate rp->r_unlname = NULL; 15907c478bd9Sstevel@tonic-gate unlcred = rp->r_unlcred; 15917c478bd9Sstevel@tonic-gate rp->r_unlcred = NULL; 15927c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 15937c478bd9Sstevel@tonic-gate 15947c478bd9Sstevel@tonic-gate /* 15957c478bd9Sstevel@tonic-gate * If there are any dirty pages left, then flush 15967c478bd9Sstevel@tonic-gate * them. This is unfortunate because they just 15977c478bd9Sstevel@tonic-gate * may get thrown away during the remove operation, 15987c478bd9Sstevel@tonic-gate * but we have to do this for correctness. 15997c478bd9Sstevel@tonic-gate */ 16007c478bd9Sstevel@tonic-gate if (vn_has_cached_data(vp) && 16017c478bd9Sstevel@tonic-gate ((rp->r_flags & RDIRTY) || rp->r_count > 0)) { 16027c478bd9Sstevel@tonic-gate ASSERT(vp->v_type != VCHR); 1603da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, 0, 1604da6c28aaSamw cr, ct); 16057c478bd9Sstevel@tonic-gate if (error) { 16067c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 16077c478bd9Sstevel@tonic-gate if (!rp->r_error) 16087c478bd9Sstevel@tonic-gate rp->r_error = error; 16097c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 16107c478bd9Sstevel@tonic-gate } 16117c478bd9Sstevel@tonic-gate } 16127c478bd9Sstevel@tonic-gate 16137c478bd9Sstevel@tonic-gate /* 16147c478bd9Sstevel@tonic-gate * Do the remove operation on the renamed file 16157c478bd9Sstevel@tonic-gate */ 16167c478bd9Sstevel@tonic-gate setdiropargs(&da, unlname, unldvp); 16177c478bd9Sstevel@tonic-gate 16187c478bd9Sstevel@tonic-gate douprintf = 1; 16197c478bd9Sstevel@tonic-gate 16207c478bd9Sstevel@tonic-gate (void) rfs2call(VTOMI(unldvp), RFS_REMOVE, 16217c478bd9Sstevel@tonic-gate xdr_diropargs, (caddr_t)&da, 16227c478bd9Sstevel@tonic-gate xdr_enum, (caddr_t)&status, unlcred, 16237c478bd9Sstevel@tonic-gate &douprintf, &status, 0, NULL); 16247c478bd9Sstevel@tonic-gate 16257c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(VTOR(unldvp))) 16267c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(unldvp); 16277c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(unldvp); 16287c478bd9Sstevel@tonic-gate 16297c478bd9Sstevel@tonic-gate /* 16307c478bd9Sstevel@tonic-gate * Release stuff held for the remove 16317c478bd9Sstevel@tonic-gate */ 16327c478bd9Sstevel@tonic-gate VN_RELE(unldvp); 16337c478bd9Sstevel@tonic-gate kmem_free(unlname, MAXNAMELEN); 16347c478bd9Sstevel@tonic-gate crfree(unlcred); 16357c478bd9Sstevel@tonic-gate goto redo; 16367c478bd9Sstevel@tonic-gate } 16377c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 16387c478bd9Sstevel@tonic-gate } 16397c478bd9Sstevel@tonic-gate 16407c478bd9Sstevel@tonic-gate rp_addfree(rp, cr); 16417c478bd9Sstevel@tonic-gate } 16427c478bd9Sstevel@tonic-gate 16437c478bd9Sstevel@tonic-gate /* 16447c478bd9Sstevel@tonic-gate * Remote file system operations having to do with directory manipulation. 16457c478bd9Sstevel@tonic-gate */ 16467c478bd9Sstevel@tonic-gate 1647da6c28aaSamw /* ARGSUSED */ 16487c478bd9Sstevel@tonic-gate static int 16497c478bd9Sstevel@tonic-gate nfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1650da6c28aaSamw int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1651da6c28aaSamw int *direntflags, pathname_t *realpnp) 16527c478bd9Sstevel@tonic-gate { 16537c478bd9Sstevel@tonic-gate int error; 16547c478bd9Sstevel@tonic-gate vnode_t *vp; 16557c478bd9Sstevel@tonic-gate vnode_t *avp = NULL; 16567c478bd9Sstevel@tonic-gate rnode_t *drp; 16577c478bd9Sstevel@tonic-gate 1658108322fbScarlsonj if (nfs_zone() != VTOMI(dvp)->mi_zone) 16597c478bd9Sstevel@tonic-gate return (EPERM); 16607c478bd9Sstevel@tonic-gate 16617c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 16627c478bd9Sstevel@tonic-gate 16637c478bd9Sstevel@tonic-gate /* 16647c478bd9Sstevel@tonic-gate * Are we looking up extended attributes? If so, "dvp" is 16657c478bd9Sstevel@tonic-gate * the file or directory for which we want attributes, and 16667c478bd9Sstevel@tonic-gate * we need a lookup of the hidden attribute directory 16677c478bd9Sstevel@tonic-gate * before we lookup the rest of the path. 16687c478bd9Sstevel@tonic-gate */ 16697c478bd9Sstevel@tonic-gate if (flags & LOOKUP_XATTR) { 16707c478bd9Sstevel@tonic-gate bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0); 16717c478bd9Sstevel@tonic-gate mntinfo_t *mi; 16727c478bd9Sstevel@tonic-gate 16737c478bd9Sstevel@tonic-gate mi = VTOMI(dvp); 16747c478bd9Sstevel@tonic-gate if (!(mi->mi_flags & MI_EXTATTR)) 16757c478bd9Sstevel@tonic-gate return (EINVAL); 16767c478bd9Sstevel@tonic-gate 16777c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) 16787c478bd9Sstevel@tonic-gate return (EINTR); 16797c478bd9Sstevel@tonic-gate 16807c478bd9Sstevel@tonic-gate (void) nfslookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr); 16817c478bd9Sstevel@tonic-gate if (avp == NULL) 16827c478bd9Sstevel@tonic-gate error = acl_getxattrdir2(dvp, &avp, cflag, cr, 0); 16837c478bd9Sstevel@tonic-gate else 16847c478bd9Sstevel@tonic-gate error = 0; 16857c478bd9Sstevel@tonic-gate 16867c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 16877c478bd9Sstevel@tonic-gate 16887c478bd9Sstevel@tonic-gate if (error) { 16897c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_EXTATTR) 16907c478bd9Sstevel@tonic-gate return (error); 16917c478bd9Sstevel@tonic-gate return (EINVAL); 16927c478bd9Sstevel@tonic-gate } 16937c478bd9Sstevel@tonic-gate dvp = avp; 16947c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 16957c478bd9Sstevel@tonic-gate } 16967c478bd9Sstevel@tonic-gate 16977c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) { 16987c478bd9Sstevel@tonic-gate error = EINTR; 16997c478bd9Sstevel@tonic-gate goto out; 17007c478bd9Sstevel@tonic-gate } 17017c478bd9Sstevel@tonic-gate 17027c478bd9Sstevel@tonic-gate error = nfslookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0); 17037c478bd9Sstevel@tonic-gate 17047c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 17057c478bd9Sstevel@tonic-gate 17067c478bd9Sstevel@tonic-gate /* 17077c478bd9Sstevel@tonic-gate * If vnode is a device, create special vnode. 17087c478bd9Sstevel@tonic-gate */ 17097c478bd9Sstevel@tonic-gate if (!error && IS_DEVVP(*vpp)) { 17107c478bd9Sstevel@tonic-gate vp = *vpp; 17117c478bd9Sstevel@tonic-gate *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 17127c478bd9Sstevel@tonic-gate VN_RELE(vp); 17137c478bd9Sstevel@tonic-gate } 17147c478bd9Sstevel@tonic-gate 17157c478bd9Sstevel@tonic-gate out: 17167c478bd9Sstevel@tonic-gate if (avp != NULL) 17177c478bd9Sstevel@tonic-gate VN_RELE(avp); 17187c478bd9Sstevel@tonic-gate 17197c478bd9Sstevel@tonic-gate return (error); 17207c478bd9Sstevel@tonic-gate } 17217c478bd9Sstevel@tonic-gate 17227c478bd9Sstevel@tonic-gate static int nfs_lookup_neg_cache = 1; 17237c478bd9Sstevel@tonic-gate 17247c478bd9Sstevel@tonic-gate #ifdef DEBUG 17257c478bd9Sstevel@tonic-gate static int nfs_lookup_dnlc_hits = 0; 17267c478bd9Sstevel@tonic-gate static int nfs_lookup_dnlc_misses = 0; 17277c478bd9Sstevel@tonic-gate static int nfs_lookup_dnlc_neg_hits = 0; 17287c478bd9Sstevel@tonic-gate static int nfs_lookup_dnlc_disappears = 0; 17297c478bd9Sstevel@tonic-gate static int nfs_lookup_dnlc_lookups = 0; 17307c478bd9Sstevel@tonic-gate #endif 17317c478bd9Sstevel@tonic-gate 17327c478bd9Sstevel@tonic-gate /* ARGSUSED */ 17337c478bd9Sstevel@tonic-gate int 17347c478bd9Sstevel@tonic-gate nfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 17357c478bd9Sstevel@tonic-gate int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags) 17367c478bd9Sstevel@tonic-gate { 17377c478bd9Sstevel@tonic-gate int error; 17387c478bd9Sstevel@tonic-gate 1739108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 17407c478bd9Sstevel@tonic-gate 17417c478bd9Sstevel@tonic-gate /* 17427c478bd9Sstevel@tonic-gate * If lookup is for "", just return dvp. Don't need 17437c478bd9Sstevel@tonic-gate * to send it over the wire, look it up in the dnlc, 17447c478bd9Sstevel@tonic-gate * or perform any access checks. 17457c478bd9Sstevel@tonic-gate */ 17467c478bd9Sstevel@tonic-gate if (*nm == '\0') { 17477c478bd9Sstevel@tonic-gate VN_HOLD(dvp); 17487c478bd9Sstevel@tonic-gate *vpp = dvp; 17497c478bd9Sstevel@tonic-gate return (0); 17507c478bd9Sstevel@tonic-gate } 17517c478bd9Sstevel@tonic-gate 17527c478bd9Sstevel@tonic-gate /* 17537c478bd9Sstevel@tonic-gate * Can't do lookups in non-directories. 17547c478bd9Sstevel@tonic-gate */ 17557c478bd9Sstevel@tonic-gate if (dvp->v_type != VDIR) 17567c478bd9Sstevel@tonic-gate return (ENOTDIR); 17577c478bd9Sstevel@tonic-gate 17587c478bd9Sstevel@tonic-gate /* 17597c478bd9Sstevel@tonic-gate * If we're called with RFSCALL_SOFT, it's important that 17607c478bd9Sstevel@tonic-gate * the only rfscall is one we make directly; if we permit 17617c478bd9Sstevel@tonic-gate * an access call because we're looking up "." or validating 17627c478bd9Sstevel@tonic-gate * a dnlc hit, we'll deadlock because that rfscall will not 17637c478bd9Sstevel@tonic-gate * have the RFSCALL_SOFT set. 17647c478bd9Sstevel@tonic-gate */ 17657c478bd9Sstevel@tonic-gate if (rfscall_flags & RFSCALL_SOFT) 17667c478bd9Sstevel@tonic-gate goto callit; 17677c478bd9Sstevel@tonic-gate 17687c478bd9Sstevel@tonic-gate /* 17697c478bd9Sstevel@tonic-gate * If lookup is for ".", just return dvp. Don't need 17707c478bd9Sstevel@tonic-gate * to send it over the wire or look it up in the dnlc, 17717c478bd9Sstevel@tonic-gate * just need to check access. 17727c478bd9Sstevel@tonic-gate */ 17737c478bd9Sstevel@tonic-gate if (strcmp(nm, ".") == 0) { 1774da6c28aaSamw error = nfs_access(dvp, VEXEC, 0, cr, NULL); 17757c478bd9Sstevel@tonic-gate if (error) 17767c478bd9Sstevel@tonic-gate return (error); 17777c478bd9Sstevel@tonic-gate VN_HOLD(dvp); 17787c478bd9Sstevel@tonic-gate *vpp = dvp; 17797c478bd9Sstevel@tonic-gate return (0); 17807c478bd9Sstevel@tonic-gate } 17817c478bd9Sstevel@tonic-gate 17827c478bd9Sstevel@tonic-gate /* 17837c478bd9Sstevel@tonic-gate * Lookup this name in the DNLC. If there was a valid entry, 17847c478bd9Sstevel@tonic-gate * then return the results of the lookup. 17857c478bd9Sstevel@tonic-gate */ 17867c478bd9Sstevel@tonic-gate error = nfslookup_dnlc(dvp, nm, vpp, cr); 17877c478bd9Sstevel@tonic-gate if (error || *vpp != NULL) 17887c478bd9Sstevel@tonic-gate return (error); 17897c478bd9Sstevel@tonic-gate 17907c478bd9Sstevel@tonic-gate callit: 17917c478bd9Sstevel@tonic-gate error = nfslookup_otw(dvp, nm, vpp, cr, rfscall_flags); 17927c478bd9Sstevel@tonic-gate 17937c478bd9Sstevel@tonic-gate return (error); 17947c478bd9Sstevel@tonic-gate } 17957c478bd9Sstevel@tonic-gate 17967c478bd9Sstevel@tonic-gate static int 17977c478bd9Sstevel@tonic-gate nfslookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr) 17987c478bd9Sstevel@tonic-gate { 17997c478bd9Sstevel@tonic-gate int error; 18007c478bd9Sstevel@tonic-gate vnode_t *vp; 18017c478bd9Sstevel@tonic-gate 18027c478bd9Sstevel@tonic-gate ASSERT(*nm != '\0'); 1803108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 18047c478bd9Sstevel@tonic-gate 18057c478bd9Sstevel@tonic-gate /* 18067c478bd9Sstevel@tonic-gate * Lookup this name in the DNLC. If successful, then validate 18077c478bd9Sstevel@tonic-gate * the caches and then recheck the DNLC. The DNLC is rechecked 18087c478bd9Sstevel@tonic-gate * just in case this entry got invalidated during the call 18097c478bd9Sstevel@tonic-gate * to nfs_validate_caches. 18107c478bd9Sstevel@tonic-gate * 18117c478bd9Sstevel@tonic-gate * An assumption is being made that it is safe to say that a 18127c478bd9Sstevel@tonic-gate * file exists which may not on the server. Any operations to 18137c478bd9Sstevel@tonic-gate * the server will fail with ESTALE. 18147c478bd9Sstevel@tonic-gate */ 18157c478bd9Sstevel@tonic-gate #ifdef DEBUG 18167c478bd9Sstevel@tonic-gate nfs_lookup_dnlc_lookups++; 18177c478bd9Sstevel@tonic-gate #endif 18187c478bd9Sstevel@tonic-gate vp = dnlc_lookup(dvp, nm); 18197c478bd9Sstevel@tonic-gate if (vp != NULL) { 18207c478bd9Sstevel@tonic-gate VN_RELE(vp); 18217c478bd9Sstevel@tonic-gate if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) { 18227c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(dvp); 18237c478bd9Sstevel@tonic-gate } 18247c478bd9Sstevel@tonic-gate error = nfs_validate_caches(dvp, cr); 18257c478bd9Sstevel@tonic-gate if (error) 18267c478bd9Sstevel@tonic-gate return (error); 18277c478bd9Sstevel@tonic-gate vp = dnlc_lookup(dvp, nm); 18287c478bd9Sstevel@tonic-gate if (vp != NULL) { 1829da6c28aaSamw error = nfs_access(dvp, VEXEC, 0, cr, NULL); 18307c478bd9Sstevel@tonic-gate if (error) { 18317c478bd9Sstevel@tonic-gate VN_RELE(vp); 18327c478bd9Sstevel@tonic-gate return (error); 18337c478bd9Sstevel@tonic-gate } 18347c478bd9Sstevel@tonic-gate if (vp == DNLC_NO_VNODE) { 18357c478bd9Sstevel@tonic-gate VN_RELE(vp); 18367c478bd9Sstevel@tonic-gate #ifdef DEBUG 18377c478bd9Sstevel@tonic-gate nfs_lookup_dnlc_neg_hits++; 18387c478bd9Sstevel@tonic-gate #endif 18397c478bd9Sstevel@tonic-gate return (ENOENT); 18407c478bd9Sstevel@tonic-gate } 18417c478bd9Sstevel@tonic-gate *vpp = vp; 18427c478bd9Sstevel@tonic-gate #ifdef DEBUG 18437c478bd9Sstevel@tonic-gate nfs_lookup_dnlc_hits++; 18447c478bd9Sstevel@tonic-gate #endif 18457c478bd9Sstevel@tonic-gate return (0); 18467c478bd9Sstevel@tonic-gate } 18477c478bd9Sstevel@tonic-gate #ifdef DEBUG 18487c478bd9Sstevel@tonic-gate nfs_lookup_dnlc_disappears++; 18497c478bd9Sstevel@tonic-gate #endif 18507c478bd9Sstevel@tonic-gate } 18517c478bd9Sstevel@tonic-gate #ifdef DEBUG 18527c478bd9Sstevel@tonic-gate else 18537c478bd9Sstevel@tonic-gate nfs_lookup_dnlc_misses++; 18547c478bd9Sstevel@tonic-gate #endif 18557c478bd9Sstevel@tonic-gate 18567c478bd9Sstevel@tonic-gate *vpp = NULL; 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate return (0); 18597c478bd9Sstevel@tonic-gate } 18607c478bd9Sstevel@tonic-gate 18617c478bd9Sstevel@tonic-gate static int 18627c478bd9Sstevel@tonic-gate nfslookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr, 18637c478bd9Sstevel@tonic-gate int rfscall_flags) 18647c478bd9Sstevel@tonic-gate { 18657c478bd9Sstevel@tonic-gate int error; 18667c478bd9Sstevel@tonic-gate struct nfsdiropargs da; 18677c478bd9Sstevel@tonic-gate struct nfsdiropres dr; 18687c478bd9Sstevel@tonic-gate int douprintf; 18697c478bd9Sstevel@tonic-gate failinfo_t fi; 18707c478bd9Sstevel@tonic-gate hrtime_t t; 18717c478bd9Sstevel@tonic-gate 18727c478bd9Sstevel@tonic-gate ASSERT(*nm != '\0'); 18737c478bd9Sstevel@tonic-gate ASSERT(dvp->v_type == VDIR); 1874108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone); 18757c478bd9Sstevel@tonic-gate 18767c478bd9Sstevel@tonic-gate setdiropargs(&da, nm, dvp); 18777c478bd9Sstevel@tonic-gate 18787c478bd9Sstevel@tonic-gate fi.vp = dvp; 18797c478bd9Sstevel@tonic-gate fi.fhp = NULL; /* no need to update, filehandle not copied */ 18807c478bd9Sstevel@tonic-gate fi.copyproc = nfscopyfh; 18817c478bd9Sstevel@tonic-gate fi.lookupproc = nfslookup; 18827c478bd9Sstevel@tonic-gate fi.xattrdirproc = acl_getxattrdir2; 18837c478bd9Sstevel@tonic-gate 18847c478bd9Sstevel@tonic-gate douprintf = 1; 18857c478bd9Sstevel@tonic-gate 18867c478bd9Sstevel@tonic-gate t = gethrtime(); 18877c478bd9Sstevel@tonic-gate 18887c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(dvp), RFS_LOOKUP, 18897c478bd9Sstevel@tonic-gate xdr_diropargs, (caddr_t)&da, 18907c478bd9Sstevel@tonic-gate xdr_diropres, (caddr_t)&dr, cr, 18917c478bd9Sstevel@tonic-gate &douprintf, &dr.dr_status, rfscall_flags, &fi); 18927c478bd9Sstevel@tonic-gate 18937c478bd9Sstevel@tonic-gate if (!error) { 18947c478bd9Sstevel@tonic-gate error = geterrno(dr.dr_status); 18957c478bd9Sstevel@tonic-gate if (!error) { 18967c478bd9Sstevel@tonic-gate *vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr, 18977c478bd9Sstevel@tonic-gate dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm); 18987c478bd9Sstevel@tonic-gate /* 18997c478bd9Sstevel@tonic-gate * If NFS_ACL is supported on the server, then the 19007c478bd9Sstevel@tonic-gate * attributes returned by server may have minimal 19017c478bd9Sstevel@tonic-gate * permissions sometimes denying access to users having 19027c478bd9Sstevel@tonic-gate * proper access. To get the proper attributes, mark 19037c478bd9Sstevel@tonic-gate * the attributes as expired so that they will be 19047c478bd9Sstevel@tonic-gate * regotten via the NFS_ACL GETATTR2 procedure. 19057c478bd9Sstevel@tonic-gate */ 19067c478bd9Sstevel@tonic-gate if (VTOMI(*vpp)->mi_flags & MI_ACL) { 19077c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(*vpp); 19087c478bd9Sstevel@tonic-gate } 19097c478bd9Sstevel@tonic-gate if (!(rfscall_flags & RFSCALL_SOFT)) 19107c478bd9Sstevel@tonic-gate dnlc_update(dvp, nm, *vpp); 19117c478bd9Sstevel@tonic-gate } else { 19127c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, dvp, cr); 19137c478bd9Sstevel@tonic-gate if (error == ENOENT && nfs_lookup_neg_cache) 19147c478bd9Sstevel@tonic-gate dnlc_enter(dvp, nm, DNLC_NO_VNODE); 19157c478bd9Sstevel@tonic-gate } 19167c478bd9Sstevel@tonic-gate } 19177c478bd9Sstevel@tonic-gate 19187c478bd9Sstevel@tonic-gate return (error); 19197c478bd9Sstevel@tonic-gate } 19207c478bd9Sstevel@tonic-gate 19217c478bd9Sstevel@tonic-gate /* ARGSUSED */ 19227c478bd9Sstevel@tonic-gate static int 19237c478bd9Sstevel@tonic-gate nfs_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive, 1924da6c28aaSamw int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct, 1925da6c28aaSamw vsecattr_t *vsecp) 19267c478bd9Sstevel@tonic-gate { 19277c478bd9Sstevel@tonic-gate int error; 19287c478bd9Sstevel@tonic-gate struct nfscreatargs args; 19297c478bd9Sstevel@tonic-gate struct nfsdiropres dr; 19307c478bd9Sstevel@tonic-gate int douprintf; 19317c478bd9Sstevel@tonic-gate vnode_t *vp; 19327c478bd9Sstevel@tonic-gate rnode_t *rp; 19337c478bd9Sstevel@tonic-gate struct vattr vattr; 19347c478bd9Sstevel@tonic-gate rnode_t *drp; 19357c478bd9Sstevel@tonic-gate vnode_t *tempvp; 19367c478bd9Sstevel@tonic-gate hrtime_t t; 19377c478bd9Sstevel@tonic-gate 19387c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 19397c478bd9Sstevel@tonic-gate 1940108322fbScarlsonj if (nfs_zone() != VTOMI(dvp)->mi_zone) 19417c478bd9Sstevel@tonic-gate return (EPERM); 19427c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 19437c478bd9Sstevel@tonic-gate return (EINTR); 19447c478bd9Sstevel@tonic-gate 19457c478bd9Sstevel@tonic-gate /* 19467c478bd9Sstevel@tonic-gate * We make a copy of the attributes because the caller does not 19477c478bd9Sstevel@tonic-gate * expect us to change what va points to. 19487c478bd9Sstevel@tonic-gate */ 19497c478bd9Sstevel@tonic-gate vattr = *va; 19507c478bd9Sstevel@tonic-gate 19517c478bd9Sstevel@tonic-gate /* 19527c478bd9Sstevel@tonic-gate * If the pathname is "", just use dvp. Don't need 19537c478bd9Sstevel@tonic-gate * to send it over the wire, look it up in the dnlc, 19547c478bd9Sstevel@tonic-gate * or perform any access checks. 19557c478bd9Sstevel@tonic-gate */ 19567c478bd9Sstevel@tonic-gate if (*nm == '\0') { 19577c478bd9Sstevel@tonic-gate error = 0; 19587c478bd9Sstevel@tonic-gate VN_HOLD(dvp); 19597c478bd9Sstevel@tonic-gate vp = dvp; 19607c478bd9Sstevel@tonic-gate /* 19617c478bd9Sstevel@tonic-gate * If the pathname is ".", just use dvp. Don't need 19627c478bd9Sstevel@tonic-gate * to send it over the wire or look it up in the dnlc, 19637c478bd9Sstevel@tonic-gate * just need to check access. 19647c478bd9Sstevel@tonic-gate */ 19657c478bd9Sstevel@tonic-gate } else if (strcmp(nm, ".") == 0) { 1966da6c28aaSamw error = nfs_access(dvp, VEXEC, 0, cr, ct); 19677c478bd9Sstevel@tonic-gate if (error) { 19687c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 19697c478bd9Sstevel@tonic-gate return (error); 19707c478bd9Sstevel@tonic-gate } 19717c478bd9Sstevel@tonic-gate VN_HOLD(dvp); 19727c478bd9Sstevel@tonic-gate vp = dvp; 19737c478bd9Sstevel@tonic-gate /* 19747c478bd9Sstevel@tonic-gate * We need to go over the wire, just to be sure whether the 19757c478bd9Sstevel@tonic-gate * file exists or not. Using the DNLC can be dangerous in 19767c478bd9Sstevel@tonic-gate * this case when making a decision regarding existence. 19777c478bd9Sstevel@tonic-gate */ 19787c478bd9Sstevel@tonic-gate } else { 19797c478bd9Sstevel@tonic-gate error = nfslookup_otw(dvp, nm, &vp, cr, 0); 19807c478bd9Sstevel@tonic-gate } 19817c478bd9Sstevel@tonic-gate if (!error) { 19827c478bd9Sstevel@tonic-gate if (exclusive == EXCL) 19837c478bd9Sstevel@tonic-gate error = EEXIST; 19847c478bd9Sstevel@tonic-gate else if (vp->v_type == VDIR && (mode & VWRITE)) 19857c478bd9Sstevel@tonic-gate error = EISDIR; 19867c478bd9Sstevel@tonic-gate else { 19877c478bd9Sstevel@tonic-gate /* 19887c478bd9Sstevel@tonic-gate * If vnode is a device, create special vnode. 19897c478bd9Sstevel@tonic-gate */ 19907c478bd9Sstevel@tonic-gate if (IS_DEVVP(vp)) { 19917c478bd9Sstevel@tonic-gate tempvp = vp; 19927c478bd9Sstevel@tonic-gate vp = specvp(vp, vp->v_rdev, vp->v_type, cr); 19937c478bd9Sstevel@tonic-gate VN_RELE(tempvp); 19947c478bd9Sstevel@tonic-gate } 1995da6c28aaSamw if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) { 19967c478bd9Sstevel@tonic-gate if ((vattr.va_mask & AT_SIZE) && 19977c478bd9Sstevel@tonic-gate vp->v_type == VREG) { 19987c478bd9Sstevel@tonic-gate vattr.va_mask = AT_SIZE; 19997c478bd9Sstevel@tonic-gate error = nfssetattr(vp, &vattr, 0, cr); 20007c478bd9Sstevel@tonic-gate } 20017c478bd9Sstevel@tonic-gate } 20027c478bd9Sstevel@tonic-gate } 20037c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 20047c478bd9Sstevel@tonic-gate if (error) { 20057c478bd9Sstevel@tonic-gate VN_RELE(vp); 2006df2381bfSpraks } else { 2007df2381bfSpraks /* 2008df2381bfSpraks * existing file got truncated, notify. 2009df2381bfSpraks */ 2010da6c28aaSamw vnevent_create(vp, ct); 20117c478bd9Sstevel@tonic-gate *vpp = vp; 2012df2381bfSpraks } 20137c478bd9Sstevel@tonic-gate return (error); 20147c478bd9Sstevel@tonic-gate } 20157c478bd9Sstevel@tonic-gate 20167c478bd9Sstevel@tonic-gate ASSERT(vattr.va_mask & AT_TYPE); 20177c478bd9Sstevel@tonic-gate if (vattr.va_type == VREG) { 20187c478bd9Sstevel@tonic-gate ASSERT(vattr.va_mask & AT_MODE); 20197c478bd9Sstevel@tonic-gate if (MANDMODE(vattr.va_mode)) { 20207c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 20217c478bd9Sstevel@tonic-gate return (EACCES); 20227c478bd9Sstevel@tonic-gate } 20237c478bd9Sstevel@tonic-gate } 20247c478bd9Sstevel@tonic-gate 20257c478bd9Sstevel@tonic-gate dnlc_remove(dvp, nm); 20267c478bd9Sstevel@tonic-gate 20277c478bd9Sstevel@tonic-gate setdiropargs(&args.ca_da, nm, dvp); 20287c478bd9Sstevel@tonic-gate 20297c478bd9Sstevel@tonic-gate /* 20307c478bd9Sstevel@tonic-gate * Decide what the group-id of the created file should be. 20317c478bd9Sstevel@tonic-gate * Set it in attribute list as advisory...then do a setattr 20327c478bd9Sstevel@tonic-gate * if the server didn't get it right the first time. 20337c478bd9Sstevel@tonic-gate */ 20347c478bd9Sstevel@tonic-gate error = setdirgid(dvp, &vattr.va_gid, cr); 20357c478bd9Sstevel@tonic-gate if (error) { 20367c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 20377c478bd9Sstevel@tonic-gate return (error); 20387c478bd9Sstevel@tonic-gate } 20397c478bd9Sstevel@tonic-gate vattr.va_mask |= AT_GID; 20407c478bd9Sstevel@tonic-gate 20417c478bd9Sstevel@tonic-gate /* 20427c478bd9Sstevel@tonic-gate * This is a completely gross hack to make mknod 20437c478bd9Sstevel@tonic-gate * work over the wire until we can wack the protocol 20447c478bd9Sstevel@tonic-gate */ 20457c478bd9Sstevel@tonic-gate #define IFCHR 0020000 /* character special */ 20467c478bd9Sstevel@tonic-gate #define IFBLK 0060000 /* block special */ 20477c478bd9Sstevel@tonic-gate #define IFSOCK 0140000 /* socket */ 20487c478bd9Sstevel@tonic-gate 20497c478bd9Sstevel@tonic-gate /* 20507c478bd9Sstevel@tonic-gate * dev_t is uint_t in 5.x and short in 4.x. Both 4.x 20517c478bd9Sstevel@tonic-gate * supports 8 bit majors. 5.x supports 14 bit majors. 5.x supports 18 20527c478bd9Sstevel@tonic-gate * bits in the minor number where 4.x supports 8 bits. If the 5.x 20537c478bd9Sstevel@tonic-gate * minor/major numbers <= 8 bits long, compress the device 20547c478bd9Sstevel@tonic-gate * number before sending it. Otherwise, the 4.x server will not 20557c478bd9Sstevel@tonic-gate * create the device with the correct device number and nothing can be 20567c478bd9Sstevel@tonic-gate * done about this. 20577c478bd9Sstevel@tonic-gate */ 20587c478bd9Sstevel@tonic-gate if (vattr.va_type == VCHR || vattr.va_type == VBLK) { 20597c478bd9Sstevel@tonic-gate dev_t d = vattr.va_rdev; 20607c478bd9Sstevel@tonic-gate dev32_t dev32; 20617c478bd9Sstevel@tonic-gate 20627c478bd9Sstevel@tonic-gate if (vattr.va_type == VCHR) 20637c478bd9Sstevel@tonic-gate vattr.va_mode |= IFCHR; 20647c478bd9Sstevel@tonic-gate else 20657c478bd9Sstevel@tonic-gate vattr.va_mode |= IFBLK; 20667c478bd9Sstevel@tonic-gate 20677c478bd9Sstevel@tonic-gate (void) cmpldev(&dev32, d); 20687c478bd9Sstevel@tonic-gate if (dev32 & ~((SO4_MAXMAJ << L_BITSMINOR32) | SO4_MAXMIN)) 20697c478bd9Sstevel@tonic-gate vattr.va_size = (u_offset_t)dev32; 20707c478bd9Sstevel@tonic-gate else 20717c478bd9Sstevel@tonic-gate vattr.va_size = (u_offset_t)nfsv2_cmpdev(d); 20727c478bd9Sstevel@tonic-gate 20737c478bd9Sstevel@tonic-gate vattr.va_mask |= AT_MODE|AT_SIZE; 20747c478bd9Sstevel@tonic-gate } else if (vattr.va_type == VFIFO) { 20757c478bd9Sstevel@tonic-gate vattr.va_mode |= IFCHR; /* xtra kludge for namedpipe */ 20767c478bd9Sstevel@tonic-gate vattr.va_size = (u_offset_t)NFS_FIFO_DEV; /* blech */ 20777c478bd9Sstevel@tonic-gate vattr.va_mask |= AT_MODE|AT_SIZE; 20787c478bd9Sstevel@tonic-gate } else if (vattr.va_type == VSOCK) { 20797c478bd9Sstevel@tonic-gate vattr.va_mode |= IFSOCK; 20807c478bd9Sstevel@tonic-gate /* 20817c478bd9Sstevel@tonic-gate * To avoid triggering bugs in the servers set AT_SIZE 20827c478bd9Sstevel@tonic-gate * (all other RFS_CREATE calls set this). 20837c478bd9Sstevel@tonic-gate */ 20847c478bd9Sstevel@tonic-gate vattr.va_size = 0; 20857c478bd9Sstevel@tonic-gate vattr.va_mask |= AT_MODE|AT_SIZE; 20867c478bd9Sstevel@tonic-gate } 20877c478bd9Sstevel@tonic-gate 20887c478bd9Sstevel@tonic-gate args.ca_sa = &args.ca_sa_buf; 20897c478bd9Sstevel@tonic-gate error = vattr_to_sattr(&vattr, args.ca_sa); 20907c478bd9Sstevel@tonic-gate if (error) { 20917c478bd9Sstevel@tonic-gate /* req time field(s) overflow - return immediately */ 20927c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 20937c478bd9Sstevel@tonic-gate return (error); 20947c478bd9Sstevel@tonic-gate } 20957c478bd9Sstevel@tonic-gate 20967c478bd9Sstevel@tonic-gate douprintf = 1; 20977c478bd9Sstevel@tonic-gate 20987c478bd9Sstevel@tonic-gate t = gethrtime(); 20997c478bd9Sstevel@tonic-gate 21007c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(dvp), RFS_CREATE, 21017c478bd9Sstevel@tonic-gate xdr_creatargs, (caddr_t)&args, 21027c478bd9Sstevel@tonic-gate xdr_diropres, (caddr_t)&dr, cr, 21037c478bd9Sstevel@tonic-gate &douprintf, &dr.dr_status, 0, NULL); 21047c478bd9Sstevel@tonic-gate 21057c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(dvp); /* mod time changed */ 21067c478bd9Sstevel@tonic-gate 21077c478bd9Sstevel@tonic-gate if (!error) { 21087c478bd9Sstevel@tonic-gate error = geterrno(dr.dr_status); 21097c478bd9Sstevel@tonic-gate if (!error) { 21107c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(drp)) 21117c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(dvp); 21127c478bd9Sstevel@tonic-gate vp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr, 21137c478bd9Sstevel@tonic-gate dvp->v_vfsp, t, cr, NULL, NULL); 21147c478bd9Sstevel@tonic-gate /* 21157c478bd9Sstevel@tonic-gate * If NFS_ACL is supported on the server, then the 21167c478bd9Sstevel@tonic-gate * attributes returned by server may have minimal 21177c478bd9Sstevel@tonic-gate * permissions sometimes denying access to users having 21187c478bd9Sstevel@tonic-gate * proper access. To get the proper attributes, mark 21197c478bd9Sstevel@tonic-gate * the attributes as expired so that they will be 21207c478bd9Sstevel@tonic-gate * regotten via the NFS_ACL GETATTR2 procedure. 21217c478bd9Sstevel@tonic-gate */ 21227c478bd9Sstevel@tonic-gate if (VTOMI(vp)->mi_flags & MI_ACL) { 21237c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); 21247c478bd9Sstevel@tonic-gate } 21257c478bd9Sstevel@tonic-gate dnlc_update(dvp, nm, vp); 21267c478bd9Sstevel@tonic-gate rp = VTOR(vp); 21277c478bd9Sstevel@tonic-gate if (vattr.va_size == 0) { 21287c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 21297c478bd9Sstevel@tonic-gate rp->r_size = 0; 21307c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 21317c478bd9Sstevel@tonic-gate if (vn_has_cached_data(vp)) { 21327c478bd9Sstevel@tonic-gate ASSERT(vp->v_type != VCHR); 21337c478bd9Sstevel@tonic-gate nfs_invalidate_pages(vp, 21347c478bd9Sstevel@tonic-gate (u_offset_t)0, cr); 21357c478bd9Sstevel@tonic-gate } 21367c478bd9Sstevel@tonic-gate } 21377c478bd9Sstevel@tonic-gate 21387c478bd9Sstevel@tonic-gate /* 21397c478bd9Sstevel@tonic-gate * Make sure the gid was set correctly. 21407c478bd9Sstevel@tonic-gate * If not, try to set it (but don't lose 21417c478bd9Sstevel@tonic-gate * any sleep over it). 21427c478bd9Sstevel@tonic-gate */ 21437c478bd9Sstevel@tonic-gate if (vattr.va_gid != rp->r_attr.va_gid) { 21447c478bd9Sstevel@tonic-gate vattr.va_mask = AT_GID; 21457c478bd9Sstevel@tonic-gate (void) nfssetattr(vp, &vattr, 0, cr); 21467c478bd9Sstevel@tonic-gate } 21477c478bd9Sstevel@tonic-gate 21487c478bd9Sstevel@tonic-gate /* 21497c478bd9Sstevel@tonic-gate * If vnode is a device create special vnode 21507c478bd9Sstevel@tonic-gate */ 21517c478bd9Sstevel@tonic-gate if (IS_DEVVP(vp)) { 21527c478bd9Sstevel@tonic-gate *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr); 21537c478bd9Sstevel@tonic-gate VN_RELE(vp); 21547c478bd9Sstevel@tonic-gate } else 21557c478bd9Sstevel@tonic-gate *vpp = vp; 21567c478bd9Sstevel@tonic-gate } else { 21577c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, dvp, cr); 21587c478bd9Sstevel@tonic-gate } 21597c478bd9Sstevel@tonic-gate } 21607c478bd9Sstevel@tonic-gate 21617c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 21627c478bd9Sstevel@tonic-gate 21637c478bd9Sstevel@tonic-gate return (error); 21647c478bd9Sstevel@tonic-gate } 21657c478bd9Sstevel@tonic-gate 21667c478bd9Sstevel@tonic-gate /* 21677c478bd9Sstevel@tonic-gate * Weirdness: if the vnode to be removed is open 21687c478bd9Sstevel@tonic-gate * we rename it instead of removing it and nfs_inactive 21697c478bd9Sstevel@tonic-gate * will remove the new name. 21707c478bd9Sstevel@tonic-gate */ 2171da6c28aaSamw /* ARGSUSED */ 21727c478bd9Sstevel@tonic-gate static int 2173da6c28aaSamw nfs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags) 21747c478bd9Sstevel@tonic-gate { 21757c478bd9Sstevel@tonic-gate int error; 21767c478bd9Sstevel@tonic-gate struct nfsdiropargs da; 21777c478bd9Sstevel@tonic-gate enum nfsstat status; 21787c478bd9Sstevel@tonic-gate vnode_t *vp; 21797c478bd9Sstevel@tonic-gate char *tmpname; 21807c478bd9Sstevel@tonic-gate int douprintf; 21817c478bd9Sstevel@tonic-gate rnode_t *rp; 21827c478bd9Sstevel@tonic-gate rnode_t *drp; 21837c478bd9Sstevel@tonic-gate 2184108322fbScarlsonj if (nfs_zone() != VTOMI(dvp)->mi_zone) 21857c478bd9Sstevel@tonic-gate return (EPERM); 21867c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 21877c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 21887c478bd9Sstevel@tonic-gate return (EINTR); 21897c478bd9Sstevel@tonic-gate 21907c478bd9Sstevel@tonic-gate error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 21917c478bd9Sstevel@tonic-gate if (error) { 21927c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 21937c478bd9Sstevel@tonic-gate return (error); 21947c478bd9Sstevel@tonic-gate } 21957c478bd9Sstevel@tonic-gate 21967c478bd9Sstevel@tonic-gate if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) { 21977c478bd9Sstevel@tonic-gate VN_RELE(vp); 21987c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 21997c478bd9Sstevel@tonic-gate return (EPERM); 22007c478bd9Sstevel@tonic-gate } 22017c478bd9Sstevel@tonic-gate 22027c478bd9Sstevel@tonic-gate /* 22037c478bd9Sstevel@tonic-gate * First just remove the entry from the name cache, as it 22047c478bd9Sstevel@tonic-gate * is most likely the only entry for this vp. 22057c478bd9Sstevel@tonic-gate */ 22067c478bd9Sstevel@tonic-gate dnlc_remove(dvp, nm); 22077c478bd9Sstevel@tonic-gate 22087c478bd9Sstevel@tonic-gate /* 22097c478bd9Sstevel@tonic-gate * If the file has a v_count > 1 then there may be more than one 22107c478bd9Sstevel@tonic-gate * entry in the name cache due multiple links or an open file, 22117c478bd9Sstevel@tonic-gate * but we don't have the real reference count so flush all 22127c478bd9Sstevel@tonic-gate * possible entries. 22137c478bd9Sstevel@tonic-gate */ 22147c478bd9Sstevel@tonic-gate if (vp->v_count > 1) 22157c478bd9Sstevel@tonic-gate dnlc_purge_vp(vp); 22167c478bd9Sstevel@tonic-gate 22177c478bd9Sstevel@tonic-gate /* 22187c478bd9Sstevel@tonic-gate * Now we have the real reference count on the vnode 22197c478bd9Sstevel@tonic-gate */ 22207c478bd9Sstevel@tonic-gate rp = VTOR(vp); 22217c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 22227c478bd9Sstevel@tonic-gate if (vp->v_count > 1 && 22237c478bd9Sstevel@tonic-gate (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) { 22247c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 22257c478bd9Sstevel@tonic-gate tmpname = newname(); 2226da6c28aaSamw error = nfsrename(dvp, nm, dvp, tmpname, cr, ct); 22277c478bd9Sstevel@tonic-gate if (error) 22287c478bd9Sstevel@tonic-gate kmem_free(tmpname, MAXNAMELEN); 22297c478bd9Sstevel@tonic-gate else { 22307c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 22317c478bd9Sstevel@tonic-gate if (rp->r_unldvp == NULL) { 22327c478bd9Sstevel@tonic-gate VN_HOLD(dvp); 22337c478bd9Sstevel@tonic-gate rp->r_unldvp = dvp; 22347c478bd9Sstevel@tonic-gate if (rp->r_unlcred != NULL) 22357c478bd9Sstevel@tonic-gate crfree(rp->r_unlcred); 22367c478bd9Sstevel@tonic-gate crhold(cr); 22377c478bd9Sstevel@tonic-gate rp->r_unlcred = cr; 22387c478bd9Sstevel@tonic-gate rp->r_unlname = tmpname; 22397c478bd9Sstevel@tonic-gate } else { 22407c478bd9Sstevel@tonic-gate kmem_free(rp->r_unlname, MAXNAMELEN); 22417c478bd9Sstevel@tonic-gate rp->r_unlname = tmpname; 22427c478bd9Sstevel@tonic-gate } 22437c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 22447c478bd9Sstevel@tonic-gate } 22457c478bd9Sstevel@tonic-gate } else { 22467c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 22477c478bd9Sstevel@tonic-gate /* 22487c478bd9Sstevel@tonic-gate * We need to flush any dirty pages which happen to 22497c478bd9Sstevel@tonic-gate * be hanging around before removing the file. This 22507c478bd9Sstevel@tonic-gate * shouldn't happen very often and mostly on file 22517c478bd9Sstevel@tonic-gate * systems mounted "nocto". 22527c478bd9Sstevel@tonic-gate */ 22537c478bd9Sstevel@tonic-gate if (vn_has_cached_data(vp) && 22547c478bd9Sstevel@tonic-gate ((rp->r_flags & RDIRTY) || rp->r_count > 0)) { 2255da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct); 22567c478bd9Sstevel@tonic-gate if (error && (error == ENOSPC || error == EDQUOT)) { 22577c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 22587c478bd9Sstevel@tonic-gate if (!rp->r_error) 22597c478bd9Sstevel@tonic-gate rp->r_error = error; 22607c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 22617c478bd9Sstevel@tonic-gate } 22627c478bd9Sstevel@tonic-gate } 22637c478bd9Sstevel@tonic-gate 22647c478bd9Sstevel@tonic-gate setdiropargs(&da, nm, dvp); 22657c478bd9Sstevel@tonic-gate 22667c478bd9Sstevel@tonic-gate douprintf = 1; 22677c478bd9Sstevel@tonic-gate 22687c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(dvp), RFS_REMOVE, 22697c478bd9Sstevel@tonic-gate xdr_diropargs, (caddr_t)&da, 22707c478bd9Sstevel@tonic-gate xdr_enum, (caddr_t)&status, cr, 22717c478bd9Sstevel@tonic-gate &douprintf, &status, 0, NULL); 22727c478bd9Sstevel@tonic-gate 22737c478bd9Sstevel@tonic-gate /* 22747c478bd9Sstevel@tonic-gate * The xattr dir may be gone after last attr is removed, 22757c478bd9Sstevel@tonic-gate * so flush it from dnlc. 22767c478bd9Sstevel@tonic-gate */ 22777c478bd9Sstevel@tonic-gate if (dvp->v_flag & V_XATTRDIR) 22787c478bd9Sstevel@tonic-gate dnlc_purge_vp(dvp); 22797c478bd9Sstevel@tonic-gate 22807c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(dvp); /* mod time changed */ 22817c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(vp); /* link count changed */ 22827c478bd9Sstevel@tonic-gate 22837c478bd9Sstevel@tonic-gate if (!error) { 22847c478bd9Sstevel@tonic-gate error = geterrno(status); 22857c478bd9Sstevel@tonic-gate if (!error) { 22867c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(drp)) 22877c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(dvp); 22887c478bd9Sstevel@tonic-gate } else { 22897c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, dvp, cr); 22907c478bd9Sstevel@tonic-gate } 22917c478bd9Sstevel@tonic-gate } 22927c478bd9Sstevel@tonic-gate } 22937c478bd9Sstevel@tonic-gate 2294df2381bfSpraks if (error == 0) { 2295da6c28aaSamw vnevent_remove(vp, dvp, nm, ct); 2296df2381bfSpraks } 22977c478bd9Sstevel@tonic-gate VN_RELE(vp); 22987c478bd9Sstevel@tonic-gate 22997c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 23007c478bd9Sstevel@tonic-gate 23017c478bd9Sstevel@tonic-gate return (error); 23027c478bd9Sstevel@tonic-gate } 23037c478bd9Sstevel@tonic-gate 2304da6c28aaSamw /* ARGSUSED */ 23057c478bd9Sstevel@tonic-gate static int 2306da6c28aaSamw nfs_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr, 2307da6c28aaSamw caller_context_t *ct, int flags) 23087c478bd9Sstevel@tonic-gate { 23097c478bd9Sstevel@tonic-gate int error; 23107c478bd9Sstevel@tonic-gate struct nfslinkargs args; 23117c478bd9Sstevel@tonic-gate enum nfsstat status; 23127c478bd9Sstevel@tonic-gate vnode_t *realvp; 23137c478bd9Sstevel@tonic-gate int douprintf; 23147c478bd9Sstevel@tonic-gate rnode_t *tdrp; 23157c478bd9Sstevel@tonic-gate 2316108322fbScarlsonj if (nfs_zone() != VTOMI(tdvp)->mi_zone) 23177c478bd9Sstevel@tonic-gate return (EPERM); 2318da6c28aaSamw if (VOP_REALVP(svp, &realvp, ct) == 0) 23197c478bd9Sstevel@tonic-gate svp = realvp; 23207c478bd9Sstevel@tonic-gate 23217c478bd9Sstevel@tonic-gate args.la_from = VTOFH(svp); 23227c478bd9Sstevel@tonic-gate setdiropargs(&args.la_to, tnm, tdvp); 23237c478bd9Sstevel@tonic-gate 23247c478bd9Sstevel@tonic-gate tdrp = VTOR(tdvp); 23257c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp))) 23267c478bd9Sstevel@tonic-gate return (EINTR); 23277c478bd9Sstevel@tonic-gate 23287c478bd9Sstevel@tonic-gate dnlc_remove(tdvp, tnm); 23297c478bd9Sstevel@tonic-gate 23307c478bd9Sstevel@tonic-gate douprintf = 1; 23317c478bd9Sstevel@tonic-gate 23327c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(svp), RFS_LINK, 23337c478bd9Sstevel@tonic-gate xdr_linkargs, (caddr_t)&args, 23347c478bd9Sstevel@tonic-gate xdr_enum, (caddr_t)&status, cr, 23357c478bd9Sstevel@tonic-gate &douprintf, &status, 0, NULL); 23367c478bd9Sstevel@tonic-gate 23377c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(tdvp); /* mod time changed */ 23387c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(svp); /* link count changed */ 23397c478bd9Sstevel@tonic-gate 23407c478bd9Sstevel@tonic-gate if (!error) { 23417c478bd9Sstevel@tonic-gate error = geterrno(status); 23427c478bd9Sstevel@tonic-gate if (!error) { 23437c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(tdrp)) 23447c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(tdvp); 23457c478bd9Sstevel@tonic-gate } 23467c478bd9Sstevel@tonic-gate } 23477c478bd9Sstevel@tonic-gate 23487c478bd9Sstevel@tonic-gate nfs_rw_exit(&tdrp->r_rwlock); 23497c478bd9Sstevel@tonic-gate 2350df2381bfSpraks if (!error) { 2351df2381bfSpraks /* 2352df2381bfSpraks * Notify the source file of this link operation. 2353df2381bfSpraks */ 2354da6c28aaSamw vnevent_link(svp, ct); 2355df2381bfSpraks } 23567c478bd9Sstevel@tonic-gate return (error); 23577c478bd9Sstevel@tonic-gate } 23587c478bd9Sstevel@tonic-gate 2359da6c28aaSamw /* ARGSUSED */ 23607c478bd9Sstevel@tonic-gate static int 2361da6c28aaSamw nfs_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr, 2362da6c28aaSamw caller_context_t *ct, int flags) 23637c478bd9Sstevel@tonic-gate { 23647c478bd9Sstevel@tonic-gate vnode_t *realvp; 23657c478bd9Sstevel@tonic-gate 2366108322fbScarlsonj if (nfs_zone() != VTOMI(odvp)->mi_zone) 23677c478bd9Sstevel@tonic-gate return (EPERM); 2368da6c28aaSamw if (VOP_REALVP(ndvp, &realvp, ct) == 0) 23697c478bd9Sstevel@tonic-gate ndvp = realvp; 23707c478bd9Sstevel@tonic-gate 2371da6c28aaSamw return (nfsrename(odvp, onm, ndvp, nnm, cr, ct)); 23727c478bd9Sstevel@tonic-gate } 23737c478bd9Sstevel@tonic-gate 23747c478bd9Sstevel@tonic-gate /* 23757c478bd9Sstevel@tonic-gate * nfsrename does the real work of renaming in NFS Version 2. 23767c478bd9Sstevel@tonic-gate */ 23777c478bd9Sstevel@tonic-gate static int 2378da6c28aaSamw nfsrename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr, 2379da6c28aaSamw caller_context_t *ct) 23807c478bd9Sstevel@tonic-gate { 23817c478bd9Sstevel@tonic-gate int error; 23827c478bd9Sstevel@tonic-gate enum nfsstat status; 23837c478bd9Sstevel@tonic-gate struct nfsrnmargs args; 23847c478bd9Sstevel@tonic-gate int douprintf; 2385df2381bfSpraks vnode_t *nvp = NULL; 23867c478bd9Sstevel@tonic-gate vnode_t *ovp = NULL; 23877c478bd9Sstevel@tonic-gate char *tmpname; 23887c478bd9Sstevel@tonic-gate rnode_t *rp; 23897c478bd9Sstevel@tonic-gate rnode_t *odrp; 23907c478bd9Sstevel@tonic-gate rnode_t *ndrp; 23917c478bd9Sstevel@tonic-gate 2392108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone); 23937c478bd9Sstevel@tonic-gate if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 || 23947c478bd9Sstevel@tonic-gate strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0) 23957c478bd9Sstevel@tonic-gate return (EINVAL); 23967c478bd9Sstevel@tonic-gate 23977c478bd9Sstevel@tonic-gate odrp = VTOR(odvp); 23987c478bd9Sstevel@tonic-gate ndrp = VTOR(ndvp); 23997c478bd9Sstevel@tonic-gate if ((intptr_t)odrp < (intptr_t)ndrp) { 24007c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) 24017c478bd9Sstevel@tonic-gate return (EINTR); 24027c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) { 24037c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 24047c478bd9Sstevel@tonic-gate return (EINTR); 24057c478bd9Sstevel@tonic-gate } 24067c478bd9Sstevel@tonic-gate } else { 24077c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) 24087c478bd9Sstevel@tonic-gate return (EINTR); 24097c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) { 24107c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 24117c478bd9Sstevel@tonic-gate return (EINTR); 24127c478bd9Sstevel@tonic-gate } 24137c478bd9Sstevel@tonic-gate } 24147c478bd9Sstevel@tonic-gate 24157c478bd9Sstevel@tonic-gate /* 24167c478bd9Sstevel@tonic-gate * Lookup the target file. If it exists, it needs to be 24177c478bd9Sstevel@tonic-gate * checked to see whether it is a mount point and whether 24187c478bd9Sstevel@tonic-gate * it is active (open). 24197c478bd9Sstevel@tonic-gate */ 24207c478bd9Sstevel@tonic-gate error = nfslookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0); 24217c478bd9Sstevel@tonic-gate if (!error) { 24227c478bd9Sstevel@tonic-gate /* 24237c478bd9Sstevel@tonic-gate * If this file has been mounted on, then just 24247c478bd9Sstevel@tonic-gate * return busy because renaming to it would remove 24257c478bd9Sstevel@tonic-gate * the mounted file system from the name space. 24267c478bd9Sstevel@tonic-gate */ 24277c478bd9Sstevel@tonic-gate if (vn_mountedvfs(nvp) != NULL) { 24287c478bd9Sstevel@tonic-gate VN_RELE(nvp); 24297c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 24307c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 24317c478bd9Sstevel@tonic-gate return (EBUSY); 24327c478bd9Sstevel@tonic-gate } 24337c478bd9Sstevel@tonic-gate 24347c478bd9Sstevel@tonic-gate /* 24357c478bd9Sstevel@tonic-gate * Purge the name cache of all references to this vnode 24367c478bd9Sstevel@tonic-gate * so that we can check the reference count to infer 24377c478bd9Sstevel@tonic-gate * whether it is active or not. 24387c478bd9Sstevel@tonic-gate */ 24397c478bd9Sstevel@tonic-gate /* 24407c478bd9Sstevel@tonic-gate * First just remove the entry from the name cache, as it 24417c478bd9Sstevel@tonic-gate * is most likely the only entry for this vp. 24427c478bd9Sstevel@tonic-gate */ 24437c478bd9Sstevel@tonic-gate dnlc_remove(ndvp, nnm); 24447c478bd9Sstevel@tonic-gate /* 24457c478bd9Sstevel@tonic-gate * If the file has a v_count > 1 then there may be more 24467c478bd9Sstevel@tonic-gate * than one entry in the name cache due multiple links 24477c478bd9Sstevel@tonic-gate * or an open file, but we don't have the real reference 24487c478bd9Sstevel@tonic-gate * count so flush all possible entries. 24497c478bd9Sstevel@tonic-gate */ 24507c478bd9Sstevel@tonic-gate if (nvp->v_count > 1) 24517c478bd9Sstevel@tonic-gate dnlc_purge_vp(nvp); 24527c478bd9Sstevel@tonic-gate 24537c478bd9Sstevel@tonic-gate /* 24547c478bd9Sstevel@tonic-gate * If the vnode is active and is not a directory, 24557c478bd9Sstevel@tonic-gate * arrange to rename it to a 24567c478bd9Sstevel@tonic-gate * temporary file so that it will continue to be 24577c478bd9Sstevel@tonic-gate * accessible. This implements the "unlink-open-file" 24587c478bd9Sstevel@tonic-gate * semantics for the target of a rename operation. 24597c478bd9Sstevel@tonic-gate * Before doing this though, make sure that the 24607c478bd9Sstevel@tonic-gate * source and target files are not already the same. 24617c478bd9Sstevel@tonic-gate */ 24627c478bd9Sstevel@tonic-gate if (nvp->v_count > 1 && nvp->v_type != VDIR) { 24637c478bd9Sstevel@tonic-gate /* 24647c478bd9Sstevel@tonic-gate * Lookup the source name. 24657c478bd9Sstevel@tonic-gate */ 24667c478bd9Sstevel@tonic-gate error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL, 24677c478bd9Sstevel@tonic-gate cr, 0); 24687c478bd9Sstevel@tonic-gate 24697c478bd9Sstevel@tonic-gate /* 24707c478bd9Sstevel@tonic-gate * The source name *should* already exist. 24717c478bd9Sstevel@tonic-gate */ 24727c478bd9Sstevel@tonic-gate if (error) { 24737c478bd9Sstevel@tonic-gate VN_RELE(nvp); 24747c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 24757c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 24767c478bd9Sstevel@tonic-gate return (error); 24777c478bd9Sstevel@tonic-gate } 24787c478bd9Sstevel@tonic-gate 24797c478bd9Sstevel@tonic-gate /* 24807c478bd9Sstevel@tonic-gate * Compare the two vnodes. If they are the same, 24817c478bd9Sstevel@tonic-gate * just release all held vnodes and return success. 24827c478bd9Sstevel@tonic-gate */ 24837c478bd9Sstevel@tonic-gate if (ovp == nvp) { 24847c478bd9Sstevel@tonic-gate VN_RELE(ovp); 24857c478bd9Sstevel@tonic-gate VN_RELE(nvp); 24867c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 24877c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 24887c478bd9Sstevel@tonic-gate return (0); 24897c478bd9Sstevel@tonic-gate } 24907c478bd9Sstevel@tonic-gate 24917c478bd9Sstevel@tonic-gate /* 24927c478bd9Sstevel@tonic-gate * Can't mix and match directories and non- 24937c478bd9Sstevel@tonic-gate * directories in rename operations. We already 24947c478bd9Sstevel@tonic-gate * know that the target is not a directory. If 24957c478bd9Sstevel@tonic-gate * the source is a directory, return an error. 24967c478bd9Sstevel@tonic-gate */ 24977c478bd9Sstevel@tonic-gate if (ovp->v_type == VDIR) { 24987c478bd9Sstevel@tonic-gate VN_RELE(ovp); 24997c478bd9Sstevel@tonic-gate VN_RELE(nvp); 25007c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 25017c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 25027c478bd9Sstevel@tonic-gate return (ENOTDIR); 25037c478bd9Sstevel@tonic-gate } 25047c478bd9Sstevel@tonic-gate 25057c478bd9Sstevel@tonic-gate /* 25067c478bd9Sstevel@tonic-gate * The target file exists, is not the same as 25077c478bd9Sstevel@tonic-gate * the source file, and is active. Link it 25087c478bd9Sstevel@tonic-gate * to a temporary filename to avoid having 25097c478bd9Sstevel@tonic-gate * the server removing the file completely. 25107c478bd9Sstevel@tonic-gate */ 25117c478bd9Sstevel@tonic-gate tmpname = newname(); 2512da6c28aaSamw error = nfs_link(ndvp, nvp, tmpname, cr, NULL, 0); 25137c478bd9Sstevel@tonic-gate if (error == EOPNOTSUPP) { 25147c478bd9Sstevel@tonic-gate error = nfs_rename(ndvp, nnm, ndvp, tmpname, 2515da6c28aaSamw cr, NULL, 0); 25167c478bd9Sstevel@tonic-gate } 25177c478bd9Sstevel@tonic-gate if (error) { 25187c478bd9Sstevel@tonic-gate kmem_free(tmpname, MAXNAMELEN); 25197c478bd9Sstevel@tonic-gate VN_RELE(ovp); 25207c478bd9Sstevel@tonic-gate VN_RELE(nvp); 25217c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 25227c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 25237c478bd9Sstevel@tonic-gate return (error); 25247c478bd9Sstevel@tonic-gate } 25257c478bd9Sstevel@tonic-gate rp = VTOR(nvp); 25267c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 25277c478bd9Sstevel@tonic-gate if (rp->r_unldvp == NULL) { 25287c478bd9Sstevel@tonic-gate VN_HOLD(ndvp); 25297c478bd9Sstevel@tonic-gate rp->r_unldvp = ndvp; 25307c478bd9Sstevel@tonic-gate if (rp->r_unlcred != NULL) 25317c478bd9Sstevel@tonic-gate crfree(rp->r_unlcred); 25327c478bd9Sstevel@tonic-gate crhold(cr); 25337c478bd9Sstevel@tonic-gate rp->r_unlcred = cr; 25347c478bd9Sstevel@tonic-gate rp->r_unlname = tmpname; 25357c478bd9Sstevel@tonic-gate } else { 25367c478bd9Sstevel@tonic-gate kmem_free(rp->r_unlname, MAXNAMELEN); 25377c478bd9Sstevel@tonic-gate rp->r_unlname = tmpname; 25387c478bd9Sstevel@tonic-gate } 25397c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 25407c478bd9Sstevel@tonic-gate } 25417c478bd9Sstevel@tonic-gate } 25427c478bd9Sstevel@tonic-gate 25437c478bd9Sstevel@tonic-gate if (ovp == NULL) { 25447c478bd9Sstevel@tonic-gate /* 25457c478bd9Sstevel@tonic-gate * When renaming directories to be a subdirectory of a 25467c478bd9Sstevel@tonic-gate * different parent, the dnlc entry for ".." will no 25477c478bd9Sstevel@tonic-gate * longer be valid, so it must be removed. 25487c478bd9Sstevel@tonic-gate * 25497c478bd9Sstevel@tonic-gate * We do a lookup here to determine whether we are renaming 25507c478bd9Sstevel@tonic-gate * a directory and we need to check if we are renaming 25517c478bd9Sstevel@tonic-gate * an unlinked file. This might have already been done 25527c478bd9Sstevel@tonic-gate * in previous code, so we check ovp == NULL to avoid 25537c478bd9Sstevel@tonic-gate * doing it twice. 25547c478bd9Sstevel@tonic-gate */ 25557c478bd9Sstevel@tonic-gate 25567c478bd9Sstevel@tonic-gate error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0); 25577c478bd9Sstevel@tonic-gate 25587c478bd9Sstevel@tonic-gate /* 25597c478bd9Sstevel@tonic-gate * The source name *should* already exist. 25607c478bd9Sstevel@tonic-gate */ 25617c478bd9Sstevel@tonic-gate if (error) { 25627c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 25637c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 2564df2381bfSpraks if (nvp) { 2565df2381bfSpraks VN_RELE(nvp); 2566df2381bfSpraks } 25677c478bd9Sstevel@tonic-gate return (error); 25687c478bd9Sstevel@tonic-gate } 25697c478bd9Sstevel@tonic-gate ASSERT(ovp != NULL); 25707c478bd9Sstevel@tonic-gate } 25717c478bd9Sstevel@tonic-gate 25727c478bd9Sstevel@tonic-gate dnlc_remove(odvp, onm); 25737c478bd9Sstevel@tonic-gate dnlc_remove(ndvp, nnm); 25747c478bd9Sstevel@tonic-gate 25757c478bd9Sstevel@tonic-gate setdiropargs(&args.rna_from, onm, odvp); 25767c478bd9Sstevel@tonic-gate setdiropargs(&args.rna_to, nnm, ndvp); 25777c478bd9Sstevel@tonic-gate 25787c478bd9Sstevel@tonic-gate douprintf = 1; 25797c478bd9Sstevel@tonic-gate 25807c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(odvp), RFS_RENAME, 25817c478bd9Sstevel@tonic-gate xdr_rnmargs, (caddr_t)&args, 25827c478bd9Sstevel@tonic-gate xdr_enum, (caddr_t)&status, cr, 25837c478bd9Sstevel@tonic-gate &douprintf, &status, 0, NULL); 25847c478bd9Sstevel@tonic-gate 25857c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(odvp); /* mod time changed */ 25867c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(ndvp); /* mod time changed */ 25877c478bd9Sstevel@tonic-gate 25887c478bd9Sstevel@tonic-gate if (!error) { 25897c478bd9Sstevel@tonic-gate error = geterrno(status); 25907c478bd9Sstevel@tonic-gate if (!error) { 25917c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(odrp)) 25927c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(odvp); 25937c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(ndrp)) 25947c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(ndvp); 25957c478bd9Sstevel@tonic-gate /* 25967c478bd9Sstevel@tonic-gate * when renaming directories to be a subdirectory of a 25977c478bd9Sstevel@tonic-gate * different parent, the dnlc entry for ".." will no 25987c478bd9Sstevel@tonic-gate * longer be valid, so it must be removed 25997c478bd9Sstevel@tonic-gate */ 26007c478bd9Sstevel@tonic-gate rp = VTOR(ovp); 26017c478bd9Sstevel@tonic-gate if (ndvp != odvp) { 26027c478bd9Sstevel@tonic-gate if (ovp->v_type == VDIR) { 26037c478bd9Sstevel@tonic-gate dnlc_remove(ovp, ".."); 26047c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(rp)) 26057c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(ovp); 26067c478bd9Sstevel@tonic-gate } 26077c478bd9Sstevel@tonic-gate } 26087c478bd9Sstevel@tonic-gate 26097c478bd9Sstevel@tonic-gate /* 26107c478bd9Sstevel@tonic-gate * If we are renaming the unlinked file, update the 26117c478bd9Sstevel@tonic-gate * r_unldvp and r_unlname as needed. 26127c478bd9Sstevel@tonic-gate */ 26137c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 26147c478bd9Sstevel@tonic-gate if (rp->r_unldvp != NULL) { 26157c478bd9Sstevel@tonic-gate if (strcmp(rp->r_unlname, onm) == 0) { 26167c478bd9Sstevel@tonic-gate (void) strncpy(rp->r_unlname, 26177c478bd9Sstevel@tonic-gate nnm, MAXNAMELEN); 26187c478bd9Sstevel@tonic-gate rp->r_unlname[MAXNAMELEN - 1] = '\0'; 26197c478bd9Sstevel@tonic-gate 26207c478bd9Sstevel@tonic-gate if (ndvp != rp->r_unldvp) { 26217c478bd9Sstevel@tonic-gate VN_RELE(rp->r_unldvp); 26227c478bd9Sstevel@tonic-gate rp->r_unldvp = ndvp; 26237c478bd9Sstevel@tonic-gate VN_HOLD(ndvp); 26247c478bd9Sstevel@tonic-gate } 26257c478bd9Sstevel@tonic-gate } 26267c478bd9Sstevel@tonic-gate } 26277c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 26287c478bd9Sstevel@tonic-gate } else { 26297c478bd9Sstevel@tonic-gate /* 26307c478bd9Sstevel@tonic-gate * System V defines rename to return EEXIST, not 26317c478bd9Sstevel@tonic-gate * ENOTEMPTY if the target directory is not empty. 26327c478bd9Sstevel@tonic-gate * Over the wire, the error is NFSERR_ENOTEMPTY 26337c478bd9Sstevel@tonic-gate * which geterrno maps to ENOTEMPTY. 26347c478bd9Sstevel@tonic-gate */ 26357c478bd9Sstevel@tonic-gate if (error == ENOTEMPTY) 26367c478bd9Sstevel@tonic-gate error = EEXIST; 26377c478bd9Sstevel@tonic-gate } 26387c478bd9Sstevel@tonic-gate } 26397c478bd9Sstevel@tonic-gate 2640df2381bfSpraks if (error == 0) { 2641df2381bfSpraks if (nvp) 2642da6c28aaSamw vnevent_rename_dest(nvp, ndvp, nnm, ct); 2643df2381bfSpraks 2644df2381bfSpraks if (odvp != ndvp) 2645da6c28aaSamw vnevent_rename_dest_dir(ndvp, ct); 2646df2381bfSpraks 2647df2381bfSpraks ASSERT(ovp != NULL); 2648da6c28aaSamw vnevent_rename_src(ovp, odvp, onm, ct); 2649df2381bfSpraks } 2650df2381bfSpraks 2651df2381bfSpraks if (nvp) { 2652df2381bfSpraks VN_RELE(nvp); 2653df2381bfSpraks } 26547c478bd9Sstevel@tonic-gate VN_RELE(ovp); 26557c478bd9Sstevel@tonic-gate 26567c478bd9Sstevel@tonic-gate nfs_rw_exit(&odrp->r_rwlock); 26577c478bd9Sstevel@tonic-gate nfs_rw_exit(&ndrp->r_rwlock); 26587c478bd9Sstevel@tonic-gate 26597c478bd9Sstevel@tonic-gate return (error); 26607c478bd9Sstevel@tonic-gate } 26617c478bd9Sstevel@tonic-gate 2662da6c28aaSamw /* ARGSUSED */ 26637c478bd9Sstevel@tonic-gate static int 2664da6c28aaSamw nfs_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr, 2665da6c28aaSamw caller_context_t *ct, int flags, vsecattr_t *vsecp) 26667c478bd9Sstevel@tonic-gate { 26677c478bd9Sstevel@tonic-gate int error; 26687c478bd9Sstevel@tonic-gate struct nfscreatargs args; 26697c478bd9Sstevel@tonic-gate struct nfsdiropres dr; 26707c478bd9Sstevel@tonic-gate int douprintf; 26717c478bd9Sstevel@tonic-gate rnode_t *drp; 26727c478bd9Sstevel@tonic-gate hrtime_t t; 26737c478bd9Sstevel@tonic-gate 2674108322fbScarlsonj if (nfs_zone() != VTOMI(dvp)->mi_zone) 26757c478bd9Sstevel@tonic-gate return (EPERM); 26767c478bd9Sstevel@tonic-gate 26777c478bd9Sstevel@tonic-gate setdiropargs(&args.ca_da, nm, dvp); 26787c478bd9Sstevel@tonic-gate 26797c478bd9Sstevel@tonic-gate /* 26807c478bd9Sstevel@tonic-gate * Decide what the group-id and set-gid bit of the created directory 26817c478bd9Sstevel@tonic-gate * should be. May have to do a setattr to get the gid right. 26827c478bd9Sstevel@tonic-gate */ 26837c478bd9Sstevel@tonic-gate error = setdirgid(dvp, &va->va_gid, cr); 26847c478bd9Sstevel@tonic-gate if (error) 26857c478bd9Sstevel@tonic-gate return (error); 26867c478bd9Sstevel@tonic-gate error = setdirmode(dvp, &va->va_mode, cr); 26877c478bd9Sstevel@tonic-gate if (error) 26887c478bd9Sstevel@tonic-gate return (error); 26897c478bd9Sstevel@tonic-gate va->va_mask |= AT_MODE|AT_GID; 26907c478bd9Sstevel@tonic-gate 26917c478bd9Sstevel@tonic-gate args.ca_sa = &args.ca_sa_buf; 26927c478bd9Sstevel@tonic-gate error = vattr_to_sattr(va, args.ca_sa); 26937c478bd9Sstevel@tonic-gate if (error) { 26947c478bd9Sstevel@tonic-gate /* req time field(s) overflow - return immediately */ 26957c478bd9Sstevel@tonic-gate return (error); 26967c478bd9Sstevel@tonic-gate } 26977c478bd9Sstevel@tonic-gate 26987c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 26997c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 27007c478bd9Sstevel@tonic-gate return (EINTR); 27017c478bd9Sstevel@tonic-gate 27027c478bd9Sstevel@tonic-gate dnlc_remove(dvp, nm); 27037c478bd9Sstevel@tonic-gate 27047c478bd9Sstevel@tonic-gate douprintf = 1; 27057c478bd9Sstevel@tonic-gate 27067c478bd9Sstevel@tonic-gate t = gethrtime(); 27077c478bd9Sstevel@tonic-gate 27087c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(dvp), RFS_MKDIR, 27097c478bd9Sstevel@tonic-gate xdr_creatargs, (caddr_t)&args, 27107c478bd9Sstevel@tonic-gate xdr_diropres, (caddr_t)&dr, cr, 27117c478bd9Sstevel@tonic-gate &douprintf, &dr.dr_status, 0, NULL); 27127c478bd9Sstevel@tonic-gate 27137c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(dvp); /* mod time changed */ 27147c478bd9Sstevel@tonic-gate 27157c478bd9Sstevel@tonic-gate if (!error) { 27167c478bd9Sstevel@tonic-gate error = geterrno(dr.dr_status); 27177c478bd9Sstevel@tonic-gate if (!error) { 27187c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(drp)) 27197c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(dvp); 27207c478bd9Sstevel@tonic-gate /* 27217c478bd9Sstevel@tonic-gate * The attributes returned by RFS_MKDIR can not 27227c478bd9Sstevel@tonic-gate * be depended upon, so mark the attribute cache 27237c478bd9Sstevel@tonic-gate * as purged. A subsequent GETATTR will get the 27247c478bd9Sstevel@tonic-gate * correct attributes from the server. 27257c478bd9Sstevel@tonic-gate */ 27267c478bd9Sstevel@tonic-gate *vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr, 27277c478bd9Sstevel@tonic-gate dvp->v_vfsp, t, cr, NULL, NULL); 27287c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(*vpp); 27297c478bd9Sstevel@tonic-gate dnlc_update(dvp, nm, *vpp); 27307c478bd9Sstevel@tonic-gate 27317c478bd9Sstevel@tonic-gate /* 27327c478bd9Sstevel@tonic-gate * Make sure the gid was set correctly. 27337c478bd9Sstevel@tonic-gate * If not, try to set it (but don't lose 27347c478bd9Sstevel@tonic-gate * any sleep over it). 27357c478bd9Sstevel@tonic-gate */ 27367c478bd9Sstevel@tonic-gate if (va->va_gid != VTOR(*vpp)->r_attr.va_gid) { 27377c478bd9Sstevel@tonic-gate va->va_mask = AT_GID; 27387c478bd9Sstevel@tonic-gate (void) nfssetattr(*vpp, va, 0, cr); 27397c478bd9Sstevel@tonic-gate } 27407c478bd9Sstevel@tonic-gate } else { 27417c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, dvp, cr); 27427c478bd9Sstevel@tonic-gate } 27437c478bd9Sstevel@tonic-gate } 27447c478bd9Sstevel@tonic-gate 27457c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 27467c478bd9Sstevel@tonic-gate 27477c478bd9Sstevel@tonic-gate return (error); 27487c478bd9Sstevel@tonic-gate } 27497c478bd9Sstevel@tonic-gate 2750da6c28aaSamw /* ARGSUSED */ 27517c478bd9Sstevel@tonic-gate static int 2752da6c28aaSamw nfs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr, 2753da6c28aaSamw caller_context_t *ct, int flags) 27547c478bd9Sstevel@tonic-gate { 27557c478bd9Sstevel@tonic-gate int error; 27567c478bd9Sstevel@tonic-gate enum nfsstat status; 27577c478bd9Sstevel@tonic-gate struct nfsdiropargs da; 27587c478bd9Sstevel@tonic-gate vnode_t *vp; 27597c478bd9Sstevel@tonic-gate int douprintf; 27607c478bd9Sstevel@tonic-gate rnode_t *drp; 27617c478bd9Sstevel@tonic-gate 2762108322fbScarlsonj if (nfs_zone() != VTOMI(dvp)->mi_zone) 27637c478bd9Sstevel@tonic-gate return (EPERM); 27647c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 27657c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 27667c478bd9Sstevel@tonic-gate return (EINTR); 27677c478bd9Sstevel@tonic-gate 27687c478bd9Sstevel@tonic-gate /* 27697c478bd9Sstevel@tonic-gate * Attempt to prevent a rmdir(".") from succeeding. 27707c478bd9Sstevel@tonic-gate */ 27717c478bd9Sstevel@tonic-gate error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0); 27727c478bd9Sstevel@tonic-gate if (error) { 27737c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 27747c478bd9Sstevel@tonic-gate return (error); 27757c478bd9Sstevel@tonic-gate } 27767c478bd9Sstevel@tonic-gate 27777c478bd9Sstevel@tonic-gate if (vp == cdir) { 27787c478bd9Sstevel@tonic-gate VN_RELE(vp); 27797c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 27807c478bd9Sstevel@tonic-gate return (EINVAL); 27817c478bd9Sstevel@tonic-gate } 27827c478bd9Sstevel@tonic-gate 27837c478bd9Sstevel@tonic-gate setdiropargs(&da, nm, dvp); 27847c478bd9Sstevel@tonic-gate 27857c478bd9Sstevel@tonic-gate /* 27867c478bd9Sstevel@tonic-gate * First just remove the entry from the name cache, as it 27877c478bd9Sstevel@tonic-gate * is most likely an entry for this vp. 27887c478bd9Sstevel@tonic-gate */ 27897c478bd9Sstevel@tonic-gate dnlc_remove(dvp, nm); 27907c478bd9Sstevel@tonic-gate 27917c478bd9Sstevel@tonic-gate /* 27927c478bd9Sstevel@tonic-gate * If there vnode reference count is greater than one, then 27937c478bd9Sstevel@tonic-gate * there may be additional references in the DNLC which will 27947c478bd9Sstevel@tonic-gate * need to be purged. First, trying removing the entry for 27957c478bd9Sstevel@tonic-gate * the parent directory and see if that removes the additional 27967c478bd9Sstevel@tonic-gate * reference(s). If that doesn't do it, then use dnlc_purge_vp 27977c478bd9Sstevel@tonic-gate * to completely remove any references to the directory which 27987c478bd9Sstevel@tonic-gate * might still exist in the DNLC. 27997c478bd9Sstevel@tonic-gate */ 28007c478bd9Sstevel@tonic-gate if (vp->v_count > 1) { 28017c478bd9Sstevel@tonic-gate dnlc_remove(vp, ".."); 28027c478bd9Sstevel@tonic-gate if (vp->v_count > 1) 28037c478bd9Sstevel@tonic-gate dnlc_purge_vp(vp); 28047c478bd9Sstevel@tonic-gate } 28057c478bd9Sstevel@tonic-gate 28067c478bd9Sstevel@tonic-gate douprintf = 1; 28077c478bd9Sstevel@tonic-gate 28087c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(dvp), RFS_RMDIR, 28097c478bd9Sstevel@tonic-gate xdr_diropargs, (caddr_t)&da, 28107c478bd9Sstevel@tonic-gate xdr_enum, (caddr_t)&status, cr, 28117c478bd9Sstevel@tonic-gate &douprintf, &status, 0, NULL); 28127c478bd9Sstevel@tonic-gate 28137c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(dvp); /* mod time changed */ 28147c478bd9Sstevel@tonic-gate 28157c478bd9Sstevel@tonic-gate if (error) { 28167c478bd9Sstevel@tonic-gate VN_RELE(vp); 28177c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 28187c478bd9Sstevel@tonic-gate return (error); 28197c478bd9Sstevel@tonic-gate } 28207c478bd9Sstevel@tonic-gate 28217c478bd9Sstevel@tonic-gate error = geterrno(status); 28227c478bd9Sstevel@tonic-gate if (!error) { 28237c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(drp)) 28247c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(dvp); 28257c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(VTOR(vp))) 28267c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(vp); 28277c478bd9Sstevel@tonic-gate } else { 28287c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, dvp, cr); 28297c478bd9Sstevel@tonic-gate /* 28307c478bd9Sstevel@tonic-gate * System V defines rmdir to return EEXIST, not 28317c478bd9Sstevel@tonic-gate * ENOTEMPTY if the directory is not empty. Over 28327c478bd9Sstevel@tonic-gate * the wire, the error is NFSERR_ENOTEMPTY which 28337c478bd9Sstevel@tonic-gate * geterrno maps to ENOTEMPTY. 28347c478bd9Sstevel@tonic-gate */ 28357c478bd9Sstevel@tonic-gate if (error == ENOTEMPTY) 28367c478bd9Sstevel@tonic-gate error = EEXIST; 28377c478bd9Sstevel@tonic-gate } 28387c478bd9Sstevel@tonic-gate 2839df2381bfSpraks if (error == 0) { 2840da6c28aaSamw vnevent_rmdir(vp, dvp, nm, ct); 2841df2381bfSpraks } 28427c478bd9Sstevel@tonic-gate VN_RELE(vp); 28437c478bd9Sstevel@tonic-gate 28447c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 28457c478bd9Sstevel@tonic-gate 28467c478bd9Sstevel@tonic-gate return (error); 28477c478bd9Sstevel@tonic-gate } 28487c478bd9Sstevel@tonic-gate 2849da6c28aaSamw /* ARGSUSED */ 28507c478bd9Sstevel@tonic-gate static int 2851da6c28aaSamw nfs_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr, 2852da6c28aaSamw caller_context_t *ct, int flags) 28537c478bd9Sstevel@tonic-gate { 28547c478bd9Sstevel@tonic-gate int error; 28557c478bd9Sstevel@tonic-gate struct nfsslargs args; 28567c478bd9Sstevel@tonic-gate enum nfsstat status; 28577c478bd9Sstevel@tonic-gate int douprintf; 28587c478bd9Sstevel@tonic-gate rnode_t *drp; 28597c478bd9Sstevel@tonic-gate 2860108322fbScarlsonj if (nfs_zone() != VTOMI(dvp)->mi_zone) 28617c478bd9Sstevel@tonic-gate return (EPERM); 28627c478bd9Sstevel@tonic-gate setdiropargs(&args.sla_from, lnm, dvp); 28637c478bd9Sstevel@tonic-gate args.sla_sa = &args.sla_sa_buf; 28647c478bd9Sstevel@tonic-gate error = vattr_to_sattr(tva, args.sla_sa); 28657c478bd9Sstevel@tonic-gate if (error) { 28667c478bd9Sstevel@tonic-gate /* req time field(s) overflow - return immediately */ 28677c478bd9Sstevel@tonic-gate return (error); 28687c478bd9Sstevel@tonic-gate } 28697c478bd9Sstevel@tonic-gate args.sla_tnm = tnm; 28707c478bd9Sstevel@tonic-gate 28717c478bd9Sstevel@tonic-gate drp = VTOR(dvp); 28727c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp))) 28737c478bd9Sstevel@tonic-gate return (EINTR); 28747c478bd9Sstevel@tonic-gate 28757c478bd9Sstevel@tonic-gate dnlc_remove(dvp, lnm); 28767c478bd9Sstevel@tonic-gate 28777c478bd9Sstevel@tonic-gate douprintf = 1; 28787c478bd9Sstevel@tonic-gate 28797c478bd9Sstevel@tonic-gate error = rfs2call(VTOMI(dvp), RFS_SYMLINK, 28807c478bd9Sstevel@tonic-gate xdr_slargs, (caddr_t)&args, 28817c478bd9Sstevel@tonic-gate xdr_enum, (caddr_t)&status, cr, 28827c478bd9Sstevel@tonic-gate &douprintf, &status, 0, NULL); 28837c478bd9Sstevel@tonic-gate 28847c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE(dvp); /* mod time changed */ 28857c478bd9Sstevel@tonic-gate 28867c478bd9Sstevel@tonic-gate if (!error) { 28877c478bd9Sstevel@tonic-gate error = geterrno(status); 28887c478bd9Sstevel@tonic-gate if (!error) { 28897c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(drp)) 28907c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(dvp); 28917c478bd9Sstevel@tonic-gate } else { 28927c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, dvp, cr); 28937c478bd9Sstevel@tonic-gate } 28947c478bd9Sstevel@tonic-gate } 28957c478bd9Sstevel@tonic-gate 28967c478bd9Sstevel@tonic-gate nfs_rw_exit(&drp->r_rwlock); 28977c478bd9Sstevel@tonic-gate 28987c478bd9Sstevel@tonic-gate return (error); 28997c478bd9Sstevel@tonic-gate } 29007c478bd9Sstevel@tonic-gate 29017c478bd9Sstevel@tonic-gate #ifdef DEBUG 29027c478bd9Sstevel@tonic-gate static int nfs_readdir_cache_hits = 0; 29037c478bd9Sstevel@tonic-gate static int nfs_readdir_cache_shorts = 0; 29047c478bd9Sstevel@tonic-gate static int nfs_readdir_cache_waits = 0; 29057c478bd9Sstevel@tonic-gate static int nfs_readdir_cache_misses = 0; 29067c478bd9Sstevel@tonic-gate static int nfs_readdir_readahead = 0; 29077c478bd9Sstevel@tonic-gate #endif 29087c478bd9Sstevel@tonic-gate 29097c478bd9Sstevel@tonic-gate static int nfs_shrinkreaddir = 0; 29107c478bd9Sstevel@tonic-gate 29117c478bd9Sstevel@tonic-gate /* 29127c478bd9Sstevel@tonic-gate * Read directory entries. 29137c478bd9Sstevel@tonic-gate * There are some weird things to look out for here. The uio_offset 29147c478bd9Sstevel@tonic-gate * field is either 0 or it is the offset returned from a previous 29157c478bd9Sstevel@tonic-gate * readdir. It is an opaque value used by the server to find the 29167c478bd9Sstevel@tonic-gate * correct directory block to read. The count field is the number 29177c478bd9Sstevel@tonic-gate * of blocks to read on the server. This is advisory only, the server 29187c478bd9Sstevel@tonic-gate * may return only one block's worth of entries. Entries may be compressed 29197c478bd9Sstevel@tonic-gate * on the server. 29207c478bd9Sstevel@tonic-gate */ 2921da6c28aaSamw /* ARGSUSED */ 29227c478bd9Sstevel@tonic-gate static int 2923da6c28aaSamw nfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp, 2924da6c28aaSamw caller_context_t *ct, int flags) 29257c478bd9Sstevel@tonic-gate { 29267c478bd9Sstevel@tonic-gate int error; 29277c478bd9Sstevel@tonic-gate size_t count; 29287c478bd9Sstevel@tonic-gate rnode_t *rp; 29297c478bd9Sstevel@tonic-gate rddir_cache *rdc; 29307c478bd9Sstevel@tonic-gate rddir_cache *nrdc; 29317c478bd9Sstevel@tonic-gate rddir_cache *rrdc; 29327c478bd9Sstevel@tonic-gate #ifdef DEBUG 29337c478bd9Sstevel@tonic-gate int missed; 29347c478bd9Sstevel@tonic-gate #endif 29357c478bd9Sstevel@tonic-gate rddir_cache srdc; 29367c478bd9Sstevel@tonic-gate avl_index_t where; 29377c478bd9Sstevel@tonic-gate 29387c478bd9Sstevel@tonic-gate rp = VTOR(vp); 29397c478bd9Sstevel@tonic-gate 29407c478bd9Sstevel@tonic-gate ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER)); 2941108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 29427c478bd9Sstevel@tonic-gate return (EIO); 29437c478bd9Sstevel@tonic-gate /* 29447c478bd9Sstevel@tonic-gate * Make sure that the directory cache is valid. 29457c478bd9Sstevel@tonic-gate */ 29467c478bd9Sstevel@tonic-gate if (HAVE_RDDIR_CACHE(rp)) { 29477c478bd9Sstevel@tonic-gate if (nfs_disable_rddir_cache) { 29487c478bd9Sstevel@tonic-gate /* 29497c478bd9Sstevel@tonic-gate * Setting nfs_disable_rddir_cache in /etc/system 29507c478bd9Sstevel@tonic-gate * allows interoperability with servers that do not 29517c478bd9Sstevel@tonic-gate * properly update the attributes of directories. 29527c478bd9Sstevel@tonic-gate * Any cached information gets purged before an 29537c478bd9Sstevel@tonic-gate * access is made to it. 29547c478bd9Sstevel@tonic-gate */ 29557c478bd9Sstevel@tonic-gate nfs_purge_rddir_cache(vp); 29567c478bd9Sstevel@tonic-gate } else { 29577c478bd9Sstevel@tonic-gate error = nfs_validate_caches(vp, cr); 29587c478bd9Sstevel@tonic-gate if (error) 29597c478bd9Sstevel@tonic-gate return (error); 29607c478bd9Sstevel@tonic-gate } 29617c478bd9Sstevel@tonic-gate } 29627c478bd9Sstevel@tonic-gate 29637c478bd9Sstevel@tonic-gate /* 29647c478bd9Sstevel@tonic-gate * UGLINESS: SunOS 3.2 servers apparently cannot always handle an 29657c478bd9Sstevel@tonic-gate * RFS_READDIR request with rda_count set to more than 0x400. So 29667c478bd9Sstevel@tonic-gate * we reduce the request size here purely for compatibility. 29677c478bd9Sstevel@tonic-gate * 29687c478bd9Sstevel@tonic-gate * In general, this is no longer required. However, if a server 29697c478bd9Sstevel@tonic-gate * is discovered which can not handle requests larger than 1024, 29707c478bd9Sstevel@tonic-gate * nfs_shrinkreaddir can be set to 1 to enable this backwards 29717c478bd9Sstevel@tonic-gate * compatibility. 29727c478bd9Sstevel@tonic-gate * 29737c478bd9Sstevel@tonic-gate * In any case, the request size is limited to NFS_MAXDATA bytes. 29747c478bd9Sstevel@tonic-gate */ 29757c478bd9Sstevel@tonic-gate count = MIN(uiop->uio_iov->iov_len, 29767c478bd9Sstevel@tonic-gate nfs_shrinkreaddir ? 0x400 : NFS_MAXDATA); 29777c478bd9Sstevel@tonic-gate 29787c478bd9Sstevel@tonic-gate nrdc = NULL; 29797c478bd9Sstevel@tonic-gate #ifdef DEBUG 29807c478bd9Sstevel@tonic-gate missed = 0; 29817c478bd9Sstevel@tonic-gate #endif 29827c478bd9Sstevel@tonic-gate top: 29837c478bd9Sstevel@tonic-gate /* 29847c478bd9Sstevel@tonic-gate * Short circuit last readdir which always returns 0 bytes. 29857c478bd9Sstevel@tonic-gate * This can be done after the directory has been read through 29867c478bd9Sstevel@tonic-gate * completely at least once. This will set r_direof which 29877c478bd9Sstevel@tonic-gate * can be used to find the value of the last cookie. 29887c478bd9Sstevel@tonic-gate */ 29897c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 29907c478bd9Sstevel@tonic-gate if (rp->r_direof != NULL && 29917c478bd9Sstevel@tonic-gate uiop->uio_offset == rp->r_direof->nfs_ncookie) { 29927c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 29937c478bd9Sstevel@tonic-gate #ifdef DEBUG 29947c478bd9Sstevel@tonic-gate nfs_readdir_cache_shorts++; 29957c478bd9Sstevel@tonic-gate #endif 29967c478bd9Sstevel@tonic-gate if (eofp) 29977c478bd9Sstevel@tonic-gate *eofp = 1; 29987c478bd9Sstevel@tonic-gate if (nrdc != NULL) 29997c478bd9Sstevel@tonic-gate rddir_cache_rele(nrdc); 30007c478bd9Sstevel@tonic-gate return (0); 30017c478bd9Sstevel@tonic-gate } 30027c478bd9Sstevel@tonic-gate /* 30037c478bd9Sstevel@tonic-gate * Look for a cache entry. Cache entries are identified 30047c478bd9Sstevel@tonic-gate * by the NFS cookie value and the byte count requested. 30057c478bd9Sstevel@tonic-gate */ 30067c478bd9Sstevel@tonic-gate srdc.nfs_cookie = uiop->uio_offset; 30077c478bd9Sstevel@tonic-gate srdc.buflen = count; 30087c478bd9Sstevel@tonic-gate rdc = avl_find(&rp->r_dir, &srdc, &where); 30097c478bd9Sstevel@tonic-gate if (rdc != NULL) { 30107c478bd9Sstevel@tonic-gate rddir_cache_hold(rdc); 30117c478bd9Sstevel@tonic-gate /* 30127c478bd9Sstevel@tonic-gate * If the cache entry is in the process of being 30137c478bd9Sstevel@tonic-gate * filled in, wait until this completes. The 30147c478bd9Sstevel@tonic-gate * RDDIRWAIT bit is set to indicate that someone 30157c478bd9Sstevel@tonic-gate * is waiting and then the thread currently 30167c478bd9Sstevel@tonic-gate * filling the entry is done, it should do a 30177c478bd9Sstevel@tonic-gate * cv_broadcast to wakeup all of the threads 30187c478bd9Sstevel@tonic-gate * waiting for it to finish. 30197c478bd9Sstevel@tonic-gate */ 30207c478bd9Sstevel@tonic-gate if (rdc->flags & RDDIR) { 30217c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_rwlock); 30227c478bd9Sstevel@tonic-gate rdc->flags |= RDDIRWAIT; 30237c478bd9Sstevel@tonic-gate #ifdef DEBUG 30247c478bd9Sstevel@tonic-gate nfs_readdir_cache_waits++; 30257c478bd9Sstevel@tonic-gate #endif 30267c478bd9Sstevel@tonic-gate if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) { 30277c478bd9Sstevel@tonic-gate /* 30287c478bd9Sstevel@tonic-gate * We got interrupted, probably 30297c478bd9Sstevel@tonic-gate * the user typed ^C or an alarm 30307c478bd9Sstevel@tonic-gate * fired. We free the new entry 30317c478bd9Sstevel@tonic-gate * if we allocated one. 30327c478bd9Sstevel@tonic-gate */ 30337c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 30347c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&rp->r_rwlock, 30357c478bd9Sstevel@tonic-gate RW_READER, FALSE); 30367c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 30377c478bd9Sstevel@tonic-gate if (nrdc != NULL) 30387c478bd9Sstevel@tonic-gate rddir_cache_rele(nrdc); 30397c478bd9Sstevel@tonic-gate return (EINTR); 30407c478bd9Sstevel@tonic-gate } 30417c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 30427c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&rp->r_rwlock, 30437c478bd9Sstevel@tonic-gate RW_READER, FALSE); 30447c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 30457c478bd9Sstevel@tonic-gate goto top; 30467c478bd9Sstevel@tonic-gate } 30477c478bd9Sstevel@tonic-gate /* 30487c478bd9Sstevel@tonic-gate * Check to see if a readdir is required to 30497c478bd9Sstevel@tonic-gate * fill the entry. If so, mark this entry 30507c478bd9Sstevel@tonic-gate * as being filled, remove our reference, 30517c478bd9Sstevel@tonic-gate * and branch to the code to fill the entry. 30527c478bd9Sstevel@tonic-gate */ 30537c478bd9Sstevel@tonic-gate if (rdc->flags & RDDIRREQ) { 30547c478bd9Sstevel@tonic-gate rdc->flags &= ~RDDIRREQ; 30557c478bd9Sstevel@tonic-gate rdc->flags |= RDDIR; 30567c478bd9Sstevel@tonic-gate if (nrdc != NULL) 30577c478bd9Sstevel@tonic-gate rddir_cache_rele(nrdc); 30587c478bd9Sstevel@tonic-gate nrdc = rdc; 30597c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 30607c478bd9Sstevel@tonic-gate goto bottom; 30617c478bd9Sstevel@tonic-gate } 30627c478bd9Sstevel@tonic-gate #ifdef DEBUG 30637c478bd9Sstevel@tonic-gate if (!missed) 30647c478bd9Sstevel@tonic-gate nfs_readdir_cache_hits++; 30657c478bd9Sstevel@tonic-gate #endif 30667c478bd9Sstevel@tonic-gate /* 30677c478bd9Sstevel@tonic-gate * If an error occurred while attempting 30687c478bd9Sstevel@tonic-gate * to fill the cache entry, just return it. 30697c478bd9Sstevel@tonic-gate */ 30707c478bd9Sstevel@tonic-gate if (rdc->error) { 30717c478bd9Sstevel@tonic-gate error = rdc->error; 30727c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 30737c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 30747c478bd9Sstevel@tonic-gate if (nrdc != NULL) 30757c478bd9Sstevel@tonic-gate rddir_cache_rele(nrdc); 30767c478bd9Sstevel@tonic-gate return (error); 30777c478bd9Sstevel@tonic-gate } 30787c478bd9Sstevel@tonic-gate 30797c478bd9Sstevel@tonic-gate /* 30807c478bd9Sstevel@tonic-gate * The cache entry is complete and good, 30817c478bd9Sstevel@tonic-gate * copyout the dirent structs to the calling 30827c478bd9Sstevel@tonic-gate * thread. 30837c478bd9Sstevel@tonic-gate */ 30847c478bd9Sstevel@tonic-gate error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop); 30857c478bd9Sstevel@tonic-gate 30867c478bd9Sstevel@tonic-gate /* 30877c478bd9Sstevel@tonic-gate * If no error occurred during the copyout, 30887c478bd9Sstevel@tonic-gate * update the offset in the uio struct to 30897c478bd9Sstevel@tonic-gate * contain the value of the next cookie 30907c478bd9Sstevel@tonic-gate * and set the eof value appropriately. 30917c478bd9Sstevel@tonic-gate */ 30927c478bd9Sstevel@tonic-gate if (!error) { 30937c478bd9Sstevel@tonic-gate uiop->uio_offset = rdc->nfs_ncookie; 30947c478bd9Sstevel@tonic-gate if (eofp) 30957c478bd9Sstevel@tonic-gate *eofp = rdc->eof; 30967c478bd9Sstevel@tonic-gate } 30977c478bd9Sstevel@tonic-gate 30987c478bd9Sstevel@tonic-gate /* 30997c478bd9Sstevel@tonic-gate * Decide whether to do readahead. Don't if 31007c478bd9Sstevel@tonic-gate * have already read to the end of directory. 31017c478bd9Sstevel@tonic-gate */ 31027c478bd9Sstevel@tonic-gate if (rdc->eof) { 31037c478bd9Sstevel@tonic-gate rp->r_direof = rdc; 31047c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 31057c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 31067c478bd9Sstevel@tonic-gate if (nrdc != NULL) 31077c478bd9Sstevel@tonic-gate rddir_cache_rele(nrdc); 31087c478bd9Sstevel@tonic-gate return (error); 31097c478bd9Sstevel@tonic-gate } 31107c478bd9Sstevel@tonic-gate 31117c478bd9Sstevel@tonic-gate /* 31127c478bd9Sstevel@tonic-gate * Check to see whether we found an entry 31137c478bd9Sstevel@tonic-gate * for the readahead. If so, we don't need 31147c478bd9Sstevel@tonic-gate * to do anything further, so free the new 31157c478bd9Sstevel@tonic-gate * entry if one was allocated. Otherwise, 31167c478bd9Sstevel@tonic-gate * allocate a new entry, add it to the cache, 31177c478bd9Sstevel@tonic-gate * and then initiate an asynchronous readdir 31187c478bd9Sstevel@tonic-gate * operation to fill it. 31197c478bd9Sstevel@tonic-gate */ 31207c478bd9Sstevel@tonic-gate srdc.nfs_cookie = rdc->nfs_ncookie; 31217c478bd9Sstevel@tonic-gate srdc.buflen = count; 31227c478bd9Sstevel@tonic-gate rrdc = avl_find(&rp->r_dir, &srdc, &where); 31237c478bd9Sstevel@tonic-gate if (rrdc != NULL) { 31247c478bd9Sstevel@tonic-gate if (nrdc != NULL) 31257c478bd9Sstevel@tonic-gate rddir_cache_rele(nrdc); 31267c478bd9Sstevel@tonic-gate } else { 31277c478bd9Sstevel@tonic-gate if (nrdc != NULL) 31287c478bd9Sstevel@tonic-gate rrdc = nrdc; 31297c478bd9Sstevel@tonic-gate else { 31307c478bd9Sstevel@tonic-gate rrdc = rddir_cache_alloc(KM_NOSLEEP); 31317c478bd9Sstevel@tonic-gate } 31327c478bd9Sstevel@tonic-gate if (rrdc != NULL) { 31337c478bd9Sstevel@tonic-gate rrdc->nfs_cookie = rdc->nfs_ncookie; 31347c478bd9Sstevel@tonic-gate rrdc->buflen = count; 31357c478bd9Sstevel@tonic-gate avl_insert(&rp->r_dir, rrdc, where); 31367c478bd9Sstevel@tonic-gate rddir_cache_hold(rrdc); 31377c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 31387c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 31397c478bd9Sstevel@tonic-gate #ifdef DEBUG 31407c478bd9Sstevel@tonic-gate nfs_readdir_readahead++; 31417c478bd9Sstevel@tonic-gate #endif 31427c478bd9Sstevel@tonic-gate nfs_async_readdir(vp, rrdc, cr, nfsreaddir); 31437c478bd9Sstevel@tonic-gate return (error); 31447c478bd9Sstevel@tonic-gate } 31457c478bd9Sstevel@tonic-gate } 31467c478bd9Sstevel@tonic-gate 31477c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 31487c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 31497c478bd9Sstevel@tonic-gate return (error); 31507c478bd9Sstevel@tonic-gate } 31517c478bd9Sstevel@tonic-gate 31527c478bd9Sstevel@tonic-gate /* 31537c478bd9Sstevel@tonic-gate * Didn't find an entry in the cache. Construct a new empty 31547c478bd9Sstevel@tonic-gate * entry and link it into the cache. Other processes attempting 31557c478bd9Sstevel@tonic-gate * to access this entry will need to wait until it is filled in. 31567c478bd9Sstevel@tonic-gate * 31577c478bd9Sstevel@tonic-gate * Since kmem_alloc may block, another pass through the cache 31587c478bd9Sstevel@tonic-gate * will need to be taken to make sure that another process 31597c478bd9Sstevel@tonic-gate * hasn't already added an entry to the cache for this request. 31607c478bd9Sstevel@tonic-gate */ 31617c478bd9Sstevel@tonic-gate if (nrdc == NULL) { 31627c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 31637c478bd9Sstevel@tonic-gate nrdc = rddir_cache_alloc(KM_SLEEP); 31647c478bd9Sstevel@tonic-gate nrdc->nfs_cookie = uiop->uio_offset; 31657c478bd9Sstevel@tonic-gate nrdc->buflen = count; 31667c478bd9Sstevel@tonic-gate goto top; 31677c478bd9Sstevel@tonic-gate } 31687c478bd9Sstevel@tonic-gate 31697c478bd9Sstevel@tonic-gate /* 31707c478bd9Sstevel@tonic-gate * Add this entry to the cache. 31717c478bd9Sstevel@tonic-gate */ 31727c478bd9Sstevel@tonic-gate avl_insert(&rp->r_dir, nrdc, where); 31737c478bd9Sstevel@tonic-gate rddir_cache_hold(nrdc); 31747c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 31757c478bd9Sstevel@tonic-gate 31767c478bd9Sstevel@tonic-gate bottom: 31777c478bd9Sstevel@tonic-gate #ifdef DEBUG 31787c478bd9Sstevel@tonic-gate missed = 1; 31797c478bd9Sstevel@tonic-gate nfs_readdir_cache_misses++; 31807c478bd9Sstevel@tonic-gate #endif 31817c478bd9Sstevel@tonic-gate /* 31827c478bd9Sstevel@tonic-gate * Do the readdir. 31837c478bd9Sstevel@tonic-gate */ 31847c478bd9Sstevel@tonic-gate error = nfsreaddir(vp, nrdc, cr); 31857c478bd9Sstevel@tonic-gate 31867c478bd9Sstevel@tonic-gate /* 31877c478bd9Sstevel@tonic-gate * If this operation failed, just return the error which occurred. 31887c478bd9Sstevel@tonic-gate */ 31897c478bd9Sstevel@tonic-gate if (error != 0) 31907c478bd9Sstevel@tonic-gate return (error); 31917c478bd9Sstevel@tonic-gate 31927c478bd9Sstevel@tonic-gate /* 31937c478bd9Sstevel@tonic-gate * Since the RPC operation will have taken sometime and blocked 31947c478bd9Sstevel@tonic-gate * this process, another pass through the cache will need to be 31957c478bd9Sstevel@tonic-gate * taken to find the correct cache entry. It is possible that 31967c478bd9Sstevel@tonic-gate * the correct cache entry will not be there (although one was 31977c478bd9Sstevel@tonic-gate * added) because the directory changed during the RPC operation 31987c478bd9Sstevel@tonic-gate * and the readdir cache was flushed. In this case, just start 31997c478bd9Sstevel@tonic-gate * over. It is hoped that this will not happen too often... :-) 32007c478bd9Sstevel@tonic-gate */ 32017c478bd9Sstevel@tonic-gate nrdc = NULL; 32027c478bd9Sstevel@tonic-gate goto top; 32037c478bd9Sstevel@tonic-gate /* NOTREACHED */ 32047c478bd9Sstevel@tonic-gate } 32057c478bd9Sstevel@tonic-gate 32067c478bd9Sstevel@tonic-gate static int 32077c478bd9Sstevel@tonic-gate nfsreaddir(vnode_t *vp, rddir_cache *rdc, cred_t *cr) 32087c478bd9Sstevel@tonic-gate { 32097c478bd9Sstevel@tonic-gate int error; 32107c478bd9Sstevel@tonic-gate struct nfsrddirargs rda; 32117c478bd9Sstevel@tonic-gate struct nfsrddirres rd; 32127c478bd9Sstevel@tonic-gate rnode_t *rp; 32137c478bd9Sstevel@tonic-gate mntinfo_t *mi; 32147c478bd9Sstevel@tonic-gate uint_t count; 32157c478bd9Sstevel@tonic-gate int douprintf; 32167c478bd9Sstevel@tonic-gate failinfo_t fi, *fip; 32177c478bd9Sstevel@tonic-gate 3218108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 32197c478bd9Sstevel@tonic-gate count = rdc->buflen; 32207c478bd9Sstevel@tonic-gate 32217c478bd9Sstevel@tonic-gate rp = VTOR(vp); 32227c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 32237c478bd9Sstevel@tonic-gate 32247c478bd9Sstevel@tonic-gate rda.rda_fh = *VTOFH(vp); 32257c478bd9Sstevel@tonic-gate rda.rda_offset = rdc->nfs_cookie; 32267c478bd9Sstevel@tonic-gate 32277c478bd9Sstevel@tonic-gate /* 32287c478bd9Sstevel@tonic-gate * NFS client failover support 32297c478bd9Sstevel@tonic-gate * suppress failover unless we have a zero cookie 32307c478bd9Sstevel@tonic-gate */ 32317c478bd9Sstevel@tonic-gate if (rdc->nfs_cookie == (off_t)0) { 32327c478bd9Sstevel@tonic-gate fi.vp = vp; 32337c478bd9Sstevel@tonic-gate fi.fhp = (caddr_t)&rda.rda_fh; 32347c478bd9Sstevel@tonic-gate fi.copyproc = nfscopyfh; 32357c478bd9Sstevel@tonic-gate fi.lookupproc = nfslookup; 32367c478bd9Sstevel@tonic-gate fi.xattrdirproc = acl_getxattrdir2; 32377c478bd9Sstevel@tonic-gate fip = &fi; 32387c478bd9Sstevel@tonic-gate } else { 32397c478bd9Sstevel@tonic-gate fip = NULL; 32407c478bd9Sstevel@tonic-gate } 32417c478bd9Sstevel@tonic-gate 32427c478bd9Sstevel@tonic-gate rd.rd_entries = kmem_alloc(rdc->buflen, KM_SLEEP); 32437c478bd9Sstevel@tonic-gate rd.rd_size = count; 32447c478bd9Sstevel@tonic-gate rd.rd_offset = rda.rda_offset; 32457c478bd9Sstevel@tonic-gate 32467c478bd9Sstevel@tonic-gate douprintf = 1; 32477c478bd9Sstevel@tonic-gate 32487c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 32497c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 32507c478bd9Sstevel@tonic-gate kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 32517c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 32527c478bd9Sstevel@tonic-gate } 32537c478bd9Sstevel@tonic-gate 32547c478bd9Sstevel@tonic-gate do { 32557c478bd9Sstevel@tonic-gate rda.rda_count = MIN(count, mi->mi_curread); 32567c478bd9Sstevel@tonic-gate error = rfs2call(mi, RFS_READDIR, 32577c478bd9Sstevel@tonic-gate xdr_rddirargs, (caddr_t)&rda, 32587c478bd9Sstevel@tonic-gate xdr_getrddirres, (caddr_t)&rd, cr, 32597c478bd9Sstevel@tonic-gate &douprintf, &rd.rd_status, 0, fip); 32607c478bd9Sstevel@tonic-gate } while (error == ENFS_TRYAGAIN); 32617c478bd9Sstevel@tonic-gate 32627c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 32637c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 32647c478bd9Sstevel@tonic-gate kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 32657c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 32667c478bd9Sstevel@tonic-gate } 32677c478bd9Sstevel@tonic-gate 32687c478bd9Sstevel@tonic-gate /* 32697c478bd9Sstevel@tonic-gate * Since we are actually doing a READDIR RPC, we must have 32707c478bd9Sstevel@tonic-gate * exclusive access to the cache entry being filled. Thus, 32717c478bd9Sstevel@tonic-gate * it is safe to update all fields except for the flags 32727c478bd9Sstevel@tonic-gate * field. The r_statelock in the rnode must be held to 32737c478bd9Sstevel@tonic-gate * prevent two different threads from simultaneously 32747c478bd9Sstevel@tonic-gate * attempting to update the flags field. This can happen 32757c478bd9Sstevel@tonic-gate * if we are turning off RDDIR and the other thread is 32767c478bd9Sstevel@tonic-gate * trying to set RDDIRWAIT. 32777c478bd9Sstevel@tonic-gate */ 32787c478bd9Sstevel@tonic-gate ASSERT(rdc->flags & RDDIR); 32797c478bd9Sstevel@tonic-gate if (!error) { 32807c478bd9Sstevel@tonic-gate error = geterrno(rd.rd_status); 32817c478bd9Sstevel@tonic-gate if (!error) { 32827c478bd9Sstevel@tonic-gate rdc->nfs_ncookie = rd.rd_offset; 32837c478bd9Sstevel@tonic-gate rdc->eof = rd.rd_eof ? 1 : 0; 32847c478bd9Sstevel@tonic-gate rdc->entlen = rd.rd_size; 32857c478bd9Sstevel@tonic-gate ASSERT(rdc->entlen <= rdc->buflen); 32867c478bd9Sstevel@tonic-gate #ifdef DEBUG 32877c478bd9Sstevel@tonic-gate rdc->entries = rddir_cache_buf_alloc(rdc->buflen, 32887c478bd9Sstevel@tonic-gate KM_SLEEP); 32897c478bd9Sstevel@tonic-gate #else 32907c478bd9Sstevel@tonic-gate rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP); 32917c478bd9Sstevel@tonic-gate #endif 32927c478bd9Sstevel@tonic-gate bcopy(rd.rd_entries, rdc->entries, rdc->entlen); 32937c478bd9Sstevel@tonic-gate rdc->error = 0; 32947c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 32957c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 32967c478bd9Sstevel@tonic-gate KSTAT_IO_PTR(mi->mi_io_kstats)->reads++; 32977c478bd9Sstevel@tonic-gate KSTAT_IO_PTR(mi->mi_io_kstats)->nread += 32987c478bd9Sstevel@tonic-gate rd.rd_size; 32997c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 33007c478bd9Sstevel@tonic-gate } 33017c478bd9Sstevel@tonic-gate } else { 33027c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, vp, cr); 33037c478bd9Sstevel@tonic-gate } 33047c478bd9Sstevel@tonic-gate } 33057c478bd9Sstevel@tonic-gate if (error) { 33067c478bd9Sstevel@tonic-gate rdc->entries = NULL; 33077c478bd9Sstevel@tonic-gate rdc->error = error; 33087c478bd9Sstevel@tonic-gate } 33097c478bd9Sstevel@tonic-gate kmem_free(rd.rd_entries, rdc->buflen); 33107c478bd9Sstevel@tonic-gate 33117c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 33127c478bd9Sstevel@tonic-gate rdc->flags &= ~RDDIR; 33137c478bd9Sstevel@tonic-gate if (rdc->flags & RDDIRWAIT) { 33147c478bd9Sstevel@tonic-gate rdc->flags &= ~RDDIRWAIT; 33157c478bd9Sstevel@tonic-gate cv_broadcast(&rdc->cv); 33167c478bd9Sstevel@tonic-gate } 33177c478bd9Sstevel@tonic-gate if (error) 33187c478bd9Sstevel@tonic-gate rdc->flags |= RDDIRREQ; 33197c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 33207c478bd9Sstevel@tonic-gate 33217c478bd9Sstevel@tonic-gate rddir_cache_rele(rdc); 33227c478bd9Sstevel@tonic-gate 33237c478bd9Sstevel@tonic-gate return (error); 33247c478bd9Sstevel@tonic-gate } 33257c478bd9Sstevel@tonic-gate 33267c478bd9Sstevel@tonic-gate #ifdef DEBUG 33277c478bd9Sstevel@tonic-gate static int nfs_bio_do_stop = 0; 33287c478bd9Sstevel@tonic-gate #endif 33297c478bd9Sstevel@tonic-gate 33307c478bd9Sstevel@tonic-gate static int 33317c478bd9Sstevel@tonic-gate nfs_bio(struct buf *bp, cred_t *cr) 33327c478bd9Sstevel@tonic-gate { 33337c478bd9Sstevel@tonic-gate rnode_t *rp = VTOR(bp->b_vp); 33347c478bd9Sstevel@tonic-gate int count; 33357c478bd9Sstevel@tonic-gate int error; 33367c478bd9Sstevel@tonic-gate cred_t *cred; 33377c478bd9Sstevel@tonic-gate uint_t offset; 33387c478bd9Sstevel@tonic-gate 33397c478bd9Sstevel@tonic-gate DTRACE_IO1(start, struct buf *, bp); 33407c478bd9Sstevel@tonic-gate 3341108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone); 33427c478bd9Sstevel@tonic-gate offset = dbtob(bp->b_blkno); 33437c478bd9Sstevel@tonic-gate 33447c478bd9Sstevel@tonic-gate if (bp->b_flags & B_READ) { 33457c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 33467c478bd9Sstevel@tonic-gate if (rp->r_cred != NULL) { 33477c478bd9Sstevel@tonic-gate cred = rp->r_cred; 33487c478bd9Sstevel@tonic-gate crhold(cred); 33497c478bd9Sstevel@tonic-gate } else { 33507c478bd9Sstevel@tonic-gate rp->r_cred = cr; 33517c478bd9Sstevel@tonic-gate crhold(cr); 33527c478bd9Sstevel@tonic-gate cred = cr; 33537c478bd9Sstevel@tonic-gate crhold(cred); 33547c478bd9Sstevel@tonic-gate } 33557c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 33567c478bd9Sstevel@tonic-gate read_again: 33577c478bd9Sstevel@tonic-gate error = bp->b_error = nfsread(bp->b_vp, bp->b_un.b_addr, 33587c478bd9Sstevel@tonic-gate offset, bp->b_bcount, &bp->b_resid, cred); 33597c478bd9Sstevel@tonic-gate crfree(cred); 33607c478bd9Sstevel@tonic-gate if (!error) { 33617c478bd9Sstevel@tonic-gate if (bp->b_resid) { 33627c478bd9Sstevel@tonic-gate /* 33637c478bd9Sstevel@tonic-gate * Didn't get it all because we hit EOF, 33647c478bd9Sstevel@tonic-gate * zero all the memory beyond the EOF. 33657c478bd9Sstevel@tonic-gate */ 33667c478bd9Sstevel@tonic-gate /* bzero(rdaddr + */ 33677c478bd9Sstevel@tonic-gate bzero(bp->b_un.b_addr + 33687c478bd9Sstevel@tonic-gate bp->b_bcount - bp->b_resid, bp->b_resid); 33697c478bd9Sstevel@tonic-gate } 33707c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 33717c478bd9Sstevel@tonic-gate if (bp->b_resid == bp->b_bcount && 33727c478bd9Sstevel@tonic-gate offset >= rp->r_size) { 33737c478bd9Sstevel@tonic-gate /* 33747c478bd9Sstevel@tonic-gate * We didn't read anything at all as we are 33757c478bd9Sstevel@tonic-gate * past EOF. Return an error indicator back 33767c478bd9Sstevel@tonic-gate * but don't destroy the pages (yet). 33777c478bd9Sstevel@tonic-gate */ 33787c478bd9Sstevel@tonic-gate error = NFS_EOF; 33797c478bd9Sstevel@tonic-gate } 33807c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 33817c478bd9Sstevel@tonic-gate } else if (error == EACCES) { 33827c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 33837c478bd9Sstevel@tonic-gate if (cred != cr) { 33847c478bd9Sstevel@tonic-gate if (rp->r_cred != NULL) 33857c478bd9Sstevel@tonic-gate crfree(rp->r_cred); 33867c478bd9Sstevel@tonic-gate rp->r_cred = cr; 33877c478bd9Sstevel@tonic-gate crhold(cr); 33887c478bd9Sstevel@tonic-gate cred = cr; 33897c478bd9Sstevel@tonic-gate crhold(cred); 33907c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 33917c478bd9Sstevel@tonic-gate goto read_again; 33927c478bd9Sstevel@tonic-gate } 33937c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 33947c478bd9Sstevel@tonic-gate } 33957c478bd9Sstevel@tonic-gate } else { 33967c478bd9Sstevel@tonic-gate if (!(rp->r_flags & RSTALE)) { 33977c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 33987c478bd9Sstevel@tonic-gate if (rp->r_cred != NULL) { 33997c478bd9Sstevel@tonic-gate cred = rp->r_cred; 34007c478bd9Sstevel@tonic-gate crhold(cred); 34017c478bd9Sstevel@tonic-gate } else { 34027c478bd9Sstevel@tonic-gate rp->r_cred = cr; 34037c478bd9Sstevel@tonic-gate crhold(cr); 34047c478bd9Sstevel@tonic-gate cred = cr; 34057c478bd9Sstevel@tonic-gate crhold(cred); 34067c478bd9Sstevel@tonic-gate } 34077c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 34087c478bd9Sstevel@tonic-gate write_again: 34097c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 34107c478bd9Sstevel@tonic-gate count = MIN(bp->b_bcount, rp->r_size - offset); 34117c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 34127c478bd9Sstevel@tonic-gate if (count < 0) 34137c478bd9Sstevel@tonic-gate cmn_err(CE_PANIC, "nfs_bio: write count < 0"); 34147c478bd9Sstevel@tonic-gate #ifdef DEBUG 34157c478bd9Sstevel@tonic-gate if (count == 0) { 34167c478bd9Sstevel@tonic-gate zcmn_err(getzoneid(), CE_WARN, 34177c478bd9Sstevel@tonic-gate "nfs_bio: zero length write at %d", 34187c478bd9Sstevel@tonic-gate offset); 34197c478bd9Sstevel@tonic-gate nfs_printfhandle(&rp->r_fh); 34207c478bd9Sstevel@tonic-gate if (nfs_bio_do_stop) 34217c478bd9Sstevel@tonic-gate debug_enter("nfs_bio"); 34227c478bd9Sstevel@tonic-gate } 34237c478bd9Sstevel@tonic-gate #endif 34247c478bd9Sstevel@tonic-gate error = nfswrite(bp->b_vp, bp->b_un.b_addr, offset, 34257c478bd9Sstevel@tonic-gate count, cred); 34267c478bd9Sstevel@tonic-gate if (error == EACCES) { 34277c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 34287c478bd9Sstevel@tonic-gate if (cred != cr) { 34297c478bd9Sstevel@tonic-gate if (rp->r_cred != NULL) 34307c478bd9Sstevel@tonic-gate crfree(rp->r_cred); 34317c478bd9Sstevel@tonic-gate rp->r_cred = cr; 34327c478bd9Sstevel@tonic-gate crhold(cr); 34337c478bd9Sstevel@tonic-gate crfree(cred); 34347c478bd9Sstevel@tonic-gate cred = cr; 34357c478bd9Sstevel@tonic-gate crhold(cred); 34367c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 34377c478bd9Sstevel@tonic-gate goto write_again; 34387c478bd9Sstevel@tonic-gate } 34397c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 34407c478bd9Sstevel@tonic-gate } 34417c478bd9Sstevel@tonic-gate bp->b_error = error; 34427c478bd9Sstevel@tonic-gate if (error && error != EINTR) { 34437c478bd9Sstevel@tonic-gate /* 34447c478bd9Sstevel@tonic-gate * Don't print EDQUOT errors on the console. 34457c478bd9Sstevel@tonic-gate * Don't print asynchronous EACCES errors. 34467c478bd9Sstevel@tonic-gate * Don't print EFBIG errors. 34477c478bd9Sstevel@tonic-gate * Print all other write errors. 34487c478bd9Sstevel@tonic-gate */ 34497c478bd9Sstevel@tonic-gate if (error != EDQUOT && error != EFBIG && 34507c478bd9Sstevel@tonic-gate (error != EACCES || 34517c478bd9Sstevel@tonic-gate !(bp->b_flags & B_ASYNC))) 34527c478bd9Sstevel@tonic-gate nfs_write_error(bp->b_vp, error, cred); 34537c478bd9Sstevel@tonic-gate /* 34547c478bd9Sstevel@tonic-gate * Update r_error and r_flags as appropriate. 34557c478bd9Sstevel@tonic-gate * If the error was ESTALE, then mark the 34567c478bd9Sstevel@tonic-gate * rnode as not being writeable and save 34577c478bd9Sstevel@tonic-gate * the error status. Otherwise, save any 34587c478bd9Sstevel@tonic-gate * errors which occur from asynchronous 34597c478bd9Sstevel@tonic-gate * page invalidations. Any errors occurring 34607c478bd9Sstevel@tonic-gate * from other operations should be saved 34617c478bd9Sstevel@tonic-gate * by the caller. 34627c478bd9Sstevel@tonic-gate */ 34637c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 34647c478bd9Sstevel@tonic-gate if (error == ESTALE) { 34657c478bd9Sstevel@tonic-gate rp->r_flags |= RSTALE; 34667c478bd9Sstevel@tonic-gate if (!rp->r_error) 34677c478bd9Sstevel@tonic-gate rp->r_error = error; 34687c478bd9Sstevel@tonic-gate } else if (!rp->r_error && 34697c478bd9Sstevel@tonic-gate (bp->b_flags & 34707c478bd9Sstevel@tonic-gate (B_INVAL|B_FORCE|B_ASYNC)) == 34717c478bd9Sstevel@tonic-gate (B_INVAL|B_FORCE|B_ASYNC)) { 34727c478bd9Sstevel@tonic-gate rp->r_error = error; 34737c478bd9Sstevel@tonic-gate } 34747c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 34757c478bd9Sstevel@tonic-gate } 34767c478bd9Sstevel@tonic-gate crfree(cred); 34777c478bd9Sstevel@tonic-gate } else 34787c478bd9Sstevel@tonic-gate error = rp->r_error; 34797c478bd9Sstevel@tonic-gate } 34807c478bd9Sstevel@tonic-gate 34817c478bd9Sstevel@tonic-gate if (error != 0 && error != NFS_EOF) 34827c478bd9Sstevel@tonic-gate bp->b_flags |= B_ERROR; 34837c478bd9Sstevel@tonic-gate 34847c478bd9Sstevel@tonic-gate DTRACE_IO1(done, struct buf *, bp); 34857c478bd9Sstevel@tonic-gate 34867c478bd9Sstevel@tonic-gate return (error); 34877c478bd9Sstevel@tonic-gate } 34887c478bd9Sstevel@tonic-gate 3489da6c28aaSamw /* ARGSUSED */ 34907c478bd9Sstevel@tonic-gate static int 3491da6c28aaSamw nfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 34927c478bd9Sstevel@tonic-gate { 34937c478bd9Sstevel@tonic-gate struct nfs_fid *fp; 34947c478bd9Sstevel@tonic-gate rnode_t *rp; 34957c478bd9Sstevel@tonic-gate 34967c478bd9Sstevel@tonic-gate rp = VTOR(vp); 34977c478bd9Sstevel@tonic-gate 34987c478bd9Sstevel@tonic-gate if (fidp->fid_len < (sizeof (struct nfs_fid) - sizeof (short))) { 34997c478bd9Sstevel@tonic-gate fidp->fid_len = sizeof (struct nfs_fid) - sizeof (short); 35007c478bd9Sstevel@tonic-gate return (ENOSPC); 35017c478bd9Sstevel@tonic-gate } 35027c478bd9Sstevel@tonic-gate fp = (struct nfs_fid *)fidp; 35037c478bd9Sstevel@tonic-gate fp->nf_pad = 0; 35047c478bd9Sstevel@tonic-gate fp->nf_len = sizeof (struct nfs_fid) - sizeof (short); 35057c478bd9Sstevel@tonic-gate bcopy(rp->r_fh.fh_buf, fp->nf_data, NFS_FHSIZE); 35067c478bd9Sstevel@tonic-gate return (0); 35077c478bd9Sstevel@tonic-gate } 35087c478bd9Sstevel@tonic-gate 35097c478bd9Sstevel@tonic-gate /* ARGSUSED2 */ 35107c478bd9Sstevel@tonic-gate static int 35117c478bd9Sstevel@tonic-gate nfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp) 35127c478bd9Sstevel@tonic-gate { 35137c478bd9Sstevel@tonic-gate rnode_t *rp = VTOR(vp); 35147c478bd9Sstevel@tonic-gate 35157c478bd9Sstevel@tonic-gate if (!write_lock) { 35167c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE); 35177c478bd9Sstevel@tonic-gate return (V_WRITELOCK_FALSE); 35187c478bd9Sstevel@tonic-gate } 35197c478bd9Sstevel@tonic-gate 35207c478bd9Sstevel@tonic-gate if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) { 35217c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE); 35227c478bd9Sstevel@tonic-gate if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp)) 35237c478bd9Sstevel@tonic-gate return (V_WRITELOCK_FALSE); 35247c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_rwlock); 35257c478bd9Sstevel@tonic-gate } 35267c478bd9Sstevel@tonic-gate 35277c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE); 35287c478bd9Sstevel@tonic-gate return (V_WRITELOCK_TRUE); 35297c478bd9Sstevel@tonic-gate } 35307c478bd9Sstevel@tonic-gate 35317c478bd9Sstevel@tonic-gate /* ARGSUSED */ 35327c478bd9Sstevel@tonic-gate static void 35337c478bd9Sstevel@tonic-gate nfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp) 35347c478bd9Sstevel@tonic-gate { 35357c478bd9Sstevel@tonic-gate rnode_t *rp = VTOR(vp); 35367c478bd9Sstevel@tonic-gate 35377c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_rwlock); 35387c478bd9Sstevel@tonic-gate } 35397c478bd9Sstevel@tonic-gate 35407c478bd9Sstevel@tonic-gate /* ARGSUSED */ 35417c478bd9Sstevel@tonic-gate static int 3542da6c28aaSamw nfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct) 35437c478bd9Sstevel@tonic-gate { 35447c478bd9Sstevel@tonic-gate 35457c478bd9Sstevel@tonic-gate /* 35467c478bd9Sstevel@tonic-gate * Because we stuff the readdir cookie into the offset field 35477c478bd9Sstevel@tonic-gate * someone may attempt to do an lseek with the cookie which 35487c478bd9Sstevel@tonic-gate * we want to succeed. 35497c478bd9Sstevel@tonic-gate */ 35507c478bd9Sstevel@tonic-gate if (vp->v_type == VDIR) 35517c478bd9Sstevel@tonic-gate return (0); 35527c478bd9Sstevel@tonic-gate if (*noffp < 0 || *noffp > MAXOFF32_T) 35537c478bd9Sstevel@tonic-gate return (EINVAL); 35547c478bd9Sstevel@tonic-gate return (0); 35557c478bd9Sstevel@tonic-gate } 35567c478bd9Sstevel@tonic-gate 35577c478bd9Sstevel@tonic-gate /* 35587c478bd9Sstevel@tonic-gate * number of NFS_MAXDATA blocks to read ahead 35597c478bd9Sstevel@tonic-gate * optimized for 100 base-T. 35607c478bd9Sstevel@tonic-gate */ 35617c478bd9Sstevel@tonic-gate static int nfs_nra = 4; 35627c478bd9Sstevel@tonic-gate 35637c478bd9Sstevel@tonic-gate #ifdef DEBUG 35647c478bd9Sstevel@tonic-gate static int nfs_lostpage = 0; /* number of times we lost original page */ 35657c478bd9Sstevel@tonic-gate #endif 35667c478bd9Sstevel@tonic-gate 35677c478bd9Sstevel@tonic-gate /* 35687c478bd9Sstevel@tonic-gate * Return all the pages from [off..off+len) in file 35697c478bd9Sstevel@tonic-gate */ 3570da6c28aaSamw /* ARGSUSED */ 35717c478bd9Sstevel@tonic-gate static int 35727c478bd9Sstevel@tonic-gate nfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 35737c478bd9Sstevel@tonic-gate page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3574da6c28aaSamw enum seg_rw rw, cred_t *cr, caller_context_t *ct) 35757c478bd9Sstevel@tonic-gate { 35767c478bd9Sstevel@tonic-gate rnode_t *rp; 35777c478bd9Sstevel@tonic-gate int error; 35787c478bd9Sstevel@tonic-gate mntinfo_t *mi; 35797c478bd9Sstevel@tonic-gate 35807c478bd9Sstevel@tonic-gate if (vp->v_flag & VNOMAP) 35817c478bd9Sstevel@tonic-gate return (ENOSYS); 35827c478bd9Sstevel@tonic-gate 35837c478bd9Sstevel@tonic-gate ASSERT(off <= MAXOFF32_T); 3584108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 35857c478bd9Sstevel@tonic-gate return (EIO); 35867c478bd9Sstevel@tonic-gate if (protp != NULL) 35877c478bd9Sstevel@tonic-gate *protp = PROT_ALL; 35887c478bd9Sstevel@tonic-gate 35897c478bd9Sstevel@tonic-gate /* 35907c478bd9Sstevel@tonic-gate * Now valididate that the caches are up to date. 35917c478bd9Sstevel@tonic-gate */ 35927c478bd9Sstevel@tonic-gate error = nfs_validate_caches(vp, cr); 35937c478bd9Sstevel@tonic-gate if (error) 35947c478bd9Sstevel@tonic-gate return (error); 35957c478bd9Sstevel@tonic-gate 35967c478bd9Sstevel@tonic-gate rp = VTOR(vp); 35977c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 35987c478bd9Sstevel@tonic-gate retry: 35997c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 36007c478bd9Sstevel@tonic-gate 36017c478bd9Sstevel@tonic-gate /* 36027c478bd9Sstevel@tonic-gate * Don't create dirty pages faster than they 36037c478bd9Sstevel@tonic-gate * can be cleaned so that the system doesn't 36047c478bd9Sstevel@tonic-gate * get imbalanced. If the async queue is 36057c478bd9Sstevel@tonic-gate * maxed out, then wait for it to drain before 36067c478bd9Sstevel@tonic-gate * creating more dirty pages. Also, wait for 36077c478bd9Sstevel@tonic-gate * any threads doing pagewalks in the vop_getattr 36087c478bd9Sstevel@tonic-gate * entry points so that they don't block for 36097c478bd9Sstevel@tonic-gate * long periods. 36107c478bd9Sstevel@tonic-gate */ 36117c478bd9Sstevel@tonic-gate if (rw == S_CREATE) { 36127c478bd9Sstevel@tonic-gate while ((mi->mi_max_threads != 0 && 36137c478bd9Sstevel@tonic-gate rp->r_awcount > 2 * mi->mi_max_threads) || 36147c478bd9Sstevel@tonic-gate rp->r_gcount > 0) 36157c478bd9Sstevel@tonic-gate cv_wait(&rp->r_cv, &rp->r_statelock); 36167c478bd9Sstevel@tonic-gate } 36177c478bd9Sstevel@tonic-gate 36187c478bd9Sstevel@tonic-gate /* 36197c478bd9Sstevel@tonic-gate * If we are getting called as a side effect of an nfs_write() 36207c478bd9Sstevel@tonic-gate * operation the local file size might not be extended yet. 36217c478bd9Sstevel@tonic-gate * In this case we want to be able to return pages of zeroes. 36227c478bd9Sstevel@tonic-gate */ 36237c478bd9Sstevel@tonic-gate if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) { 36247c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 36257c478bd9Sstevel@tonic-gate return (EFAULT); /* beyond EOF */ 36267c478bd9Sstevel@tonic-gate } 36277c478bd9Sstevel@tonic-gate 36287c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 36297c478bd9Sstevel@tonic-gate 36307c478bd9Sstevel@tonic-gate if (len <= PAGESIZE) { 36317c478bd9Sstevel@tonic-gate error = nfs_getapage(vp, off, len, protp, pl, plsz, 36327c478bd9Sstevel@tonic-gate seg, addr, rw, cr); 36337c478bd9Sstevel@tonic-gate } else { 36347c478bd9Sstevel@tonic-gate error = pvn_getpages(nfs_getapage, vp, off, len, protp, 36357c478bd9Sstevel@tonic-gate pl, plsz, seg, addr, rw, cr); 36367c478bd9Sstevel@tonic-gate } 36377c478bd9Sstevel@tonic-gate 36387c478bd9Sstevel@tonic-gate switch (error) { 36397c478bd9Sstevel@tonic-gate case NFS_EOF: 36407c478bd9Sstevel@tonic-gate nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr); 36417c478bd9Sstevel@tonic-gate goto retry; 36427c478bd9Sstevel@tonic-gate case ESTALE: 36437c478bd9Sstevel@tonic-gate PURGE_STALE_FH(error, vp, cr); 36447c478bd9Sstevel@tonic-gate } 36457c478bd9Sstevel@tonic-gate 36467c478bd9Sstevel@tonic-gate return (error); 36477c478bd9Sstevel@tonic-gate } 36487c478bd9Sstevel@tonic-gate 36497c478bd9Sstevel@tonic-gate /* 36507c478bd9Sstevel@tonic-gate * Called from pvn_getpages or nfs_getpage to get a particular page. 36517c478bd9Sstevel@tonic-gate */ 36527c478bd9Sstevel@tonic-gate /* ARGSUSED */ 36537c478bd9Sstevel@tonic-gate static int 36547c478bd9Sstevel@tonic-gate nfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp, 36557c478bd9Sstevel@tonic-gate page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 36567c478bd9Sstevel@tonic-gate enum seg_rw rw, cred_t *cr) 36577c478bd9Sstevel@tonic-gate { 36587c478bd9Sstevel@tonic-gate rnode_t *rp; 36597c478bd9Sstevel@tonic-gate uint_t bsize; 36607c478bd9Sstevel@tonic-gate struct buf *bp; 36617c478bd9Sstevel@tonic-gate page_t *pp; 36627c478bd9Sstevel@tonic-gate u_offset_t lbn; 36637c478bd9Sstevel@tonic-gate u_offset_t io_off; 36647c478bd9Sstevel@tonic-gate u_offset_t blkoff; 36657c478bd9Sstevel@tonic-gate u_offset_t rablkoff; 36667c478bd9Sstevel@tonic-gate size_t io_len; 36677c478bd9Sstevel@tonic-gate uint_t blksize; 36687c478bd9Sstevel@tonic-gate int error; 36697c478bd9Sstevel@tonic-gate int readahead; 36707c478bd9Sstevel@tonic-gate int readahead_issued = 0; 36717c478bd9Sstevel@tonic-gate int ra_window; /* readahead window */ 36727c478bd9Sstevel@tonic-gate page_t *pagefound; 36737c478bd9Sstevel@tonic-gate 3674108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 36757c478bd9Sstevel@tonic-gate return (EIO); 36767c478bd9Sstevel@tonic-gate rp = VTOR(vp); 36777c478bd9Sstevel@tonic-gate bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 36787c478bd9Sstevel@tonic-gate 36797c478bd9Sstevel@tonic-gate reread: 36807c478bd9Sstevel@tonic-gate bp = NULL; 36817c478bd9Sstevel@tonic-gate pp = NULL; 36827c478bd9Sstevel@tonic-gate pagefound = NULL; 36837c478bd9Sstevel@tonic-gate 36847c478bd9Sstevel@tonic-gate if (pl != NULL) 36857c478bd9Sstevel@tonic-gate pl[0] = NULL; 36867c478bd9Sstevel@tonic-gate 36877c478bd9Sstevel@tonic-gate error = 0; 36887c478bd9Sstevel@tonic-gate lbn = off / bsize; 36897c478bd9Sstevel@tonic-gate blkoff = lbn * bsize; 36907c478bd9Sstevel@tonic-gate 36917c478bd9Sstevel@tonic-gate /* 36927c478bd9Sstevel@tonic-gate * Queueing up the readahead before doing the synchronous read 36937c478bd9Sstevel@tonic-gate * results in a significant increase in read throughput because 36947c478bd9Sstevel@tonic-gate * of the increased parallelism between the async threads and 36957c478bd9Sstevel@tonic-gate * the process context. 36967c478bd9Sstevel@tonic-gate */ 36977c478bd9Sstevel@tonic-gate if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 && 36987c478bd9Sstevel@tonic-gate rw != S_CREATE && 36997c478bd9Sstevel@tonic-gate !(vp->v_flag & VNOCACHE)) { 37007c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 37017c478bd9Sstevel@tonic-gate 37027c478bd9Sstevel@tonic-gate /* 37037c478bd9Sstevel@tonic-gate * Calculate the number of readaheads to do. 37047c478bd9Sstevel@tonic-gate * a) No readaheads at offset = 0. 37057c478bd9Sstevel@tonic-gate * b) Do maximum(nfs_nra) readaheads when the readahead 37067c478bd9Sstevel@tonic-gate * window is closed. 37077c478bd9Sstevel@tonic-gate * c) Do readaheads between 1 to (nfs_nra - 1) depending 37087c478bd9Sstevel@tonic-gate * upon how far the readahead window is open or close. 37097c478bd9Sstevel@tonic-gate * d) No readaheads if rp->r_nextr is not within the scope 37107c478bd9Sstevel@tonic-gate * of the readahead window (random i/o). 37117c478bd9Sstevel@tonic-gate */ 37127c478bd9Sstevel@tonic-gate 37137c478bd9Sstevel@tonic-gate if (off == 0) 37147c478bd9Sstevel@tonic-gate readahead = 0; 37157c478bd9Sstevel@tonic-gate else if (blkoff == rp->r_nextr) 37167c478bd9Sstevel@tonic-gate readahead = nfs_nra; 37177c478bd9Sstevel@tonic-gate else if (rp->r_nextr > blkoff && 37187c478bd9Sstevel@tonic-gate ((ra_window = (rp->r_nextr - blkoff) / bsize) 37197c478bd9Sstevel@tonic-gate <= (nfs_nra - 1))) 37207c478bd9Sstevel@tonic-gate readahead = nfs_nra - ra_window; 37217c478bd9Sstevel@tonic-gate else 37227c478bd9Sstevel@tonic-gate readahead = 0; 37237c478bd9Sstevel@tonic-gate 37247c478bd9Sstevel@tonic-gate rablkoff = rp->r_nextr; 37257c478bd9Sstevel@tonic-gate while (readahead > 0 && rablkoff + bsize < rp->r_size) { 37267c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 37277c478bd9Sstevel@tonic-gate if (nfs_async_readahead(vp, rablkoff + bsize, 37287c478bd9Sstevel@tonic-gate addr + (rablkoff + bsize - off), seg, cr, 37297c478bd9Sstevel@tonic-gate nfs_readahead) < 0) { 37307c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 37317c478bd9Sstevel@tonic-gate break; 37327c478bd9Sstevel@tonic-gate } 37337c478bd9Sstevel@tonic-gate readahead--; 37347c478bd9Sstevel@tonic-gate rablkoff += bsize; 37357c478bd9Sstevel@tonic-gate /* 37367c478bd9Sstevel@tonic-gate * Indicate that we did a readahead so 37377c478bd9Sstevel@tonic-gate * readahead offset is not updated 37387c478bd9Sstevel@tonic-gate * by the synchronous read below. 37397c478bd9Sstevel@tonic-gate */ 37407c478bd9Sstevel@tonic-gate readahead_issued = 1; 37417c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 37427c478bd9Sstevel@tonic-gate /* 37437c478bd9Sstevel@tonic-gate * set readahead offset to 37447c478bd9Sstevel@tonic-gate * offset of last async readahead 37457c478bd9Sstevel@tonic-gate * request. 37467c478bd9Sstevel@tonic-gate */ 37477c478bd9Sstevel@tonic-gate rp->r_nextr = rablkoff; 37487c478bd9Sstevel@tonic-gate } 37497c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 37507c478bd9Sstevel@tonic-gate } 37517c478bd9Sstevel@tonic-gate 37527c478bd9Sstevel@tonic-gate again: 37537c478bd9Sstevel@tonic-gate if ((pagefound = page_exists(vp, off)) == NULL) { 37547c478bd9Sstevel@tonic-gate if (pl == NULL) { 37557c478bd9Sstevel@tonic-gate (void) nfs_async_readahead(vp, blkoff, addr, seg, cr, 37567c478bd9Sstevel@tonic-gate nfs_readahead); 37577c478bd9Sstevel@tonic-gate } else if (rw == S_CREATE) { 37587c478bd9Sstevel@tonic-gate /* 37597c478bd9Sstevel@tonic-gate * Block for this page is not allocated, or the offset 37607c478bd9Sstevel@tonic-gate * is beyond the current allocation size, or we're 37617c478bd9Sstevel@tonic-gate * allocating a swap slot and the page was not found, 37627c478bd9Sstevel@tonic-gate * so allocate it and return a zero page. 37637c478bd9Sstevel@tonic-gate */ 37647c478bd9Sstevel@tonic-gate if ((pp = page_create_va(vp, off, 37657c478bd9Sstevel@tonic-gate PAGESIZE, PG_WAIT, seg, addr)) == NULL) 37667c478bd9Sstevel@tonic-gate cmn_err(CE_PANIC, "nfs_getapage: page_create"); 37677c478bd9Sstevel@tonic-gate io_len = PAGESIZE; 37687c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 37697c478bd9Sstevel@tonic-gate rp->r_nextr = off + PAGESIZE; 37707c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 37717c478bd9Sstevel@tonic-gate } else { 37727c478bd9Sstevel@tonic-gate /* 37737c478bd9Sstevel@tonic-gate * Need to go to server to get a BLOCK, exception to 37747c478bd9Sstevel@tonic-gate * that being while reading at offset = 0 or doing 37757c478bd9Sstevel@tonic-gate * random i/o, in that case read only a PAGE. 37767c478bd9Sstevel@tonic-gate */ 37777c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 37787c478bd9Sstevel@tonic-gate if (blkoff < rp->r_size && 37797c478bd9Sstevel@tonic-gate blkoff + bsize >= rp->r_size) { 37807c478bd9Sstevel@tonic-gate /* 37817c478bd9Sstevel@tonic-gate * If only a block or less is left in 37827c478bd9Sstevel@tonic-gate * the file, read all that is remaining. 37837c478bd9Sstevel@tonic-gate */ 37847c478bd9Sstevel@tonic-gate if (rp->r_size <= off) { 37857c478bd9Sstevel@tonic-gate /* 37867c478bd9Sstevel@tonic-gate * Trying to access beyond EOF, 37877c478bd9Sstevel@tonic-gate * set up to get at least one page. 37887c478bd9Sstevel@tonic-gate */ 37897c478bd9Sstevel@tonic-gate blksize = off + PAGESIZE - blkoff; 37907c478bd9Sstevel@tonic-gate } else 37917c478bd9Sstevel@tonic-gate blksize = rp->r_size - blkoff; 37927c478bd9Sstevel@tonic-gate } else if ((off == 0) || 37937c478bd9Sstevel@tonic-gate (off != rp->r_nextr && !readahead_issued)) { 37947c478bd9Sstevel@tonic-gate blksize = PAGESIZE; 37957c478bd9Sstevel@tonic-gate blkoff = off; /* block = page here */ 37967c478bd9Sstevel@tonic-gate } else 37977c478bd9Sstevel@tonic-gate blksize = bsize; 37987c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 37997c478bd9Sstevel@tonic-gate 38007c478bd9Sstevel@tonic-gate pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 38017c478bd9Sstevel@tonic-gate &io_len, blkoff, blksize, 0); 38027c478bd9Sstevel@tonic-gate 38037c478bd9Sstevel@tonic-gate /* 38047c478bd9Sstevel@tonic-gate * Some other thread has entered the page, 38057c478bd9Sstevel@tonic-gate * so just use it. 38067c478bd9Sstevel@tonic-gate */ 38077c478bd9Sstevel@tonic-gate if (pp == NULL) 38087c478bd9Sstevel@tonic-gate goto again; 38097c478bd9Sstevel@tonic-gate 38107c478bd9Sstevel@tonic-gate /* 38117c478bd9Sstevel@tonic-gate * Now round the request size up to page boundaries. 38127c478bd9Sstevel@tonic-gate * This ensures that the entire page will be 38137c478bd9Sstevel@tonic-gate * initialized to zeroes if EOF is encountered. 38147c478bd9Sstevel@tonic-gate */ 38157c478bd9Sstevel@tonic-gate io_len = ptob(btopr(io_len)); 38167c478bd9Sstevel@tonic-gate 38177c478bd9Sstevel@tonic-gate bp = pageio_setup(pp, io_len, vp, B_READ); 38187c478bd9Sstevel@tonic-gate ASSERT(bp != NULL); 38197c478bd9Sstevel@tonic-gate 38207c478bd9Sstevel@tonic-gate /* 38217c478bd9Sstevel@tonic-gate * pageio_setup should have set b_addr to 0. This 38227c478bd9Sstevel@tonic-gate * is correct since we want to do I/O on a page 38237c478bd9Sstevel@tonic-gate * boundary. bp_mapin will use this addr to calculate 38247c478bd9Sstevel@tonic-gate * an offset, and then set b_addr to the kernel virtual 38257c478bd9Sstevel@tonic-gate * address it allocated for us. 38267c478bd9Sstevel@tonic-gate */ 38277c478bd9Sstevel@tonic-gate ASSERT(bp->b_un.b_addr == 0); 38287c478bd9Sstevel@tonic-gate 38297c478bd9Sstevel@tonic-gate bp->b_edev = 0; 38307c478bd9Sstevel@tonic-gate bp->b_dev = 0; 38317c478bd9Sstevel@tonic-gate bp->b_lblkno = lbtodb(io_off); 38327c478bd9Sstevel@tonic-gate bp->b_file = vp; 38337c478bd9Sstevel@tonic-gate bp->b_offset = (offset_t)off; 38347c478bd9Sstevel@tonic-gate bp_mapin(bp); 38357c478bd9Sstevel@tonic-gate 38367c478bd9Sstevel@tonic-gate /* 38377c478bd9Sstevel@tonic-gate * If doing a write beyond what we believe is EOF, 38387c478bd9Sstevel@tonic-gate * don't bother trying to read the pages from the 38397c478bd9Sstevel@tonic-gate * server, we'll just zero the pages here. We 38407c478bd9Sstevel@tonic-gate * don't check that the rw flag is S_WRITE here 38417c478bd9Sstevel@tonic-gate * because some implementations may attempt a 38427c478bd9Sstevel@tonic-gate * read access to the buffer before copying data. 38437c478bd9Sstevel@tonic-gate */ 38447c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 38457c478bd9Sstevel@tonic-gate if (io_off >= rp->r_size && seg == segkmap) { 38467c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 38477c478bd9Sstevel@tonic-gate bzero(bp->b_un.b_addr, io_len); 38487c478bd9Sstevel@tonic-gate } else { 38497c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 38507c478bd9Sstevel@tonic-gate error = nfs_bio(bp, cr); 38517c478bd9Sstevel@tonic-gate } 38527c478bd9Sstevel@tonic-gate 38537c478bd9Sstevel@tonic-gate /* 38547c478bd9Sstevel@tonic-gate * Unmap the buffer before freeing it. 38557c478bd9Sstevel@tonic-gate */ 38567c478bd9Sstevel@tonic-gate bp_mapout(bp); 38577c478bd9Sstevel@tonic-gate pageio_done(bp); 38587c478bd9Sstevel@tonic-gate 38597c478bd9Sstevel@tonic-gate if (error == NFS_EOF) { 38607c478bd9Sstevel@tonic-gate /* 38617c478bd9Sstevel@tonic-gate * If doing a write system call just return 38627c478bd9Sstevel@tonic-gate * zeroed pages, else user tried to get pages 38637c478bd9Sstevel@tonic-gate * beyond EOF, return error. We don't check 38647c478bd9Sstevel@tonic-gate * that the rw flag is S_WRITE here because 38657c478bd9Sstevel@tonic-gate * some implementations may attempt a read 38667c478bd9Sstevel@tonic-gate * access to the buffer before copying data. 38677c478bd9Sstevel@tonic-gate */ 38687c478bd9Sstevel@tonic-gate if (seg == segkmap) 38697c478bd9Sstevel@tonic-gate error = 0; 38707c478bd9Sstevel@tonic-gate else 38717c478bd9Sstevel@tonic-gate error = EFAULT; 38727c478bd9Sstevel@tonic-gate } 38737c478bd9Sstevel@tonic-gate 38747c478bd9Sstevel@tonic-gate if (!readahead_issued && !error) { 38757c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 38767c478bd9Sstevel@tonic-gate rp->r_nextr = io_off + io_len; 38777c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 38787c478bd9Sstevel@tonic-gate } 38797c478bd9Sstevel@tonic-gate } 38807c478bd9Sstevel@tonic-gate } 38817c478bd9Sstevel@tonic-gate 38827c478bd9Sstevel@tonic-gate out: 38837c478bd9Sstevel@tonic-gate if (pl == NULL) 38847c478bd9Sstevel@tonic-gate return (error); 38857c478bd9Sstevel@tonic-gate 38867c478bd9Sstevel@tonic-gate if (error) { 38877c478bd9Sstevel@tonic-gate if (pp != NULL) 38887c478bd9Sstevel@tonic-gate pvn_read_done(pp, B_ERROR); 38897c478bd9Sstevel@tonic-gate return (error); 38907c478bd9Sstevel@tonic-gate } 38917c478bd9Sstevel@tonic-gate 38927c478bd9Sstevel@tonic-gate if (pagefound) { 38937c478bd9Sstevel@tonic-gate se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED); 38947c478bd9Sstevel@tonic-gate 38957c478bd9Sstevel@tonic-gate /* 38967c478bd9Sstevel@tonic-gate * Page exists in the cache, acquire the appropriate lock. 38977c478bd9Sstevel@tonic-gate * If this fails, start all over again. 38987c478bd9Sstevel@tonic-gate */ 38997c478bd9Sstevel@tonic-gate if ((pp = page_lookup(vp, off, se)) == NULL) { 39007c478bd9Sstevel@tonic-gate #ifdef DEBUG 39017c478bd9Sstevel@tonic-gate nfs_lostpage++; 39027c478bd9Sstevel@tonic-gate #endif 39037c478bd9Sstevel@tonic-gate goto reread; 39047c478bd9Sstevel@tonic-gate } 39057c478bd9Sstevel@tonic-gate pl[0] = pp; 39067c478bd9Sstevel@tonic-gate pl[1] = NULL; 39077c478bd9Sstevel@tonic-gate return (0); 39087c478bd9Sstevel@tonic-gate } 39097c478bd9Sstevel@tonic-gate 39107c478bd9Sstevel@tonic-gate if (pp != NULL) 39117c478bd9Sstevel@tonic-gate pvn_plist_init(pp, pl, plsz, off, io_len, rw); 39127c478bd9Sstevel@tonic-gate 39137c478bd9Sstevel@tonic-gate return (error); 39147c478bd9Sstevel@tonic-gate } 39157c478bd9Sstevel@tonic-gate 39167c478bd9Sstevel@tonic-gate static void 39177c478bd9Sstevel@tonic-gate nfs_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg, 39187c478bd9Sstevel@tonic-gate cred_t *cr) 39197c478bd9Sstevel@tonic-gate { 39207c478bd9Sstevel@tonic-gate int error; 39217c478bd9Sstevel@tonic-gate page_t *pp; 39227c478bd9Sstevel@tonic-gate u_offset_t io_off; 39237c478bd9Sstevel@tonic-gate size_t io_len; 39247c478bd9Sstevel@tonic-gate struct buf *bp; 39257c478bd9Sstevel@tonic-gate uint_t bsize, blksize; 39267c478bd9Sstevel@tonic-gate rnode_t *rp = VTOR(vp); 39277c478bd9Sstevel@tonic-gate 3928108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 39297c478bd9Sstevel@tonic-gate 39307c478bd9Sstevel@tonic-gate bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 39317c478bd9Sstevel@tonic-gate 39327c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 39337c478bd9Sstevel@tonic-gate if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) { 39347c478bd9Sstevel@tonic-gate /* 39357c478bd9Sstevel@tonic-gate * If less than a block left in file read less 39367c478bd9Sstevel@tonic-gate * than a block. 39377c478bd9Sstevel@tonic-gate */ 39387c478bd9Sstevel@tonic-gate blksize = rp->r_size - blkoff; 39397c478bd9Sstevel@tonic-gate } else 39407c478bd9Sstevel@tonic-gate blksize = bsize; 39417c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 39427c478bd9Sstevel@tonic-gate 39437c478bd9Sstevel@tonic-gate pp = pvn_read_kluster(vp, blkoff, segkmap, addr, 39447c478bd9Sstevel@tonic-gate &io_off, &io_len, blkoff, blksize, 1); 39457c478bd9Sstevel@tonic-gate /* 39467c478bd9Sstevel@tonic-gate * The isra flag passed to the kluster function is 1, we may have 39477c478bd9Sstevel@tonic-gate * gotten a return value of NULL for a variety of reasons (# of free 39487c478bd9Sstevel@tonic-gate * pages < minfree, someone entered the page on the vnode etc). In all 39497c478bd9Sstevel@tonic-gate * cases, we want to punt on the readahead. 39507c478bd9Sstevel@tonic-gate */ 39517c478bd9Sstevel@tonic-gate if (pp == NULL) 39527c478bd9Sstevel@tonic-gate return; 39537c478bd9Sstevel@tonic-gate 39547c478bd9Sstevel@tonic-gate /* 39557c478bd9Sstevel@tonic-gate * Now round the request size up to page boundaries. 39567c478bd9Sstevel@tonic-gate * This ensures that the entire page will be 39577c478bd9Sstevel@tonic-gate * initialized to zeroes if EOF is encountered. 39587c478bd9Sstevel@tonic-gate */ 39597c478bd9Sstevel@tonic-gate io_len = ptob(btopr(io_len)); 39607c478bd9Sstevel@tonic-gate 39617c478bd9Sstevel@tonic-gate bp = pageio_setup(pp, io_len, vp, B_READ); 39627c478bd9Sstevel@tonic-gate ASSERT(bp != NULL); 39637c478bd9Sstevel@tonic-gate 39647c478bd9Sstevel@tonic-gate /* 39657c478bd9Sstevel@tonic-gate * pageio_setup should have set b_addr to 0. This is correct since 39667c478bd9Sstevel@tonic-gate * we want to do I/O on a page boundary. bp_mapin() will use this addr 39677c478bd9Sstevel@tonic-gate * to calculate an offset, and then set b_addr to the kernel virtual 39687c478bd9Sstevel@tonic-gate * address it allocated for us. 39697c478bd9Sstevel@tonic-gate */ 39707c478bd9Sstevel@tonic-gate ASSERT(bp->b_un.b_addr == 0); 39717c478bd9Sstevel@tonic-gate 39727c478bd9Sstevel@tonic-gate bp->b_edev = 0; 39737c478bd9Sstevel@tonic-gate bp->b_dev = 0; 39747c478bd9Sstevel@tonic-gate bp->b_lblkno = lbtodb(io_off); 39757c478bd9Sstevel@tonic-gate bp->b_file = vp; 39767c478bd9Sstevel@tonic-gate bp->b_offset = (offset_t)blkoff; 39777c478bd9Sstevel@tonic-gate bp_mapin(bp); 39787c478bd9Sstevel@tonic-gate 39797c478bd9Sstevel@tonic-gate /* 39807c478bd9Sstevel@tonic-gate * If doing a write beyond what we believe is EOF, don't bother trying 39817c478bd9Sstevel@tonic-gate * to read the pages from the server, we'll just zero the pages here. 39827c478bd9Sstevel@tonic-gate * We don't check that the rw flag is S_WRITE here because some 39837c478bd9Sstevel@tonic-gate * implementations may attempt a read access to the buffer before 39847c478bd9Sstevel@tonic-gate * copying data. 39857c478bd9Sstevel@tonic-gate */ 39867c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 39877c478bd9Sstevel@tonic-gate if (io_off >= rp->r_size && seg == segkmap) { 39887c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 39897c478bd9Sstevel@tonic-gate bzero(bp->b_un.b_addr, io_len); 39907c478bd9Sstevel@tonic-gate error = 0; 39917c478bd9Sstevel@tonic-gate } else { 39927c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 39937c478bd9Sstevel@tonic-gate error = nfs_bio(bp, cr); 39947c478bd9Sstevel@tonic-gate if (error == NFS_EOF) 39957c478bd9Sstevel@tonic-gate error = 0; 39967c478bd9Sstevel@tonic-gate } 39977c478bd9Sstevel@tonic-gate 39987c478bd9Sstevel@tonic-gate /* 39997c478bd9Sstevel@tonic-gate * Unmap the buffer before freeing it. 40007c478bd9Sstevel@tonic-gate */ 40017c478bd9Sstevel@tonic-gate bp_mapout(bp); 40027c478bd9Sstevel@tonic-gate pageio_done(bp); 40037c478bd9Sstevel@tonic-gate 40047c478bd9Sstevel@tonic-gate pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ); 40057c478bd9Sstevel@tonic-gate 40067c478bd9Sstevel@tonic-gate /* 40077c478bd9Sstevel@tonic-gate * In case of error set readahead offset 40087c478bd9Sstevel@tonic-gate * to the lowest offset. 40097c478bd9Sstevel@tonic-gate * pvn_read_done() calls VN_DISPOSE to destroy the pages 40107c478bd9Sstevel@tonic-gate */ 40117c478bd9Sstevel@tonic-gate if (error && rp->r_nextr > io_off) { 40127c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 40137c478bd9Sstevel@tonic-gate if (rp->r_nextr > io_off) 40147c478bd9Sstevel@tonic-gate rp->r_nextr = io_off; 40157c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 40167c478bd9Sstevel@tonic-gate } 40177c478bd9Sstevel@tonic-gate } 40187c478bd9Sstevel@tonic-gate 40197c478bd9Sstevel@tonic-gate /* 40207c478bd9Sstevel@tonic-gate * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE} 40217c478bd9Sstevel@tonic-gate * If len == 0, do from off to EOF. 40227c478bd9Sstevel@tonic-gate * 40237c478bd9Sstevel@tonic-gate * The normal cases should be len == 0 && off == 0 (entire vp list), 40247c478bd9Sstevel@tonic-gate * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE 40257c478bd9Sstevel@tonic-gate * (from pageout). 40267c478bd9Sstevel@tonic-gate */ 4027da6c28aaSamw /* ARGSUSED */ 40287c478bd9Sstevel@tonic-gate static int 4029da6c28aaSamw nfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4030da6c28aaSamw caller_context_t *ct) 40317c478bd9Sstevel@tonic-gate { 40327c478bd9Sstevel@tonic-gate int error; 40337c478bd9Sstevel@tonic-gate rnode_t *rp; 40347c478bd9Sstevel@tonic-gate 40357c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 40367c478bd9Sstevel@tonic-gate 40377c478bd9Sstevel@tonic-gate /* 40387c478bd9Sstevel@tonic-gate * XXX - Why should this check be made here? 40397c478bd9Sstevel@tonic-gate */ 40407c478bd9Sstevel@tonic-gate if (vp->v_flag & VNOMAP) 40417c478bd9Sstevel@tonic-gate return (ENOSYS); 40427c478bd9Sstevel@tonic-gate 40437c478bd9Sstevel@tonic-gate if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp)) 40447c478bd9Sstevel@tonic-gate return (0); 40457c478bd9Sstevel@tonic-gate 4046108322fbScarlsonj if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone) 40477c478bd9Sstevel@tonic-gate return (EIO); 40487c478bd9Sstevel@tonic-gate ASSERT(off <= MAXOFF32_T); 40497c478bd9Sstevel@tonic-gate 40507c478bd9Sstevel@tonic-gate rp = VTOR(vp); 40517c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 40527c478bd9Sstevel@tonic-gate rp->r_count++; 40537c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 40547c478bd9Sstevel@tonic-gate error = nfs_putpages(vp, off, len, flags, cr); 40557c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 40567c478bd9Sstevel@tonic-gate rp->r_count--; 40577c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 40587c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 40597c478bd9Sstevel@tonic-gate 40607c478bd9Sstevel@tonic-gate return (error); 40617c478bd9Sstevel@tonic-gate } 40627c478bd9Sstevel@tonic-gate 40637c478bd9Sstevel@tonic-gate /* 40647c478bd9Sstevel@tonic-gate * Write out a single page, possibly klustering adjacent dirty pages. 40657c478bd9Sstevel@tonic-gate */ 40667c478bd9Sstevel@tonic-gate int 40677c478bd9Sstevel@tonic-gate nfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 40687c478bd9Sstevel@tonic-gate int flags, cred_t *cr) 40697c478bd9Sstevel@tonic-gate { 40707c478bd9Sstevel@tonic-gate u_offset_t io_off; 40717c478bd9Sstevel@tonic-gate u_offset_t lbn_off; 40727c478bd9Sstevel@tonic-gate u_offset_t lbn; 40737c478bd9Sstevel@tonic-gate size_t io_len; 40747c478bd9Sstevel@tonic-gate uint_t bsize; 40757c478bd9Sstevel@tonic-gate int error; 40767c478bd9Sstevel@tonic-gate rnode_t *rp; 40777c478bd9Sstevel@tonic-gate 40787c478bd9Sstevel@tonic-gate ASSERT(!vn_is_readonly(vp)); 40797c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 40807c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 4081108322fbScarlsonj ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone); 40827c478bd9Sstevel@tonic-gate 40837c478bd9Sstevel@tonic-gate rp = VTOR(vp); 40847c478bd9Sstevel@tonic-gate ASSERT(rp->r_count > 0); 40857c478bd9Sstevel@tonic-gate 40867c478bd9Sstevel@tonic-gate ASSERT(pp->p_offset <= MAXOFF32_T); 40877c478bd9Sstevel@tonic-gate 40887c478bd9Sstevel@tonic-gate bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE); 40897c478bd9Sstevel@tonic-gate lbn = pp->p_offset / bsize; 40907c478bd9Sstevel@tonic-gate lbn_off = lbn * bsize; 40917c478bd9Sstevel@tonic-gate 40927c478bd9Sstevel@tonic-gate /* 40937c478bd9Sstevel@tonic-gate * Find a kluster that fits in one block, or in 40947c478bd9Sstevel@tonic-gate * one page if pages are bigger than blocks. If 40957c478bd9Sstevel@tonic-gate * there is less file space allocated than a whole 40967c478bd9Sstevel@tonic-gate * page, we'll shorten the i/o request below. 40977c478bd9Sstevel@tonic-gate */ 40987c478bd9Sstevel@tonic-gate pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off, 40997c478bd9Sstevel@tonic-gate roundup(bsize, PAGESIZE), flags); 41007c478bd9Sstevel@tonic-gate 41017c478bd9Sstevel@tonic-gate /* 41027c478bd9Sstevel@tonic-gate * pvn_write_kluster shouldn't have returned a page with offset 41037c478bd9Sstevel@tonic-gate * behind the original page we were given. Verify that. 41047c478bd9Sstevel@tonic-gate */ 41057c478bd9Sstevel@tonic-gate ASSERT((pp->p_offset / bsize) >= lbn); 41067c478bd9Sstevel@tonic-gate 41077c478bd9Sstevel@tonic-gate /* 41087c478bd9Sstevel@tonic-gate * Now pp will have the list of kept dirty pages marked for 41097c478bd9Sstevel@tonic-gate * write back. It will also handle invalidation and freeing 41107c478bd9Sstevel@tonic-gate * of pages that are not dirty. Check for page length rounding 41117c478bd9Sstevel@tonic-gate * problems. 41127c478bd9Sstevel@tonic-gate */ 41137c478bd9Sstevel@tonic-gate if (io_off + io_len > lbn_off + bsize) { 41147c478bd9Sstevel@tonic-gate ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE); 41157c478bd9Sstevel@tonic-gate io_len = lbn_off + bsize - io_off; 41167c478bd9Sstevel@tonic-gate } 41177c478bd9Sstevel@tonic-gate /* 41187c478bd9Sstevel@tonic-gate * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a 41197c478bd9Sstevel@tonic-gate * consistent value of r_size. RMODINPROGRESS is set in writerp(). 41207c478bd9Sstevel@tonic-gate * When RMODINPROGRESS is set it indicates that a uiomove() is in 41217c478bd9Sstevel@tonic-gate * progress and the r_size has not been made consistent with the 41227c478bd9Sstevel@tonic-gate * new size of the file. When the uiomove() completes the r_size is 41237c478bd9Sstevel@tonic-gate * updated and the RMODINPROGRESS flag is cleared. 41247c478bd9Sstevel@tonic-gate * 41257c478bd9Sstevel@tonic-gate * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a 41267c478bd9Sstevel@tonic-gate * consistent value of r_size. Without this handshaking, it is 41277c478bd9Sstevel@tonic-gate * possible that nfs(3)_bio() picks up the old value of r_size 41287c478bd9Sstevel@tonic-gate * before the uiomove() in writerp() completes. This will result 41297c478bd9Sstevel@tonic-gate * in the write through nfs(3)_bio() being dropped. 41307c478bd9Sstevel@tonic-gate * 41317c478bd9Sstevel@tonic-gate * More precisely, there is a window between the time the uiomove() 41327c478bd9Sstevel@tonic-gate * completes and the time the r_size is updated. If a VOP_PUTPAGE() 41337c478bd9Sstevel@tonic-gate * operation intervenes in this window, the page will be picked up, 41347c478bd9Sstevel@tonic-gate * because it is dirty (it will be unlocked, unless it was 41357c478bd9Sstevel@tonic-gate * pagecreate'd). When the page is picked up as dirty, the dirty 41367c478bd9Sstevel@tonic-gate * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is 41377c478bd9Sstevel@tonic-gate * checked. This will still be the old size. Therefore the page will 41387c478bd9Sstevel@tonic-gate * not be written out. When segmap_release() calls VOP_PUTPAGE(), 41397c478bd9Sstevel@tonic-gate * the page will be found to be clean and the write will be dropped. 41407c478bd9Sstevel@tonic-gate */ 41417c478bd9Sstevel@tonic-gate if (rp->r_flags & RMODINPROGRESS) { 41427c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 41437c478bd9Sstevel@tonic-gate if ((rp->r_flags & RMODINPROGRESS) && 41447c478bd9Sstevel@tonic-gate rp->r_modaddr + MAXBSIZE > io_off && 41457c478bd9Sstevel@tonic-gate rp->r_modaddr < io_off + io_len) { 41467c478bd9Sstevel@tonic-gate page_t *plist; 41477c478bd9Sstevel@tonic-gate /* 41487c478bd9Sstevel@tonic-gate * A write is in progress for this region of the file. 41497c478bd9Sstevel@tonic-gate * If we did not detect RMODINPROGRESS here then this 41507c478bd9Sstevel@tonic-gate * path through nfs_putapage() would eventually go to 41517c478bd9Sstevel@tonic-gate * nfs(3)_bio() and may not write out all of the data 41527c478bd9Sstevel@tonic-gate * in the pages. We end up losing data. So we decide 41537c478bd9Sstevel@tonic-gate * to set the modified bit on each page in the page 41547c478bd9Sstevel@tonic-gate * list and mark the rnode with RDIRTY. This write 41557c478bd9Sstevel@tonic-gate * will be restarted at some later time. 41567c478bd9Sstevel@tonic-gate */ 41577c478bd9Sstevel@tonic-gate plist = pp; 41587c478bd9Sstevel@tonic-gate while (plist != NULL) { 41597c478bd9Sstevel@tonic-gate pp = plist; 41607c478bd9Sstevel@tonic-gate page_sub(&plist, pp); 41617c478bd9Sstevel@tonic-gate hat_setmod(pp); 41627c478bd9Sstevel@tonic-gate page_io_unlock(pp); 41637c478bd9Sstevel@tonic-gate page_unlock(pp); 41647c478bd9Sstevel@tonic-gate } 41657c478bd9Sstevel@tonic-gate rp->r_flags |= RDIRTY; 41667c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 41677c478bd9Sstevel@tonic-gate if (offp) 41687c478bd9Sstevel@tonic-gate *offp = io_off; 41697c478bd9Sstevel@tonic-gate if (lenp) 41707c478bd9Sstevel@tonic-gate *lenp = io_len; 41717c478bd9Sstevel@tonic-gate return (0); 41727c478bd9Sstevel@tonic-gate } 41737c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 41747c478bd9Sstevel@tonic-gate } 41757c478bd9Sstevel@tonic-gate 41767c478bd9Sstevel@tonic-gate if (flags & B_ASYNC) { 41777c478bd9Sstevel@tonic-gate error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr, 41787c478bd9Sstevel@tonic-gate nfs_sync_putapage); 41797c478bd9Sstevel@tonic-gate } else 41807c478bd9Sstevel@tonic-gate error = nfs_sync_putapage(vp, pp, io_off, io_len, flags, cr); 41817c478bd9Sstevel@tonic-gate 41827c478bd9Sstevel@tonic-gate if (offp) 41837c478bd9Sstevel@tonic-gate *offp = io_off; 41847c478bd9Sstevel@tonic-gate if (lenp) 41857c478bd9Sstevel@tonic-gate *lenp = io_len; 41867c478bd9Sstevel@tonic-gate return (error); 41877c478bd9Sstevel@tonic-gate } 41887c478bd9Sstevel@tonic-gate 41897c478bd9Sstevel@tonic-gate static int 41907c478bd9Sstevel@tonic-gate nfs_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 41917c478bd9Sstevel@tonic-gate int flags, cred_t *cr) 41927c478bd9Sstevel@tonic-gate { 41937c478bd9Sstevel@tonic-gate int error; 41947c478bd9Sstevel@tonic-gate rnode_t *rp; 41957c478bd9Sstevel@tonic-gate 41967c478bd9Sstevel@tonic-gate flags |= B_WRITE; 41977c478bd9Sstevel@tonic-gate 4198108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 41997c478bd9Sstevel@tonic-gate error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 42007c478bd9Sstevel@tonic-gate 42017c478bd9Sstevel@tonic-gate rp = VTOR(vp); 42027c478bd9Sstevel@tonic-gate 42037c478bd9Sstevel@tonic-gate if ((error == ENOSPC || error == EDQUOT || error == EACCES) && 42047c478bd9Sstevel@tonic-gate (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) { 42057c478bd9Sstevel@tonic-gate if (!(rp->r_flags & ROUTOFSPACE)) { 42067c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 42077c478bd9Sstevel@tonic-gate rp->r_flags |= ROUTOFSPACE; 42087c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 42097c478bd9Sstevel@tonic-gate } 42107c478bd9Sstevel@tonic-gate flags |= B_ERROR; 42117c478bd9Sstevel@tonic-gate pvn_write_done(pp, flags); 42127c478bd9Sstevel@tonic-gate /* 42137c478bd9Sstevel@tonic-gate * If this was not an async thread, then try again to 42147c478bd9Sstevel@tonic-gate * write out the pages, but this time, also destroy 42157c478bd9Sstevel@tonic-gate * them whether or not the write is successful. This 42167c478bd9Sstevel@tonic-gate * will prevent memory from filling up with these 42177c478bd9Sstevel@tonic-gate * pages and destroying them is the only alternative 42187c478bd9Sstevel@tonic-gate * if they can't be written out. 42197c478bd9Sstevel@tonic-gate * 42207c478bd9Sstevel@tonic-gate * Don't do this if this is an async thread because 42217c478bd9Sstevel@tonic-gate * when the pages are unlocked in pvn_write_done, 42227c478bd9Sstevel@tonic-gate * some other thread could have come along, locked 42237c478bd9Sstevel@tonic-gate * them, and queued for an async thread. It would be 42247c478bd9Sstevel@tonic-gate * possible for all of the async threads to be tied 42257c478bd9Sstevel@tonic-gate * up waiting to lock the pages again and they would 42267c478bd9Sstevel@tonic-gate * all already be locked and waiting for an async 42277c478bd9Sstevel@tonic-gate * thread to handle them. Deadlock. 42287c478bd9Sstevel@tonic-gate */ 42297c478bd9Sstevel@tonic-gate if (!(flags & B_ASYNC)) { 42307c478bd9Sstevel@tonic-gate error = nfs_putpage(vp, io_off, io_len, 4231da6c28aaSamw B_INVAL | B_FORCE, cr, NULL); 42327c478bd9Sstevel@tonic-gate } 42337c478bd9Sstevel@tonic-gate } else { 42347c478bd9Sstevel@tonic-gate if (error) 42357c478bd9Sstevel@tonic-gate flags |= B_ERROR; 42367c478bd9Sstevel@tonic-gate else if (rp->r_flags & ROUTOFSPACE) { 42377c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 42387c478bd9Sstevel@tonic-gate rp->r_flags &= ~ROUTOFSPACE; 42397c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 42407c478bd9Sstevel@tonic-gate } 42417c478bd9Sstevel@tonic-gate pvn_write_done(pp, flags); 42427c478bd9Sstevel@tonic-gate } 42437c478bd9Sstevel@tonic-gate 42447c478bd9Sstevel@tonic-gate return (error); 42457c478bd9Sstevel@tonic-gate } 42467c478bd9Sstevel@tonic-gate 4247da6c28aaSamw /* ARGSUSED */ 42487c478bd9Sstevel@tonic-gate static int 42497c478bd9Sstevel@tonic-gate nfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4250da6c28aaSamw size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4251da6c28aaSamw caller_context_t *ct) 42527c478bd9Sstevel@tonic-gate { 42537c478bd9Sstevel@tonic-gate struct segvn_crargs vn_a; 42547c478bd9Sstevel@tonic-gate int error; 42557c478bd9Sstevel@tonic-gate rnode_t *rp; 42567c478bd9Sstevel@tonic-gate struct vattr va; 42577c478bd9Sstevel@tonic-gate 4258108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 42597c478bd9Sstevel@tonic-gate return (EIO); 42607c478bd9Sstevel@tonic-gate 42617c478bd9Sstevel@tonic-gate if (vp->v_flag & VNOMAP) 42627c478bd9Sstevel@tonic-gate return (ENOSYS); 42637c478bd9Sstevel@tonic-gate 42647c478bd9Sstevel@tonic-gate if (off > MAXOFF32_T) 42657c478bd9Sstevel@tonic-gate return (EFBIG); 42667c478bd9Sstevel@tonic-gate 42677c478bd9Sstevel@tonic-gate if (off < 0 || off + len < 0) 42687c478bd9Sstevel@tonic-gate return (ENXIO); 42697c478bd9Sstevel@tonic-gate 42707c478bd9Sstevel@tonic-gate if (vp->v_type != VREG) 42717c478bd9Sstevel@tonic-gate return (ENODEV); 42727c478bd9Sstevel@tonic-gate 42737c478bd9Sstevel@tonic-gate /* 42747c478bd9Sstevel@tonic-gate * If there is cached data and if close-to-open consistency 42757c478bd9Sstevel@tonic-gate * checking is not turned off and if the file system is not 42767c478bd9Sstevel@tonic-gate * mounted readonly, then force an over the wire getattr. 42777c478bd9Sstevel@tonic-gate * Otherwise, just invoke nfsgetattr to get a copy of the 42787c478bd9Sstevel@tonic-gate * attributes. The attribute cache will be used unless it 42797c478bd9Sstevel@tonic-gate * is timed out and if it is, then an over the wire getattr 42807c478bd9Sstevel@tonic-gate * will be issued. 42817c478bd9Sstevel@tonic-gate */ 42827c478bd9Sstevel@tonic-gate va.va_mask = AT_ALL; 42837c478bd9Sstevel@tonic-gate if (vn_has_cached_data(vp) && 42847c478bd9Sstevel@tonic-gate !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp)) 42857c478bd9Sstevel@tonic-gate error = nfs_getattr_otw(vp, &va, cr); 42867c478bd9Sstevel@tonic-gate else 42877c478bd9Sstevel@tonic-gate error = nfsgetattr(vp, &va, cr); 42887c478bd9Sstevel@tonic-gate if (error) 42897c478bd9Sstevel@tonic-gate return (error); 42907c478bd9Sstevel@tonic-gate 42917c478bd9Sstevel@tonic-gate /* 42927c478bd9Sstevel@tonic-gate * Check to see if the vnode is currently marked as not cachable. 42937c478bd9Sstevel@tonic-gate * This means portions of the file are locked (through VOP_FRLOCK). 42947c478bd9Sstevel@tonic-gate * In this case the map request must be refused. We use 42957c478bd9Sstevel@tonic-gate * rp->r_lkserlock to avoid a race with concurrent lock requests. 42967c478bd9Sstevel@tonic-gate */ 42977c478bd9Sstevel@tonic-gate rp = VTOR(vp); 42987c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) 42997c478bd9Sstevel@tonic-gate return (EINTR); 43007c478bd9Sstevel@tonic-gate 43017c478bd9Sstevel@tonic-gate if (vp->v_flag & VNOCACHE) { 43027c478bd9Sstevel@tonic-gate error = EAGAIN; 43037c478bd9Sstevel@tonic-gate goto done; 43047c478bd9Sstevel@tonic-gate } 43057c478bd9Sstevel@tonic-gate 43067c478bd9Sstevel@tonic-gate /* 43077c478bd9Sstevel@tonic-gate * Don't allow concurrent locks and mapping if mandatory locking is 43087c478bd9Sstevel@tonic-gate * enabled. 43097c478bd9Sstevel@tonic-gate */ 43107c478bd9Sstevel@tonic-gate if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) && 43117c478bd9Sstevel@tonic-gate MANDLOCK(vp, va.va_mode)) { 43127c478bd9Sstevel@tonic-gate error = EAGAIN; 43137c478bd9Sstevel@tonic-gate goto done; 43147c478bd9Sstevel@tonic-gate } 43157c478bd9Sstevel@tonic-gate 43167c478bd9Sstevel@tonic-gate as_rangelock(as); 431760946fe0Smec error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 431860946fe0Smec if (error != 0) { 43197c478bd9Sstevel@tonic-gate as_rangeunlock(as); 43207c478bd9Sstevel@tonic-gate goto done; 43217c478bd9Sstevel@tonic-gate } 43227c478bd9Sstevel@tonic-gate 43237c478bd9Sstevel@tonic-gate vn_a.vp = vp; 43247c478bd9Sstevel@tonic-gate vn_a.offset = off; 43257c478bd9Sstevel@tonic-gate vn_a.type = (flags & MAP_TYPE); 43267c478bd9Sstevel@tonic-gate vn_a.prot = (uchar_t)prot; 43277c478bd9Sstevel@tonic-gate vn_a.maxprot = (uchar_t)maxprot; 43287c478bd9Sstevel@tonic-gate vn_a.flags = (flags & ~MAP_TYPE); 43297c478bd9Sstevel@tonic-gate vn_a.cred = cr; 43307c478bd9Sstevel@tonic-gate vn_a.amp = NULL; 43317c478bd9Sstevel@tonic-gate vn_a.szc = 0; 43327c478bd9Sstevel@tonic-gate vn_a.lgrp_mem_policy_flags = 0; 43337c478bd9Sstevel@tonic-gate 43347c478bd9Sstevel@tonic-gate error = as_map(as, *addrp, len, segvn_create, &vn_a); 43357c478bd9Sstevel@tonic-gate as_rangeunlock(as); 43367c478bd9Sstevel@tonic-gate 43377c478bd9Sstevel@tonic-gate done: 43387c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_lkserlock); 43397c478bd9Sstevel@tonic-gate return (error); 43407c478bd9Sstevel@tonic-gate } 43417c478bd9Sstevel@tonic-gate 43427c478bd9Sstevel@tonic-gate /* ARGSUSED */ 43437c478bd9Sstevel@tonic-gate static int 43447c478bd9Sstevel@tonic-gate nfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4345da6c28aaSamw size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4346da6c28aaSamw caller_context_t *ct) 43477c478bd9Sstevel@tonic-gate { 43487c478bd9Sstevel@tonic-gate rnode_t *rp; 43497c478bd9Sstevel@tonic-gate 43507c478bd9Sstevel@tonic-gate if (vp->v_flag & VNOMAP) 43517c478bd9Sstevel@tonic-gate return (ENOSYS); 4352108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 43537c478bd9Sstevel@tonic-gate return (EIO); 43547c478bd9Sstevel@tonic-gate 43557c478bd9Sstevel@tonic-gate /* 43567c478bd9Sstevel@tonic-gate * Need to hold rwlock while incrementing the mapcnt so that 43577c478bd9Sstevel@tonic-gate * mmap'ing can be serialized with writes so that the caching 43587c478bd9Sstevel@tonic-gate * can be handled correctly. 43597c478bd9Sstevel@tonic-gate */ 43607c478bd9Sstevel@tonic-gate rp = VTOR(vp); 43617c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp))) 43627c478bd9Sstevel@tonic-gate return (EINTR); 43637c478bd9Sstevel@tonic-gate atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len)); 43647c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_rwlock); 43657c478bd9Sstevel@tonic-gate 43667c478bd9Sstevel@tonic-gate return (0); 43677c478bd9Sstevel@tonic-gate } 43687c478bd9Sstevel@tonic-gate 4369da6c28aaSamw /* ARGSUSED */ 43707c478bd9Sstevel@tonic-gate static int 4371da6c28aaSamw nfs_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, offset_t offset, 4372da6c28aaSamw struct flk_callback *flk_cbp, cred_t *cr, caller_context_t *ct) 43737c478bd9Sstevel@tonic-gate { 43747c478bd9Sstevel@tonic-gate netobj lm_fh; 43757c478bd9Sstevel@tonic-gate int rc; 43767c478bd9Sstevel@tonic-gate u_offset_t start, end; 43777c478bd9Sstevel@tonic-gate rnode_t *rp; 43787c478bd9Sstevel@tonic-gate int error = 0, intr = INTR(vp); 43797c478bd9Sstevel@tonic-gate 43807c478bd9Sstevel@tonic-gate /* check for valid cmd parameter */ 43817c478bd9Sstevel@tonic-gate if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW) 43827c478bd9Sstevel@tonic-gate return (EINVAL); 4383108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 43847c478bd9Sstevel@tonic-gate return (EIO); 43857c478bd9Sstevel@tonic-gate 43867c478bd9Sstevel@tonic-gate /* Verify l_type. */ 43877c478bd9Sstevel@tonic-gate switch (bfp->l_type) { 43887c478bd9Sstevel@tonic-gate case F_RDLCK: 43897c478bd9Sstevel@tonic-gate if (cmd != F_GETLK && !(flag & FREAD)) 43907c478bd9Sstevel@tonic-gate return (EBADF); 43917c478bd9Sstevel@tonic-gate break; 43927c478bd9Sstevel@tonic-gate case F_WRLCK: 43937c478bd9Sstevel@tonic-gate if (cmd != F_GETLK && !(flag & FWRITE)) 43947c478bd9Sstevel@tonic-gate return (EBADF); 43957c478bd9Sstevel@tonic-gate break; 43967c478bd9Sstevel@tonic-gate case F_UNLCK: 43977c478bd9Sstevel@tonic-gate intr = 0; 43987c478bd9Sstevel@tonic-gate break; 43997c478bd9Sstevel@tonic-gate 44007c478bd9Sstevel@tonic-gate default: 44017c478bd9Sstevel@tonic-gate return (EINVAL); 44027c478bd9Sstevel@tonic-gate } 44037c478bd9Sstevel@tonic-gate 44047c478bd9Sstevel@tonic-gate /* check the validity of the lock range */ 44057c478bd9Sstevel@tonic-gate if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset)) 44067c478bd9Sstevel@tonic-gate return (rc); 44077c478bd9Sstevel@tonic-gate if (rc = flk_check_lock_data(start, end, MAXOFF32_T)) 44087c478bd9Sstevel@tonic-gate return (rc); 44097c478bd9Sstevel@tonic-gate 44107c478bd9Sstevel@tonic-gate /* 44117c478bd9Sstevel@tonic-gate * If the filesystem is mounted using local locking, pass the 44127c478bd9Sstevel@tonic-gate * request off to the local locking code. 44137c478bd9Sstevel@tonic-gate */ 44147c478bd9Sstevel@tonic-gate if (VTOMI(vp)->mi_flags & MI_LLOCK) { 44157c478bd9Sstevel@tonic-gate if (offset > MAXOFF32_T) 44167c478bd9Sstevel@tonic-gate return (EFBIG); 44177c478bd9Sstevel@tonic-gate if (cmd == F_SETLK || cmd == F_SETLKW) { 44187c478bd9Sstevel@tonic-gate /* 44197c478bd9Sstevel@tonic-gate * For complete safety, we should be holding 44207c478bd9Sstevel@tonic-gate * r_lkserlock. However, we can't call 44217c478bd9Sstevel@tonic-gate * lm_safelock and then fs_frlock while 44227c478bd9Sstevel@tonic-gate * holding r_lkserlock, so just invoke 44237c478bd9Sstevel@tonic-gate * lm_safelock and expect that this will 44247c478bd9Sstevel@tonic-gate * catch enough of the cases. 44257c478bd9Sstevel@tonic-gate */ 44267c478bd9Sstevel@tonic-gate if (!lm_safelock(vp, bfp, cr)) 44277c478bd9Sstevel@tonic-gate return (EAGAIN); 44287c478bd9Sstevel@tonic-gate } 4429da6c28aaSamw return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 44307c478bd9Sstevel@tonic-gate } 44317c478bd9Sstevel@tonic-gate 44327c478bd9Sstevel@tonic-gate rp = VTOR(vp); 44337c478bd9Sstevel@tonic-gate 44347c478bd9Sstevel@tonic-gate /* 44357c478bd9Sstevel@tonic-gate * Check whether the given lock request can proceed, given the 44367c478bd9Sstevel@tonic-gate * current file mappings. 44377c478bd9Sstevel@tonic-gate */ 44387c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr)) 44397c478bd9Sstevel@tonic-gate return (EINTR); 44407c478bd9Sstevel@tonic-gate if (cmd == F_SETLK || cmd == F_SETLKW) { 44417c478bd9Sstevel@tonic-gate if (!lm_safelock(vp, bfp, cr)) { 44427c478bd9Sstevel@tonic-gate rc = EAGAIN; 44437c478bd9Sstevel@tonic-gate goto done; 44447c478bd9Sstevel@tonic-gate } 44457c478bd9Sstevel@tonic-gate } 44467c478bd9Sstevel@tonic-gate 44477c478bd9Sstevel@tonic-gate /* 44487c478bd9Sstevel@tonic-gate * Flush the cache after waiting for async I/O to finish. For new 44497c478bd9Sstevel@tonic-gate * locks, this is so that the process gets the latest bits from the 44507c478bd9Sstevel@tonic-gate * server. For unlocks, this is so that other clients see the 44517c478bd9Sstevel@tonic-gate * latest bits once the file has been unlocked. If currently dirty 44527c478bd9Sstevel@tonic-gate * pages can't be flushed, then don't allow a lock to be set. But 44537c478bd9Sstevel@tonic-gate * allow unlocks to succeed, to avoid having orphan locks on the 44547c478bd9Sstevel@tonic-gate * server. 44557c478bd9Sstevel@tonic-gate */ 44567c478bd9Sstevel@tonic-gate if (cmd != F_GETLK) { 44577c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 44587c478bd9Sstevel@tonic-gate while (rp->r_count > 0) { 44597c478bd9Sstevel@tonic-gate if (intr) { 44607c478bd9Sstevel@tonic-gate klwp_t *lwp = ttolwp(curthread); 44617c478bd9Sstevel@tonic-gate 44627c478bd9Sstevel@tonic-gate if (lwp != NULL) 44637c478bd9Sstevel@tonic-gate lwp->lwp_nostop++; 4464da6c28aaSamw if (cv_wait_sig(&rp->r_cv, &rp->r_statelock) 4465da6c28aaSamw == 0) { 44667c478bd9Sstevel@tonic-gate if (lwp != NULL) 44677c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 44687c478bd9Sstevel@tonic-gate rc = EINTR; 44697c478bd9Sstevel@tonic-gate break; 44707c478bd9Sstevel@tonic-gate } 44717c478bd9Sstevel@tonic-gate if (lwp != NULL) 44727c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 44737c478bd9Sstevel@tonic-gate } else 44747c478bd9Sstevel@tonic-gate cv_wait(&rp->r_cv, &rp->r_statelock); 44757c478bd9Sstevel@tonic-gate } 44767c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 44777c478bd9Sstevel@tonic-gate if (rc != 0) 44787c478bd9Sstevel@tonic-gate goto done; 4479da6c28aaSamw error = nfs_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct); 44807c478bd9Sstevel@tonic-gate if (error) { 44817c478bd9Sstevel@tonic-gate if (error == ENOSPC || error == EDQUOT) { 44827c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 44837c478bd9Sstevel@tonic-gate if (!rp->r_error) 44847c478bd9Sstevel@tonic-gate rp->r_error = error; 44857c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 44867c478bd9Sstevel@tonic-gate } 44877c478bd9Sstevel@tonic-gate if (bfp->l_type != F_UNLCK) { 44887c478bd9Sstevel@tonic-gate rc = ENOLCK; 44897c478bd9Sstevel@tonic-gate goto done; 44907c478bd9Sstevel@tonic-gate } 44917c478bd9Sstevel@tonic-gate } 44927c478bd9Sstevel@tonic-gate } 44937c478bd9Sstevel@tonic-gate 44947c478bd9Sstevel@tonic-gate lm_fh.n_len = sizeof (fhandle_t); 44957c478bd9Sstevel@tonic-gate lm_fh.n_bytes = (char *)VTOFH(vp); 44967c478bd9Sstevel@tonic-gate 44977c478bd9Sstevel@tonic-gate /* 44987c478bd9Sstevel@tonic-gate * Call the lock manager to do the real work of contacting 44997c478bd9Sstevel@tonic-gate * the server and obtaining the lock. 45007c478bd9Sstevel@tonic-gate */ 45017c478bd9Sstevel@tonic-gate rc = lm_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh, flk_cbp); 45027c478bd9Sstevel@tonic-gate 45037c478bd9Sstevel@tonic-gate if (rc == 0) 45047c478bd9Sstevel@tonic-gate nfs_lockcompletion(vp, cmd); 45057c478bd9Sstevel@tonic-gate 45067c478bd9Sstevel@tonic-gate done: 45077c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_lkserlock); 45087c478bd9Sstevel@tonic-gate return (rc); 45097c478bd9Sstevel@tonic-gate } 45107c478bd9Sstevel@tonic-gate 45117c478bd9Sstevel@tonic-gate /* 45127c478bd9Sstevel@tonic-gate * Free storage space associated with the specified vnode. The portion 45137c478bd9Sstevel@tonic-gate * to be freed is specified by bfp->l_start and bfp->l_len (already 45147c478bd9Sstevel@tonic-gate * normalized to a "whence" of 0). 45157c478bd9Sstevel@tonic-gate * 45167c478bd9Sstevel@tonic-gate * This is an experimental facility whose continued existence is not 45177c478bd9Sstevel@tonic-gate * guaranteed. Currently, we only support the special case 45187c478bd9Sstevel@tonic-gate * of l_len == 0, meaning free to end of file. 45197c478bd9Sstevel@tonic-gate */ 45207c478bd9Sstevel@tonic-gate /* ARGSUSED */ 45217c478bd9Sstevel@tonic-gate static int 45227c478bd9Sstevel@tonic-gate nfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, 45237c478bd9Sstevel@tonic-gate offset_t offset, cred_t *cr, caller_context_t *ct) 45247c478bd9Sstevel@tonic-gate { 45257c478bd9Sstevel@tonic-gate int error; 45267c478bd9Sstevel@tonic-gate 45277c478bd9Sstevel@tonic-gate ASSERT(vp->v_type == VREG); 45287c478bd9Sstevel@tonic-gate if (cmd != F_FREESP) 45297c478bd9Sstevel@tonic-gate return (EINVAL); 45307c478bd9Sstevel@tonic-gate 45317c478bd9Sstevel@tonic-gate if (offset > MAXOFF32_T) 45327c478bd9Sstevel@tonic-gate return (EFBIG); 45337c478bd9Sstevel@tonic-gate 45347c478bd9Sstevel@tonic-gate if ((bfp->l_start > MAXOFF32_T) || (bfp->l_end > MAXOFF32_T) || 45357c478bd9Sstevel@tonic-gate (bfp->l_len > MAXOFF32_T)) 45367c478bd9Sstevel@tonic-gate return (EFBIG); 45377c478bd9Sstevel@tonic-gate 4538108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 45397c478bd9Sstevel@tonic-gate return (EIO); 45407c478bd9Sstevel@tonic-gate 45417c478bd9Sstevel@tonic-gate error = convoff(vp, bfp, 0, offset); 45427c478bd9Sstevel@tonic-gate if (!error) { 45437c478bd9Sstevel@tonic-gate ASSERT(bfp->l_start >= 0); 45447c478bd9Sstevel@tonic-gate if (bfp->l_len == 0) { 45457c478bd9Sstevel@tonic-gate struct vattr va; 45467c478bd9Sstevel@tonic-gate 45477c478bd9Sstevel@tonic-gate /* 45487c478bd9Sstevel@tonic-gate * ftruncate should not change the ctime and 45497c478bd9Sstevel@tonic-gate * mtime if we truncate the file to its 45507c478bd9Sstevel@tonic-gate * previous size. 45517c478bd9Sstevel@tonic-gate */ 45527c478bd9Sstevel@tonic-gate va.va_mask = AT_SIZE; 45537c478bd9Sstevel@tonic-gate error = nfsgetattr(vp, &va, cr); 45547c478bd9Sstevel@tonic-gate if (error || va.va_size == bfp->l_start) 45557c478bd9Sstevel@tonic-gate return (error); 45567c478bd9Sstevel@tonic-gate va.va_mask = AT_SIZE; 45577c478bd9Sstevel@tonic-gate va.va_size = bfp->l_start; 45587c478bd9Sstevel@tonic-gate error = nfssetattr(vp, &va, 0, cr); 45597c478bd9Sstevel@tonic-gate } else 45607c478bd9Sstevel@tonic-gate error = EINVAL; 45617c478bd9Sstevel@tonic-gate } 45627c478bd9Sstevel@tonic-gate 45637c478bd9Sstevel@tonic-gate return (error); 45647c478bd9Sstevel@tonic-gate } 45657c478bd9Sstevel@tonic-gate 45667c478bd9Sstevel@tonic-gate /* ARGSUSED */ 45677c478bd9Sstevel@tonic-gate static int 4568da6c28aaSamw nfs_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct) 45697c478bd9Sstevel@tonic-gate { 45707c478bd9Sstevel@tonic-gate 45717c478bd9Sstevel@tonic-gate return (EINVAL); 45727c478bd9Sstevel@tonic-gate } 45737c478bd9Sstevel@tonic-gate 45747c478bd9Sstevel@tonic-gate /* 45757c478bd9Sstevel@tonic-gate * Setup and add an address space callback to do the work of the delmap call. 45767c478bd9Sstevel@tonic-gate * The callback will (and must be) deleted in the actual callback function. 45777c478bd9Sstevel@tonic-gate * 45787c478bd9Sstevel@tonic-gate * This is done in order to take care of the problem that we have with holding 45797c478bd9Sstevel@tonic-gate * the address space's a_lock for a long period of time (e.g. if the NFS server 45807c478bd9Sstevel@tonic-gate * is down). Callbacks will be executed in the address space code while the 45817c478bd9Sstevel@tonic-gate * a_lock is not held. Holding the address space's a_lock causes things such 45827c478bd9Sstevel@tonic-gate * as ps and fork to hang because they are trying to acquire this lock as well. 45837c478bd9Sstevel@tonic-gate */ 45847c478bd9Sstevel@tonic-gate /* ARGSUSED */ 45857c478bd9Sstevel@tonic-gate static int 45867c478bd9Sstevel@tonic-gate nfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4587da6c28aaSamw size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4588da6c28aaSamw caller_context_t *ct) 45897c478bd9Sstevel@tonic-gate { 45907c478bd9Sstevel@tonic-gate int caller_found; 45917c478bd9Sstevel@tonic-gate int error; 45927c478bd9Sstevel@tonic-gate rnode_t *rp; 45937c478bd9Sstevel@tonic-gate nfs_delmap_args_t *dmapp; 45947c478bd9Sstevel@tonic-gate nfs_delmapcall_t *delmap_call; 45957c478bd9Sstevel@tonic-gate 45967c478bd9Sstevel@tonic-gate if (vp->v_flag & VNOMAP) 45977c478bd9Sstevel@tonic-gate return (ENOSYS); 45987c478bd9Sstevel@tonic-gate /* 45997c478bd9Sstevel@tonic-gate * A process may not change zones if it has NFS pages mmap'ed 46007c478bd9Sstevel@tonic-gate * in, so we can't legitimately get here from the wrong zone. 46017c478bd9Sstevel@tonic-gate */ 4602108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 46037c478bd9Sstevel@tonic-gate 46047c478bd9Sstevel@tonic-gate rp = VTOR(vp); 46057c478bd9Sstevel@tonic-gate 46067c478bd9Sstevel@tonic-gate /* 46077c478bd9Sstevel@tonic-gate * The way that the address space of this process deletes its mapping 46087c478bd9Sstevel@tonic-gate * of this file is via the following call chains: 46097c478bd9Sstevel@tonic-gate * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap() 46107c478bd9Sstevel@tonic-gate * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap() 46117c478bd9Sstevel@tonic-gate * 46127c478bd9Sstevel@tonic-gate * With the use of address space callbacks we are allowed to drop the 46137c478bd9Sstevel@tonic-gate * address space lock, a_lock, while executing the NFS operations that 46147c478bd9Sstevel@tonic-gate * need to go over the wire. Returning EAGAIN to the caller of this 46157c478bd9Sstevel@tonic-gate * function is what drives the execution of the callback that we add 46167c478bd9Sstevel@tonic-gate * below. The callback will be executed by the address space code 46177c478bd9Sstevel@tonic-gate * after dropping the a_lock. When the callback is finished, since 46187c478bd9Sstevel@tonic-gate * we dropped the a_lock, it must be re-acquired and segvn_unmap() 46197c478bd9Sstevel@tonic-gate * is called again on the same segment to finish the rest of the work 46207c478bd9Sstevel@tonic-gate * that needs to happen during unmapping. 46217c478bd9Sstevel@tonic-gate * 46227c478bd9Sstevel@tonic-gate * This action of calling back into the segment driver causes 46237c478bd9Sstevel@tonic-gate * nfs_delmap() to get called again, but since the callback was 46247c478bd9Sstevel@tonic-gate * already executed at this point, it already did the work and there 46257c478bd9Sstevel@tonic-gate * is nothing left for us to do. 46267c478bd9Sstevel@tonic-gate * 46277c478bd9Sstevel@tonic-gate * To Summarize: 46287c478bd9Sstevel@tonic-gate * - The first time nfs_delmap is called by the current thread is when 46297c478bd9Sstevel@tonic-gate * we add the caller associated with this delmap to the delmap caller 46307c478bd9Sstevel@tonic-gate * list, add the callback, and return EAGAIN. 46317c478bd9Sstevel@tonic-gate * - The second time in this call chain when nfs_delmap is called we 46327c478bd9Sstevel@tonic-gate * will find this caller in the delmap caller list and realize there 46337c478bd9Sstevel@tonic-gate * is no more work to do thus removing this caller from the list and 46347c478bd9Sstevel@tonic-gate * returning the error that was set in the callback execution. 46357c478bd9Sstevel@tonic-gate */ 46367c478bd9Sstevel@tonic-gate caller_found = nfs_find_and_delete_delmapcall(rp, &error); 46377c478bd9Sstevel@tonic-gate if (caller_found) { 46387c478bd9Sstevel@tonic-gate /* 46397c478bd9Sstevel@tonic-gate * 'error' is from the actual delmap operations. To avoid 46407c478bd9Sstevel@tonic-gate * hangs, we need to handle the return of EAGAIN differently 46417c478bd9Sstevel@tonic-gate * since this is what drives the callback execution. 46427c478bd9Sstevel@tonic-gate * In this case, we don't want to return EAGAIN and do the 46437c478bd9Sstevel@tonic-gate * callback execution because there are none to execute. 46447c478bd9Sstevel@tonic-gate */ 46457c478bd9Sstevel@tonic-gate if (error == EAGAIN) 46467c478bd9Sstevel@tonic-gate return (0); 46477c478bd9Sstevel@tonic-gate else 46487c478bd9Sstevel@tonic-gate return (error); 46497c478bd9Sstevel@tonic-gate } 46507c478bd9Sstevel@tonic-gate 46517c478bd9Sstevel@tonic-gate /* current caller was not in the list */ 46527c478bd9Sstevel@tonic-gate delmap_call = nfs_init_delmapcall(); 46537c478bd9Sstevel@tonic-gate 46547c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 46557c478bd9Sstevel@tonic-gate list_insert_tail(&rp->r_indelmap, delmap_call); 46567c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 46577c478bd9Sstevel@tonic-gate 46587c478bd9Sstevel@tonic-gate dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP); 46597c478bd9Sstevel@tonic-gate 46607c478bd9Sstevel@tonic-gate dmapp->vp = vp; 46617c478bd9Sstevel@tonic-gate dmapp->off = off; 46627c478bd9Sstevel@tonic-gate dmapp->addr = addr; 46637c478bd9Sstevel@tonic-gate dmapp->len = len; 46647c478bd9Sstevel@tonic-gate dmapp->prot = prot; 46657c478bd9Sstevel@tonic-gate dmapp->maxprot = maxprot; 46667c478bd9Sstevel@tonic-gate dmapp->flags = flags; 46677c478bd9Sstevel@tonic-gate dmapp->cr = cr; 46687c478bd9Sstevel@tonic-gate dmapp->caller = delmap_call; 46697c478bd9Sstevel@tonic-gate 46707c478bd9Sstevel@tonic-gate error = as_add_callback(as, nfs_delmap_callback, dmapp, 46717c478bd9Sstevel@tonic-gate AS_UNMAP_EVENT, addr, len, KM_SLEEP); 46727c478bd9Sstevel@tonic-gate 46737c478bd9Sstevel@tonic-gate return (error ? error : EAGAIN); 46747c478bd9Sstevel@tonic-gate } 46757c478bd9Sstevel@tonic-gate 46767c478bd9Sstevel@tonic-gate /* 46777c478bd9Sstevel@tonic-gate * Remove some pages from an mmap'd vnode. Just update the 46787c478bd9Sstevel@tonic-gate * count of pages. If doing close-to-open, then flush all 46797c478bd9Sstevel@tonic-gate * of the pages associated with this file. Otherwise, start 46807c478bd9Sstevel@tonic-gate * an asynchronous page flush to write out any dirty pages. 46817c478bd9Sstevel@tonic-gate * This will also associate a credential with the rnode which 46827c478bd9Sstevel@tonic-gate * can be used to write the pages. 46837c478bd9Sstevel@tonic-gate */ 46847c478bd9Sstevel@tonic-gate /* ARGSUSED */ 46857c478bd9Sstevel@tonic-gate static void 46867c478bd9Sstevel@tonic-gate nfs_delmap_callback(struct as *as, void *arg, uint_t event) 46877c478bd9Sstevel@tonic-gate { 46887c478bd9Sstevel@tonic-gate int error; 46897c478bd9Sstevel@tonic-gate rnode_t *rp; 46907c478bd9Sstevel@tonic-gate mntinfo_t *mi; 46917c478bd9Sstevel@tonic-gate nfs_delmap_args_t *dmapp = (nfs_delmap_args_t *)arg; 46927c478bd9Sstevel@tonic-gate 46937c478bd9Sstevel@tonic-gate rp = VTOR(dmapp->vp); 46947c478bd9Sstevel@tonic-gate mi = VTOMI(dmapp->vp); 46957c478bd9Sstevel@tonic-gate 46967c478bd9Sstevel@tonic-gate atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len)); 46977c478bd9Sstevel@tonic-gate ASSERT(rp->r_mapcnt >= 0); 46987c478bd9Sstevel@tonic-gate 46997c478bd9Sstevel@tonic-gate /* 47007c478bd9Sstevel@tonic-gate * Initiate a page flush if there are pages, the file system 47017c478bd9Sstevel@tonic-gate * was not mounted readonly, the segment was mapped shared, and 47027c478bd9Sstevel@tonic-gate * the pages themselves were writeable. 47037c478bd9Sstevel@tonic-gate */ 47047c478bd9Sstevel@tonic-gate if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) && 47057c478bd9Sstevel@tonic-gate dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) { 47067c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 47077c478bd9Sstevel@tonic-gate rp->r_flags |= RDIRTY; 47087c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 47097c478bd9Sstevel@tonic-gate /* 47107c478bd9Sstevel@tonic-gate * If this is a cross-zone access a sync putpage won't work, so 47117c478bd9Sstevel@tonic-gate * the best we can do is try an async putpage. That seems 47127c478bd9Sstevel@tonic-gate * better than something more draconian such as discarding the 47137c478bd9Sstevel@tonic-gate * dirty pages. 47147c478bd9Sstevel@tonic-gate */ 47157c478bd9Sstevel@tonic-gate if ((mi->mi_flags & MI_NOCTO) || 4716108322fbScarlsonj nfs_zone() != mi->mi_zone) 47177c478bd9Sstevel@tonic-gate error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len, 4718da6c28aaSamw B_ASYNC, dmapp->cr, NULL); 47197c478bd9Sstevel@tonic-gate else 47207c478bd9Sstevel@tonic-gate error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len, 4721da6c28aaSamw 0, dmapp->cr, NULL); 47227c478bd9Sstevel@tonic-gate if (!error) { 47237c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 47247c478bd9Sstevel@tonic-gate error = rp->r_error; 47257c478bd9Sstevel@tonic-gate rp->r_error = 0; 47267c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 47277c478bd9Sstevel@tonic-gate } 47287c478bd9Sstevel@tonic-gate } else 47297c478bd9Sstevel@tonic-gate error = 0; 47307c478bd9Sstevel@tonic-gate 47317c478bd9Sstevel@tonic-gate if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) 47327c478bd9Sstevel@tonic-gate (void) nfs_putpage(dmapp->vp, dmapp->off, dmapp->len, 4733da6c28aaSamw B_INVAL, dmapp->cr, NULL); 47347c478bd9Sstevel@tonic-gate 47357c478bd9Sstevel@tonic-gate dmapp->caller->error = error; 47367c478bd9Sstevel@tonic-gate (void) as_delete_callback(as, arg); 47377c478bd9Sstevel@tonic-gate kmem_free(dmapp, sizeof (nfs_delmap_args_t)); 47387c478bd9Sstevel@tonic-gate } 47397c478bd9Sstevel@tonic-gate 47407c478bd9Sstevel@tonic-gate /* ARGSUSED */ 47417c478bd9Sstevel@tonic-gate static int 4742da6c28aaSamw nfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4743da6c28aaSamw caller_context_t *ct) 47447c478bd9Sstevel@tonic-gate { 47457c478bd9Sstevel@tonic-gate int error = 0; 47467c478bd9Sstevel@tonic-gate 4747108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 47487c478bd9Sstevel@tonic-gate return (EIO); 47497c478bd9Sstevel@tonic-gate /* 47507c478bd9Sstevel@tonic-gate * This looks a little weird because it's written in a general 47517c478bd9Sstevel@tonic-gate * manner but we make little use of cases. If cntl() ever gets 47527c478bd9Sstevel@tonic-gate * widely used, the outer switch will make more sense. 47537c478bd9Sstevel@tonic-gate */ 47547c478bd9Sstevel@tonic-gate 47557c478bd9Sstevel@tonic-gate switch (cmd) { 47567c478bd9Sstevel@tonic-gate 47577c478bd9Sstevel@tonic-gate /* 47587c478bd9Sstevel@tonic-gate * Large file spec - need to base answer new query with 47597c478bd9Sstevel@tonic-gate * hardcoded constant based on the protocol. 47607c478bd9Sstevel@tonic-gate */ 47617c478bd9Sstevel@tonic-gate case _PC_FILESIZEBITS: 47627c478bd9Sstevel@tonic-gate *valp = 32; 47637c478bd9Sstevel@tonic-gate return (0); 47647c478bd9Sstevel@tonic-gate 47657c478bd9Sstevel@tonic-gate case _PC_LINK_MAX: 47667c478bd9Sstevel@tonic-gate case _PC_NAME_MAX: 47677c478bd9Sstevel@tonic-gate case _PC_PATH_MAX: 47687c478bd9Sstevel@tonic-gate case _PC_SYMLINK_MAX: 47697c478bd9Sstevel@tonic-gate case _PC_CHOWN_RESTRICTED: 47707c478bd9Sstevel@tonic-gate case _PC_NO_TRUNC: { 47717c478bd9Sstevel@tonic-gate mntinfo_t *mi; 47727c478bd9Sstevel@tonic-gate struct pathcnf *pc; 47737c478bd9Sstevel@tonic-gate 47747c478bd9Sstevel@tonic-gate if ((mi = VTOMI(vp)) == NULL || (pc = mi->mi_pathconf) == NULL) 47757c478bd9Sstevel@tonic-gate return (EINVAL); 47767c478bd9Sstevel@tonic-gate error = _PC_ISSET(cmd, pc->pc_mask); /* error or bool */ 47777c478bd9Sstevel@tonic-gate switch (cmd) { 47787c478bd9Sstevel@tonic-gate case _PC_LINK_MAX: 47797c478bd9Sstevel@tonic-gate *valp = pc->pc_link_max; 47807c478bd9Sstevel@tonic-gate break; 47817c478bd9Sstevel@tonic-gate case _PC_NAME_MAX: 47827c478bd9Sstevel@tonic-gate *valp = pc->pc_name_max; 47837c478bd9Sstevel@tonic-gate break; 47847c478bd9Sstevel@tonic-gate case _PC_PATH_MAX: 47857c478bd9Sstevel@tonic-gate case _PC_SYMLINK_MAX: 47867c478bd9Sstevel@tonic-gate *valp = pc->pc_path_max; 47877c478bd9Sstevel@tonic-gate break; 47887c478bd9Sstevel@tonic-gate case _PC_CHOWN_RESTRICTED: 47897c478bd9Sstevel@tonic-gate /* 47907c478bd9Sstevel@tonic-gate * if we got here, error is really a boolean which 47917c478bd9Sstevel@tonic-gate * indicates whether cmd is set or not. 47927c478bd9Sstevel@tonic-gate */ 47937c478bd9Sstevel@tonic-gate *valp = error ? 1 : 0; /* see above */ 47947c478bd9Sstevel@tonic-gate error = 0; 47957c478bd9Sstevel@tonic-gate break; 47967c478bd9Sstevel@tonic-gate case _PC_NO_TRUNC: 47977c478bd9Sstevel@tonic-gate /* 47987c478bd9Sstevel@tonic-gate * if we got here, error is really a boolean which 47997c478bd9Sstevel@tonic-gate * indicates whether cmd is set or not. 48007c478bd9Sstevel@tonic-gate */ 48017c478bd9Sstevel@tonic-gate *valp = error ? 1 : 0; /* see above */ 48027c478bd9Sstevel@tonic-gate error = 0; 48037c478bd9Sstevel@tonic-gate break; 48047c478bd9Sstevel@tonic-gate } 48057c478bd9Sstevel@tonic-gate return (error ? EINVAL : 0); 48067c478bd9Sstevel@tonic-gate } 48077c478bd9Sstevel@tonic-gate 48087c478bd9Sstevel@tonic-gate case _PC_XATTR_EXISTS: 48097c478bd9Sstevel@tonic-gate *valp = 0; 48107c478bd9Sstevel@tonic-gate if (vp->v_vfsp->vfs_flag & VFS_XATTR) { 48117c478bd9Sstevel@tonic-gate vnode_t *avp; 48127c478bd9Sstevel@tonic-gate rnode_t *rp; 48137c478bd9Sstevel@tonic-gate mntinfo_t *mi = VTOMI(vp); 48147c478bd9Sstevel@tonic-gate 48157c478bd9Sstevel@tonic-gate if (!(mi->mi_flags & MI_EXTATTR)) 48167c478bd9Sstevel@tonic-gate return (0); 48177c478bd9Sstevel@tonic-gate 48187c478bd9Sstevel@tonic-gate rp = VTOR(vp); 48197c478bd9Sstevel@tonic-gate if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, 48207c478bd9Sstevel@tonic-gate INTR(vp))) 48217c478bd9Sstevel@tonic-gate return (EINTR); 48227c478bd9Sstevel@tonic-gate 48237c478bd9Sstevel@tonic-gate error = nfslookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr); 48247c478bd9Sstevel@tonic-gate if (error || avp == NULL) 48257c478bd9Sstevel@tonic-gate error = acl_getxattrdir2(vp, &avp, 0, cr, 0); 48267c478bd9Sstevel@tonic-gate 48277c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_rwlock); 48287c478bd9Sstevel@tonic-gate 48297c478bd9Sstevel@tonic-gate if (error == 0 && avp != NULL) { 4830*93aeed83Smarks error = do_xattr_exists_check(avp, valp, cr); 48317c478bd9Sstevel@tonic-gate VN_RELE(avp); 48327c478bd9Sstevel@tonic-gate } 48337c478bd9Sstevel@tonic-gate } 48347c478bd9Sstevel@tonic-gate return (error ? EINVAL : 0); 48357c478bd9Sstevel@tonic-gate 48367c478bd9Sstevel@tonic-gate case _PC_ACL_ENABLED: 48377c478bd9Sstevel@tonic-gate *valp = _ACL_ACLENT_ENABLED; 48387c478bd9Sstevel@tonic-gate return (0); 48397c478bd9Sstevel@tonic-gate 48407c478bd9Sstevel@tonic-gate default: 48417c478bd9Sstevel@tonic-gate return (EINVAL); 48427c478bd9Sstevel@tonic-gate } 48437c478bd9Sstevel@tonic-gate } 48447c478bd9Sstevel@tonic-gate 48457c478bd9Sstevel@tonic-gate /* 48467c478bd9Sstevel@tonic-gate * Called by async thread to do synchronous pageio. Do the i/o, wait 48477c478bd9Sstevel@tonic-gate * for it to complete, and cleanup the page list when done. 48487c478bd9Sstevel@tonic-gate */ 48497c478bd9Sstevel@tonic-gate static int 48507c478bd9Sstevel@tonic-gate nfs_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 48517c478bd9Sstevel@tonic-gate int flags, cred_t *cr) 48527c478bd9Sstevel@tonic-gate { 48537c478bd9Sstevel@tonic-gate int error; 48547c478bd9Sstevel@tonic-gate 4855108322fbScarlsonj ASSERT(nfs_zone() == VTOMI(vp)->mi_zone); 48567c478bd9Sstevel@tonic-gate error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 48577c478bd9Sstevel@tonic-gate if (flags & B_READ) 48587c478bd9Sstevel@tonic-gate pvn_read_done(pp, (error ? B_ERROR : 0) | flags); 48597c478bd9Sstevel@tonic-gate else 48607c478bd9Sstevel@tonic-gate pvn_write_done(pp, (error ? B_ERROR : 0) | flags); 48617c478bd9Sstevel@tonic-gate return (error); 48627c478bd9Sstevel@tonic-gate } 48637c478bd9Sstevel@tonic-gate 4864da6c28aaSamw /* ARGSUSED */ 48657c478bd9Sstevel@tonic-gate static int 48667c478bd9Sstevel@tonic-gate nfs_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 4867da6c28aaSamw int flags, cred_t *cr, caller_context_t *ct) 48687c478bd9Sstevel@tonic-gate { 48697c478bd9Sstevel@tonic-gate int error; 48707c478bd9Sstevel@tonic-gate rnode_t *rp; 48717c478bd9Sstevel@tonic-gate 48727c478bd9Sstevel@tonic-gate if (pp == NULL) 48737c478bd9Sstevel@tonic-gate return (EINVAL); 48747c478bd9Sstevel@tonic-gate 48757c478bd9Sstevel@tonic-gate if (io_off > MAXOFF32_T) 48767c478bd9Sstevel@tonic-gate return (EFBIG); 4877108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 48787c478bd9Sstevel@tonic-gate return (EIO); 48797c478bd9Sstevel@tonic-gate rp = VTOR(vp); 48807c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 48817c478bd9Sstevel@tonic-gate rp->r_count++; 48827c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 48837c478bd9Sstevel@tonic-gate 48847c478bd9Sstevel@tonic-gate if (flags & B_ASYNC) { 48857c478bd9Sstevel@tonic-gate error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr, 48867c478bd9Sstevel@tonic-gate nfs_sync_pageio); 48877c478bd9Sstevel@tonic-gate } else 48887c478bd9Sstevel@tonic-gate error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr); 48897c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 48907c478bd9Sstevel@tonic-gate rp->r_count--; 48917c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 48927c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 48937c478bd9Sstevel@tonic-gate return (error); 48947c478bd9Sstevel@tonic-gate } 48957c478bd9Sstevel@tonic-gate 4896da6c28aaSamw /* ARGSUSED */ 48977c478bd9Sstevel@tonic-gate static int 4898da6c28aaSamw nfs_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr, 4899da6c28aaSamw caller_context_t *ct) 49007c478bd9Sstevel@tonic-gate { 49017c478bd9Sstevel@tonic-gate int error; 49027c478bd9Sstevel@tonic-gate mntinfo_t *mi; 49037c478bd9Sstevel@tonic-gate 49047c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 49057c478bd9Sstevel@tonic-gate 4906108322fbScarlsonj if (nfs_zone() != mi->mi_zone) 49077c478bd9Sstevel@tonic-gate return (EIO); 49087c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) { 49097c478bd9Sstevel@tonic-gate error = acl_setacl2(vp, vsecattr, flag, cr); 49107c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) 49117c478bd9Sstevel@tonic-gate return (error); 49127c478bd9Sstevel@tonic-gate } 49137c478bd9Sstevel@tonic-gate 49147c478bd9Sstevel@tonic-gate return (ENOSYS); 49157c478bd9Sstevel@tonic-gate } 49167c478bd9Sstevel@tonic-gate 4917da6c28aaSamw /* ARGSUSED */ 49187c478bd9Sstevel@tonic-gate static int 4919da6c28aaSamw nfs_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr, 4920da6c28aaSamw caller_context_t *ct) 49217c478bd9Sstevel@tonic-gate { 49227c478bd9Sstevel@tonic-gate int error; 49237c478bd9Sstevel@tonic-gate mntinfo_t *mi; 49247c478bd9Sstevel@tonic-gate 49257c478bd9Sstevel@tonic-gate mi = VTOMI(vp); 49267c478bd9Sstevel@tonic-gate 4927108322fbScarlsonj if (nfs_zone() != mi->mi_zone) 49287c478bd9Sstevel@tonic-gate return (EIO); 49297c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) { 49307c478bd9Sstevel@tonic-gate error = acl_getacl2(vp, vsecattr, flag, cr); 49317c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI_ACL) 49327c478bd9Sstevel@tonic-gate return (error); 49337c478bd9Sstevel@tonic-gate } 49347c478bd9Sstevel@tonic-gate 4935da6c28aaSamw return (fs_fab_acl(vp, vsecattr, flag, cr, ct)); 49367c478bd9Sstevel@tonic-gate } 49377c478bd9Sstevel@tonic-gate 4938da6c28aaSamw /* ARGSUSED */ 49397c478bd9Sstevel@tonic-gate static int 4940da6c28aaSamw nfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr, 4941da6c28aaSamw caller_context_t *ct) 49427c478bd9Sstevel@tonic-gate { 49437c478bd9Sstevel@tonic-gate int error; 49447c478bd9Sstevel@tonic-gate struct shrlock nshr; 49457c478bd9Sstevel@tonic-gate struct nfs_owner nfs_owner; 49467c478bd9Sstevel@tonic-gate netobj lm_fh; 49477c478bd9Sstevel@tonic-gate 4948108322fbScarlsonj if (nfs_zone() != VTOMI(vp)->mi_zone) 49497c478bd9Sstevel@tonic-gate return (EIO); 49507c478bd9Sstevel@tonic-gate 49517c478bd9Sstevel@tonic-gate /* 49527c478bd9Sstevel@tonic-gate * check for valid cmd parameter 49537c478bd9Sstevel@tonic-gate */ 49547c478bd9Sstevel@tonic-gate if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS) 49557c478bd9Sstevel@tonic-gate return (EINVAL); 49567c478bd9Sstevel@tonic-gate 49577c478bd9Sstevel@tonic-gate /* 49587c478bd9Sstevel@tonic-gate * Check access permissions 49597c478bd9Sstevel@tonic-gate */ 49607c478bd9Sstevel@tonic-gate if (cmd == F_SHARE && 49617c478bd9Sstevel@tonic-gate (((shr->s_access & F_RDACC) && !(flag & FREAD)) || 49627c478bd9Sstevel@tonic-gate ((shr->s_access & F_WRACC) && !(flag & FWRITE)))) 49637c478bd9Sstevel@tonic-gate return (EBADF); 49647c478bd9Sstevel@tonic-gate 49657c478bd9Sstevel@tonic-gate /* 49667c478bd9Sstevel@tonic-gate * If the filesystem is mounted using local locking, pass the 49677c478bd9Sstevel@tonic-gate * request off to the local share code. 49687c478bd9Sstevel@tonic-gate */ 49697c478bd9Sstevel@tonic-gate if (VTOMI(vp)->mi_flags & MI_LLOCK) 4970da6c28aaSamw return (fs_shrlock(vp, cmd, shr, flag, cr, ct)); 49717c478bd9Sstevel@tonic-gate 49727c478bd9Sstevel@tonic-gate switch (cmd) { 49737c478bd9Sstevel@tonic-gate case F_SHARE: 49747c478bd9Sstevel@tonic-gate case F_UNSHARE: 49757c478bd9Sstevel@tonic-gate lm_fh.n_len = sizeof (fhandle_t); 49767c478bd9Sstevel@tonic-gate lm_fh.n_bytes = (char *)VTOFH(vp); 49777c478bd9Sstevel@tonic-gate 49787c478bd9Sstevel@tonic-gate /* 49797c478bd9Sstevel@tonic-gate * If passed an owner that is too large to fit in an 49807c478bd9Sstevel@tonic-gate * nfs_owner it is likely a recursive call from the 49817c478bd9Sstevel@tonic-gate * lock manager client and pass it straight through. If 49827c478bd9Sstevel@tonic-gate * it is not a nfs_owner then simply return an error. 49837c478bd9Sstevel@tonic-gate */ 49847c478bd9Sstevel@tonic-gate if (shr->s_own_len > sizeof (nfs_owner.lowner)) { 49857c478bd9Sstevel@tonic-gate if (((struct nfs_owner *)shr->s_owner)->magic != 49867c478bd9Sstevel@tonic-gate NFS_OWNER_MAGIC) 49877c478bd9Sstevel@tonic-gate return (EINVAL); 49887c478bd9Sstevel@tonic-gate 49897c478bd9Sstevel@tonic-gate if (error = lm_shrlock(vp, cmd, shr, flag, &lm_fh)) { 49907c478bd9Sstevel@tonic-gate error = set_errno(error); 49917c478bd9Sstevel@tonic-gate } 49927c478bd9Sstevel@tonic-gate return (error); 49937c478bd9Sstevel@tonic-gate } 49947c478bd9Sstevel@tonic-gate /* 49957c478bd9Sstevel@tonic-gate * Remote share reservations owner is a combination of 49967c478bd9Sstevel@tonic-gate * a magic number, hostname, and the local owner 49977c478bd9Sstevel@tonic-gate */ 49987c478bd9Sstevel@tonic-gate bzero(&nfs_owner, sizeof (nfs_owner)); 49997c478bd9Sstevel@tonic-gate nfs_owner.magic = NFS_OWNER_MAGIC; 50007c478bd9Sstevel@tonic-gate (void) strncpy(nfs_owner.hname, uts_nodename(), 50017c478bd9Sstevel@tonic-gate sizeof (nfs_owner.hname)); 50027c478bd9Sstevel@tonic-gate bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len); 50037c478bd9Sstevel@tonic-gate nshr.s_access = shr->s_access; 50047c478bd9Sstevel@tonic-gate nshr.s_deny = shr->s_deny; 50057c478bd9Sstevel@tonic-gate nshr.s_sysid = 0; 50067c478bd9Sstevel@tonic-gate nshr.s_pid = ttoproc(curthread)->p_pid; 50077c478bd9Sstevel@tonic-gate nshr.s_own_len = sizeof (nfs_owner); 50087c478bd9Sstevel@tonic-gate nshr.s_owner = (caddr_t)&nfs_owner; 50097c478bd9Sstevel@tonic-gate 50107c478bd9Sstevel@tonic-gate if (error = lm_shrlock(vp, cmd, &nshr, flag, &lm_fh)) { 50117c478bd9Sstevel@tonic-gate error = set_errno(error); 50127c478bd9Sstevel@tonic-gate } 50137c478bd9Sstevel@tonic-gate 50147c478bd9Sstevel@tonic-gate break; 50157c478bd9Sstevel@tonic-gate 50167c478bd9Sstevel@tonic-gate case F_HASREMOTELOCKS: 50177c478bd9Sstevel@tonic-gate /* 50187c478bd9Sstevel@tonic-gate * NFS client can't store remote locks itself 50197c478bd9Sstevel@tonic-gate */ 50207c478bd9Sstevel@tonic-gate shr->s_access = 0; 50217c478bd9Sstevel@tonic-gate error = 0; 50227c478bd9Sstevel@tonic-gate break; 50237c478bd9Sstevel@tonic-gate 50247c478bd9Sstevel@tonic-gate default: 50257c478bd9Sstevel@tonic-gate error = EINVAL; 50267c478bd9Sstevel@tonic-gate break; 50277c478bd9Sstevel@tonic-gate } 50287c478bd9Sstevel@tonic-gate 50297c478bd9Sstevel@tonic-gate return (error); 50307c478bd9Sstevel@tonic-gate } 5031