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 550a83466Sjwahlig * Common Development and Distribution License (the "License"). 650a83466Sjwahlig * 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 /* 22b9238976Sth * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate 267c478bd9Sstevel@tonic-gate /* 277c478bd9Sstevel@tonic-gate * Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T. 287c478bd9Sstevel@tonic-gate * All Rights Reserved 297c478bd9Sstevel@tonic-gate */ 307c478bd9Sstevel@tonic-gate 317c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 327c478bd9Sstevel@tonic-gate 337c478bd9Sstevel@tonic-gate #include <sys/param.h> 347c478bd9Sstevel@tonic-gate #include <sys/types.h> 357c478bd9Sstevel@tonic-gate #include <sys/systm.h> 367c478bd9Sstevel@tonic-gate #include <sys/thread.h> 377c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 387c478bd9Sstevel@tonic-gate #include <sys/time.h> 397c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 407c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 417c478bd9Sstevel@tonic-gate #include <sys/errno.h> 427c478bd9Sstevel@tonic-gate #include <sys/buf.h> 437c478bd9Sstevel@tonic-gate #include <sys/stat.h> 447c478bd9Sstevel@tonic-gate #include <sys/cred.h> 457c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 467c478bd9Sstevel@tonic-gate #include <sys/debug.h> 477c478bd9Sstevel@tonic-gate #include <sys/dnlc.h> 487c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 497c478bd9Sstevel@tonic-gate #include <sys/flock.h> 507c478bd9Sstevel@tonic-gate #include <sys/share.h> 517c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 527c478bd9Sstevel@tonic-gate #include <sys/tiuser.h> 537c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 547c478bd9Sstevel@tonic-gate #include <sys/callb.h> 557c478bd9Sstevel@tonic-gate #include <sys/acl.h> 567c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 577c478bd9Sstevel@tonic-gate #include <sys/signal.h> 587c478bd9Sstevel@tonic-gate #include <sys/disp.h> 597c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 607c478bd9Sstevel@tonic-gate #include <sys/list.h> 617c478bd9Sstevel@tonic-gate #include <sys/sdt.h> 627c478bd9Sstevel@tonic-gate 637c478bd9Sstevel@tonic-gate #include <rpc/types.h> 647c478bd9Sstevel@tonic-gate #include <rpc/xdr.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/nfs_acl.h> 717c478bd9Sstevel@tonic-gate 727c478bd9Sstevel@tonic-gate #include <nfs/nfs4.h> 737c478bd9Sstevel@tonic-gate #include <nfs/rnode4.h> 747c478bd9Sstevel@tonic-gate #include <nfs/nfs4_clnt.h> 757c478bd9Sstevel@tonic-gate 767c478bd9Sstevel@tonic-gate #include <vm/hat.h> 777c478bd9Sstevel@tonic-gate #include <vm/as.h> 787c478bd9Sstevel@tonic-gate #include <vm/page.h> 797c478bd9Sstevel@tonic-gate #include <vm/pvn.h> 807c478bd9Sstevel@tonic-gate #include <vm/seg.h> 817c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 827c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 837c478bd9Sstevel@tonic-gate 847c478bd9Sstevel@tonic-gate #include <sys/ddi.h> 857c478bd9Sstevel@tonic-gate 867c478bd9Sstevel@tonic-gate /* 877c478bd9Sstevel@tonic-gate * Arguments to page-flush thread. 887c478bd9Sstevel@tonic-gate */ 897c478bd9Sstevel@tonic-gate typedef struct { 907c478bd9Sstevel@tonic-gate vnode_t *vp; 917c478bd9Sstevel@tonic-gate cred_t *cr; 927c478bd9Sstevel@tonic-gate } pgflush_t; 937c478bd9Sstevel@tonic-gate 947c478bd9Sstevel@tonic-gate #ifdef DEBUG 957c478bd9Sstevel@tonic-gate int nfs4_client_lease_debug; 967c478bd9Sstevel@tonic-gate int nfs4_sharedfh_debug; 977c478bd9Sstevel@tonic-gate int nfs4_fname_debug; 987c478bd9Sstevel@tonic-gate 997c478bd9Sstevel@tonic-gate /* temporary: panic if v_type is inconsistent with r_attr va_type */ 1007c478bd9Sstevel@tonic-gate int nfs4_vtype_debug; 1017c478bd9Sstevel@tonic-gate 1027c478bd9Sstevel@tonic-gate uint_t nfs4_tsd_key; 1037c478bd9Sstevel@tonic-gate #endif 1047c478bd9Sstevel@tonic-gate 1057c478bd9Sstevel@tonic-gate static time_t nfs4_client_resumed = 0; 1067c478bd9Sstevel@tonic-gate static callb_id_t cid = 0; 1077c478bd9Sstevel@tonic-gate 1087c478bd9Sstevel@tonic-gate static int nfs4renew(nfs4_server_t *); 1097c478bd9Sstevel@tonic-gate static void nfs4_attrcache_va(vnode_t *, nfs4_ga_res_t *, int); 1107c478bd9Sstevel@tonic-gate static void nfs4_pgflush_thread(pgflush_t *); 1117c478bd9Sstevel@tonic-gate static void flush_pages(vnode_t *, cred_t *); 1127c478bd9Sstevel@tonic-gate 1137c478bd9Sstevel@tonic-gate static boolean_t nfs4_client_cpr_callb(void *, int); 1147c478bd9Sstevel@tonic-gate 1157c478bd9Sstevel@tonic-gate struct mi4_globals { 1167c478bd9Sstevel@tonic-gate kmutex_t mig_lock; /* lock protecting mig_list */ 1177c478bd9Sstevel@tonic-gate list_t mig_list; /* list of NFS v4 mounts in zone */ 1187c478bd9Sstevel@tonic-gate boolean_t mig_destructor_called; 1197c478bd9Sstevel@tonic-gate }; 1207c478bd9Sstevel@tonic-gate 1217c478bd9Sstevel@tonic-gate static zone_key_t mi4_list_key; 1227c478bd9Sstevel@tonic-gate 1237c478bd9Sstevel@tonic-gate /* 1247c478bd9Sstevel@tonic-gate * Attributes caching: 1257c478bd9Sstevel@tonic-gate * 1267c478bd9Sstevel@tonic-gate * Attributes are cached in the rnode in struct vattr form. 1277c478bd9Sstevel@tonic-gate * There is a time associated with the cached attributes (r_time_attr_inval) 1287c478bd9Sstevel@tonic-gate * which tells whether the attributes are valid. The time is initialized 1297c478bd9Sstevel@tonic-gate * to the difference between current time and the modify time of the vnode 1307c478bd9Sstevel@tonic-gate * when new attributes are cached. This allows the attributes for 1317c478bd9Sstevel@tonic-gate * files that have changed recently to be timed out sooner than for files 1327c478bd9Sstevel@tonic-gate * that have not changed for a long time. There are minimum and maximum 1337c478bd9Sstevel@tonic-gate * timeout values that can be set per mount point. 1347c478bd9Sstevel@tonic-gate */ 1357c478bd9Sstevel@tonic-gate 1367c478bd9Sstevel@tonic-gate /* 1377c478bd9Sstevel@tonic-gate * If a cache purge is in progress, wait for it to finish. 1387c478bd9Sstevel@tonic-gate * 1397c478bd9Sstevel@tonic-gate * The current thread must not be in the middle of an 1407c478bd9Sstevel@tonic-gate * nfs4_start_op/nfs4_end_op region. Otherwise, there could be a deadlock 1417c478bd9Sstevel@tonic-gate * between this thread, a recovery thread, and the page flush thread. 1427c478bd9Sstevel@tonic-gate */ 1437c478bd9Sstevel@tonic-gate int 1447c478bd9Sstevel@tonic-gate nfs4_waitfor_purge_complete(vnode_t *vp) 1457c478bd9Sstevel@tonic-gate { 1467c478bd9Sstevel@tonic-gate rnode4_t *rp; 1477c478bd9Sstevel@tonic-gate k_sigset_t smask; 1487c478bd9Sstevel@tonic-gate 1497c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 1507c478bd9Sstevel@tonic-gate if ((rp->r_serial != NULL && rp->r_serial != curthread) || 1517c478bd9Sstevel@tonic-gate ((rp->r_flags & R4PGFLUSH) && rp->r_pgflush != curthread)) { 1527c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 1537c478bd9Sstevel@tonic-gate sigintr(&smask, VTOMI4(vp)->mi_flags & MI4_INT); 1547c478bd9Sstevel@tonic-gate while ((rp->r_serial != NULL && rp->r_serial != curthread) || 1557c478bd9Sstevel@tonic-gate ((rp->r_flags & R4PGFLUSH) && 1567c478bd9Sstevel@tonic-gate rp->r_pgflush != curthread)) { 1577c478bd9Sstevel@tonic-gate if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) { 1587c478bd9Sstevel@tonic-gate sigunintr(&smask); 1597c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 1607c478bd9Sstevel@tonic-gate return (EINTR); 1617c478bd9Sstevel@tonic-gate } 1627c478bd9Sstevel@tonic-gate } 1637c478bd9Sstevel@tonic-gate sigunintr(&smask); 1647c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 1657c478bd9Sstevel@tonic-gate } 1667c478bd9Sstevel@tonic-gate return (0); 1677c478bd9Sstevel@tonic-gate } 1687c478bd9Sstevel@tonic-gate 1697c478bd9Sstevel@tonic-gate /* 1707c478bd9Sstevel@tonic-gate * Validate caches by checking cached attributes. If they have timed out, 1717c478bd9Sstevel@tonic-gate * then get new attributes from the server. As a side effect, cache 1727c478bd9Sstevel@tonic-gate * invalidation is done if the attributes have changed. 1737c478bd9Sstevel@tonic-gate * 1747c478bd9Sstevel@tonic-gate * If the attributes have not timed out and if there is a cache 1757c478bd9Sstevel@tonic-gate * invalidation being done by some other thread, then wait until that 1767c478bd9Sstevel@tonic-gate * thread has completed the cache invalidation. 1777c478bd9Sstevel@tonic-gate */ 1787c478bd9Sstevel@tonic-gate int 1797c478bd9Sstevel@tonic-gate nfs4_validate_caches(vnode_t *vp, cred_t *cr) 1807c478bd9Sstevel@tonic-gate { 1817c478bd9Sstevel@tonic-gate int error; 1827c478bd9Sstevel@tonic-gate nfs4_ga_res_t gar; 1837c478bd9Sstevel@tonic-gate 1847c478bd9Sstevel@tonic-gate if (ATTRCACHE4_VALID(vp)) { 1857c478bd9Sstevel@tonic-gate error = nfs4_waitfor_purge_complete(vp); 1867c478bd9Sstevel@tonic-gate if (error) 1877c478bd9Sstevel@tonic-gate return (error); 1887c478bd9Sstevel@tonic-gate return (0); 1897c478bd9Sstevel@tonic-gate } 1907c478bd9Sstevel@tonic-gate 1917c478bd9Sstevel@tonic-gate gar.n4g_va.va_mask = AT_ALL; 1927c478bd9Sstevel@tonic-gate return (nfs4_getattr_otw(vp, &gar, cr, 0)); 1937c478bd9Sstevel@tonic-gate } 1947c478bd9Sstevel@tonic-gate 1957c478bd9Sstevel@tonic-gate /* 1967c478bd9Sstevel@tonic-gate * Fill in attribute from the cache. 1977c478bd9Sstevel@tonic-gate * If valid, then return 0 to indicate that no error occurred, 1987c478bd9Sstevel@tonic-gate * otherwise return 1 to indicate that an error occurred. 1997c478bd9Sstevel@tonic-gate */ 2007c478bd9Sstevel@tonic-gate static int 2017c478bd9Sstevel@tonic-gate nfs4_getattr_cache(vnode_t *vp, struct vattr *vap) 2027c478bd9Sstevel@tonic-gate { 2037c478bd9Sstevel@tonic-gate rnode4_t *rp; 2047c478bd9Sstevel@tonic-gate 2057c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 2067c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 2077c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statev4_lock); 2087c478bd9Sstevel@tonic-gate if (ATTRCACHE4_VALID(vp)) { 2097c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statev4_lock); 2107c478bd9Sstevel@tonic-gate /* 2117c478bd9Sstevel@tonic-gate * Cached attributes are valid 2127c478bd9Sstevel@tonic-gate */ 2137c478bd9Sstevel@tonic-gate *vap = rp->r_attr; 2147c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 2157c478bd9Sstevel@tonic-gate return (0); 2167c478bd9Sstevel@tonic-gate } 2177c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statev4_lock); 2187c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 2197c478bd9Sstevel@tonic-gate return (1); 2207c478bd9Sstevel@tonic-gate } 2217c478bd9Sstevel@tonic-gate 2227c478bd9Sstevel@tonic-gate 2237c478bd9Sstevel@tonic-gate /* 2247c478bd9Sstevel@tonic-gate * If returned error is ESTALE flush all caches. The nfs4_purge_caches() 2257c478bd9Sstevel@tonic-gate * call is synchronous because all the pages were invalidated by the 2267c478bd9Sstevel@tonic-gate * nfs4_invalidate_pages() call. 2277c478bd9Sstevel@tonic-gate */ 2287c478bd9Sstevel@tonic-gate void 2297c478bd9Sstevel@tonic-gate nfs4_purge_stale_fh(int errno, vnode_t *vp, cred_t *cr) 2307c478bd9Sstevel@tonic-gate { 2317c478bd9Sstevel@tonic-gate struct rnode4 *rp = VTOR4(vp); 2327c478bd9Sstevel@tonic-gate 2337c478bd9Sstevel@tonic-gate /* Ensure that the ..._end_op() call has been done */ 2347c478bd9Sstevel@tonic-gate ASSERT(tsd_get(nfs4_tsd_key) == NULL); 2357c478bd9Sstevel@tonic-gate 2367c478bd9Sstevel@tonic-gate if (errno != ESTALE) 2377c478bd9Sstevel@tonic-gate return; 2387c478bd9Sstevel@tonic-gate 2397c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 2407c478bd9Sstevel@tonic-gate rp->r_flags |= R4STALE; 2417c478bd9Sstevel@tonic-gate if (!rp->r_error) 2427c478bd9Sstevel@tonic-gate rp->r_error = errno; 2437c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 2447c478bd9Sstevel@tonic-gate if (nfs4_has_pages(vp)) 2457c478bd9Sstevel@tonic-gate nfs4_invalidate_pages(vp, (u_offset_t)0, cr); 2467c478bd9Sstevel@tonic-gate nfs4_purge_caches(vp, NFS4_PURGE_DNLC, cr, FALSE); 2477c478bd9Sstevel@tonic-gate } 2487c478bd9Sstevel@tonic-gate 2497c478bd9Sstevel@tonic-gate /* 2507c478bd9Sstevel@tonic-gate * Purge all of the various NFS `data' caches. If "asyncpg" is TRUE, the 2517c478bd9Sstevel@tonic-gate * page purge is done asynchronously. 2527c478bd9Sstevel@tonic-gate */ 2537c478bd9Sstevel@tonic-gate void 2547c478bd9Sstevel@tonic-gate nfs4_purge_caches(vnode_t *vp, int purge_dnlc, cred_t *cr, int asyncpg) 2557c478bd9Sstevel@tonic-gate { 2567c478bd9Sstevel@tonic-gate rnode4_t *rp; 2577c478bd9Sstevel@tonic-gate char *contents; 2587c478bd9Sstevel@tonic-gate vnode_t *xattr; 2597c478bd9Sstevel@tonic-gate int size; 2607c478bd9Sstevel@tonic-gate int pgflush; /* are we the page flush thread? */ 2617c478bd9Sstevel@tonic-gate 2627c478bd9Sstevel@tonic-gate /* 2637c478bd9Sstevel@tonic-gate * Purge the DNLC for any entries which refer to this file. 2647c478bd9Sstevel@tonic-gate */ 2657c478bd9Sstevel@tonic-gate if (vp->v_count > 1 && 2667c478bd9Sstevel@tonic-gate (vp->v_type == VDIR || purge_dnlc == NFS4_PURGE_DNLC)) 2677c478bd9Sstevel@tonic-gate dnlc_purge_vp(vp); 2687c478bd9Sstevel@tonic-gate 2697c478bd9Sstevel@tonic-gate /* 2707c478bd9Sstevel@tonic-gate * Clear any readdir state bits and purge the readlink response cache. 2717c478bd9Sstevel@tonic-gate */ 2727c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 2737c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 2747c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4LOOKUP; 2757c478bd9Sstevel@tonic-gate contents = rp->r_symlink.contents; 2767c478bd9Sstevel@tonic-gate size = rp->r_symlink.size; 2777c478bd9Sstevel@tonic-gate rp->r_symlink.contents = NULL; 2787c478bd9Sstevel@tonic-gate 2797c478bd9Sstevel@tonic-gate xattr = rp->r_xattr_dir; 2807c478bd9Sstevel@tonic-gate rp->r_xattr_dir = NULL; 2817c478bd9Sstevel@tonic-gate 2827c478bd9Sstevel@tonic-gate /* 2837c478bd9Sstevel@tonic-gate * Purge pathconf cache too. 2847c478bd9Sstevel@tonic-gate */ 2857c478bd9Sstevel@tonic-gate rp->r_pathconf.pc4_xattr_valid = 0; 2867c478bd9Sstevel@tonic-gate rp->r_pathconf.pc4_cache_valid = 0; 2877c478bd9Sstevel@tonic-gate 2887c478bd9Sstevel@tonic-gate pgflush = (curthread == rp->r_pgflush); 2897c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 2907c478bd9Sstevel@tonic-gate 2917c478bd9Sstevel@tonic-gate if (contents != NULL) { 2927c478bd9Sstevel@tonic-gate 2937c478bd9Sstevel@tonic-gate kmem_free((void *)contents, size); 2947c478bd9Sstevel@tonic-gate } 2957c478bd9Sstevel@tonic-gate 2967c478bd9Sstevel@tonic-gate if (xattr != NULL) 2977c478bd9Sstevel@tonic-gate VN_RELE(xattr); 2987c478bd9Sstevel@tonic-gate 2997c478bd9Sstevel@tonic-gate /* 3007c478bd9Sstevel@tonic-gate * Flush the page cache. If the current thread is the page flush 3017c478bd9Sstevel@tonic-gate * thread, don't initiate a new page flush. There's no need for 3027c478bd9Sstevel@tonic-gate * it, and doing it correctly is hard. 3037c478bd9Sstevel@tonic-gate */ 3047c478bd9Sstevel@tonic-gate if (nfs4_has_pages(vp) && !pgflush) { 3057c478bd9Sstevel@tonic-gate if (!asyncpg) { 3067c478bd9Sstevel@tonic-gate (void) nfs4_waitfor_purge_complete(vp); 3077c478bd9Sstevel@tonic-gate flush_pages(vp, cr); 3087c478bd9Sstevel@tonic-gate } else { 3097c478bd9Sstevel@tonic-gate pgflush_t *args; 3107c478bd9Sstevel@tonic-gate 3117c478bd9Sstevel@tonic-gate /* 3127c478bd9Sstevel@tonic-gate * We don't hold r_statelock while creating the 3137c478bd9Sstevel@tonic-gate * thread, in case the call blocks. So we use a 3147c478bd9Sstevel@tonic-gate * flag to indicate that a page flush thread is 3157c478bd9Sstevel@tonic-gate * active. 3167c478bd9Sstevel@tonic-gate */ 3177c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 3187c478bd9Sstevel@tonic-gate if (rp->r_flags & R4PGFLUSH) { 3197c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3207c478bd9Sstevel@tonic-gate } else { 3217c478bd9Sstevel@tonic-gate rp->r_flags |= R4PGFLUSH; 3227c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3237c478bd9Sstevel@tonic-gate 3247c478bd9Sstevel@tonic-gate args = kmem_alloc(sizeof (pgflush_t), 325b9238976Sth KM_SLEEP); 3267c478bd9Sstevel@tonic-gate args->vp = vp; 3277c478bd9Sstevel@tonic-gate VN_HOLD(args->vp); 3287c478bd9Sstevel@tonic-gate args->cr = cr; 3297c478bd9Sstevel@tonic-gate crhold(args->cr); 3307c478bd9Sstevel@tonic-gate (void) zthread_create(NULL, 0, 331b9238976Sth nfs4_pgflush_thread, args, 0, 332b9238976Sth minclsyspri); 3337c478bd9Sstevel@tonic-gate } 3347c478bd9Sstevel@tonic-gate } 3357c478bd9Sstevel@tonic-gate } 3367c478bd9Sstevel@tonic-gate 3377c478bd9Sstevel@tonic-gate /* 3387c478bd9Sstevel@tonic-gate * Flush the readdir response cache. 3397c478bd9Sstevel@tonic-gate */ 3407c478bd9Sstevel@tonic-gate nfs4_purge_rddir_cache(vp); 3417c478bd9Sstevel@tonic-gate } 3427c478bd9Sstevel@tonic-gate 3437c478bd9Sstevel@tonic-gate /* 3447c478bd9Sstevel@tonic-gate * Invalidate all pages for the given file, after writing back the dirty 3457c478bd9Sstevel@tonic-gate * ones. 3467c478bd9Sstevel@tonic-gate */ 3477c478bd9Sstevel@tonic-gate 3487c478bd9Sstevel@tonic-gate static void 3497c478bd9Sstevel@tonic-gate flush_pages(vnode_t *vp, cred_t *cr) 3507c478bd9Sstevel@tonic-gate { 3517c478bd9Sstevel@tonic-gate int error; 3527c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(vp); 3537c478bd9Sstevel@tonic-gate 354da6c28aaSamw error = VOP_PUTPAGE(vp, (u_offset_t)0, 0, B_INVAL, cr, NULL); 3557c478bd9Sstevel@tonic-gate if (error == ENOSPC || error == EDQUOT) { 3567c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 3577c478bd9Sstevel@tonic-gate if (!rp->r_error) 3587c478bd9Sstevel@tonic-gate rp->r_error = error; 3597c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3607c478bd9Sstevel@tonic-gate } 3617c478bd9Sstevel@tonic-gate } 3627c478bd9Sstevel@tonic-gate 3637c478bd9Sstevel@tonic-gate /* 3647c478bd9Sstevel@tonic-gate * Page flush thread. 3657c478bd9Sstevel@tonic-gate */ 3667c478bd9Sstevel@tonic-gate 3677c478bd9Sstevel@tonic-gate static void 3687c478bd9Sstevel@tonic-gate nfs4_pgflush_thread(pgflush_t *args) 3697c478bd9Sstevel@tonic-gate { 3707c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(args->vp); 3717c478bd9Sstevel@tonic-gate 3727c478bd9Sstevel@tonic-gate /* remember which thread we are, so we don't deadlock ourselves */ 3737c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 3747c478bd9Sstevel@tonic-gate ASSERT(rp->r_pgflush == NULL); 3757c478bd9Sstevel@tonic-gate rp->r_pgflush = curthread; 3767c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3777c478bd9Sstevel@tonic-gate 3787c478bd9Sstevel@tonic-gate flush_pages(args->vp, args->cr); 3797c478bd9Sstevel@tonic-gate 3807c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 3817c478bd9Sstevel@tonic-gate rp->r_pgflush = NULL; 3827c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4PGFLUSH; 3837c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 3847c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 3857c478bd9Sstevel@tonic-gate 3867c478bd9Sstevel@tonic-gate VN_RELE(args->vp); 3877c478bd9Sstevel@tonic-gate crfree(args->cr); 3887c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (pgflush_t)); 3897c478bd9Sstevel@tonic-gate zthread_exit(); 3907c478bd9Sstevel@tonic-gate } 3917c478bd9Sstevel@tonic-gate 3927c478bd9Sstevel@tonic-gate /* 3937c478bd9Sstevel@tonic-gate * Purge the readdir cache of all entries which are not currently 3947c478bd9Sstevel@tonic-gate * being filled. 3957c478bd9Sstevel@tonic-gate */ 3967c478bd9Sstevel@tonic-gate void 3977c478bd9Sstevel@tonic-gate nfs4_purge_rddir_cache(vnode_t *vp) 3987c478bd9Sstevel@tonic-gate { 3997c478bd9Sstevel@tonic-gate rnode4_t *rp; 4007c478bd9Sstevel@tonic-gate 4017c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 4027c478bd9Sstevel@tonic-gate 4037c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 4047c478bd9Sstevel@tonic-gate rp->r_direof = NULL; 4057c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4LOOKUP; 4067c478bd9Sstevel@tonic-gate rp->r_flags |= R4READDIRWATTR; 4077c478bd9Sstevel@tonic-gate rddir4_cache_purge(rp); 4087c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 4097c478bd9Sstevel@tonic-gate } 4107c478bd9Sstevel@tonic-gate 4117c478bd9Sstevel@tonic-gate /* 4127c478bd9Sstevel@tonic-gate * Set attributes cache for given vnode using virtual attributes. There is 4137c478bd9Sstevel@tonic-gate * no cache validation, but if the attributes are deemed to be stale, they 4147c478bd9Sstevel@tonic-gate * are ignored. This corresponds to nfs3_attrcache(). 4157c478bd9Sstevel@tonic-gate * 4167c478bd9Sstevel@tonic-gate * Set the timeout value on the attribute cache and fill it 4177c478bd9Sstevel@tonic-gate * with the passed in attributes. 4187c478bd9Sstevel@tonic-gate */ 4197c478bd9Sstevel@tonic-gate void 4207c478bd9Sstevel@tonic-gate nfs4_attrcache_noinval(vnode_t *vp, nfs4_ga_res_t *garp, hrtime_t t) 4217c478bd9Sstevel@tonic-gate { 4227c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(vp); 4237c478bd9Sstevel@tonic-gate 4247c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 4257c478bd9Sstevel@tonic-gate if (rp->r_time_attr_saved <= t) 4267c478bd9Sstevel@tonic-gate nfs4_attrcache_va(vp, garp, FALSE); 4277c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 4287c478bd9Sstevel@tonic-gate } 4297c478bd9Sstevel@tonic-gate 4307c478bd9Sstevel@tonic-gate /* 4317c478bd9Sstevel@tonic-gate * Use the passed in virtual attributes to check to see whether the 4327c478bd9Sstevel@tonic-gate * data and metadata caches are valid, cache the new attributes, and 4337c478bd9Sstevel@tonic-gate * then do the cache invalidation if required. 4347c478bd9Sstevel@tonic-gate * 4357c478bd9Sstevel@tonic-gate * The cache validation and caching of the new attributes is done 4367c478bd9Sstevel@tonic-gate * atomically via the use of the mutex, r_statelock. If required, 4377c478bd9Sstevel@tonic-gate * the cache invalidation is done atomically w.r.t. the cache 4387c478bd9Sstevel@tonic-gate * validation and caching of the attributes via the pseudo lock, 4397c478bd9Sstevel@tonic-gate * r_serial. 4407c478bd9Sstevel@tonic-gate * 4417c478bd9Sstevel@tonic-gate * This routine is used to do cache validation and attributes caching 4427c478bd9Sstevel@tonic-gate * for operations with a single set of post operation attributes. 4437c478bd9Sstevel@tonic-gate */ 4447c478bd9Sstevel@tonic-gate 4457c478bd9Sstevel@tonic-gate void 4467c478bd9Sstevel@tonic-gate nfs4_attr_cache(vnode_t *vp, nfs4_ga_res_t *garp, 447b9238976Sth hrtime_t t, cred_t *cr, int async, 448b9238976Sth change_info4 *cinfo) 4497c478bd9Sstevel@tonic-gate { 4507c478bd9Sstevel@tonic-gate rnode4_t *rp; 451*5e4df02aSvv int mtime_changed = 0; 452*5e4df02aSvv int ctime_changed = 0; 4537c478bd9Sstevel@tonic-gate vsecattr_t *vsp; 4547c478bd9Sstevel@tonic-gate int was_serial, set_time_cache_inval, recov; 4557c478bd9Sstevel@tonic-gate vattr_t *vap = &garp->n4g_va; 4567c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VTOMI4(vp); 457*5e4df02aSvv len_t preattr_rsize; 458*5e4df02aSvv boolean_t writemodify_set = B_FALSE; 459*5e4df02aSvv boolean_t cachepurge_set = B_FALSE; 4607c478bd9Sstevel@tonic-gate 4617c478bd9Sstevel@tonic-gate ASSERT(mi->mi_vfsp->vfs_dev == garp->n4g_va.va_fsid); 4627c478bd9Sstevel@tonic-gate 4637c478bd9Sstevel@tonic-gate /* Is curthread the recovery thread? */ 4647c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 4657c478bd9Sstevel@tonic-gate recov = (VTOMI4(vp)->mi_recovthread == curthread); 4667c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 4677c478bd9Sstevel@tonic-gate 4687c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 4697c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 4707c478bd9Sstevel@tonic-gate was_serial = (rp->r_serial == curthread); 4717c478bd9Sstevel@tonic-gate if (rp->r_serial && !was_serial) { 4727c478bd9Sstevel@tonic-gate klwp_t *lwp = ttolwp(curthread); 4737c478bd9Sstevel@tonic-gate 4747c478bd9Sstevel@tonic-gate /* 4757c478bd9Sstevel@tonic-gate * If we're the recovery thread, then purge current attrs 4767c478bd9Sstevel@tonic-gate * and bail out to avoid potential deadlock between another 4777c478bd9Sstevel@tonic-gate * thread caching attrs (r_serial thread), recov thread, 4787c478bd9Sstevel@tonic-gate * and an async writer thread. 4797c478bd9Sstevel@tonic-gate */ 4807c478bd9Sstevel@tonic-gate if (recov) { 4817c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE4_LOCKED(rp); 4827c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 4837c478bd9Sstevel@tonic-gate return; 4847c478bd9Sstevel@tonic-gate } 4857c478bd9Sstevel@tonic-gate 4867c478bd9Sstevel@tonic-gate if (lwp != NULL) 4877c478bd9Sstevel@tonic-gate lwp->lwp_nostop++; 4887c478bd9Sstevel@tonic-gate while (rp->r_serial != NULL) { 4897c478bd9Sstevel@tonic-gate if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) { 4907c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 4917c478bd9Sstevel@tonic-gate if (lwp != NULL) 4927c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 4937c478bd9Sstevel@tonic-gate return; 4947c478bd9Sstevel@tonic-gate } 4957c478bd9Sstevel@tonic-gate } 4967c478bd9Sstevel@tonic-gate if (lwp != NULL) 4977c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 4987c478bd9Sstevel@tonic-gate } 4997c478bd9Sstevel@tonic-gate 5007c478bd9Sstevel@tonic-gate /* 5017c478bd9Sstevel@tonic-gate * If there is a page flush thread, the current thread needs to 5027c478bd9Sstevel@tonic-gate * bail out, to prevent a possible deadlock between the current 5037c478bd9Sstevel@tonic-gate * thread (which might be in a start_op/end_op region), the 5047c478bd9Sstevel@tonic-gate * recovery thread, and the page flush thread. Expire the 5057c478bd9Sstevel@tonic-gate * attribute cache, so that any attributes the current thread was 5067c478bd9Sstevel@tonic-gate * going to set are not lost. 5077c478bd9Sstevel@tonic-gate */ 5087c478bd9Sstevel@tonic-gate if ((rp->r_flags & R4PGFLUSH) && rp->r_pgflush != curthread) { 5097c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE4_LOCKED(rp); 5107c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 5117c478bd9Sstevel@tonic-gate return; 5127c478bd9Sstevel@tonic-gate } 5137c478bd9Sstevel@tonic-gate 5147c478bd9Sstevel@tonic-gate if (rp->r_time_attr_saved > t) { 5157c478bd9Sstevel@tonic-gate /* 5167c478bd9Sstevel@tonic-gate * Attributes have been cached since these attributes were 51700fdf600Smaheshvs * probably made. If there is an inconsistency in what is 51800fdf600Smaheshvs * cached, mark them invalid. If not, don't act on them. 5197c478bd9Sstevel@tonic-gate */ 52000fdf600Smaheshvs if (!CACHE4_VALID(rp, vap->va_mtime, vap->va_size)) 52100fdf600Smaheshvs PURGE_ATTRCACHE4_LOCKED(rp); 5227c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 5237c478bd9Sstevel@tonic-gate return; 5247c478bd9Sstevel@tonic-gate } 5257c478bd9Sstevel@tonic-gate set_time_cache_inval = 0; 5267c478bd9Sstevel@tonic-gate if (cinfo) { 5277c478bd9Sstevel@tonic-gate /* 5287c478bd9Sstevel@tonic-gate * Only directory modifying callers pass non-NULL cinfo. 5297c478bd9Sstevel@tonic-gate */ 5307c478bd9Sstevel@tonic-gate ASSERT(vp->v_type == VDIR); 5317c478bd9Sstevel@tonic-gate /* 5327c478bd9Sstevel@tonic-gate * If the cache timeout either doesn't exist or hasn't expired, 5337c478bd9Sstevel@tonic-gate * and dir didn't changed on server before dirmod op 5347c478bd9Sstevel@tonic-gate * and dir didn't change after dirmod op but before getattr 5357c478bd9Sstevel@tonic-gate * then there's a chance that the client's cached data for 5367c478bd9Sstevel@tonic-gate * this object is current (not stale). No immediate cache 5377c478bd9Sstevel@tonic-gate * flush is required. 5387c478bd9Sstevel@tonic-gate * 5397c478bd9Sstevel@tonic-gate */ 5407c478bd9Sstevel@tonic-gate if ((! rp->r_time_cache_inval || t < rp->r_time_cache_inval) && 5417c478bd9Sstevel@tonic-gate cinfo->before == rp->r_change && 5427c478bd9Sstevel@tonic-gate (garp->n4g_change_valid && 5437c478bd9Sstevel@tonic-gate cinfo->after == garp->n4g_change)) { 5447c478bd9Sstevel@tonic-gate 5457c478bd9Sstevel@tonic-gate /* 5467c478bd9Sstevel@tonic-gate * If atomic isn't set, then the before/after info 5477c478bd9Sstevel@tonic-gate * cannot be blindly trusted. For this case, we tell 5487c478bd9Sstevel@tonic-gate * nfs4_attrcache_va to cache the attrs but also 5497c478bd9Sstevel@tonic-gate * establish an absolute maximum cache timeout. When 5507c478bd9Sstevel@tonic-gate * the timeout is reached, caches will be flushed. 5517c478bd9Sstevel@tonic-gate */ 5527c478bd9Sstevel@tonic-gate if (! cinfo->atomic) 5537c478bd9Sstevel@tonic-gate set_time_cache_inval = 1; 5547c478bd9Sstevel@tonic-gate } else { 5557c478bd9Sstevel@tonic-gate 5567c478bd9Sstevel@tonic-gate /* 5577c478bd9Sstevel@tonic-gate * We're not sure exactly what changed, but we know 5587c478bd9Sstevel@tonic-gate * what to do. flush all caches for dir. remove the 5597c478bd9Sstevel@tonic-gate * attr timeout. 5607c478bd9Sstevel@tonic-gate * 5617c478bd9Sstevel@tonic-gate * a) timeout expired. flush all caches. 5627c478bd9Sstevel@tonic-gate * b) r_change != cinfo.before. flush all caches. 5637c478bd9Sstevel@tonic-gate * c) r_change == cinfo.before, but cinfo.after != 5647c478bd9Sstevel@tonic-gate * post-op getattr(change). flush all caches. 5657c478bd9Sstevel@tonic-gate * d) post-op getattr(change) not provided by server. 5667c478bd9Sstevel@tonic-gate * flush all caches. 5677c478bd9Sstevel@tonic-gate */ 5687c478bd9Sstevel@tonic-gate mtime_changed = 1; 5697c478bd9Sstevel@tonic-gate ctime_changed = 1; 5707c478bd9Sstevel@tonic-gate rp->r_time_cache_inval = 0; 5717c478bd9Sstevel@tonic-gate } 5727c478bd9Sstevel@tonic-gate } else { 573*5e4df02aSvv /* 574*5e4df02aSvv * Write thread after writing data to file on remote server, 575*5e4df02aSvv * will always set R4WRITEMODIFIED to indicate that file on 576*5e4df02aSvv * remote server was modified with a WRITE operation and would 577*5e4df02aSvv * have marked attribute cache as timed out. If R4WRITEMODIFIED 578*5e4df02aSvv * is set, then do not check for mtime and ctime change. 579*5e4df02aSvv */ 5807c478bd9Sstevel@tonic-gate if (!(rp->r_flags & R4WRITEMODIFIED)) { 5817c478bd9Sstevel@tonic-gate if (!CACHE4_VALID(rp, vap->va_mtime, vap->va_size)) 5827c478bd9Sstevel@tonic-gate mtime_changed = 1; 583*5e4df02aSvv 5847c478bd9Sstevel@tonic-gate if (rp->r_attr.va_ctime.tv_sec != 5857c478bd9Sstevel@tonic-gate vap->va_ctime.tv_sec || 5867c478bd9Sstevel@tonic-gate rp->r_attr.va_ctime.tv_nsec != 5877c478bd9Sstevel@tonic-gate vap->va_ctime.tv_nsec) 5887c478bd9Sstevel@tonic-gate ctime_changed = 1; 5897c478bd9Sstevel@tonic-gate } else { 590*5e4df02aSvv writemodify_set = B_TRUE; 5917c478bd9Sstevel@tonic-gate } 5927c478bd9Sstevel@tonic-gate } 5937c478bd9Sstevel@tonic-gate 594*5e4df02aSvv preattr_rsize = rp->r_size; 595*5e4df02aSvv 5967c478bd9Sstevel@tonic-gate nfs4_attrcache_va(vp, garp, set_time_cache_inval); 5977c478bd9Sstevel@tonic-gate 598*5e4df02aSvv /* 599*5e4df02aSvv * If we have updated filesize in nfs4_attrcache_va, as soon as we 600*5e4df02aSvv * drop statelock we will be in transition of purging all 601*5e4df02aSvv * our caches and updating them. It is possible for another 602*5e4df02aSvv * thread to pick this new file size and read in zeroed data. 603*5e4df02aSvv * stall other threads till cache purge is complete. 604*5e4df02aSvv */ 605*5e4df02aSvv if ((!cinfo) && (rp->r_size != preattr_rsize)) { 606*5e4df02aSvv /* 607*5e4df02aSvv * If R4WRITEMODIFIED was set and we have updated the file 608*5e4df02aSvv * size, Server's returned file size need not necessarily 609*5e4df02aSvv * be because of this Client's WRITE. We need to purge 610*5e4df02aSvv * all caches. 611*5e4df02aSvv */ 612*5e4df02aSvv if (writemodify_set) 613*5e4df02aSvv mtime_changed = 1; 614*5e4df02aSvv 615*5e4df02aSvv if (mtime_changed && !(rp->r_flags & R4INCACHEPURGE)) { 616*5e4df02aSvv rp->r_flags |= R4INCACHEPURGE; 617*5e4df02aSvv cachepurge_set = B_TRUE; 618*5e4df02aSvv } 619*5e4df02aSvv } 620*5e4df02aSvv 6217c478bd9Sstevel@tonic-gate if (!mtime_changed && !ctime_changed) { 6227c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 6237c478bd9Sstevel@tonic-gate return; 6247c478bd9Sstevel@tonic-gate } 6257c478bd9Sstevel@tonic-gate 6267c478bd9Sstevel@tonic-gate rp->r_serial = curthread; 6277c478bd9Sstevel@tonic-gate 6287c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 6297c478bd9Sstevel@tonic-gate 6307c478bd9Sstevel@tonic-gate /* 6317c478bd9Sstevel@tonic-gate * If we're the recov thread, then force async nfs4_purge_caches 6327c478bd9Sstevel@tonic-gate * to avoid potential deadlock. 6337c478bd9Sstevel@tonic-gate */ 6347c478bd9Sstevel@tonic-gate if (mtime_changed) 6357c478bd9Sstevel@tonic-gate nfs4_purge_caches(vp, NFS4_NOPURGE_DNLC, cr, recov ? 1 : async); 6367c478bd9Sstevel@tonic-gate 637*5e4df02aSvv if ((rp->r_flags & R4INCACHEPURGE) && cachepurge_set) { 638*5e4df02aSvv mutex_enter(&rp->r_statelock); 639*5e4df02aSvv rp->r_flags &= ~R4INCACHEPURGE; 640*5e4df02aSvv cv_broadcast(&rp->r_cv); 641*5e4df02aSvv mutex_exit(&rp->r_statelock); 642*5e4df02aSvv cachepurge_set = B_FALSE; 643*5e4df02aSvv } 644*5e4df02aSvv 6457c478bd9Sstevel@tonic-gate if (ctime_changed) { 6467c478bd9Sstevel@tonic-gate (void) nfs4_access_purge_rp(rp); 6477c478bd9Sstevel@tonic-gate if (rp->r_secattr != NULL) { 6487c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 6497c478bd9Sstevel@tonic-gate vsp = rp->r_secattr; 6507c478bd9Sstevel@tonic-gate rp->r_secattr = NULL; 6517c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 6527c478bd9Sstevel@tonic-gate if (vsp != NULL) 6537c478bd9Sstevel@tonic-gate nfs4_acl_free_cache(vsp); 6547c478bd9Sstevel@tonic-gate } 6557c478bd9Sstevel@tonic-gate } 6567c478bd9Sstevel@tonic-gate 6577c478bd9Sstevel@tonic-gate if (!was_serial) { 6587c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 6597c478bd9Sstevel@tonic-gate rp->r_serial = NULL; 6607c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 6617c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 6627c478bd9Sstevel@tonic-gate } 6637c478bd9Sstevel@tonic-gate } 6647c478bd9Sstevel@tonic-gate 6657c478bd9Sstevel@tonic-gate /* 6667c478bd9Sstevel@tonic-gate * Set attributes cache for given vnode using virtual attributes. 6677c478bd9Sstevel@tonic-gate * 6687c478bd9Sstevel@tonic-gate * Set the timeout value on the attribute cache and fill it 6697c478bd9Sstevel@tonic-gate * with the passed in attributes. 6707c478bd9Sstevel@tonic-gate * 6717c478bd9Sstevel@tonic-gate * The caller must be holding r_statelock. 6727c478bd9Sstevel@tonic-gate */ 6737c478bd9Sstevel@tonic-gate static void 6747c478bd9Sstevel@tonic-gate nfs4_attrcache_va(vnode_t *vp, nfs4_ga_res_t *garp, int set_cache_timeout) 6757c478bd9Sstevel@tonic-gate { 6767c478bd9Sstevel@tonic-gate rnode4_t *rp; 6777c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 6787c478bd9Sstevel@tonic-gate hrtime_t delta; 6797c478bd9Sstevel@tonic-gate hrtime_t now; 6807c478bd9Sstevel@tonic-gate vattr_t *vap = &garp->n4g_va; 6817c478bd9Sstevel@tonic-gate 6827c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 6837c478bd9Sstevel@tonic-gate 6847c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(&rp->r_statelock)); 6857c478bd9Sstevel@tonic-gate ASSERT(vap->va_mask == AT_ALL); 6867c478bd9Sstevel@tonic-gate 6877c478bd9Sstevel@tonic-gate /* Switch to master before checking v_flag */ 6887c478bd9Sstevel@tonic-gate if (IS_SHADOW(vp, rp)) 6897c478bd9Sstevel@tonic-gate vp = RTOV4(rp); 6907c478bd9Sstevel@tonic-gate 6917c478bd9Sstevel@tonic-gate now = gethrtime(); 6927c478bd9Sstevel@tonic-gate 6937c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 6947c478bd9Sstevel@tonic-gate 6957c478bd9Sstevel@tonic-gate /* 6967c478bd9Sstevel@tonic-gate * Only establish a new cache timeout (if requested). Never 6977c478bd9Sstevel@tonic-gate * extend a timeout. Never clear a timeout. Clearing a timeout 6987c478bd9Sstevel@tonic-gate * is done by nfs4_update_dircaches (ancestor in our call chain) 6997c478bd9Sstevel@tonic-gate */ 7007c478bd9Sstevel@tonic-gate if (set_cache_timeout && ! rp->r_time_cache_inval) 7017c478bd9Sstevel@tonic-gate rp->r_time_cache_inval = now + mi->mi_acdirmax; 7027c478bd9Sstevel@tonic-gate 7037c478bd9Sstevel@tonic-gate /* 7047c478bd9Sstevel@tonic-gate * Delta is the number of nanoseconds that we will 7057c478bd9Sstevel@tonic-gate * cache the attributes of the file. It is based on 7067c478bd9Sstevel@tonic-gate * the number of nanoseconds since the last time that 7077c478bd9Sstevel@tonic-gate * we detected a change. The assumption is that files 7087c478bd9Sstevel@tonic-gate * that changed recently are likely to change again. 7097c478bd9Sstevel@tonic-gate * There is a minimum and a maximum for regular files 7107c478bd9Sstevel@tonic-gate * and for directories which is enforced though. 7117c478bd9Sstevel@tonic-gate * 7127c478bd9Sstevel@tonic-gate * Using the time since last change was detected 7137c478bd9Sstevel@tonic-gate * eliminates direct comparison or calculation 7147c478bd9Sstevel@tonic-gate * using mixed client and server times. NFS does 7157c478bd9Sstevel@tonic-gate * not make any assumptions regarding the client 7167c478bd9Sstevel@tonic-gate * and server clocks being synchronized. 7177c478bd9Sstevel@tonic-gate */ 7187c478bd9Sstevel@tonic-gate if (vap->va_mtime.tv_sec != rp->r_attr.va_mtime.tv_sec || 7197c478bd9Sstevel@tonic-gate vap->va_mtime.tv_nsec != rp->r_attr.va_mtime.tv_nsec || 7207c478bd9Sstevel@tonic-gate vap->va_size != rp->r_attr.va_size) { 7217c478bd9Sstevel@tonic-gate rp->r_time_attr_saved = now; 7227c478bd9Sstevel@tonic-gate } 7237c478bd9Sstevel@tonic-gate 7247c478bd9Sstevel@tonic-gate if ((mi->mi_flags & MI4_NOAC) || (vp->v_flag & VNOCACHE)) 7257c478bd9Sstevel@tonic-gate delta = 0; 7267c478bd9Sstevel@tonic-gate else { 7277c478bd9Sstevel@tonic-gate delta = now - rp->r_time_attr_saved; 7287c478bd9Sstevel@tonic-gate if (vp->v_type == VDIR) { 7297c478bd9Sstevel@tonic-gate if (delta < mi->mi_acdirmin) 7307c478bd9Sstevel@tonic-gate delta = mi->mi_acdirmin; 7317c478bd9Sstevel@tonic-gate else if (delta > mi->mi_acdirmax) 7327c478bd9Sstevel@tonic-gate delta = mi->mi_acdirmax; 7337c478bd9Sstevel@tonic-gate } else { 7347c478bd9Sstevel@tonic-gate if (delta < mi->mi_acregmin) 7357c478bd9Sstevel@tonic-gate delta = mi->mi_acregmin; 7367c478bd9Sstevel@tonic-gate else if (delta > mi->mi_acregmax) 7377c478bd9Sstevel@tonic-gate delta = mi->mi_acregmax; 7387c478bd9Sstevel@tonic-gate } 7397c478bd9Sstevel@tonic-gate } 7407c478bd9Sstevel@tonic-gate rp->r_time_attr_inval = now + delta; 7417c478bd9Sstevel@tonic-gate 7427c478bd9Sstevel@tonic-gate rp->r_attr = *vap; 7437c478bd9Sstevel@tonic-gate if (garp->n4g_change_valid) 7447c478bd9Sstevel@tonic-gate rp->r_change = garp->n4g_change; 7457c478bd9Sstevel@tonic-gate 7467c478bd9Sstevel@tonic-gate /* 7477c478bd9Sstevel@tonic-gate * The attributes that were returned may be valid and can 7487c478bd9Sstevel@tonic-gate * be used, but they may not be allowed to be cached. 7497c478bd9Sstevel@tonic-gate * Reset the timers to cause immediate invalidation and 7507c478bd9Sstevel@tonic-gate * clear r_change so no VERIFY operations will suceed 7517c478bd9Sstevel@tonic-gate */ 7527c478bd9Sstevel@tonic-gate if (garp->n4g_attrwhy == NFS4_GETATTR_NOCACHE_OK) { 7537c478bd9Sstevel@tonic-gate rp->r_time_attr_inval = now; 7547c478bd9Sstevel@tonic-gate rp->r_time_attr_saved = now; 7557c478bd9Sstevel@tonic-gate rp->r_change = 0; 7567c478bd9Sstevel@tonic-gate } 7577c478bd9Sstevel@tonic-gate 7587c478bd9Sstevel@tonic-gate /* 7597c478bd9Sstevel@tonic-gate * If mounted_on_fileid returned AND the object is a stub, 7607c478bd9Sstevel@tonic-gate * then set object's va_nodeid to the mounted over fid 7617c478bd9Sstevel@tonic-gate * returned by server. 7627c478bd9Sstevel@tonic-gate * 7637c478bd9Sstevel@tonic-gate * If mounted_on_fileid not provided/supported, then 7647c478bd9Sstevel@tonic-gate * just set it to 0 for now. Eventually it would be 7657c478bd9Sstevel@tonic-gate * better to set it to a hashed version of FH. This 7667c478bd9Sstevel@tonic-gate * would probably be good enough to provide a unique 7677c478bd9Sstevel@tonic-gate * fid/d_ino within a dir. 7687c478bd9Sstevel@tonic-gate * 7697c478bd9Sstevel@tonic-gate * We don't need to carry mounted_on_fileid in the 7707c478bd9Sstevel@tonic-gate * rnode as long as the client never requests fileid 7717c478bd9Sstevel@tonic-gate * without also requesting mounted_on_fileid. For 7727c478bd9Sstevel@tonic-gate * now, it stays. 7737c478bd9Sstevel@tonic-gate */ 7747c478bd9Sstevel@tonic-gate if (garp->n4g_mon_fid_valid) { 7757c478bd9Sstevel@tonic-gate rp->r_mntd_fid = garp->n4g_mon_fid; 7767c478bd9Sstevel@tonic-gate 777b9238976Sth if (RP_ISSTUB(rp)) 7787c478bd9Sstevel@tonic-gate rp->r_attr.va_nodeid = rp->r_mntd_fid; 7797c478bd9Sstevel@tonic-gate } 7807c478bd9Sstevel@tonic-gate 7817c478bd9Sstevel@tonic-gate /* 7827c478bd9Sstevel@tonic-gate * Check to see if there are valid pathconf bits to 7837c478bd9Sstevel@tonic-gate * cache in the rnode. 7847c478bd9Sstevel@tonic-gate */ 7857c478bd9Sstevel@tonic-gate if (garp->n4g_ext_res) { 7867c478bd9Sstevel@tonic-gate if (garp->n4g_ext_res->n4g_pc4.pc4_cache_valid) { 7877c478bd9Sstevel@tonic-gate rp->r_pathconf = garp->n4g_ext_res->n4g_pc4; 7887c478bd9Sstevel@tonic-gate } else { 7897c478bd9Sstevel@tonic-gate if (garp->n4g_ext_res->n4g_pc4.pc4_xattr_valid) { 7907c478bd9Sstevel@tonic-gate rp->r_pathconf.pc4_xattr_valid = TRUE; 7917c478bd9Sstevel@tonic-gate rp->r_pathconf.pc4_xattr_exists = 7927c478bd9Sstevel@tonic-gate garp->n4g_ext_res->n4g_pc4.pc4_xattr_exists; 7937c478bd9Sstevel@tonic-gate } 7947c478bd9Sstevel@tonic-gate } 7957c478bd9Sstevel@tonic-gate } 7967c478bd9Sstevel@tonic-gate /* 7977c478bd9Sstevel@tonic-gate * Update the size of the file if there is no cached data or if 7987c478bd9Sstevel@tonic-gate * the cached data is clean and there is no data being written 7997c478bd9Sstevel@tonic-gate * out. 8007c478bd9Sstevel@tonic-gate */ 8017c478bd9Sstevel@tonic-gate if (rp->r_size != vap->va_size && 8027c478bd9Sstevel@tonic-gate (!vn_has_cached_data(vp) || 8037c478bd9Sstevel@tonic-gate (!(rp->r_flags & R4DIRTY) && rp->r_count == 0))) { 8047c478bd9Sstevel@tonic-gate rp->r_size = vap->va_size; 8057c478bd9Sstevel@tonic-gate } 8067c478bd9Sstevel@tonic-gate nfs_setswaplike(vp, vap); 8077c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4WRITEMODIFIED; 8087c478bd9Sstevel@tonic-gate } 8097c478bd9Sstevel@tonic-gate 8107c478bd9Sstevel@tonic-gate /* 8117c478bd9Sstevel@tonic-gate * Get attributes over-the-wire and update attributes cache 8127c478bd9Sstevel@tonic-gate * if no error occurred in the over-the-wire operation. 8137c478bd9Sstevel@tonic-gate * Return 0 if successful, otherwise error. 8147c478bd9Sstevel@tonic-gate */ 8157c478bd9Sstevel@tonic-gate int 8167c478bd9Sstevel@tonic-gate nfs4_getattr_otw(vnode_t *vp, nfs4_ga_res_t *garp, cred_t *cr, int get_acl) 8177c478bd9Sstevel@tonic-gate { 8187c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VTOMI4(vp); 8197c478bd9Sstevel@tonic-gate hrtime_t t; 8207c478bd9Sstevel@tonic-gate nfs4_recov_state_t recov_state; 8217c478bd9Sstevel@tonic-gate nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS }; 8227c478bd9Sstevel@tonic-gate 8237c478bd9Sstevel@tonic-gate recov_state.rs_flags = 0; 8247c478bd9Sstevel@tonic-gate recov_state.rs_num_retry_despite_err = 0; 8257c478bd9Sstevel@tonic-gate 8267c478bd9Sstevel@tonic-gate /* Save the original mount point security flavor */ 8277c478bd9Sstevel@tonic-gate (void) save_mnt_secinfo(mi->mi_curr_serv); 8287c478bd9Sstevel@tonic-gate 8297c478bd9Sstevel@tonic-gate recov_retry: 830b9238976Sth 8317c478bd9Sstevel@tonic-gate if ((e.error = nfs4_start_fop(mi, vp, NULL, OH_GETATTR, 832b9238976Sth &recov_state, NULL))) { 8337c478bd9Sstevel@tonic-gate (void) check_mnt_secinfo(mi->mi_curr_serv, vp); 8347c478bd9Sstevel@tonic-gate return (e.error); 8357c478bd9Sstevel@tonic-gate } 8367c478bd9Sstevel@tonic-gate 8377c478bd9Sstevel@tonic-gate t = gethrtime(); 8387c478bd9Sstevel@tonic-gate 8397c478bd9Sstevel@tonic-gate nfs4_getattr_otw_norecovery(vp, garp, &e, cr, get_acl); 8407c478bd9Sstevel@tonic-gate 8417c478bd9Sstevel@tonic-gate if (nfs4_needs_recovery(&e, FALSE, vp->v_vfsp)) { 8427c478bd9Sstevel@tonic-gate if (nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL, 8437c478bd9Sstevel@tonic-gate NULL, OP_GETATTR, NULL) == FALSE) { 8447c478bd9Sstevel@tonic-gate nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, 845b9238976Sth &recov_state, 1); 8467c478bd9Sstevel@tonic-gate goto recov_retry; 8477c478bd9Sstevel@tonic-gate } 8487c478bd9Sstevel@tonic-gate } 8497c478bd9Sstevel@tonic-gate 8507c478bd9Sstevel@tonic-gate nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state, 0); 8517c478bd9Sstevel@tonic-gate 8527c478bd9Sstevel@tonic-gate if (!e.error) { 8537c478bd9Sstevel@tonic-gate if (e.stat == NFS4_OK) { 8547c478bd9Sstevel@tonic-gate nfs4_attr_cache(vp, garp, t, cr, FALSE, NULL); 8557c478bd9Sstevel@tonic-gate } else { 8567c478bd9Sstevel@tonic-gate e.error = geterrno4(e.stat); 8577c478bd9Sstevel@tonic-gate 8587c478bd9Sstevel@tonic-gate nfs4_purge_stale_fh(e.error, vp, cr); 8597c478bd9Sstevel@tonic-gate } 8607c478bd9Sstevel@tonic-gate } 8617c478bd9Sstevel@tonic-gate 8627c478bd9Sstevel@tonic-gate /* 8637c478bd9Sstevel@tonic-gate * If getattr a node that is a stub for a crossed 8647c478bd9Sstevel@tonic-gate * mount point, keep the original secinfo flavor for 8657c478bd9Sstevel@tonic-gate * the current file system, not the crossed one. 8667c478bd9Sstevel@tonic-gate */ 8677c478bd9Sstevel@tonic-gate (void) check_mnt_secinfo(mi->mi_curr_serv, vp); 8687c478bd9Sstevel@tonic-gate 8697c478bd9Sstevel@tonic-gate return (e.error); 8707c478bd9Sstevel@tonic-gate } 8717c478bd9Sstevel@tonic-gate 8727c478bd9Sstevel@tonic-gate /* 8737c478bd9Sstevel@tonic-gate * Generate a compound to get attributes over-the-wire. 8747c478bd9Sstevel@tonic-gate */ 8757c478bd9Sstevel@tonic-gate void 8767c478bd9Sstevel@tonic-gate nfs4_getattr_otw_norecovery(vnode_t *vp, nfs4_ga_res_t *garp, 877b9238976Sth nfs4_error_t *ep, cred_t *cr, int get_acl) 8787c478bd9Sstevel@tonic-gate { 8797c478bd9Sstevel@tonic-gate COMPOUND4args_clnt args; 8807c478bd9Sstevel@tonic-gate COMPOUND4res_clnt res; 8817c478bd9Sstevel@tonic-gate int doqueue; 8827c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(vp); 8837c478bd9Sstevel@tonic-gate nfs_argop4 argop[2]; 8847c478bd9Sstevel@tonic-gate 8857c478bd9Sstevel@tonic-gate args.ctag = TAG_GETATTR; 8867c478bd9Sstevel@tonic-gate 8877c478bd9Sstevel@tonic-gate args.array_len = 2; 8887c478bd9Sstevel@tonic-gate args.array = argop; 8897c478bd9Sstevel@tonic-gate 8907c478bd9Sstevel@tonic-gate /* putfh */ 8917c478bd9Sstevel@tonic-gate argop[0].argop = OP_CPUTFH; 8927c478bd9Sstevel@tonic-gate argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh; 8937c478bd9Sstevel@tonic-gate 8947c478bd9Sstevel@tonic-gate /* getattr */ 8957c478bd9Sstevel@tonic-gate /* 8967c478bd9Sstevel@tonic-gate * Unlike nfs version 2 and 3, where getattr returns all the 897da6c28aaSamw * attributes, nfs version 4 returns only the ones explicitly 8987c478bd9Sstevel@tonic-gate * asked for. This creates problems, as some system functions 8997c478bd9Sstevel@tonic-gate * (e.g. cache check) require certain attributes and if the 9007c478bd9Sstevel@tonic-gate * cached node lacks some attributes such as uid/gid, it can 9017c478bd9Sstevel@tonic-gate * affect system utilities (e.g. "ls") that rely on the information 9027c478bd9Sstevel@tonic-gate * to be there. This can lead to anything from system crashes to 9037c478bd9Sstevel@tonic-gate * corrupted information processed by user apps. 9047c478bd9Sstevel@tonic-gate * So to ensure that all bases are covered, request at least 9057c478bd9Sstevel@tonic-gate * the AT_ALL attribute mask. 9067c478bd9Sstevel@tonic-gate */ 9077c478bd9Sstevel@tonic-gate argop[1].argop = OP_GETATTR; 9087c478bd9Sstevel@tonic-gate argop[1].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK; 9097c478bd9Sstevel@tonic-gate if (get_acl) 9107c478bd9Sstevel@tonic-gate argop[1].nfs_argop4_u.opgetattr.attr_request |= FATTR4_ACL_MASK; 9117c478bd9Sstevel@tonic-gate argop[1].nfs_argop4_u.opgetattr.mi = VTOMI4(vp); 9127c478bd9Sstevel@tonic-gate 9137c478bd9Sstevel@tonic-gate doqueue = 1; 9147c478bd9Sstevel@tonic-gate 9157c478bd9Sstevel@tonic-gate rfs4call(VTOMI4(vp), &args, &res, cr, &doqueue, 0, ep); 9167c478bd9Sstevel@tonic-gate 9177c478bd9Sstevel@tonic-gate if (ep->error) 9187c478bd9Sstevel@tonic-gate return; 9197c478bd9Sstevel@tonic-gate 9207c478bd9Sstevel@tonic-gate if (res.status != NFS4_OK) { 9217c478bd9Sstevel@tonic-gate (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 9227c478bd9Sstevel@tonic-gate return; 9237c478bd9Sstevel@tonic-gate } 9247c478bd9Sstevel@tonic-gate 9257c478bd9Sstevel@tonic-gate *garp = res.array[1].nfs_resop4_u.opgetattr.ga_res; 9267c478bd9Sstevel@tonic-gate 9277c478bd9Sstevel@tonic-gate (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 9287c478bd9Sstevel@tonic-gate } 9297c478bd9Sstevel@tonic-gate 9307c478bd9Sstevel@tonic-gate /* 9317c478bd9Sstevel@tonic-gate * Return either cached or remote attributes. If get remote attr 9327c478bd9Sstevel@tonic-gate * use them to check and invalidate caches, then cache the new attributes. 9337c478bd9Sstevel@tonic-gate */ 9347c478bd9Sstevel@tonic-gate int 9357c478bd9Sstevel@tonic-gate nfs4getattr(vnode_t *vp, vattr_t *vap, cred_t *cr) 9367c478bd9Sstevel@tonic-gate { 9377c478bd9Sstevel@tonic-gate int error; 9387c478bd9Sstevel@tonic-gate rnode4_t *rp; 9397c478bd9Sstevel@tonic-gate nfs4_ga_res_t gar; 9407c478bd9Sstevel@tonic-gate 9417c478bd9Sstevel@tonic-gate ASSERT(nfs4_consistent_type(vp)); 9427c478bd9Sstevel@tonic-gate 9437c478bd9Sstevel@tonic-gate /* 9447c478bd9Sstevel@tonic-gate * If we've got cached attributes, we're done, otherwise go 9457c478bd9Sstevel@tonic-gate * to the server to get attributes, which will update the cache 946b9238976Sth * in the process. Either way, use the cached attributes for 947b9238976Sth * the caller's vattr_t. 948b9238976Sth * 949b9238976Sth * Note that we ignore the gar set by the OTW call: the attr caching 950b9238976Sth * code may make adjustments when storing to the rnode, and we want 951b9238976Sth * to see those changes here. 9527c478bd9Sstevel@tonic-gate */ 9537c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 954b9238976Sth error = 0; 9557c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 956b9238976Sth if (!ATTRCACHE4_VALID(vp)) { 9577c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 958b9238976Sth error = nfs4_getattr_otw(vp, &gar, cr, 0); 959b9238976Sth mutex_enter(&rp->r_statelock); 9607c478bd9Sstevel@tonic-gate } 9617c478bd9Sstevel@tonic-gate 9627c478bd9Sstevel@tonic-gate if (!error) 963b9238976Sth *vap = rp->r_attr; 9647c478bd9Sstevel@tonic-gate 9657c478bd9Sstevel@tonic-gate /* Return the client's view of file size */ 9667c478bd9Sstevel@tonic-gate vap->va_size = rp->r_size; 967b9238976Sth 9687c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 9697c478bd9Sstevel@tonic-gate 9707c478bd9Sstevel@tonic-gate ASSERT(nfs4_consistent_type(vp)); 9717c478bd9Sstevel@tonic-gate 9727c478bd9Sstevel@tonic-gate return (error); 9737c478bd9Sstevel@tonic-gate } 9747c478bd9Sstevel@tonic-gate 9757c478bd9Sstevel@tonic-gate int 9767c478bd9Sstevel@tonic-gate nfs4_attr_otw(vnode_t *vp, nfs4_tag_type_t tag_type, 977b9238976Sth nfs4_ga_res_t *garp, bitmap4 reqbitmap, cred_t *cr) 9787c478bd9Sstevel@tonic-gate { 9797c478bd9Sstevel@tonic-gate COMPOUND4args_clnt args; 9807c478bd9Sstevel@tonic-gate COMPOUND4res_clnt res; 9817c478bd9Sstevel@tonic-gate int doqueue; 9827c478bd9Sstevel@tonic-gate nfs_argop4 argop[2]; 9837c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VTOMI4(vp); 9847c478bd9Sstevel@tonic-gate bool_t needrecov = FALSE; 9857c478bd9Sstevel@tonic-gate nfs4_recov_state_t recov_state; 9867c478bd9Sstevel@tonic-gate nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS }; 9877c478bd9Sstevel@tonic-gate nfs4_ga_ext_res_t *gerp; 9887c478bd9Sstevel@tonic-gate 9897c478bd9Sstevel@tonic-gate recov_state.rs_flags = 0; 9907c478bd9Sstevel@tonic-gate recov_state.rs_num_retry_despite_err = 0; 9917c478bd9Sstevel@tonic-gate 9927c478bd9Sstevel@tonic-gate recov_retry: 9937c478bd9Sstevel@tonic-gate args.ctag = tag_type; 9947c478bd9Sstevel@tonic-gate 9957c478bd9Sstevel@tonic-gate args.array_len = 2; 9967c478bd9Sstevel@tonic-gate args.array = argop; 9977c478bd9Sstevel@tonic-gate 9987c478bd9Sstevel@tonic-gate e.error = nfs4_start_fop(mi, vp, NULL, OH_GETATTR, &recov_state, NULL); 9997c478bd9Sstevel@tonic-gate if (e.error) 10007c478bd9Sstevel@tonic-gate return (e.error); 10017c478bd9Sstevel@tonic-gate 10027c478bd9Sstevel@tonic-gate /* putfh */ 10037c478bd9Sstevel@tonic-gate argop[0].argop = OP_CPUTFH; 10047c478bd9Sstevel@tonic-gate argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(vp)->r_fh; 10057c478bd9Sstevel@tonic-gate 10067c478bd9Sstevel@tonic-gate /* getattr */ 10077c478bd9Sstevel@tonic-gate argop[1].argop = OP_GETATTR; 10087c478bd9Sstevel@tonic-gate argop[1].nfs_argop4_u.opgetattr.attr_request = reqbitmap; 10097c478bd9Sstevel@tonic-gate argop[1].nfs_argop4_u.opgetattr.mi = mi; 10107c478bd9Sstevel@tonic-gate 10117c478bd9Sstevel@tonic-gate doqueue = 1; 10127c478bd9Sstevel@tonic-gate 10137c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE, 10147c478bd9Sstevel@tonic-gate "nfs4_attr_otw: %s call, rp %s", needrecov ? "recov" : "first", 10157c478bd9Sstevel@tonic-gate rnode4info(VTOR4(vp)))); 10167c478bd9Sstevel@tonic-gate 10177c478bd9Sstevel@tonic-gate rfs4call(mi, &args, &res, cr, &doqueue, 0, &e); 10187c478bd9Sstevel@tonic-gate 10197c478bd9Sstevel@tonic-gate needrecov = nfs4_needs_recovery(&e, FALSE, vp->v_vfsp); 10207c478bd9Sstevel@tonic-gate if (!needrecov && e.error) { 10217c478bd9Sstevel@tonic-gate nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state, 1022b9238976Sth needrecov); 10237c478bd9Sstevel@tonic-gate return (e.error); 10247c478bd9Sstevel@tonic-gate } 10257c478bd9Sstevel@tonic-gate 10267c478bd9Sstevel@tonic-gate if (needrecov) { 10277c478bd9Sstevel@tonic-gate bool_t abort; 10287c478bd9Sstevel@tonic-gate 10297c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE, 10307c478bd9Sstevel@tonic-gate "nfs4_attr_otw: initiating recovery\n")); 10317c478bd9Sstevel@tonic-gate 10327c478bd9Sstevel@tonic-gate abort = nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL, 1033b9238976Sth NULL, OP_GETATTR, NULL); 10347c478bd9Sstevel@tonic-gate nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state, 1035b9238976Sth needrecov); 10367c478bd9Sstevel@tonic-gate if (!e.error) { 10377c478bd9Sstevel@tonic-gate (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 10387c478bd9Sstevel@tonic-gate e.error = geterrno4(res.status); 10397c478bd9Sstevel@tonic-gate } 10407c478bd9Sstevel@tonic-gate if (abort == FALSE) 10417c478bd9Sstevel@tonic-gate goto recov_retry; 10427c478bd9Sstevel@tonic-gate return (e.error); 10437c478bd9Sstevel@tonic-gate } 10447c478bd9Sstevel@tonic-gate 10457c478bd9Sstevel@tonic-gate if (res.status) { 10467c478bd9Sstevel@tonic-gate e.error = geterrno4(res.status); 10477c478bd9Sstevel@tonic-gate } else { 10487c478bd9Sstevel@tonic-gate gerp = garp->n4g_ext_res; 10497c478bd9Sstevel@tonic-gate bcopy(&res.array[1].nfs_resop4_u.opgetattr.ga_res, 1050b9238976Sth garp, sizeof (nfs4_ga_res_t)); 10517c478bd9Sstevel@tonic-gate garp->n4g_ext_res = gerp; 10527c478bd9Sstevel@tonic-gate if (garp->n4g_ext_res && 10537c478bd9Sstevel@tonic-gate res.array[1].nfs_resop4_u.opgetattr.ga_res.n4g_ext_res) 10547c478bd9Sstevel@tonic-gate bcopy(res.array[1].nfs_resop4_u.opgetattr. 1055b9238976Sth ga_res.n4g_ext_res, 1056b9238976Sth garp->n4g_ext_res, sizeof (nfs4_ga_ext_res_t)); 10577c478bd9Sstevel@tonic-gate } 10587c478bd9Sstevel@tonic-gate (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 10597c478bd9Sstevel@tonic-gate nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state, 1060b9238976Sth needrecov); 10617c478bd9Sstevel@tonic-gate return (e.error); 10627c478bd9Sstevel@tonic-gate } 10637c478bd9Sstevel@tonic-gate 10647c478bd9Sstevel@tonic-gate /* 10657c478bd9Sstevel@tonic-gate * Asynchronous I/O parameters. nfs_async_threads is the high-water mark 10667c478bd9Sstevel@tonic-gate * for the demand-based allocation of async threads per-mount. The 10677c478bd9Sstevel@tonic-gate * nfs_async_timeout is the amount of time a thread will live after it 10687c478bd9Sstevel@tonic-gate * becomes idle, unless new I/O requests are received before the thread 10697c478bd9Sstevel@tonic-gate * dies. See nfs4_async_putpage and nfs4_async_start. 10707c478bd9Sstevel@tonic-gate */ 10717c478bd9Sstevel@tonic-gate 10727c478bd9Sstevel@tonic-gate static void nfs4_async_start(struct vfs *); 10737c478bd9Sstevel@tonic-gate 10747c478bd9Sstevel@tonic-gate static void 10757c478bd9Sstevel@tonic-gate free_async_args4(struct nfs4_async_reqs *args) 10767c478bd9Sstevel@tonic-gate { 10777c478bd9Sstevel@tonic-gate rnode4_t *rp; 10787c478bd9Sstevel@tonic-gate 10797c478bd9Sstevel@tonic-gate if (args->a_io != NFS4_INACTIVE) { 10807c478bd9Sstevel@tonic-gate rp = VTOR4(args->a_vp); 10817c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 10827c478bd9Sstevel@tonic-gate rp->r_count--; 10837c478bd9Sstevel@tonic-gate if (args->a_io == NFS4_PUTAPAGE || 10847c478bd9Sstevel@tonic-gate args->a_io == NFS4_PAGEIO) 10857c478bd9Sstevel@tonic-gate rp->r_awcount--; 10867c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 10877c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 10887c478bd9Sstevel@tonic-gate VN_RELE(args->a_vp); 10897c478bd9Sstevel@tonic-gate } 10907c478bd9Sstevel@tonic-gate crfree(args->a_cred); 10917c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 10927c478bd9Sstevel@tonic-gate } 10937c478bd9Sstevel@tonic-gate 10947c478bd9Sstevel@tonic-gate /* 10957c478bd9Sstevel@tonic-gate * Cross-zone thread creation and NFS access is disallowed, yet fsflush() and 10967c478bd9Sstevel@tonic-gate * pageout(), running in the global zone, have legitimate reasons to do 10977c478bd9Sstevel@tonic-gate * VOP_PUTPAGE(B_ASYNC) on other zones' NFS mounts. We avoid the problem by 10987c478bd9Sstevel@tonic-gate * use of a a per-mount "asynchronous requests manager thread" which is 10997c478bd9Sstevel@tonic-gate * signaled by the various asynchronous work routines when there is 11007c478bd9Sstevel@tonic-gate * asynchronous work to be done. It is responsible for creating new 11017c478bd9Sstevel@tonic-gate * worker threads if necessary, and notifying existing worker threads 11027c478bd9Sstevel@tonic-gate * that there is work to be done. 11037c478bd9Sstevel@tonic-gate * 11047c478bd9Sstevel@tonic-gate * In other words, it will "take the specifications from the customers and 11057c478bd9Sstevel@tonic-gate * give them to the engineers." 11067c478bd9Sstevel@tonic-gate * 11077c478bd9Sstevel@tonic-gate * Worker threads die off of their own accord if they are no longer 11087c478bd9Sstevel@tonic-gate * needed. 11097c478bd9Sstevel@tonic-gate * 11107c478bd9Sstevel@tonic-gate * This thread is killed when the zone is going away or the filesystem 11117c478bd9Sstevel@tonic-gate * is being unmounted. 11127c478bd9Sstevel@tonic-gate */ 11137c478bd9Sstevel@tonic-gate void 11147c478bd9Sstevel@tonic-gate nfs4_async_manager(vfs_t *vfsp) 11157c478bd9Sstevel@tonic-gate { 11167c478bd9Sstevel@tonic-gate callb_cpr_t cprinfo; 11177c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 11187c478bd9Sstevel@tonic-gate uint_t max_threads; 11197c478bd9Sstevel@tonic-gate 11207c478bd9Sstevel@tonic-gate mi = VFTOMI4(vfsp); 11217c478bd9Sstevel@tonic-gate 11227c478bd9Sstevel@tonic-gate CALLB_CPR_INIT(&cprinfo, &mi->mi_async_lock, callb_generic_cpr, 1123b9238976Sth "nfs4_async_manager"); 11247c478bd9Sstevel@tonic-gate 11257c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 11267c478bd9Sstevel@tonic-gate /* 11277c478bd9Sstevel@tonic-gate * We want to stash the max number of threads that this mount was 11287c478bd9Sstevel@tonic-gate * allowed so we can use it later when the variable is set to zero as 11297c478bd9Sstevel@tonic-gate * part of the zone/mount going away. 11307c478bd9Sstevel@tonic-gate * 11317c478bd9Sstevel@tonic-gate * We want to be able to create at least one thread to handle 11327c478bd9Sstevel@tonic-gate * asyncrhonous inactive calls. 11337c478bd9Sstevel@tonic-gate */ 11347c478bd9Sstevel@tonic-gate max_threads = MAX(mi->mi_max_threads, 1); 11357c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 11367c478bd9Sstevel@tonic-gate /* 11377c478bd9Sstevel@tonic-gate * We don't want to wait for mi_max_threads to go to zero, since that 11387c478bd9Sstevel@tonic-gate * happens as part of a failed unmount, but this thread should only 11397c478bd9Sstevel@tonic-gate * exit when the mount is really going away. 11407c478bd9Sstevel@tonic-gate * 11417c478bd9Sstevel@tonic-gate * Once MI4_ASYNC_MGR_STOP is set, no more async operations will be 11427c478bd9Sstevel@tonic-gate * attempted: the various _async_*() functions know to do things 11437c478bd9Sstevel@tonic-gate * inline if mi_max_threads == 0. Henceforth we just drain out the 11447c478bd9Sstevel@tonic-gate * outstanding requests. 11457c478bd9Sstevel@tonic-gate * 11467c478bd9Sstevel@tonic-gate * Note that we still create zthreads even if we notice the zone is 11477c478bd9Sstevel@tonic-gate * shutting down (MI4_ASYNC_MGR_STOP is set); this may cause the zone 11487c478bd9Sstevel@tonic-gate * shutdown sequence to take slightly longer in some cases, but 11497c478bd9Sstevel@tonic-gate * doesn't violate the protocol, as all threads will exit as soon as 11507c478bd9Sstevel@tonic-gate * they're done processing the remaining requests. 11517c478bd9Sstevel@tonic-gate */ 11527c478bd9Sstevel@tonic-gate while (!(mi->mi_flags & MI4_ASYNC_MGR_STOP) || 11537c478bd9Sstevel@tonic-gate mi->mi_async_req_count > 0) { 11547c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 11557c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_BEGIN(&cprinfo); 11567c478bd9Sstevel@tonic-gate cv_wait(&mi->mi_async_reqs_cv, &mi->mi_async_lock); 11577c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_END(&cprinfo, &mi->mi_async_lock); 11587c478bd9Sstevel@tonic-gate while (mi->mi_async_req_count > 0) { 11597c478bd9Sstevel@tonic-gate /* 11607c478bd9Sstevel@tonic-gate * Paranoia: If the mount started out having 11617c478bd9Sstevel@tonic-gate * (mi->mi_max_threads == 0), and the value was 11627c478bd9Sstevel@tonic-gate * later changed (via a debugger or somesuch), 11637c478bd9Sstevel@tonic-gate * we could be confused since we will think we 11647c478bd9Sstevel@tonic-gate * can't create any threads, and the calling 11657c478bd9Sstevel@tonic-gate * code (which looks at the current value of 11667c478bd9Sstevel@tonic-gate * mi->mi_max_threads, now non-zero) thinks we 11677c478bd9Sstevel@tonic-gate * can. 11687c478bd9Sstevel@tonic-gate * 11697c478bd9Sstevel@tonic-gate * So, because we're paranoid, we create threads 11707c478bd9Sstevel@tonic-gate * up to the maximum of the original and the 11717c478bd9Sstevel@tonic-gate * current value. This means that future 11727c478bd9Sstevel@tonic-gate * (debugger-induced) alterations of 11737c478bd9Sstevel@tonic-gate * mi->mi_max_threads are ignored for our 11747c478bd9Sstevel@tonic-gate * purposes, but who told them they could change 11757c478bd9Sstevel@tonic-gate * random values on a live kernel anyhow? 11767c478bd9Sstevel@tonic-gate */ 11777c478bd9Sstevel@tonic-gate if (mi->mi_threads < 11787c478bd9Sstevel@tonic-gate MAX(mi->mi_max_threads, max_threads)) { 11797c478bd9Sstevel@tonic-gate mi->mi_threads++; 11807c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 118150a83466Sjwahlig MI4_HOLD(mi); 11827c478bd9Sstevel@tonic-gate VFS_HOLD(vfsp); /* hold for new thread */ 11837c478bd9Sstevel@tonic-gate (void) zthread_create(NULL, 0, nfs4_async_start, 11847c478bd9Sstevel@tonic-gate vfsp, 0, minclsyspri); 11857c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 11867c478bd9Sstevel@tonic-gate } 11877c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_work_cv); 11887c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 11897c478bd9Sstevel@tonic-gate mi->mi_async_req_count--; 11907c478bd9Sstevel@tonic-gate } 11917c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 11927c478bd9Sstevel@tonic-gate } 11937c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 11947c478bd9Sstevel@tonic-gate 11957c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE, 11967c478bd9Sstevel@tonic-gate "nfs4_async_manager exiting for vfs %p\n", (void *)mi->mi_vfsp)); 11977c478bd9Sstevel@tonic-gate /* 11987c478bd9Sstevel@tonic-gate * Let everyone know we're done. 11997c478bd9Sstevel@tonic-gate */ 12007c478bd9Sstevel@tonic-gate mi->mi_manager_thread = NULL; 12017c478bd9Sstevel@tonic-gate /* 12027c478bd9Sstevel@tonic-gate * Wake up the inactive thread. 12037c478bd9Sstevel@tonic-gate */ 12047c478bd9Sstevel@tonic-gate cv_broadcast(&mi->mi_inact_req_cv); 12057c478bd9Sstevel@tonic-gate /* 12067c478bd9Sstevel@tonic-gate * Wake up anyone sitting in nfs4_async_manager_stop() 12077c478bd9Sstevel@tonic-gate */ 12087c478bd9Sstevel@tonic-gate cv_broadcast(&mi->mi_async_cv); 12097c478bd9Sstevel@tonic-gate /* 12107c478bd9Sstevel@tonic-gate * There is no explicit call to mutex_exit(&mi->mi_async_lock) 12117c478bd9Sstevel@tonic-gate * since CALLB_CPR_EXIT is actually responsible for releasing 12127c478bd9Sstevel@tonic-gate * 'mi_async_lock'. 12137c478bd9Sstevel@tonic-gate */ 12147c478bd9Sstevel@tonic-gate CALLB_CPR_EXIT(&cprinfo); 12157c478bd9Sstevel@tonic-gate VFS_RELE(vfsp); /* release thread's hold */ 121650a83466Sjwahlig MI4_RELE(mi); 12177c478bd9Sstevel@tonic-gate zthread_exit(); 12187c478bd9Sstevel@tonic-gate } 12197c478bd9Sstevel@tonic-gate 12207c478bd9Sstevel@tonic-gate /* 12217c478bd9Sstevel@tonic-gate * Signal (and wait for) the async manager thread to clean up and go away. 12227c478bd9Sstevel@tonic-gate */ 12237c478bd9Sstevel@tonic-gate void 12247c478bd9Sstevel@tonic-gate nfs4_async_manager_stop(vfs_t *vfsp) 12257c478bd9Sstevel@tonic-gate { 12267c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VFTOMI4(vfsp); 12277c478bd9Sstevel@tonic-gate 12287c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 12297c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 12307c478bd9Sstevel@tonic-gate mi->mi_flags |= MI4_ASYNC_MGR_STOP; 12317c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 12327c478bd9Sstevel@tonic-gate cv_broadcast(&mi->mi_async_reqs_cv); 12337c478bd9Sstevel@tonic-gate /* 12347c478bd9Sstevel@tonic-gate * Wait for the async manager thread to die. 12357c478bd9Sstevel@tonic-gate */ 12367c478bd9Sstevel@tonic-gate while (mi->mi_manager_thread != NULL) 12377c478bd9Sstevel@tonic-gate cv_wait(&mi->mi_async_cv, &mi->mi_async_lock); 12387c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 12397c478bd9Sstevel@tonic-gate } 12407c478bd9Sstevel@tonic-gate 12417c478bd9Sstevel@tonic-gate int 12427c478bd9Sstevel@tonic-gate nfs4_async_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, 1243b9238976Sth struct seg *seg, cred_t *cr, void (*readahead)(vnode_t *, 1244b9238976Sth u_offset_t, caddr_t, struct seg *, cred_t *)) 12457c478bd9Sstevel@tonic-gate { 12467c478bd9Sstevel@tonic-gate rnode4_t *rp; 12477c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 12487c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 12497c478bd9Sstevel@tonic-gate 12507c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 12517c478bd9Sstevel@tonic-gate ASSERT(rp->r_freef == NULL); 12527c478bd9Sstevel@tonic-gate 12537c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 12547c478bd9Sstevel@tonic-gate 12557c478bd9Sstevel@tonic-gate /* 12567c478bd9Sstevel@tonic-gate * If addr falls in a different segment, don't bother doing readahead. 12577c478bd9Sstevel@tonic-gate */ 12587c478bd9Sstevel@tonic-gate if (addr >= seg->s_base + seg->s_size) 12597c478bd9Sstevel@tonic-gate return (-1); 12607c478bd9Sstevel@tonic-gate 12617c478bd9Sstevel@tonic-gate /* 12627c478bd9Sstevel@tonic-gate * If we can't allocate a request structure, punt on the readahead. 12637c478bd9Sstevel@tonic-gate */ 12647c478bd9Sstevel@tonic-gate if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL) 12657c478bd9Sstevel@tonic-gate return (-1); 12667c478bd9Sstevel@tonic-gate 12677c478bd9Sstevel@tonic-gate /* 12687c478bd9Sstevel@tonic-gate * If a lock operation is pending, don't initiate any new 12697c478bd9Sstevel@tonic-gate * readaheads. Otherwise, bump r_count to indicate the new 12707c478bd9Sstevel@tonic-gate * asynchronous I/O. 12717c478bd9Sstevel@tonic-gate */ 12727c478bd9Sstevel@tonic-gate if (!nfs_rw_tryenter(&rp->r_lkserlock, RW_READER)) { 12737c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 12747c478bd9Sstevel@tonic-gate return (-1); 12757c478bd9Sstevel@tonic-gate } 12767c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 12777c478bd9Sstevel@tonic-gate rp->r_count++; 12787c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 12797c478bd9Sstevel@tonic-gate nfs_rw_exit(&rp->r_lkserlock); 12807c478bd9Sstevel@tonic-gate 12817c478bd9Sstevel@tonic-gate args->a_next = NULL; 12827c478bd9Sstevel@tonic-gate #ifdef DEBUG 12837c478bd9Sstevel@tonic-gate args->a_queuer = curthread; 12847c478bd9Sstevel@tonic-gate #endif 12857c478bd9Sstevel@tonic-gate VN_HOLD(vp); 12867c478bd9Sstevel@tonic-gate args->a_vp = vp; 12877c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 12887c478bd9Sstevel@tonic-gate crhold(cr); 12897c478bd9Sstevel@tonic-gate args->a_cred = cr; 12907c478bd9Sstevel@tonic-gate args->a_io = NFS4_READ_AHEAD; 12917c478bd9Sstevel@tonic-gate args->a_nfs4_readahead = readahead; 12927c478bd9Sstevel@tonic-gate args->a_nfs4_blkoff = blkoff; 12937c478bd9Sstevel@tonic-gate args->a_nfs4_seg = seg; 12947c478bd9Sstevel@tonic-gate args->a_nfs4_addr = addr; 12957c478bd9Sstevel@tonic-gate 12967c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 12977c478bd9Sstevel@tonic-gate 12987c478bd9Sstevel@tonic-gate /* 12997c478bd9Sstevel@tonic-gate * If asyncio has been disabled, don't bother readahead. 13007c478bd9Sstevel@tonic-gate */ 13017c478bd9Sstevel@tonic-gate if (mi->mi_max_threads == 0) { 13027c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 13037c478bd9Sstevel@tonic-gate goto noasync; 13047c478bd9Sstevel@tonic-gate } 13057c478bd9Sstevel@tonic-gate 13067c478bd9Sstevel@tonic-gate /* 13077c478bd9Sstevel@tonic-gate * Link request structure into the async list and 13087c478bd9Sstevel@tonic-gate * wakeup async thread to do the i/o. 13097c478bd9Sstevel@tonic-gate */ 13107c478bd9Sstevel@tonic-gate if (mi->mi_async_reqs[NFS4_READ_AHEAD] == NULL) { 13117c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_READ_AHEAD] = args; 13127c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_READ_AHEAD] = args; 13137c478bd9Sstevel@tonic-gate } else { 13147c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_READ_AHEAD]->a_next = args; 13157c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_READ_AHEAD] = args; 13167c478bd9Sstevel@tonic-gate } 13177c478bd9Sstevel@tonic-gate 13187c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 13197c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 13207c478bd9Sstevel@tonic-gate kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 13217c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 13227c478bd9Sstevel@tonic-gate } 13237c478bd9Sstevel@tonic-gate 13247c478bd9Sstevel@tonic-gate mi->mi_async_req_count++; 13257c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 13267c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_reqs_cv); 13277c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 13287c478bd9Sstevel@tonic-gate return (0); 13297c478bd9Sstevel@tonic-gate 13307c478bd9Sstevel@tonic-gate noasync: 13317c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 13327c478bd9Sstevel@tonic-gate rp->r_count--; 13337c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 13347c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 13357c478bd9Sstevel@tonic-gate VN_RELE(vp); 13367c478bd9Sstevel@tonic-gate crfree(cr); 13377c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 13387c478bd9Sstevel@tonic-gate return (-1); 13397c478bd9Sstevel@tonic-gate } 13407c478bd9Sstevel@tonic-gate 13417c478bd9Sstevel@tonic-gate /* 13427c478bd9Sstevel@tonic-gate * The async queues for each mounted file system are arranged as a 13437c478bd9Sstevel@tonic-gate * set of queues, one for each async i/o type. Requests are taken 13447c478bd9Sstevel@tonic-gate * from the queues in a round-robin fashion. A number of consecutive 13457c478bd9Sstevel@tonic-gate * requests are taken from each queue before moving on to the next 13467c478bd9Sstevel@tonic-gate * queue. This functionality may allow the NFS Version 2 server to do 13477c478bd9Sstevel@tonic-gate * write clustering, even if the client is mixing writes and reads 13487c478bd9Sstevel@tonic-gate * because it will take multiple write requests from the queue 13497c478bd9Sstevel@tonic-gate * before processing any of the other async i/o types. 13507c478bd9Sstevel@tonic-gate * 13517c478bd9Sstevel@tonic-gate * XXX The nfs4_async_start thread is unsafe in the light of the present 13527c478bd9Sstevel@tonic-gate * model defined by cpr to suspend the system. Specifically over the 13537c478bd9Sstevel@tonic-gate * wire calls are cpr-unsafe. The thread should be reevaluated in 13547c478bd9Sstevel@tonic-gate * case of future updates to the cpr model. 13557c478bd9Sstevel@tonic-gate */ 13567c478bd9Sstevel@tonic-gate static void 13577c478bd9Sstevel@tonic-gate nfs4_async_start(struct vfs *vfsp) 13587c478bd9Sstevel@tonic-gate { 13597c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 13607c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VFTOMI4(vfsp); 13617c478bd9Sstevel@tonic-gate clock_t time_left = 1; 13627c478bd9Sstevel@tonic-gate callb_cpr_t cprinfo; 13637c478bd9Sstevel@tonic-gate int i; 13647c478bd9Sstevel@tonic-gate extern int nfs_async_timeout; 13657c478bd9Sstevel@tonic-gate 13667c478bd9Sstevel@tonic-gate /* 13677c478bd9Sstevel@tonic-gate * Dynamic initialization of nfs_async_timeout to allow nfs to be 13687c478bd9Sstevel@tonic-gate * built in an implementation independent manner. 13697c478bd9Sstevel@tonic-gate */ 13707c478bd9Sstevel@tonic-gate if (nfs_async_timeout == -1) 13717c478bd9Sstevel@tonic-gate nfs_async_timeout = NFS_ASYNC_TIMEOUT; 13727c478bd9Sstevel@tonic-gate 13737c478bd9Sstevel@tonic-gate CALLB_CPR_INIT(&cprinfo, &mi->mi_async_lock, callb_generic_cpr, "nas"); 13747c478bd9Sstevel@tonic-gate 13757c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 13767c478bd9Sstevel@tonic-gate for (;;) { 13777c478bd9Sstevel@tonic-gate /* 13787c478bd9Sstevel@tonic-gate * Find the next queue containing an entry. We start 13797c478bd9Sstevel@tonic-gate * at the current queue pointer and then round robin 13807c478bd9Sstevel@tonic-gate * through all of them until we either find a non-empty 13817c478bd9Sstevel@tonic-gate * queue or have looked through all of them. 13827c478bd9Sstevel@tonic-gate */ 13837c478bd9Sstevel@tonic-gate for (i = 0; i < NFS4_ASYNC_TYPES; i++) { 13847c478bd9Sstevel@tonic-gate args = *mi->mi_async_curr; 13857c478bd9Sstevel@tonic-gate if (args != NULL) 13867c478bd9Sstevel@tonic-gate break; 13877c478bd9Sstevel@tonic-gate mi->mi_async_curr++; 13887c478bd9Sstevel@tonic-gate if (mi->mi_async_curr == 13897c478bd9Sstevel@tonic-gate &mi->mi_async_reqs[NFS4_ASYNC_TYPES]) 13907c478bd9Sstevel@tonic-gate mi->mi_async_curr = &mi->mi_async_reqs[0]; 13917c478bd9Sstevel@tonic-gate } 13927c478bd9Sstevel@tonic-gate /* 13937c478bd9Sstevel@tonic-gate * If we didn't find a entry, then block until woken up 13947c478bd9Sstevel@tonic-gate * again and then look through the queues again. 13957c478bd9Sstevel@tonic-gate */ 13967c478bd9Sstevel@tonic-gate if (args == NULL) { 13977c478bd9Sstevel@tonic-gate /* 13987c478bd9Sstevel@tonic-gate * Exiting is considered to be safe for CPR as well 13997c478bd9Sstevel@tonic-gate */ 14007c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_BEGIN(&cprinfo); 14017c478bd9Sstevel@tonic-gate 14027c478bd9Sstevel@tonic-gate /* 14037c478bd9Sstevel@tonic-gate * Wakeup thread waiting to unmount the file 14047c478bd9Sstevel@tonic-gate * system only if all async threads are inactive. 14057c478bd9Sstevel@tonic-gate * 14067c478bd9Sstevel@tonic-gate * If we've timed-out and there's nothing to do, 14077c478bd9Sstevel@tonic-gate * then get rid of this thread. 14087c478bd9Sstevel@tonic-gate */ 14097c478bd9Sstevel@tonic-gate if (mi->mi_max_threads == 0 || time_left <= 0) { 14107c478bd9Sstevel@tonic-gate if (--mi->mi_threads == 0) 14117c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_cv); 14127c478bd9Sstevel@tonic-gate CALLB_CPR_EXIT(&cprinfo); 14137c478bd9Sstevel@tonic-gate VFS_RELE(vfsp); /* release thread's hold */ 141450a83466Sjwahlig MI4_RELE(mi); 14157c478bd9Sstevel@tonic-gate zthread_exit(); 14167c478bd9Sstevel@tonic-gate /* NOTREACHED */ 14177c478bd9Sstevel@tonic-gate } 14187c478bd9Sstevel@tonic-gate time_left = cv_timedwait(&mi->mi_async_work_cv, 14197c478bd9Sstevel@tonic-gate &mi->mi_async_lock, nfs_async_timeout + lbolt); 14207c478bd9Sstevel@tonic-gate 14217c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_END(&cprinfo, &mi->mi_async_lock); 14227c478bd9Sstevel@tonic-gate 14237c478bd9Sstevel@tonic-gate continue; 14247c478bd9Sstevel@tonic-gate } else { 14257c478bd9Sstevel@tonic-gate time_left = 1; 14267c478bd9Sstevel@tonic-gate } 14277c478bd9Sstevel@tonic-gate 14287c478bd9Sstevel@tonic-gate /* 14297c478bd9Sstevel@tonic-gate * Remove the request from the async queue and then 14307c478bd9Sstevel@tonic-gate * update the current async request queue pointer. If 14317c478bd9Sstevel@tonic-gate * the current queue is empty or we have removed enough 14327c478bd9Sstevel@tonic-gate * consecutive entries from it, then reset the counter 14337c478bd9Sstevel@tonic-gate * for this queue and then move the current pointer to 14347c478bd9Sstevel@tonic-gate * the next queue. 14357c478bd9Sstevel@tonic-gate */ 14367c478bd9Sstevel@tonic-gate *mi->mi_async_curr = args->a_next; 14377c478bd9Sstevel@tonic-gate if (*mi->mi_async_curr == NULL || 14387c478bd9Sstevel@tonic-gate --mi->mi_async_clusters[args->a_io] == 0) { 14397c478bd9Sstevel@tonic-gate mi->mi_async_clusters[args->a_io] = 1440b9238976Sth mi->mi_async_init_clusters; 14417c478bd9Sstevel@tonic-gate mi->mi_async_curr++; 14427c478bd9Sstevel@tonic-gate if (mi->mi_async_curr == 14437c478bd9Sstevel@tonic-gate &mi->mi_async_reqs[NFS4_ASYNC_TYPES]) 14447c478bd9Sstevel@tonic-gate mi->mi_async_curr = &mi->mi_async_reqs[0]; 14457c478bd9Sstevel@tonic-gate } 14467c478bd9Sstevel@tonic-gate 14477c478bd9Sstevel@tonic-gate if (args->a_io != NFS4_INACTIVE && mi->mi_io_kstats) { 14487c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 14497c478bd9Sstevel@tonic-gate kstat_waitq_exit(KSTAT_IO_PTR(mi->mi_io_kstats)); 14507c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 14517c478bd9Sstevel@tonic-gate } 14527c478bd9Sstevel@tonic-gate 14537c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 14547c478bd9Sstevel@tonic-gate 14557c478bd9Sstevel@tonic-gate /* 14567c478bd9Sstevel@tonic-gate * Obtain arguments from the async request structure. 14577c478bd9Sstevel@tonic-gate */ 14587c478bd9Sstevel@tonic-gate if (args->a_io == NFS4_READ_AHEAD && mi->mi_max_threads > 0) { 14597c478bd9Sstevel@tonic-gate (*args->a_nfs4_readahead)(args->a_vp, 1460b9238976Sth args->a_nfs4_blkoff, args->a_nfs4_addr, 1461b9238976Sth args->a_nfs4_seg, args->a_cred); 14627c478bd9Sstevel@tonic-gate } else if (args->a_io == NFS4_PUTAPAGE) { 14637c478bd9Sstevel@tonic-gate (void) (*args->a_nfs4_putapage)(args->a_vp, 1464b9238976Sth args->a_nfs4_pp, args->a_nfs4_off, 1465b9238976Sth args->a_nfs4_len, args->a_nfs4_flags, 1466b9238976Sth args->a_cred); 14677c478bd9Sstevel@tonic-gate } else if (args->a_io == NFS4_PAGEIO) { 14687c478bd9Sstevel@tonic-gate (void) (*args->a_nfs4_pageio)(args->a_vp, 1469b9238976Sth args->a_nfs4_pp, args->a_nfs4_off, 1470b9238976Sth args->a_nfs4_len, args->a_nfs4_flags, 1471b9238976Sth args->a_cred); 14727c478bd9Sstevel@tonic-gate } else if (args->a_io == NFS4_READDIR) { 14737c478bd9Sstevel@tonic-gate (void) ((*args->a_nfs4_readdir)(args->a_vp, 1474b9238976Sth args->a_nfs4_rdc, args->a_cred)); 14757c478bd9Sstevel@tonic-gate } else if (args->a_io == NFS4_COMMIT) { 14767c478bd9Sstevel@tonic-gate (*args->a_nfs4_commit)(args->a_vp, args->a_nfs4_plist, 1477b9238976Sth args->a_nfs4_offset, args->a_nfs4_count, 1478b9238976Sth args->a_cred); 14797c478bd9Sstevel@tonic-gate } else if (args->a_io == NFS4_INACTIVE) { 14807c478bd9Sstevel@tonic-gate nfs4_inactive_otw(args->a_vp, args->a_cred); 14817c478bd9Sstevel@tonic-gate } 14827c478bd9Sstevel@tonic-gate 14837c478bd9Sstevel@tonic-gate /* 14847c478bd9Sstevel@tonic-gate * Now, release the vnode and free the credentials 14857c478bd9Sstevel@tonic-gate * structure. 14867c478bd9Sstevel@tonic-gate */ 14877c478bd9Sstevel@tonic-gate free_async_args4(args); 14887c478bd9Sstevel@tonic-gate /* 14897c478bd9Sstevel@tonic-gate * Reacquire the mutex because it will be needed above. 14907c478bd9Sstevel@tonic-gate */ 14917c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 14927c478bd9Sstevel@tonic-gate } 14937c478bd9Sstevel@tonic-gate } 14947c478bd9Sstevel@tonic-gate 14957c478bd9Sstevel@tonic-gate /* 14967c478bd9Sstevel@tonic-gate * nfs4_inactive_thread - look for vnodes that need over-the-wire calls as 14977c478bd9Sstevel@tonic-gate * part of VOP_INACTIVE. 14987c478bd9Sstevel@tonic-gate */ 14997c478bd9Sstevel@tonic-gate 15007c478bd9Sstevel@tonic-gate void 15017c478bd9Sstevel@tonic-gate nfs4_inactive_thread(mntinfo4_t *mi) 15027c478bd9Sstevel@tonic-gate { 15037c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 15047c478bd9Sstevel@tonic-gate callb_cpr_t cprinfo; 15057c478bd9Sstevel@tonic-gate vfs_t *vfsp = mi->mi_vfsp; 15067c478bd9Sstevel@tonic-gate 15077c478bd9Sstevel@tonic-gate CALLB_CPR_INIT(&cprinfo, &mi->mi_async_lock, callb_generic_cpr, 1508b9238976Sth "nfs4_inactive_thread"); 15097c478bd9Sstevel@tonic-gate 15107c478bd9Sstevel@tonic-gate for (;;) { 15117c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 15127c478bd9Sstevel@tonic-gate args = mi->mi_async_reqs[NFS4_INACTIVE]; 15137c478bd9Sstevel@tonic-gate if (args == NULL) { 15147c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 15157c478bd9Sstevel@tonic-gate /* 151650a83466Sjwahlig * We don't want to exit until the async manager is done 15177c478bd9Sstevel@tonic-gate * with its work; hence the check for mi_manager_thread 15187c478bd9Sstevel@tonic-gate * being NULL. 15197c478bd9Sstevel@tonic-gate * 15207c478bd9Sstevel@tonic-gate * The async manager thread will cv_broadcast() on 15217c478bd9Sstevel@tonic-gate * mi_inact_req_cv when it's done, at which point we'll 15227c478bd9Sstevel@tonic-gate * wake up and exit. 15237c478bd9Sstevel@tonic-gate */ 152450a83466Sjwahlig if (mi->mi_manager_thread == NULL) 15257c478bd9Sstevel@tonic-gate goto die; 15267c478bd9Sstevel@tonic-gate mi->mi_flags |= MI4_INACTIVE_IDLE; 15277c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 15287c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_cv); 15297c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_BEGIN(&cprinfo); 15307c478bd9Sstevel@tonic-gate cv_wait(&mi->mi_inact_req_cv, &mi->mi_async_lock); 15317c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_END(&cprinfo, &mi->mi_async_lock); 15327c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 15337c478bd9Sstevel@tonic-gate } else { 15347c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 15357c478bd9Sstevel@tonic-gate mi->mi_flags &= ~MI4_INACTIVE_IDLE; 15367c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 15377c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_INACTIVE] = args->a_next; 15387c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 15397c478bd9Sstevel@tonic-gate nfs4_inactive_otw(args->a_vp, args->a_cred); 15407c478bd9Sstevel@tonic-gate crfree(args->a_cred); 15417c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 15427c478bd9Sstevel@tonic-gate } 15437c478bd9Sstevel@tonic-gate } 15447c478bd9Sstevel@tonic-gate die: 15457c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 15467c478bd9Sstevel@tonic-gate mi->mi_inactive_thread = NULL; 15477c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_cv); 154850a83466Sjwahlig 15497c478bd9Sstevel@tonic-gate /* 15507c478bd9Sstevel@tonic-gate * There is no explicit call to mutex_exit(&mi->mi_async_lock) since 15517c478bd9Sstevel@tonic-gate * CALLB_CPR_EXIT is actually responsible for releasing 'mi_async_lock'. 15527c478bd9Sstevel@tonic-gate */ 15537c478bd9Sstevel@tonic-gate CALLB_CPR_EXIT(&cprinfo); 155450a83466Sjwahlig 15557c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE, 15567c478bd9Sstevel@tonic-gate "nfs4_inactive_thread exiting for vfs %p\n", (void *)vfsp)); 155750a83466Sjwahlig 155850a83466Sjwahlig MI4_RELE(mi); 15597c478bd9Sstevel@tonic-gate zthread_exit(); 15607c478bd9Sstevel@tonic-gate /* NOTREACHED */ 15617c478bd9Sstevel@tonic-gate } 15627c478bd9Sstevel@tonic-gate 15637c478bd9Sstevel@tonic-gate /* 15647c478bd9Sstevel@tonic-gate * nfs_async_stop: 15657c478bd9Sstevel@tonic-gate * Wait for all outstanding putpage operations and the inactive thread to 15667c478bd9Sstevel@tonic-gate * complete; nfs4_async_stop_sig() without interruptibility. 15677c478bd9Sstevel@tonic-gate */ 15687c478bd9Sstevel@tonic-gate void 15697c478bd9Sstevel@tonic-gate nfs4_async_stop(struct vfs *vfsp) 15707c478bd9Sstevel@tonic-gate { 15717c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VFTOMI4(vfsp); 15727c478bd9Sstevel@tonic-gate 15737c478bd9Sstevel@tonic-gate /* 15747c478bd9Sstevel@tonic-gate * Wait for all outstanding async operations to complete and for 15757c478bd9Sstevel@tonic-gate * worker threads to exit. 15767c478bd9Sstevel@tonic-gate */ 15777c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 15787c478bd9Sstevel@tonic-gate mi->mi_max_threads = 0; 15797c478bd9Sstevel@tonic-gate cv_broadcast(&mi->mi_async_work_cv); 15807c478bd9Sstevel@tonic-gate while (mi->mi_threads != 0) 15817c478bd9Sstevel@tonic-gate cv_wait(&mi->mi_async_cv, &mi->mi_async_lock); 15827c478bd9Sstevel@tonic-gate 15837c478bd9Sstevel@tonic-gate /* 15847c478bd9Sstevel@tonic-gate * Wait for the inactive thread to finish doing what it's doing. It 15857c478bd9Sstevel@tonic-gate * won't exit until the last reference to the vfs_t goes away. 15867c478bd9Sstevel@tonic-gate */ 15877c478bd9Sstevel@tonic-gate if (mi->mi_inactive_thread != NULL) { 15887c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 15897c478bd9Sstevel@tonic-gate while (!(mi->mi_flags & MI4_INACTIVE_IDLE) || 15907c478bd9Sstevel@tonic-gate (mi->mi_async_reqs[NFS4_INACTIVE] != NULL)) { 15917c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 15927c478bd9Sstevel@tonic-gate cv_wait(&mi->mi_async_cv, &mi->mi_async_lock); 15937c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 15947c478bd9Sstevel@tonic-gate } 15957c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 15967c478bd9Sstevel@tonic-gate } 15977c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 15987c478bd9Sstevel@tonic-gate } 15997c478bd9Sstevel@tonic-gate 16007c478bd9Sstevel@tonic-gate /* 16017c478bd9Sstevel@tonic-gate * nfs_async_stop_sig: 16027c478bd9Sstevel@tonic-gate * Wait for all outstanding putpage operations and the inactive thread to 16037c478bd9Sstevel@tonic-gate * complete. If a signal is delivered we will abort and return non-zero; 16047c478bd9Sstevel@tonic-gate * otherwise return 0. Since this routine is called from nfs4_unmount, we 1605da6c28aaSamw * need to make it interruptible. 16067c478bd9Sstevel@tonic-gate */ 16077c478bd9Sstevel@tonic-gate int 16087c478bd9Sstevel@tonic-gate nfs4_async_stop_sig(struct vfs *vfsp) 16097c478bd9Sstevel@tonic-gate { 16107c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VFTOMI4(vfsp); 16117c478bd9Sstevel@tonic-gate ushort_t omax; 16127c478bd9Sstevel@tonic-gate bool_t intr = FALSE; 16137c478bd9Sstevel@tonic-gate 16147c478bd9Sstevel@tonic-gate /* 16157c478bd9Sstevel@tonic-gate * Wait for all outstanding putpage operations to complete and for 16167c478bd9Sstevel@tonic-gate * worker threads to exit. 16177c478bd9Sstevel@tonic-gate */ 16187c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 16197c478bd9Sstevel@tonic-gate omax = mi->mi_max_threads; 16207c478bd9Sstevel@tonic-gate mi->mi_max_threads = 0; 16217c478bd9Sstevel@tonic-gate cv_broadcast(&mi->mi_async_work_cv); 16227c478bd9Sstevel@tonic-gate while (mi->mi_threads != 0) { 16237c478bd9Sstevel@tonic-gate if (!cv_wait_sig(&mi->mi_async_cv, &mi->mi_async_lock)) { 16247c478bd9Sstevel@tonic-gate intr = TRUE; 16257c478bd9Sstevel@tonic-gate goto interrupted; 16267c478bd9Sstevel@tonic-gate } 16277c478bd9Sstevel@tonic-gate } 16287c478bd9Sstevel@tonic-gate 16297c478bd9Sstevel@tonic-gate /* 16307c478bd9Sstevel@tonic-gate * Wait for the inactive thread to finish doing what it's doing. It 16317c478bd9Sstevel@tonic-gate * won't exit until the a last reference to the vfs_t goes away. 16327c478bd9Sstevel@tonic-gate */ 16337c478bd9Sstevel@tonic-gate if (mi->mi_inactive_thread != NULL) { 16347c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 16357c478bd9Sstevel@tonic-gate while (!(mi->mi_flags & MI4_INACTIVE_IDLE) || 16367c478bd9Sstevel@tonic-gate (mi->mi_async_reqs[NFS4_INACTIVE] != NULL)) { 16377c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 16387c478bd9Sstevel@tonic-gate if (!cv_wait_sig(&mi->mi_async_cv, 16397c478bd9Sstevel@tonic-gate &mi->mi_async_lock)) { 16407c478bd9Sstevel@tonic-gate intr = TRUE; 16417c478bd9Sstevel@tonic-gate goto interrupted; 16427c478bd9Sstevel@tonic-gate } 16437c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 16447c478bd9Sstevel@tonic-gate } 16457c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 16467c478bd9Sstevel@tonic-gate } 16477c478bd9Sstevel@tonic-gate interrupted: 16487c478bd9Sstevel@tonic-gate if (intr) 16497c478bd9Sstevel@tonic-gate mi->mi_max_threads = omax; 16507c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 16517c478bd9Sstevel@tonic-gate 16527c478bd9Sstevel@tonic-gate return (intr); 16537c478bd9Sstevel@tonic-gate } 16547c478bd9Sstevel@tonic-gate 16557c478bd9Sstevel@tonic-gate int 16567c478bd9Sstevel@tonic-gate nfs4_async_putapage(vnode_t *vp, page_t *pp, u_offset_t off, size_t len, 1657b9238976Sth int flags, cred_t *cr, int (*putapage)(vnode_t *, page_t *, 1658b9238976Sth u_offset_t, size_t, int, cred_t *)) 16597c478bd9Sstevel@tonic-gate { 16607c478bd9Sstevel@tonic-gate rnode4_t *rp; 16617c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 16627c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 16637c478bd9Sstevel@tonic-gate 16647c478bd9Sstevel@tonic-gate ASSERT(flags & B_ASYNC); 16657c478bd9Sstevel@tonic-gate ASSERT(vp->v_vfsp != NULL); 16667c478bd9Sstevel@tonic-gate 16677c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 16687c478bd9Sstevel@tonic-gate ASSERT(rp->r_count > 0); 16697c478bd9Sstevel@tonic-gate 16707c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 16717c478bd9Sstevel@tonic-gate 16727c478bd9Sstevel@tonic-gate /* 16737c478bd9Sstevel@tonic-gate * If we can't allocate a request structure, do the putpage 16747c478bd9Sstevel@tonic-gate * operation synchronously in this thread's context. 16757c478bd9Sstevel@tonic-gate */ 16767c478bd9Sstevel@tonic-gate if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL) 16777c478bd9Sstevel@tonic-gate goto noasync; 16787c478bd9Sstevel@tonic-gate 16797c478bd9Sstevel@tonic-gate args->a_next = NULL; 16807c478bd9Sstevel@tonic-gate #ifdef DEBUG 16817c478bd9Sstevel@tonic-gate args->a_queuer = curthread; 16827c478bd9Sstevel@tonic-gate #endif 16837c478bd9Sstevel@tonic-gate VN_HOLD(vp); 16847c478bd9Sstevel@tonic-gate args->a_vp = vp; 16857c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 16867c478bd9Sstevel@tonic-gate crhold(cr); 16877c478bd9Sstevel@tonic-gate args->a_cred = cr; 16887c478bd9Sstevel@tonic-gate args->a_io = NFS4_PUTAPAGE; 16897c478bd9Sstevel@tonic-gate args->a_nfs4_putapage = putapage; 16907c478bd9Sstevel@tonic-gate args->a_nfs4_pp = pp; 16917c478bd9Sstevel@tonic-gate args->a_nfs4_off = off; 16927c478bd9Sstevel@tonic-gate args->a_nfs4_len = (uint_t)len; 16937c478bd9Sstevel@tonic-gate args->a_nfs4_flags = flags; 16947c478bd9Sstevel@tonic-gate 16957c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 16967c478bd9Sstevel@tonic-gate 16977c478bd9Sstevel@tonic-gate /* 16987c478bd9Sstevel@tonic-gate * If asyncio has been disabled, then make a synchronous request. 16997c478bd9Sstevel@tonic-gate * This check is done a second time in case async io was diabled 17007c478bd9Sstevel@tonic-gate * while this thread was blocked waiting for memory pressure to 17017c478bd9Sstevel@tonic-gate * reduce or for the queue to drain. 17027c478bd9Sstevel@tonic-gate */ 17037c478bd9Sstevel@tonic-gate if (mi->mi_max_threads == 0) { 17047c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 17057c478bd9Sstevel@tonic-gate 17067c478bd9Sstevel@tonic-gate VN_RELE(vp); 17077c478bd9Sstevel@tonic-gate crfree(cr); 17087c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 17097c478bd9Sstevel@tonic-gate goto noasync; 17107c478bd9Sstevel@tonic-gate } 17117c478bd9Sstevel@tonic-gate 17127c478bd9Sstevel@tonic-gate /* 17137c478bd9Sstevel@tonic-gate * Link request structure into the async list and 17147c478bd9Sstevel@tonic-gate * wakeup async thread to do the i/o. 17157c478bd9Sstevel@tonic-gate */ 17167c478bd9Sstevel@tonic-gate if (mi->mi_async_reqs[NFS4_PUTAPAGE] == NULL) { 17177c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_PUTAPAGE] = args; 17187c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_PUTAPAGE] = args; 17197c478bd9Sstevel@tonic-gate } else { 17207c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_PUTAPAGE]->a_next = args; 17217c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_PUTAPAGE] = args; 17227c478bd9Sstevel@tonic-gate } 17237c478bd9Sstevel@tonic-gate 17247c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 17257c478bd9Sstevel@tonic-gate rp->r_count++; 17267c478bd9Sstevel@tonic-gate rp->r_awcount++; 17277c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 17287c478bd9Sstevel@tonic-gate 17297c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 17307c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 17317c478bd9Sstevel@tonic-gate kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 17327c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 17337c478bd9Sstevel@tonic-gate } 17347c478bd9Sstevel@tonic-gate 17357c478bd9Sstevel@tonic-gate mi->mi_async_req_count++; 17367c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 17377c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_reqs_cv); 17387c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 17397c478bd9Sstevel@tonic-gate return (0); 17407c478bd9Sstevel@tonic-gate 17417c478bd9Sstevel@tonic-gate noasync: 17427c478bd9Sstevel@tonic-gate 17437c478bd9Sstevel@tonic-gate if (curproc == proc_pageout || curproc == proc_fsflush || 1744108322fbScarlsonj nfs_zone() == mi->mi_zone) { 17457c478bd9Sstevel@tonic-gate /* 17467c478bd9Sstevel@tonic-gate * If we get here in the context of the pageout/fsflush, 17477c478bd9Sstevel@tonic-gate * or we have run out of memory or we're attempting to 17487c478bd9Sstevel@tonic-gate * unmount we refuse to do a sync write, because this may 17497c478bd9Sstevel@tonic-gate * hang pageout/fsflush and the machine. In this case, 17507c478bd9Sstevel@tonic-gate * we just re-mark the page as dirty and punt on the page. 17517c478bd9Sstevel@tonic-gate * 17527c478bd9Sstevel@tonic-gate * Make sure B_FORCE isn't set. We can re-mark the 17537c478bd9Sstevel@tonic-gate * pages as dirty and unlock the pages in one swoop by 17547c478bd9Sstevel@tonic-gate * passing in B_ERROR to pvn_write_done(). However, 17557c478bd9Sstevel@tonic-gate * we should make sure B_FORCE isn't set - we don't 17567c478bd9Sstevel@tonic-gate * want the page tossed before it gets written out. 17577c478bd9Sstevel@tonic-gate */ 17587c478bd9Sstevel@tonic-gate if (flags & B_FORCE) 17597c478bd9Sstevel@tonic-gate flags &= ~(B_INVAL | B_FORCE); 17607c478bd9Sstevel@tonic-gate pvn_write_done(pp, flags | B_ERROR); 17617c478bd9Sstevel@tonic-gate return (0); 17627c478bd9Sstevel@tonic-gate } 17637c478bd9Sstevel@tonic-gate 17647c478bd9Sstevel@tonic-gate /* 1765108322fbScarlsonj * We'll get here only if (nfs_zone() != mi->mi_zone) 17667c478bd9Sstevel@tonic-gate * which means that this was a cross-zone sync putpage. 17677c478bd9Sstevel@tonic-gate * 17687c478bd9Sstevel@tonic-gate * We pass in B_ERROR to pvn_write_done() to re-mark the pages 17697c478bd9Sstevel@tonic-gate * as dirty and unlock them. 17707c478bd9Sstevel@tonic-gate * 17717c478bd9Sstevel@tonic-gate * We don't want to clear B_FORCE here as the caller presumably 17727c478bd9Sstevel@tonic-gate * knows what they're doing if they set it. 17737c478bd9Sstevel@tonic-gate */ 17747c478bd9Sstevel@tonic-gate pvn_write_done(pp, flags | B_ERROR); 17757c478bd9Sstevel@tonic-gate return (EPERM); 17767c478bd9Sstevel@tonic-gate } 17777c478bd9Sstevel@tonic-gate 17787c478bd9Sstevel@tonic-gate int 17797c478bd9Sstevel@tonic-gate nfs4_async_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len, 1780b9238976Sth int flags, cred_t *cr, int (*pageio)(vnode_t *, page_t *, u_offset_t, 1781b9238976Sth size_t, int, cred_t *)) 17827c478bd9Sstevel@tonic-gate { 17837c478bd9Sstevel@tonic-gate rnode4_t *rp; 17847c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 17857c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 17867c478bd9Sstevel@tonic-gate 17877c478bd9Sstevel@tonic-gate ASSERT(flags & B_ASYNC); 17887c478bd9Sstevel@tonic-gate ASSERT(vp->v_vfsp != NULL); 17897c478bd9Sstevel@tonic-gate 17907c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 17917c478bd9Sstevel@tonic-gate ASSERT(rp->r_count > 0); 17927c478bd9Sstevel@tonic-gate 17937c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 17947c478bd9Sstevel@tonic-gate 17957c478bd9Sstevel@tonic-gate /* 17967c478bd9Sstevel@tonic-gate * If we can't allocate a request structure, do the pageio 17977c478bd9Sstevel@tonic-gate * request synchronously in this thread's context. 17987c478bd9Sstevel@tonic-gate */ 17997c478bd9Sstevel@tonic-gate if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL) 18007c478bd9Sstevel@tonic-gate goto noasync; 18017c478bd9Sstevel@tonic-gate 18027c478bd9Sstevel@tonic-gate args->a_next = NULL; 18037c478bd9Sstevel@tonic-gate #ifdef DEBUG 18047c478bd9Sstevel@tonic-gate args->a_queuer = curthread; 18057c478bd9Sstevel@tonic-gate #endif 18067c478bd9Sstevel@tonic-gate VN_HOLD(vp); 18077c478bd9Sstevel@tonic-gate args->a_vp = vp; 18087c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 18097c478bd9Sstevel@tonic-gate crhold(cr); 18107c478bd9Sstevel@tonic-gate args->a_cred = cr; 18117c478bd9Sstevel@tonic-gate args->a_io = NFS4_PAGEIO; 18127c478bd9Sstevel@tonic-gate args->a_nfs4_pageio = pageio; 18137c478bd9Sstevel@tonic-gate args->a_nfs4_pp = pp; 18147c478bd9Sstevel@tonic-gate args->a_nfs4_off = io_off; 18157c478bd9Sstevel@tonic-gate args->a_nfs4_len = (uint_t)io_len; 18167c478bd9Sstevel@tonic-gate args->a_nfs4_flags = flags; 18177c478bd9Sstevel@tonic-gate 18187c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 18197c478bd9Sstevel@tonic-gate 18207c478bd9Sstevel@tonic-gate /* 18217c478bd9Sstevel@tonic-gate * If asyncio has been disabled, then make a synchronous request. 18227c478bd9Sstevel@tonic-gate * This check is done a second time in case async io was diabled 18237c478bd9Sstevel@tonic-gate * while this thread was blocked waiting for memory pressure to 18247c478bd9Sstevel@tonic-gate * reduce or for the queue to drain. 18257c478bd9Sstevel@tonic-gate */ 18267c478bd9Sstevel@tonic-gate if (mi->mi_max_threads == 0) { 18277c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 18287c478bd9Sstevel@tonic-gate 18297c478bd9Sstevel@tonic-gate VN_RELE(vp); 18307c478bd9Sstevel@tonic-gate crfree(cr); 18317c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 18327c478bd9Sstevel@tonic-gate goto noasync; 18337c478bd9Sstevel@tonic-gate } 18347c478bd9Sstevel@tonic-gate 18357c478bd9Sstevel@tonic-gate /* 18367c478bd9Sstevel@tonic-gate * Link request structure into the async list and 18377c478bd9Sstevel@tonic-gate * wakeup async thread to do the i/o. 18387c478bd9Sstevel@tonic-gate */ 18397c478bd9Sstevel@tonic-gate if (mi->mi_async_reqs[NFS4_PAGEIO] == NULL) { 18407c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_PAGEIO] = args; 18417c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_PAGEIO] = args; 18427c478bd9Sstevel@tonic-gate } else { 18437c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_PAGEIO]->a_next = args; 18447c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_PAGEIO] = args; 18457c478bd9Sstevel@tonic-gate } 18467c478bd9Sstevel@tonic-gate 18477c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 18487c478bd9Sstevel@tonic-gate rp->r_count++; 18497c478bd9Sstevel@tonic-gate rp->r_awcount++; 18507c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 18517c478bd9Sstevel@tonic-gate 18527c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 18537c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 18547c478bd9Sstevel@tonic-gate kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 18557c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 18567c478bd9Sstevel@tonic-gate } 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate mi->mi_async_req_count++; 18597c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 18607c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_reqs_cv); 18617c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 18627c478bd9Sstevel@tonic-gate return (0); 18637c478bd9Sstevel@tonic-gate 18647c478bd9Sstevel@tonic-gate noasync: 18657c478bd9Sstevel@tonic-gate /* 18667c478bd9Sstevel@tonic-gate * If we can't do it ASYNC, for reads we do nothing (but cleanup 18677c478bd9Sstevel@tonic-gate * the page list), for writes we do it synchronously, except for 18687c478bd9Sstevel@tonic-gate * proc_pageout/proc_fsflush as described below. 18697c478bd9Sstevel@tonic-gate */ 18707c478bd9Sstevel@tonic-gate if (flags & B_READ) { 18717c478bd9Sstevel@tonic-gate pvn_read_done(pp, flags | B_ERROR); 18727c478bd9Sstevel@tonic-gate return (0); 18737c478bd9Sstevel@tonic-gate } 18747c478bd9Sstevel@tonic-gate 18757c478bd9Sstevel@tonic-gate if (curproc == proc_pageout || curproc == proc_fsflush) { 18767c478bd9Sstevel@tonic-gate /* 18777c478bd9Sstevel@tonic-gate * If we get here in the context of the pageout/fsflush, 18787c478bd9Sstevel@tonic-gate * we refuse to do a sync write, because this may hang 18797c478bd9Sstevel@tonic-gate * pageout/fsflush (and the machine). In this case, we just 18807c478bd9Sstevel@tonic-gate * re-mark the page as dirty and punt on the page. 18817c478bd9Sstevel@tonic-gate * 18827c478bd9Sstevel@tonic-gate * Make sure B_FORCE isn't set. We can re-mark the 18837c478bd9Sstevel@tonic-gate * pages as dirty and unlock the pages in one swoop by 18847c478bd9Sstevel@tonic-gate * passing in B_ERROR to pvn_write_done(). However, 18857c478bd9Sstevel@tonic-gate * we should make sure B_FORCE isn't set - we don't 18867c478bd9Sstevel@tonic-gate * want the page tossed before it gets written out. 18877c478bd9Sstevel@tonic-gate */ 18887c478bd9Sstevel@tonic-gate if (flags & B_FORCE) 18897c478bd9Sstevel@tonic-gate flags &= ~(B_INVAL | B_FORCE); 18907c478bd9Sstevel@tonic-gate pvn_write_done(pp, flags | B_ERROR); 18917c478bd9Sstevel@tonic-gate return (0); 18927c478bd9Sstevel@tonic-gate } 18937c478bd9Sstevel@tonic-gate 1894108322fbScarlsonj if (nfs_zone() != mi->mi_zone) { 18957c478bd9Sstevel@tonic-gate /* 18967c478bd9Sstevel@tonic-gate * So this was a cross-zone sync pageio. We pass in B_ERROR 18977c478bd9Sstevel@tonic-gate * to pvn_write_done() to re-mark the pages as dirty and unlock 18987c478bd9Sstevel@tonic-gate * them. 18997c478bd9Sstevel@tonic-gate * 19007c478bd9Sstevel@tonic-gate * We don't want to clear B_FORCE here as the caller presumably 19017c478bd9Sstevel@tonic-gate * knows what they're doing if they set it. 19027c478bd9Sstevel@tonic-gate */ 19037c478bd9Sstevel@tonic-gate pvn_write_done(pp, flags | B_ERROR); 19047c478bd9Sstevel@tonic-gate return (EPERM); 19057c478bd9Sstevel@tonic-gate } 19067c478bd9Sstevel@tonic-gate return ((*pageio)(vp, pp, io_off, io_len, flags, cr)); 19077c478bd9Sstevel@tonic-gate } 19087c478bd9Sstevel@tonic-gate 19097c478bd9Sstevel@tonic-gate void 19107c478bd9Sstevel@tonic-gate nfs4_async_readdir(vnode_t *vp, rddir4_cache *rdc, cred_t *cr, 1911b9238976Sth int (*readdir)(vnode_t *, rddir4_cache *, cred_t *)) 19127c478bd9Sstevel@tonic-gate { 19137c478bd9Sstevel@tonic-gate rnode4_t *rp; 19147c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 19157c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 19167c478bd9Sstevel@tonic-gate 19177c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 19187c478bd9Sstevel@tonic-gate ASSERT(rp->r_freef == NULL); 19197c478bd9Sstevel@tonic-gate 19207c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 19217c478bd9Sstevel@tonic-gate 19227c478bd9Sstevel@tonic-gate /* 19237c478bd9Sstevel@tonic-gate * If we can't allocate a request structure, skip the readdir. 19247c478bd9Sstevel@tonic-gate */ 19257c478bd9Sstevel@tonic-gate if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL) 19267c478bd9Sstevel@tonic-gate goto noasync; 19277c478bd9Sstevel@tonic-gate 19287c478bd9Sstevel@tonic-gate args->a_next = NULL; 19297c478bd9Sstevel@tonic-gate #ifdef DEBUG 19307c478bd9Sstevel@tonic-gate args->a_queuer = curthread; 19317c478bd9Sstevel@tonic-gate #endif 19327c478bd9Sstevel@tonic-gate VN_HOLD(vp); 19337c478bd9Sstevel@tonic-gate args->a_vp = vp; 19347c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 19357c478bd9Sstevel@tonic-gate crhold(cr); 19367c478bd9Sstevel@tonic-gate args->a_cred = cr; 19377c478bd9Sstevel@tonic-gate args->a_io = NFS4_READDIR; 19387c478bd9Sstevel@tonic-gate args->a_nfs4_readdir = readdir; 19397c478bd9Sstevel@tonic-gate args->a_nfs4_rdc = rdc; 19407c478bd9Sstevel@tonic-gate 19417c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 19427c478bd9Sstevel@tonic-gate 19437c478bd9Sstevel@tonic-gate /* 19447c478bd9Sstevel@tonic-gate * If asyncio has been disabled, then skip this request 19457c478bd9Sstevel@tonic-gate */ 19467c478bd9Sstevel@tonic-gate if (mi->mi_max_threads == 0) { 19477c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 19487c478bd9Sstevel@tonic-gate 19497c478bd9Sstevel@tonic-gate VN_RELE(vp); 19507c478bd9Sstevel@tonic-gate crfree(cr); 19517c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 19527c478bd9Sstevel@tonic-gate goto noasync; 19537c478bd9Sstevel@tonic-gate } 19547c478bd9Sstevel@tonic-gate 19557c478bd9Sstevel@tonic-gate /* 19567c478bd9Sstevel@tonic-gate * Link request structure into the async list and 19577c478bd9Sstevel@tonic-gate * wakeup async thread to do the i/o. 19587c478bd9Sstevel@tonic-gate */ 19597c478bd9Sstevel@tonic-gate if (mi->mi_async_reqs[NFS4_READDIR] == NULL) { 19607c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_READDIR] = args; 19617c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_READDIR] = args; 19627c478bd9Sstevel@tonic-gate } else { 19637c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_READDIR]->a_next = args; 19647c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_READDIR] = args; 19657c478bd9Sstevel@tonic-gate } 19667c478bd9Sstevel@tonic-gate 19677c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 19687c478bd9Sstevel@tonic-gate rp->r_count++; 19697c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 19707c478bd9Sstevel@tonic-gate 19717c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 19727c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 19737c478bd9Sstevel@tonic-gate kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 19747c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 19757c478bd9Sstevel@tonic-gate } 19767c478bd9Sstevel@tonic-gate 19777c478bd9Sstevel@tonic-gate mi->mi_async_req_count++; 19787c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 19797c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_reqs_cv); 19807c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 19817c478bd9Sstevel@tonic-gate return; 19827c478bd9Sstevel@tonic-gate 19837c478bd9Sstevel@tonic-gate noasync: 19847c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 19857c478bd9Sstevel@tonic-gate rdc->entries = NULL; 19867c478bd9Sstevel@tonic-gate /* 19877c478bd9Sstevel@tonic-gate * Indicate that no one is trying to fill this entry and 19887c478bd9Sstevel@tonic-gate * it still needs to be filled. 19897c478bd9Sstevel@tonic-gate */ 19907c478bd9Sstevel@tonic-gate rdc->flags &= ~RDDIR; 19917c478bd9Sstevel@tonic-gate rdc->flags |= RDDIRREQ; 19927c478bd9Sstevel@tonic-gate rddir4_cache_rele(rp, rdc); 19937c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 19947c478bd9Sstevel@tonic-gate } 19957c478bd9Sstevel@tonic-gate 19967c478bd9Sstevel@tonic-gate void 19977c478bd9Sstevel@tonic-gate nfs4_async_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count, 1998b9238976Sth cred_t *cr, void (*commit)(vnode_t *, page_t *, offset3, count3, 1999b9238976Sth cred_t *)) 20007c478bd9Sstevel@tonic-gate { 20017c478bd9Sstevel@tonic-gate rnode4_t *rp; 20027c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 20037c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 20047c478bd9Sstevel@tonic-gate page_t *pp; 20057c478bd9Sstevel@tonic-gate 20067c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 20077c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 20087c478bd9Sstevel@tonic-gate 20097c478bd9Sstevel@tonic-gate /* 20107c478bd9Sstevel@tonic-gate * If we can't allocate a request structure, do the commit 20117c478bd9Sstevel@tonic-gate * operation synchronously in this thread's context. 20127c478bd9Sstevel@tonic-gate */ 20137c478bd9Sstevel@tonic-gate if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL) 20147c478bd9Sstevel@tonic-gate goto noasync; 20157c478bd9Sstevel@tonic-gate 20167c478bd9Sstevel@tonic-gate args->a_next = NULL; 20177c478bd9Sstevel@tonic-gate #ifdef DEBUG 20187c478bd9Sstevel@tonic-gate args->a_queuer = curthread; 20197c478bd9Sstevel@tonic-gate #endif 20207c478bd9Sstevel@tonic-gate VN_HOLD(vp); 20217c478bd9Sstevel@tonic-gate args->a_vp = vp; 20227c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 20237c478bd9Sstevel@tonic-gate crhold(cr); 20247c478bd9Sstevel@tonic-gate args->a_cred = cr; 20257c478bd9Sstevel@tonic-gate args->a_io = NFS4_COMMIT; 20267c478bd9Sstevel@tonic-gate args->a_nfs4_commit = commit; 20277c478bd9Sstevel@tonic-gate args->a_nfs4_plist = plist; 20287c478bd9Sstevel@tonic-gate args->a_nfs4_offset = offset; 20297c478bd9Sstevel@tonic-gate args->a_nfs4_count = count; 20307c478bd9Sstevel@tonic-gate 20317c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 20327c478bd9Sstevel@tonic-gate 20337c478bd9Sstevel@tonic-gate /* 20347c478bd9Sstevel@tonic-gate * If asyncio has been disabled, then make a synchronous request. 20357c478bd9Sstevel@tonic-gate * This check is done a second time in case async io was diabled 20367c478bd9Sstevel@tonic-gate * while this thread was blocked waiting for memory pressure to 20377c478bd9Sstevel@tonic-gate * reduce or for the queue to drain. 20387c478bd9Sstevel@tonic-gate */ 20397c478bd9Sstevel@tonic-gate if (mi->mi_max_threads == 0) { 20407c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 20417c478bd9Sstevel@tonic-gate 20427c478bd9Sstevel@tonic-gate VN_RELE(vp); 20437c478bd9Sstevel@tonic-gate crfree(cr); 20447c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 20457c478bd9Sstevel@tonic-gate goto noasync; 20467c478bd9Sstevel@tonic-gate } 20477c478bd9Sstevel@tonic-gate 20487c478bd9Sstevel@tonic-gate /* 20497c478bd9Sstevel@tonic-gate * Link request structure into the async list and 20507c478bd9Sstevel@tonic-gate * wakeup async thread to do the i/o. 20517c478bd9Sstevel@tonic-gate */ 20527c478bd9Sstevel@tonic-gate if (mi->mi_async_reqs[NFS4_COMMIT] == NULL) { 20537c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_COMMIT] = args; 20547c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_COMMIT] = args; 20557c478bd9Sstevel@tonic-gate } else { 20567c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_COMMIT]->a_next = args; 20577c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_COMMIT] = args; 20587c478bd9Sstevel@tonic-gate } 20597c478bd9Sstevel@tonic-gate 20607c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 20617c478bd9Sstevel@tonic-gate rp->r_count++; 20627c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 20637c478bd9Sstevel@tonic-gate 20647c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 20657c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 20667c478bd9Sstevel@tonic-gate kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats)); 20677c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 20687c478bd9Sstevel@tonic-gate } 20697c478bd9Sstevel@tonic-gate 20707c478bd9Sstevel@tonic-gate mi->mi_async_req_count++; 20717c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 20727c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_reqs_cv); 20737c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 20747c478bd9Sstevel@tonic-gate return; 20757c478bd9Sstevel@tonic-gate 20767c478bd9Sstevel@tonic-gate noasync: 20777c478bd9Sstevel@tonic-gate if (curproc == proc_pageout || curproc == proc_fsflush || 2078108322fbScarlsonj nfs_zone() != mi->mi_zone) { 20797c478bd9Sstevel@tonic-gate while (plist != NULL) { 20807c478bd9Sstevel@tonic-gate pp = plist; 20817c478bd9Sstevel@tonic-gate page_sub(&plist, pp); 20827c478bd9Sstevel@tonic-gate pp->p_fsdata = C_COMMIT; 20837c478bd9Sstevel@tonic-gate page_unlock(pp); 20847c478bd9Sstevel@tonic-gate } 20857c478bd9Sstevel@tonic-gate return; 20867c478bd9Sstevel@tonic-gate } 20877c478bd9Sstevel@tonic-gate (*commit)(vp, plist, offset, count, cr); 20887c478bd9Sstevel@tonic-gate } 20897c478bd9Sstevel@tonic-gate 20907c478bd9Sstevel@tonic-gate /* 20917c478bd9Sstevel@tonic-gate * nfs4_async_inactive - hand off a VOP_INACTIVE call to a thread. The 20927c478bd9Sstevel@tonic-gate * reference to the vnode is handed over to the thread; the caller should 20937c478bd9Sstevel@tonic-gate * no longer refer to the vnode. 20947c478bd9Sstevel@tonic-gate * 20957c478bd9Sstevel@tonic-gate * Unlike most of the async routines, this handoff is needed for 20967c478bd9Sstevel@tonic-gate * correctness reasons, not just performance. So doing operations in the 20977c478bd9Sstevel@tonic-gate * context of the current thread is not an option. 20987c478bd9Sstevel@tonic-gate */ 20997c478bd9Sstevel@tonic-gate void 21007c478bd9Sstevel@tonic-gate nfs4_async_inactive(vnode_t *vp, cred_t *cr) 21017c478bd9Sstevel@tonic-gate { 21027c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 21037c478bd9Sstevel@tonic-gate struct nfs4_async_reqs *args; 21047c478bd9Sstevel@tonic-gate boolean_t signal_inactive_thread = B_FALSE; 21057c478bd9Sstevel@tonic-gate 21067c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 21077c478bd9Sstevel@tonic-gate 21087c478bd9Sstevel@tonic-gate args = kmem_alloc(sizeof (*args), KM_SLEEP); 21097c478bd9Sstevel@tonic-gate args->a_next = NULL; 21107c478bd9Sstevel@tonic-gate #ifdef DEBUG 21117c478bd9Sstevel@tonic-gate args->a_queuer = curthread; 21127c478bd9Sstevel@tonic-gate #endif 21137c478bd9Sstevel@tonic-gate args->a_vp = vp; 21147c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 21157c478bd9Sstevel@tonic-gate crhold(cr); 21167c478bd9Sstevel@tonic-gate args->a_cred = cr; 21177c478bd9Sstevel@tonic-gate args->a_io = NFS4_INACTIVE; 21187c478bd9Sstevel@tonic-gate 21197c478bd9Sstevel@tonic-gate /* 21207c478bd9Sstevel@tonic-gate * Note that we don't check mi->mi_max_threads here, since we 21217c478bd9Sstevel@tonic-gate * *need* to get rid of this vnode regardless of whether someone 21227c478bd9Sstevel@tonic-gate * set nfs4_max_threads to zero in /etc/system. 21237c478bd9Sstevel@tonic-gate * 21247c478bd9Sstevel@tonic-gate * The manager thread knows about this and is willing to create 2125da6c28aaSamw * at least one thread to accommodate us. 21267c478bd9Sstevel@tonic-gate */ 21277c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 21287c478bd9Sstevel@tonic-gate if (mi->mi_inactive_thread == NULL) { 21297c478bd9Sstevel@tonic-gate rnode4_t *rp; 21307c478bd9Sstevel@tonic-gate vnode_t *unldvp = NULL; 21317c478bd9Sstevel@tonic-gate char *unlname; 21327c478bd9Sstevel@tonic-gate cred_t *unlcred; 21337c478bd9Sstevel@tonic-gate 21347c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 21357c478bd9Sstevel@tonic-gate /* 21367c478bd9Sstevel@tonic-gate * We just need to free up the memory associated with the 21377c478bd9Sstevel@tonic-gate * vnode, which can be safely done from within the current 21387c478bd9Sstevel@tonic-gate * context. 21397c478bd9Sstevel@tonic-gate */ 21407c478bd9Sstevel@tonic-gate crfree(cr); /* drop our reference */ 21417c478bd9Sstevel@tonic-gate kmem_free(args, sizeof (*args)); 21427c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 21437c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 21447c478bd9Sstevel@tonic-gate if (rp->r_unldvp != NULL) { 21457c478bd9Sstevel@tonic-gate unldvp = rp->r_unldvp; 21467c478bd9Sstevel@tonic-gate rp->r_unldvp = NULL; 21477c478bd9Sstevel@tonic-gate unlname = rp->r_unlname; 21487c478bd9Sstevel@tonic-gate rp->r_unlname = NULL; 21497c478bd9Sstevel@tonic-gate unlcred = rp->r_unlcred; 21507c478bd9Sstevel@tonic-gate rp->r_unlcred = NULL; 21517c478bd9Sstevel@tonic-gate } 21527c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 21537c478bd9Sstevel@tonic-gate /* 21547c478bd9Sstevel@tonic-gate * No need to explicitly throw away any cached pages. The 21557c478bd9Sstevel@tonic-gate * eventual r4inactive() will attempt a synchronous 21567c478bd9Sstevel@tonic-gate * VOP_PUTPAGE() which will immediately fail since the request 21577c478bd9Sstevel@tonic-gate * is coming from the wrong zone, and then will proceed to call 21587c478bd9Sstevel@tonic-gate * nfs4_invalidate_pages() which will clean things up for us. 21597c478bd9Sstevel@tonic-gate * 21607c478bd9Sstevel@tonic-gate * Throw away the delegation here so rp4_addfree()'s attempt to 21617c478bd9Sstevel@tonic-gate * return any existing delegations becomes a no-op. 21627c478bd9Sstevel@tonic-gate */ 216350a83466Sjwahlig if (rp->r_deleg_type != OPEN_DELEGATE_NONE) { 216450a83466Sjwahlig (void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 2165b9238976Sth FALSE); 21667c478bd9Sstevel@tonic-gate (void) nfs4delegreturn(rp, NFS4_DR_DISCARD); 216750a83466Sjwahlig nfs_rw_exit(&mi->mi_recovlock); 216850a83466Sjwahlig } 21697c478bd9Sstevel@tonic-gate nfs4_clear_open_streams(rp); 21707c478bd9Sstevel@tonic-gate 21717c478bd9Sstevel@tonic-gate rp4_addfree(rp, cr); 21727c478bd9Sstevel@tonic-gate if (unldvp != NULL) { 21737c478bd9Sstevel@tonic-gate kmem_free(unlname, MAXNAMELEN); 21747c478bd9Sstevel@tonic-gate VN_RELE(unldvp); 21757c478bd9Sstevel@tonic-gate crfree(unlcred); 21767c478bd9Sstevel@tonic-gate } 21777c478bd9Sstevel@tonic-gate return; 21787c478bd9Sstevel@tonic-gate } 21797c478bd9Sstevel@tonic-gate 21807c478bd9Sstevel@tonic-gate if (mi->mi_manager_thread == NULL) { 21817c478bd9Sstevel@tonic-gate /* 21827c478bd9Sstevel@tonic-gate * We want to talk to the inactive thread. 21837c478bd9Sstevel@tonic-gate */ 21847c478bd9Sstevel@tonic-gate signal_inactive_thread = B_TRUE; 21857c478bd9Sstevel@tonic-gate } 21867c478bd9Sstevel@tonic-gate 21877c478bd9Sstevel@tonic-gate /* 21887c478bd9Sstevel@tonic-gate * Enqueue the vnode and wake up either the special thread (empty 21897c478bd9Sstevel@tonic-gate * list) or an async thread. 21907c478bd9Sstevel@tonic-gate */ 21917c478bd9Sstevel@tonic-gate if (mi->mi_async_reqs[NFS4_INACTIVE] == NULL) { 21927c478bd9Sstevel@tonic-gate mi->mi_async_reqs[NFS4_INACTIVE] = args; 21937c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_INACTIVE] = args; 21947c478bd9Sstevel@tonic-gate signal_inactive_thread = B_TRUE; 21957c478bd9Sstevel@tonic-gate } else { 21967c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_INACTIVE]->a_next = args; 21977c478bd9Sstevel@tonic-gate mi->mi_async_tail[NFS4_INACTIVE] = args; 21987c478bd9Sstevel@tonic-gate } 21997c478bd9Sstevel@tonic-gate if (signal_inactive_thread) { 22007c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_inact_req_cv); 22017c478bd9Sstevel@tonic-gate } else { 22027c478bd9Sstevel@tonic-gate mi->mi_async_req_count++; 22037c478bd9Sstevel@tonic-gate ASSERT(mi->mi_async_req_count != 0); 22047c478bd9Sstevel@tonic-gate cv_signal(&mi->mi_async_reqs_cv); 22057c478bd9Sstevel@tonic-gate } 22067c478bd9Sstevel@tonic-gate 22077c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_async_lock); 22087c478bd9Sstevel@tonic-gate } 22097c478bd9Sstevel@tonic-gate 22107c478bd9Sstevel@tonic-gate int 22117c478bd9Sstevel@tonic-gate writerp4(rnode4_t *rp, caddr_t base, int tcount, struct uio *uio, int pgcreated) 22127c478bd9Sstevel@tonic-gate { 22137c478bd9Sstevel@tonic-gate int pagecreate; 22147c478bd9Sstevel@tonic-gate int n; 22157c478bd9Sstevel@tonic-gate int saved_n; 22167c478bd9Sstevel@tonic-gate caddr_t saved_base; 22177c478bd9Sstevel@tonic-gate u_offset_t offset; 22187c478bd9Sstevel@tonic-gate int error; 22197c478bd9Sstevel@tonic-gate int sm_error; 2220a5652762Spraks vnode_t *vp = RTOV(rp); 22217c478bd9Sstevel@tonic-gate 22227c478bd9Sstevel@tonic-gate ASSERT(tcount <= MAXBSIZE && tcount <= uio->uio_resid); 22237c478bd9Sstevel@tonic-gate ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_WRITER)); 2224a5652762Spraks if (!vpm_enable) { 2225a5652762Spraks ASSERT(((uintptr_t)base & MAXBOFFSET) + tcount <= MAXBSIZE); 2226a5652762Spraks } 22277c478bd9Sstevel@tonic-gate 22287c478bd9Sstevel@tonic-gate /* 22297c478bd9Sstevel@tonic-gate * Move bytes in at most PAGESIZE chunks. We must avoid 22307c478bd9Sstevel@tonic-gate * spanning pages in uiomove() because page faults may cause 22317c478bd9Sstevel@tonic-gate * the cache to be invalidated out from under us. The r_size is not 22327c478bd9Sstevel@tonic-gate * updated until after the uiomove. If we push the last page of a 22337c478bd9Sstevel@tonic-gate * file before r_size is correct, we will lose the data written past 22347c478bd9Sstevel@tonic-gate * the current (and invalid) r_size. 22357c478bd9Sstevel@tonic-gate */ 22367c478bd9Sstevel@tonic-gate do { 22377c478bd9Sstevel@tonic-gate offset = uio->uio_loffset; 22387c478bd9Sstevel@tonic-gate pagecreate = 0; 22397c478bd9Sstevel@tonic-gate 22407c478bd9Sstevel@tonic-gate /* 22417c478bd9Sstevel@tonic-gate * n is the number of bytes required to satisfy the request 22427c478bd9Sstevel@tonic-gate * or the number of bytes to fill out the page. 22437c478bd9Sstevel@tonic-gate */ 2244a5652762Spraks n = (int)MIN((PAGESIZE - (offset & PAGEOFFSET)), tcount); 22457c478bd9Sstevel@tonic-gate 22467c478bd9Sstevel@tonic-gate /* 22477c478bd9Sstevel@tonic-gate * Check to see if we can skip reading in the page 22487c478bd9Sstevel@tonic-gate * and just allocate the memory. We can do this 22497c478bd9Sstevel@tonic-gate * if we are going to rewrite the entire mapping 22507c478bd9Sstevel@tonic-gate * or if we are going to write to or beyond the current 22517c478bd9Sstevel@tonic-gate * end of file from the beginning of the mapping. 22527c478bd9Sstevel@tonic-gate * 22537c478bd9Sstevel@tonic-gate * The read of r_size is now protected by r_statelock. 22547c478bd9Sstevel@tonic-gate */ 22557c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 22567c478bd9Sstevel@tonic-gate /* 22577c478bd9Sstevel@tonic-gate * When pgcreated is nonzero the caller has already done 22587c478bd9Sstevel@tonic-gate * a segmap_getmapflt with forcefault 0 and S_WRITE. With 22597c478bd9Sstevel@tonic-gate * segkpm this means we already have at least one page 22607c478bd9Sstevel@tonic-gate * created and mapped at base. 22617c478bd9Sstevel@tonic-gate */ 22627c478bd9Sstevel@tonic-gate pagecreate = pgcreated || 2263b9238976Sth ((offset & PAGEOFFSET) == 0 && 2264b9238976Sth (n == PAGESIZE || ((offset + n) >= rp->r_size))); 22657c478bd9Sstevel@tonic-gate 22667c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 22677c478bd9Sstevel@tonic-gate 2268a5652762Spraks if (!vpm_enable && pagecreate) { 22697c478bd9Sstevel@tonic-gate /* 22707c478bd9Sstevel@tonic-gate * The last argument tells segmap_pagecreate() to 22717c478bd9Sstevel@tonic-gate * always lock the page, as opposed to sometimes 22727c478bd9Sstevel@tonic-gate * returning with the page locked. This way we avoid a 22737c478bd9Sstevel@tonic-gate * fault on the ensuing uiomove(), but also 22747c478bd9Sstevel@tonic-gate * more importantly (to fix bug 1094402) we can 22757c478bd9Sstevel@tonic-gate * call segmap_fault() to unlock the page in all 22767c478bd9Sstevel@tonic-gate * cases. An alternative would be to modify 22777c478bd9Sstevel@tonic-gate * segmap_pagecreate() to tell us when it is 22787c478bd9Sstevel@tonic-gate * locking a page, but that's a fairly major 22797c478bd9Sstevel@tonic-gate * interface change. 22807c478bd9Sstevel@tonic-gate */ 22817c478bd9Sstevel@tonic-gate if (pgcreated == 0) 22827c478bd9Sstevel@tonic-gate (void) segmap_pagecreate(segkmap, base, 2283b9238976Sth (uint_t)n, 1); 22847c478bd9Sstevel@tonic-gate saved_base = base; 22857c478bd9Sstevel@tonic-gate saved_n = n; 22867c478bd9Sstevel@tonic-gate } 22877c478bd9Sstevel@tonic-gate 22887c478bd9Sstevel@tonic-gate /* 22897c478bd9Sstevel@tonic-gate * The number of bytes of data in the last page can not 22907c478bd9Sstevel@tonic-gate * be accurately be determined while page is being 22917c478bd9Sstevel@tonic-gate * uiomove'd to and the size of the file being updated. 22927c478bd9Sstevel@tonic-gate * Thus, inform threads which need to know accurately 22937c478bd9Sstevel@tonic-gate * how much data is in the last page of the file. They 22947c478bd9Sstevel@tonic-gate * will not do the i/o immediately, but will arrange for 22957c478bd9Sstevel@tonic-gate * the i/o to happen later when this modify operation 22967c478bd9Sstevel@tonic-gate * will have finished. 22977c478bd9Sstevel@tonic-gate */ 22987c478bd9Sstevel@tonic-gate ASSERT(!(rp->r_flags & R4MODINPROGRESS)); 22997c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 23007c478bd9Sstevel@tonic-gate rp->r_flags |= R4MODINPROGRESS; 23017c478bd9Sstevel@tonic-gate rp->r_modaddr = (offset & MAXBMASK); 23027c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 23037c478bd9Sstevel@tonic-gate 2304a5652762Spraks if (vpm_enable) { 2305a5652762Spraks /* 2306a5652762Spraks * Copy data. If new pages are created, part of 2307a5652762Spraks * the page that is not written will be initizliazed 2308a5652762Spraks * with zeros. 2309a5652762Spraks */ 2310a5652762Spraks error = vpm_data_copy(vp, offset, n, uio, 2311b9238976Sth !pagecreate, NULL, 0, S_WRITE); 2312a5652762Spraks } else { 2313a5652762Spraks error = uiomove(base, n, UIO_WRITE, uio); 2314a5652762Spraks } 23157c478bd9Sstevel@tonic-gate 23167c478bd9Sstevel@tonic-gate /* 23177c478bd9Sstevel@tonic-gate * r_size is the maximum number of 23187c478bd9Sstevel@tonic-gate * bytes known to be in the file. 23197c478bd9Sstevel@tonic-gate * Make sure it is at least as high as the 23207c478bd9Sstevel@tonic-gate * first unwritten byte pointed to by uio_loffset. 23217c478bd9Sstevel@tonic-gate */ 23227c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 23237c478bd9Sstevel@tonic-gate if (rp->r_size < uio->uio_loffset) 23247c478bd9Sstevel@tonic-gate rp->r_size = uio->uio_loffset; 23257c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4MODINPROGRESS; 23267c478bd9Sstevel@tonic-gate rp->r_flags |= R4DIRTY; 23277c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 23287c478bd9Sstevel@tonic-gate 23297c478bd9Sstevel@tonic-gate /* n = # of bytes written */ 23307c478bd9Sstevel@tonic-gate n = (int)(uio->uio_loffset - offset); 2331a5652762Spraks 2332a5652762Spraks if (!vpm_enable) { 2333a5652762Spraks base += n; 2334a5652762Spraks } 2335a5652762Spraks 23367c478bd9Sstevel@tonic-gate tcount -= n; 23377c478bd9Sstevel@tonic-gate /* 23387c478bd9Sstevel@tonic-gate * If we created pages w/o initializing them completely, 23397c478bd9Sstevel@tonic-gate * we need to zero the part that wasn't set up. 23407c478bd9Sstevel@tonic-gate * This happens on a most EOF write cases and if 23417c478bd9Sstevel@tonic-gate * we had some sort of error during the uiomove. 23427c478bd9Sstevel@tonic-gate */ 2343a5652762Spraks if (!vpm_enable && pagecreate) { 23447c478bd9Sstevel@tonic-gate if ((uio->uio_loffset & PAGEOFFSET) || n == 0) 23457c478bd9Sstevel@tonic-gate (void) kzero(base, PAGESIZE - n); 23467c478bd9Sstevel@tonic-gate 23477c478bd9Sstevel@tonic-gate if (pgcreated) { 23487c478bd9Sstevel@tonic-gate /* 23497c478bd9Sstevel@tonic-gate * Caller is responsible for this page, 23507c478bd9Sstevel@tonic-gate * it was not created in this loop. 23517c478bd9Sstevel@tonic-gate */ 23527c478bd9Sstevel@tonic-gate pgcreated = 0; 23537c478bd9Sstevel@tonic-gate } else { 23547c478bd9Sstevel@tonic-gate /* 23557c478bd9Sstevel@tonic-gate * For bug 1094402: segmap_pagecreate locks 23567c478bd9Sstevel@tonic-gate * page. Unlock it. This also unlocks the 23577c478bd9Sstevel@tonic-gate * pages allocated by page_create_va() in 23587c478bd9Sstevel@tonic-gate * segmap_pagecreate(). 23597c478bd9Sstevel@tonic-gate */ 23607c478bd9Sstevel@tonic-gate sm_error = segmap_fault(kas.a_hat, segkmap, 2361b9238976Sth saved_base, saved_n, 2362b9238976Sth F_SOFTUNLOCK, S_WRITE); 23637c478bd9Sstevel@tonic-gate if (error == 0) 23647c478bd9Sstevel@tonic-gate error = sm_error; 23657c478bd9Sstevel@tonic-gate } 23667c478bd9Sstevel@tonic-gate } 23677c478bd9Sstevel@tonic-gate } while (tcount > 0 && error == 0); 23687c478bd9Sstevel@tonic-gate 23697c478bd9Sstevel@tonic-gate return (error); 23707c478bd9Sstevel@tonic-gate } 23717c478bd9Sstevel@tonic-gate 23727c478bd9Sstevel@tonic-gate int 23737c478bd9Sstevel@tonic-gate nfs4_putpages(vnode_t *vp, u_offset_t off, size_t len, int flags, cred_t *cr) 23747c478bd9Sstevel@tonic-gate { 23757c478bd9Sstevel@tonic-gate rnode4_t *rp; 23767c478bd9Sstevel@tonic-gate page_t *pp; 23777c478bd9Sstevel@tonic-gate u_offset_t eoff; 23787c478bd9Sstevel@tonic-gate u_offset_t io_off; 23797c478bd9Sstevel@tonic-gate size_t io_len; 23807c478bd9Sstevel@tonic-gate int error; 23817c478bd9Sstevel@tonic-gate int rdirty; 23827c478bd9Sstevel@tonic-gate int err; 23837c478bd9Sstevel@tonic-gate 23847c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 23857c478bd9Sstevel@tonic-gate ASSERT(rp->r_count > 0); 23867c478bd9Sstevel@tonic-gate 23877c478bd9Sstevel@tonic-gate if (!nfs4_has_pages(vp)) 23887c478bd9Sstevel@tonic-gate return (0); 23897c478bd9Sstevel@tonic-gate 23907c478bd9Sstevel@tonic-gate ASSERT(vp->v_type != VCHR); 23917c478bd9Sstevel@tonic-gate 23927c478bd9Sstevel@tonic-gate /* 23937c478bd9Sstevel@tonic-gate * If R4OUTOFSPACE is set, then all writes turn into B_INVAL 23947c478bd9Sstevel@tonic-gate * writes. B_FORCE is set to force the VM system to actually 23957c478bd9Sstevel@tonic-gate * invalidate the pages, even if the i/o failed. The pages 23967c478bd9Sstevel@tonic-gate * need to get invalidated because they can't be written out 23977c478bd9Sstevel@tonic-gate * because there isn't any space left on either the server's 23987c478bd9Sstevel@tonic-gate * file system or in the user's disk quota. The B_FREE bit 23997c478bd9Sstevel@tonic-gate * is cleared to avoid confusion as to whether this is a 24007c478bd9Sstevel@tonic-gate * request to place the page on the freelist or to destroy 24017c478bd9Sstevel@tonic-gate * it. 24027c478bd9Sstevel@tonic-gate */ 24037c478bd9Sstevel@tonic-gate if ((rp->r_flags & R4OUTOFSPACE) || 24047c478bd9Sstevel@tonic-gate (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)) 24057c478bd9Sstevel@tonic-gate flags = (flags & ~B_FREE) | B_INVAL | B_FORCE; 24067c478bd9Sstevel@tonic-gate 24077c478bd9Sstevel@tonic-gate if (len == 0) { 24087c478bd9Sstevel@tonic-gate /* 24097c478bd9Sstevel@tonic-gate * If doing a full file synchronous operation, then clear 24107c478bd9Sstevel@tonic-gate * the R4DIRTY bit. If a page gets dirtied while the flush 24117c478bd9Sstevel@tonic-gate * is happening, then R4DIRTY will get set again. The 24127c478bd9Sstevel@tonic-gate * R4DIRTY bit must get cleared before the flush so that 24137c478bd9Sstevel@tonic-gate * we don't lose this information. 241484d68d8eSthurlow * 241584d68d8eSthurlow * If there are no full file async write operations 241684d68d8eSthurlow * pending and RDIRTY bit is set, clear it. 24177c478bd9Sstevel@tonic-gate */ 24187c478bd9Sstevel@tonic-gate if (off == (u_offset_t)0 && 24197c478bd9Sstevel@tonic-gate !(flags & B_ASYNC) && 24207c478bd9Sstevel@tonic-gate (rp->r_flags & R4DIRTY)) { 24217c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 24227c478bd9Sstevel@tonic-gate rdirty = (rp->r_flags & R4DIRTY); 24237c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4DIRTY; 24247c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 242584d68d8eSthurlow } else if (flags & B_ASYNC && off == (u_offset_t)0) { 242684d68d8eSthurlow mutex_enter(&rp->r_statelock); 242784d68d8eSthurlow if (rp->r_flags & R4DIRTY && rp->r_awcount == 0) { 242884d68d8eSthurlow rdirty = (rp->r_flags & R4DIRTY); 242984d68d8eSthurlow rp->r_flags &= ~R4DIRTY; 243084d68d8eSthurlow } 243184d68d8eSthurlow mutex_exit(&rp->r_statelock); 24327c478bd9Sstevel@tonic-gate } else 24337c478bd9Sstevel@tonic-gate rdirty = 0; 24347c478bd9Sstevel@tonic-gate 24357c478bd9Sstevel@tonic-gate /* 24367c478bd9Sstevel@tonic-gate * Search the entire vp list for pages >= off, and flush 24377c478bd9Sstevel@tonic-gate * the dirty pages. 24387c478bd9Sstevel@tonic-gate */ 24397c478bd9Sstevel@tonic-gate error = pvn_vplist_dirty(vp, off, rp->r_putapage, 2440b9238976Sth flags, cr); 24417c478bd9Sstevel@tonic-gate 24427c478bd9Sstevel@tonic-gate /* 2443da6c28aaSamw * If an error occurred and the file was marked as dirty 24447c478bd9Sstevel@tonic-gate * before and we aren't forcibly invalidating pages, then 24457c478bd9Sstevel@tonic-gate * reset the R4DIRTY flag. 24467c478bd9Sstevel@tonic-gate */ 24477c478bd9Sstevel@tonic-gate if (error && rdirty && 24487c478bd9Sstevel@tonic-gate (flags & (B_INVAL | B_FORCE)) != (B_INVAL | B_FORCE)) { 24497c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 24507c478bd9Sstevel@tonic-gate rp->r_flags |= R4DIRTY; 24517c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 24527c478bd9Sstevel@tonic-gate } 24537c478bd9Sstevel@tonic-gate } else { 24547c478bd9Sstevel@tonic-gate /* 24557c478bd9Sstevel@tonic-gate * Do a range from [off...off + len) looking for pages 24567c478bd9Sstevel@tonic-gate * to deal with. 24577c478bd9Sstevel@tonic-gate */ 24587c478bd9Sstevel@tonic-gate error = 0; 24597c478bd9Sstevel@tonic-gate io_len = 0; 24607c478bd9Sstevel@tonic-gate eoff = off + len; 24617c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 24627c478bd9Sstevel@tonic-gate for (io_off = off; io_off < eoff && io_off < rp->r_size; 24637c478bd9Sstevel@tonic-gate io_off += io_len) { 24647c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 24657c478bd9Sstevel@tonic-gate /* 24667c478bd9Sstevel@tonic-gate * If we are not invalidating, synchronously 24677c478bd9Sstevel@tonic-gate * freeing or writing pages use the routine 24687c478bd9Sstevel@tonic-gate * page_lookup_nowait() to prevent reclaiming 24697c478bd9Sstevel@tonic-gate * them from the free list. 24707c478bd9Sstevel@tonic-gate */ 24717c478bd9Sstevel@tonic-gate if ((flags & B_INVAL) || !(flags & B_ASYNC)) { 24727c478bd9Sstevel@tonic-gate pp = page_lookup(vp, io_off, 24737c478bd9Sstevel@tonic-gate (flags & (B_INVAL | B_FREE)) ? 24747c478bd9Sstevel@tonic-gate SE_EXCL : SE_SHARED); 24757c478bd9Sstevel@tonic-gate } else { 24767c478bd9Sstevel@tonic-gate pp = page_lookup_nowait(vp, io_off, 24777c478bd9Sstevel@tonic-gate (flags & B_FREE) ? SE_EXCL : SE_SHARED); 24787c478bd9Sstevel@tonic-gate } 24797c478bd9Sstevel@tonic-gate 24807c478bd9Sstevel@tonic-gate if (pp == NULL || !pvn_getdirty(pp, flags)) 24817c478bd9Sstevel@tonic-gate io_len = PAGESIZE; 24827c478bd9Sstevel@tonic-gate else { 24837c478bd9Sstevel@tonic-gate err = (*rp->r_putapage)(vp, pp, &io_off, 24847c478bd9Sstevel@tonic-gate &io_len, flags, cr); 24857c478bd9Sstevel@tonic-gate if (!error) 24867c478bd9Sstevel@tonic-gate error = err; 24877c478bd9Sstevel@tonic-gate /* 24887c478bd9Sstevel@tonic-gate * "io_off" and "io_len" are returned as 24897c478bd9Sstevel@tonic-gate * the range of pages we actually wrote. 24907c478bd9Sstevel@tonic-gate * This allows us to skip ahead more quickly 24917c478bd9Sstevel@tonic-gate * since several pages may've been dealt 24927c478bd9Sstevel@tonic-gate * with by this iteration of the loop. 24937c478bd9Sstevel@tonic-gate */ 24947c478bd9Sstevel@tonic-gate } 24957c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 24967c478bd9Sstevel@tonic-gate } 24977c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 24987c478bd9Sstevel@tonic-gate } 24997c478bd9Sstevel@tonic-gate 25007c478bd9Sstevel@tonic-gate return (error); 25017c478bd9Sstevel@tonic-gate } 25027c478bd9Sstevel@tonic-gate 25037c478bd9Sstevel@tonic-gate void 25047c478bd9Sstevel@tonic-gate nfs4_invalidate_pages(vnode_t *vp, u_offset_t off, cred_t *cr) 25057c478bd9Sstevel@tonic-gate { 25067c478bd9Sstevel@tonic-gate rnode4_t *rp; 25077c478bd9Sstevel@tonic-gate 25087c478bd9Sstevel@tonic-gate rp = VTOR4(vp); 25097c478bd9Sstevel@tonic-gate if (IS_SHADOW(vp, rp)) 25107c478bd9Sstevel@tonic-gate vp = RTOV4(rp); 25117c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 25127c478bd9Sstevel@tonic-gate while (rp->r_flags & R4TRUNCATE) 25137c478bd9Sstevel@tonic-gate cv_wait(&rp->r_cv, &rp->r_statelock); 25147c478bd9Sstevel@tonic-gate rp->r_flags |= R4TRUNCATE; 25157c478bd9Sstevel@tonic-gate if (off == (u_offset_t)0) { 25167c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4DIRTY; 25177c478bd9Sstevel@tonic-gate if (!(rp->r_flags & R4STALE)) 25187c478bd9Sstevel@tonic-gate rp->r_error = 0; 25197c478bd9Sstevel@tonic-gate } 25207c478bd9Sstevel@tonic-gate rp->r_truncaddr = off; 25217c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 25227c478bd9Sstevel@tonic-gate (void) pvn_vplist_dirty(vp, off, rp->r_putapage, 2523b9238976Sth B_INVAL | B_TRUNC, cr); 25247c478bd9Sstevel@tonic-gate mutex_enter(&rp->r_statelock); 25257c478bd9Sstevel@tonic-gate rp->r_flags &= ~R4TRUNCATE; 25267c478bd9Sstevel@tonic-gate cv_broadcast(&rp->r_cv); 25277c478bd9Sstevel@tonic-gate mutex_exit(&rp->r_statelock); 25287c478bd9Sstevel@tonic-gate } 25297c478bd9Sstevel@tonic-gate 25307c478bd9Sstevel@tonic-gate static int 25317c478bd9Sstevel@tonic-gate nfs4_mnt_kstat_update(kstat_t *ksp, int rw) 25327c478bd9Sstevel@tonic-gate { 25337c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 25347c478bd9Sstevel@tonic-gate struct mntinfo_kstat *mik; 25357c478bd9Sstevel@tonic-gate vfs_t *vfsp; 25367c478bd9Sstevel@tonic-gate 25377c478bd9Sstevel@tonic-gate /* this is a read-only kstat. Bail out on a write */ 25387c478bd9Sstevel@tonic-gate if (rw == KSTAT_WRITE) 25397c478bd9Sstevel@tonic-gate return (EACCES); 25407c478bd9Sstevel@tonic-gate 25417c478bd9Sstevel@tonic-gate 25427c478bd9Sstevel@tonic-gate /* 25437c478bd9Sstevel@tonic-gate * We don't want to wait here as kstat_chain_lock could be held by 25447c478bd9Sstevel@tonic-gate * dounmount(). dounmount() takes vfs_reflock before the chain lock 25457c478bd9Sstevel@tonic-gate * and thus could lead to a deadlock. 25467c478bd9Sstevel@tonic-gate */ 25477c478bd9Sstevel@tonic-gate vfsp = (struct vfs *)ksp->ks_private; 25487c478bd9Sstevel@tonic-gate 25497c478bd9Sstevel@tonic-gate mi = VFTOMI4(vfsp); 25507c478bd9Sstevel@tonic-gate mik = (struct mntinfo_kstat *)ksp->ks_data; 25517c478bd9Sstevel@tonic-gate 25527c478bd9Sstevel@tonic-gate (void) strcpy(mik->mik_proto, mi->mi_curr_serv->sv_knconf->knc_proto); 25537c478bd9Sstevel@tonic-gate 25547c478bd9Sstevel@tonic-gate mik->mik_vers = (uint32_t)mi->mi_vers; 25557c478bd9Sstevel@tonic-gate mik->mik_flags = mi->mi_flags; 25567c478bd9Sstevel@tonic-gate /* 25577c478bd9Sstevel@tonic-gate * The sv_secdata holds the flavor the client specifies. 25587c478bd9Sstevel@tonic-gate * If the client uses default and a security negotiation 25597c478bd9Sstevel@tonic-gate * occurs, sv_currsec will point to the current flavor 25607c478bd9Sstevel@tonic-gate * selected from the server flavor list. 25617c478bd9Sstevel@tonic-gate * sv_currsec is NULL if no security negotiation takes place. 25627c478bd9Sstevel@tonic-gate */ 25637c478bd9Sstevel@tonic-gate mik->mik_secmod = mi->mi_curr_serv->sv_currsec ? 2564b9238976Sth mi->mi_curr_serv->sv_currsec->secmod : 2565b9238976Sth mi->mi_curr_serv->sv_secdata->secmod; 25667c478bd9Sstevel@tonic-gate mik->mik_curread = (uint32_t)mi->mi_curread; 25677c478bd9Sstevel@tonic-gate mik->mik_curwrite = (uint32_t)mi->mi_curwrite; 25687c478bd9Sstevel@tonic-gate mik->mik_retrans = mi->mi_retrans; 25697c478bd9Sstevel@tonic-gate mik->mik_timeo = mi->mi_timeo; 25707c478bd9Sstevel@tonic-gate mik->mik_acregmin = HR2SEC(mi->mi_acregmin); 25717c478bd9Sstevel@tonic-gate mik->mik_acregmax = HR2SEC(mi->mi_acregmax); 25727c478bd9Sstevel@tonic-gate mik->mik_acdirmin = HR2SEC(mi->mi_acdirmin); 25737c478bd9Sstevel@tonic-gate mik->mik_acdirmax = HR2SEC(mi->mi_acdirmax); 25747c478bd9Sstevel@tonic-gate mik->mik_noresponse = (uint32_t)mi->mi_noresponse; 25757c478bd9Sstevel@tonic-gate mik->mik_failover = (uint32_t)mi->mi_failover; 25767c478bd9Sstevel@tonic-gate mik->mik_remap = (uint32_t)mi->mi_remap; 25777c478bd9Sstevel@tonic-gate 25787c478bd9Sstevel@tonic-gate (void) strcpy(mik->mik_curserver, mi->mi_curr_serv->sv_hostname); 25797c478bd9Sstevel@tonic-gate 25807c478bd9Sstevel@tonic-gate return (0); 25817c478bd9Sstevel@tonic-gate } 25827c478bd9Sstevel@tonic-gate 25837c478bd9Sstevel@tonic-gate void 25847c478bd9Sstevel@tonic-gate nfs4_mnt_kstat_init(struct vfs *vfsp) 25857c478bd9Sstevel@tonic-gate { 25867c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VFTOMI4(vfsp); 25877c478bd9Sstevel@tonic-gate 25887c478bd9Sstevel@tonic-gate /* 25897c478bd9Sstevel@tonic-gate * PSARC 2001/697 Contract Private Interface 25907c478bd9Sstevel@tonic-gate * All nfs kstats are under SunMC contract 25917c478bd9Sstevel@tonic-gate * Please refer to the PSARC listed above and contact 25927c478bd9Sstevel@tonic-gate * SunMC before making any changes! 25937c478bd9Sstevel@tonic-gate * 25947c478bd9Sstevel@tonic-gate * Changes must be reviewed by Solaris File Sharing 25957c478bd9Sstevel@tonic-gate * Changes must be communicated to contract-2001-697@sun.com 25967c478bd9Sstevel@tonic-gate * 25977c478bd9Sstevel@tonic-gate */ 25987c478bd9Sstevel@tonic-gate 25997c478bd9Sstevel@tonic-gate mi->mi_io_kstats = kstat_create_zone("nfs", getminor(vfsp->vfs_dev), 26007c478bd9Sstevel@tonic-gate NULL, "nfs", KSTAT_TYPE_IO, 1, 0, mi->mi_zone->zone_id); 26017c478bd9Sstevel@tonic-gate if (mi->mi_io_kstats) { 26027c478bd9Sstevel@tonic-gate if (mi->mi_zone->zone_id != GLOBAL_ZONEID) 26037c478bd9Sstevel@tonic-gate kstat_zone_add(mi->mi_io_kstats, GLOBAL_ZONEID); 26047c478bd9Sstevel@tonic-gate mi->mi_io_kstats->ks_lock = &mi->mi_lock; 26057c478bd9Sstevel@tonic-gate kstat_install(mi->mi_io_kstats); 26067c478bd9Sstevel@tonic-gate } 26077c478bd9Sstevel@tonic-gate 26087c478bd9Sstevel@tonic-gate if ((mi->mi_ro_kstats = kstat_create_zone("nfs", 26097c478bd9Sstevel@tonic-gate getminor(vfsp->vfs_dev), "mntinfo", "misc", KSTAT_TYPE_RAW, 26107c478bd9Sstevel@tonic-gate sizeof (struct mntinfo_kstat), 0, mi->mi_zone->zone_id)) != NULL) { 26117c478bd9Sstevel@tonic-gate if (mi->mi_zone->zone_id != GLOBAL_ZONEID) 26127c478bd9Sstevel@tonic-gate kstat_zone_add(mi->mi_ro_kstats, GLOBAL_ZONEID); 26137c478bd9Sstevel@tonic-gate mi->mi_ro_kstats->ks_update = nfs4_mnt_kstat_update; 26147c478bd9Sstevel@tonic-gate mi->mi_ro_kstats->ks_private = (void *)vfsp; 26157c478bd9Sstevel@tonic-gate kstat_install(mi->mi_ro_kstats); 26167c478bd9Sstevel@tonic-gate } 26177c478bd9Sstevel@tonic-gate 26187c478bd9Sstevel@tonic-gate nfs4_mnt_recov_kstat_init(vfsp); 26197c478bd9Sstevel@tonic-gate } 26207c478bd9Sstevel@tonic-gate 26217c478bd9Sstevel@tonic-gate void 26227c478bd9Sstevel@tonic-gate nfs4_write_error(vnode_t *vp, int error, cred_t *cr) 26237c478bd9Sstevel@tonic-gate { 26247c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 26257c478bd9Sstevel@tonic-gate 26267c478bd9Sstevel@tonic-gate mi = VTOMI4(vp); 26277c478bd9Sstevel@tonic-gate /* 26287c478bd9Sstevel@tonic-gate * In case of forced unmount, do not print any messages 26297c478bd9Sstevel@tonic-gate * since it can flood the console with error messages. 26307c478bd9Sstevel@tonic-gate */ 26317c478bd9Sstevel@tonic-gate if (mi->mi_vfsp->vfs_flag & VFS_UNMOUNTED) 26327c478bd9Sstevel@tonic-gate return; 26337c478bd9Sstevel@tonic-gate 26347c478bd9Sstevel@tonic-gate /* 26357c478bd9Sstevel@tonic-gate * If the mount point is dead, not recoverable, do not 26367c478bd9Sstevel@tonic-gate * print error messages that can flood the console. 26377c478bd9Sstevel@tonic-gate */ 26387c478bd9Sstevel@tonic-gate if (mi->mi_flags & MI4_RECOV_FAIL) 26397c478bd9Sstevel@tonic-gate return; 26407c478bd9Sstevel@tonic-gate 26417c478bd9Sstevel@tonic-gate /* 26427c478bd9Sstevel@tonic-gate * No use in flooding the console with ENOSPC 26437c478bd9Sstevel@tonic-gate * messages from the same file system. 26447c478bd9Sstevel@tonic-gate */ 26457c478bd9Sstevel@tonic-gate if ((error != ENOSPC && error != EDQUOT) || 26467c478bd9Sstevel@tonic-gate lbolt - mi->mi_printftime > 0) { 26477c478bd9Sstevel@tonic-gate zoneid_t zoneid = mi->mi_zone->zone_id; 26487c478bd9Sstevel@tonic-gate 26497c478bd9Sstevel@tonic-gate #ifdef DEBUG 26507c478bd9Sstevel@tonic-gate nfs_perror(error, "NFS%ld write error on host %s: %m.\n", 26517c478bd9Sstevel@tonic-gate mi->mi_vers, VTOR4(vp)->r_server->sv_hostname, NULL); 26527c478bd9Sstevel@tonic-gate #else 26537c478bd9Sstevel@tonic-gate nfs_perror(error, "NFS write error on host %s: %m.\n", 26547c478bd9Sstevel@tonic-gate VTOR4(vp)->r_server->sv_hostname, NULL); 26557c478bd9Sstevel@tonic-gate #endif 26567c478bd9Sstevel@tonic-gate if (error == ENOSPC || error == EDQUOT) { 26577c478bd9Sstevel@tonic-gate zcmn_err(zoneid, CE_CONT, 26587c478bd9Sstevel@tonic-gate "^File: userid=%d, groupid=%d\n", 26597c478bd9Sstevel@tonic-gate crgetuid(cr), crgetgid(cr)); 26607c478bd9Sstevel@tonic-gate if (crgetuid(curthread->t_cred) != crgetuid(cr) || 26617c478bd9Sstevel@tonic-gate crgetgid(curthread->t_cred) != crgetgid(cr)) { 26627c478bd9Sstevel@tonic-gate zcmn_err(zoneid, CE_CONT, 26637c478bd9Sstevel@tonic-gate "^User: userid=%d, groupid=%d\n", 26647c478bd9Sstevel@tonic-gate crgetuid(curthread->t_cred), 26657c478bd9Sstevel@tonic-gate crgetgid(curthread->t_cred)); 26667c478bd9Sstevel@tonic-gate } 26677c478bd9Sstevel@tonic-gate mi->mi_printftime = lbolt + 26687c478bd9Sstevel@tonic-gate nfs_write_error_interval * hz; 26697c478bd9Sstevel@tonic-gate } 26707c478bd9Sstevel@tonic-gate sfh4_printfhandle(VTOR4(vp)->r_fh); 26717c478bd9Sstevel@tonic-gate #ifdef DEBUG 26727c478bd9Sstevel@tonic-gate if (error == EACCES) { 26737c478bd9Sstevel@tonic-gate zcmn_err(zoneid, CE_CONT, 26747c478bd9Sstevel@tonic-gate "nfs_bio: cred is%s kcred\n", 26757c478bd9Sstevel@tonic-gate cr == kcred ? "" : " not"); 26767c478bd9Sstevel@tonic-gate } 26777c478bd9Sstevel@tonic-gate #endif 26787c478bd9Sstevel@tonic-gate } 26797c478bd9Sstevel@tonic-gate } 26807c478bd9Sstevel@tonic-gate 26817c478bd9Sstevel@tonic-gate /* 26827c478bd9Sstevel@tonic-gate * Return non-zero if the given file can be safely memory mapped. Locks 26837c478bd9Sstevel@tonic-gate * are safe if whole-file (length and offset are both zero). 26847c478bd9Sstevel@tonic-gate */ 26857c478bd9Sstevel@tonic-gate 26867c478bd9Sstevel@tonic-gate #define SAFE_LOCK(flk) ((flk).l_start == 0 && (flk).l_len == 0) 26877c478bd9Sstevel@tonic-gate 26887c478bd9Sstevel@tonic-gate static int 26897c478bd9Sstevel@tonic-gate nfs4_safemap(const vnode_t *vp) 26907c478bd9Sstevel@tonic-gate { 26917c478bd9Sstevel@tonic-gate locklist_t *llp, *next_llp; 26927c478bd9Sstevel@tonic-gate int safe = 1; 26937c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(vp); 26947c478bd9Sstevel@tonic-gate 26957c478bd9Sstevel@tonic-gate ASSERT(nfs_rw_lock_held(&rp->r_lkserlock, RW_WRITER)); 26967c478bd9Sstevel@tonic-gate 26977c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE, "nfs4_safemap: " 2698b9238976Sth "vp = %p", (void *)vp)); 26997c478bd9Sstevel@tonic-gate 27007c478bd9Sstevel@tonic-gate /* 27017c478bd9Sstevel@tonic-gate * Review all the locks for the vnode, both ones that have been 27027c478bd9Sstevel@tonic-gate * acquired and ones that are pending. We assume that 27037c478bd9Sstevel@tonic-gate * flk_active_locks_for_vp() has merged any locks that can be 27047c478bd9Sstevel@tonic-gate * merged (so that if a process has the entire file locked, it is 27057c478bd9Sstevel@tonic-gate * represented as a single lock). 27067c478bd9Sstevel@tonic-gate * 27077c478bd9Sstevel@tonic-gate * Note that we can't bail out of the loop if we find a non-safe 27087c478bd9Sstevel@tonic-gate * lock, because we have to free all the elements in the llp list. 27097c478bd9Sstevel@tonic-gate * We might be able to speed up this code slightly by not looking 27107c478bd9Sstevel@tonic-gate * at each lock's l_start and l_len fields once we've found a 27117c478bd9Sstevel@tonic-gate * non-safe lock. 27127c478bd9Sstevel@tonic-gate */ 27137c478bd9Sstevel@tonic-gate 27147c478bd9Sstevel@tonic-gate llp = flk_active_locks_for_vp(vp); 27157c478bd9Sstevel@tonic-gate while (llp) { 27167c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE, 27177c478bd9Sstevel@tonic-gate "nfs4_safemap: active lock (%" PRId64 ", %" PRId64 ")", 27187c478bd9Sstevel@tonic-gate llp->ll_flock.l_start, llp->ll_flock.l_len)); 27197c478bd9Sstevel@tonic-gate if (!SAFE_LOCK(llp->ll_flock)) { 27207c478bd9Sstevel@tonic-gate safe = 0; 27217c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE, 27227c478bd9Sstevel@tonic-gate "nfs4_safemap: unsafe active lock (%" PRId64 27237c478bd9Sstevel@tonic-gate ", %" PRId64 ")", llp->ll_flock.l_start, 27247c478bd9Sstevel@tonic-gate llp->ll_flock.l_len)); 27257c478bd9Sstevel@tonic-gate } 27267c478bd9Sstevel@tonic-gate next_llp = llp->ll_next; 27277c478bd9Sstevel@tonic-gate VN_RELE(llp->ll_vp); 27287c478bd9Sstevel@tonic-gate kmem_free(llp, sizeof (*llp)); 27297c478bd9Sstevel@tonic-gate llp = next_llp; 27307c478bd9Sstevel@tonic-gate } 27317c478bd9Sstevel@tonic-gate 27327c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE, "nfs4_safemap: %s", 2733b9238976Sth safe ? "safe" : "unsafe")); 27347c478bd9Sstevel@tonic-gate return (safe); 27357c478bd9Sstevel@tonic-gate } 27367c478bd9Sstevel@tonic-gate 27377c478bd9Sstevel@tonic-gate /* 27387c478bd9Sstevel@tonic-gate * Return whether there is a lost LOCK or LOCKU queued up for the given 27397c478bd9Sstevel@tonic-gate * file that would make an mmap request unsafe. cf. nfs4_safemap(). 27407c478bd9Sstevel@tonic-gate */ 27417c478bd9Sstevel@tonic-gate 27427c478bd9Sstevel@tonic-gate bool_t 27437c478bd9Sstevel@tonic-gate nfs4_map_lost_lock_conflict(vnode_t *vp) 27447c478bd9Sstevel@tonic-gate { 27457c478bd9Sstevel@tonic-gate bool_t conflict = FALSE; 27467c478bd9Sstevel@tonic-gate nfs4_lost_rqst_t *lrp; 27477c478bd9Sstevel@tonic-gate mntinfo4_t *mi = VTOMI4(vp); 27487c478bd9Sstevel@tonic-gate 27497c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 27507c478bd9Sstevel@tonic-gate for (lrp = list_head(&mi->mi_lost_state); lrp != NULL; 27517c478bd9Sstevel@tonic-gate lrp = list_next(&mi->mi_lost_state, lrp)) { 27527c478bd9Sstevel@tonic-gate if (lrp->lr_op != OP_LOCK && lrp->lr_op != OP_LOCKU) 27537c478bd9Sstevel@tonic-gate continue; 27547c478bd9Sstevel@tonic-gate ASSERT(lrp->lr_vp != NULL); 2755da6c28aaSamw if (!VOP_CMP(lrp->lr_vp, vp, NULL)) 27567c478bd9Sstevel@tonic-gate continue; /* different file */ 27577c478bd9Sstevel@tonic-gate if (!SAFE_LOCK(*lrp->lr_flk)) { 27587c478bd9Sstevel@tonic-gate conflict = TRUE; 27597c478bd9Sstevel@tonic-gate break; 27607c478bd9Sstevel@tonic-gate } 27617c478bd9Sstevel@tonic-gate } 27627c478bd9Sstevel@tonic-gate 27637c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 27647c478bd9Sstevel@tonic-gate return (conflict); 27657c478bd9Sstevel@tonic-gate } 27667c478bd9Sstevel@tonic-gate 27677c478bd9Sstevel@tonic-gate /* 27687c478bd9Sstevel@tonic-gate * nfs_lockcompletion: 27697c478bd9Sstevel@tonic-gate * 27707c478bd9Sstevel@tonic-gate * If the vnode has a lock that makes it unsafe to cache the file, mark it 27717c478bd9Sstevel@tonic-gate * as non cachable (set VNOCACHE bit). 27727c478bd9Sstevel@tonic-gate */ 27737c478bd9Sstevel@tonic-gate 27747c478bd9Sstevel@tonic-gate void 27757c478bd9Sstevel@tonic-gate nfs4_lockcompletion(vnode_t *vp, int cmd) 27767c478bd9Sstevel@tonic-gate { 27777c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(vp); 27787c478bd9Sstevel@tonic-gate 27797c478bd9Sstevel@tonic-gate ASSERT(nfs_rw_lock_held(&rp->r_lkserlock, RW_WRITER)); 27807c478bd9Sstevel@tonic-gate ASSERT(!IS_SHADOW(vp, rp)); 27817c478bd9Sstevel@tonic-gate 27827c478bd9Sstevel@tonic-gate if (cmd == F_SETLK || cmd == F_SETLKW) { 27837c478bd9Sstevel@tonic-gate 27847c478bd9Sstevel@tonic-gate if (!nfs4_safemap(vp)) { 27857c478bd9Sstevel@tonic-gate mutex_enter(&vp->v_lock); 27867c478bd9Sstevel@tonic-gate vp->v_flag |= VNOCACHE; 27877c478bd9Sstevel@tonic-gate mutex_exit(&vp->v_lock); 27887c478bd9Sstevel@tonic-gate } else { 27897c478bd9Sstevel@tonic-gate mutex_enter(&vp->v_lock); 27907c478bd9Sstevel@tonic-gate vp->v_flag &= ~VNOCACHE; 27917c478bd9Sstevel@tonic-gate mutex_exit(&vp->v_lock); 27927c478bd9Sstevel@tonic-gate } 27937c478bd9Sstevel@tonic-gate } 27947c478bd9Sstevel@tonic-gate /* 27957c478bd9Sstevel@tonic-gate * The cached attributes of the file are stale after acquiring 27967c478bd9Sstevel@tonic-gate * the lock on the file. They were updated when the file was 27977c478bd9Sstevel@tonic-gate * opened, but not updated when the lock was acquired. Therefore the 27987c478bd9Sstevel@tonic-gate * cached attributes are invalidated after the lock is obtained. 27997c478bd9Sstevel@tonic-gate */ 28007c478bd9Sstevel@tonic-gate PURGE_ATTRCACHE4(vp); 28017c478bd9Sstevel@tonic-gate } 28027c478bd9Sstevel@tonic-gate 28037c478bd9Sstevel@tonic-gate /* ARGSUSED */ 28047c478bd9Sstevel@tonic-gate static void * 28057c478bd9Sstevel@tonic-gate nfs4_mi_init(zoneid_t zoneid) 28067c478bd9Sstevel@tonic-gate { 28077c478bd9Sstevel@tonic-gate struct mi4_globals *mig; 28087c478bd9Sstevel@tonic-gate 28097c478bd9Sstevel@tonic-gate mig = kmem_alloc(sizeof (*mig), KM_SLEEP); 28107c478bd9Sstevel@tonic-gate mutex_init(&mig->mig_lock, NULL, MUTEX_DEFAULT, NULL); 28117c478bd9Sstevel@tonic-gate list_create(&mig->mig_list, sizeof (mntinfo4_t), 28127c478bd9Sstevel@tonic-gate offsetof(mntinfo4_t, mi_zone_node)); 28137c478bd9Sstevel@tonic-gate mig->mig_destructor_called = B_FALSE; 28147c478bd9Sstevel@tonic-gate return (mig); 28157c478bd9Sstevel@tonic-gate } 28167c478bd9Sstevel@tonic-gate 28177c478bd9Sstevel@tonic-gate /* 28187c478bd9Sstevel@tonic-gate * Callback routine to tell all NFSv4 mounts in the zone to start tearing down 28197c478bd9Sstevel@tonic-gate * state and killing off threads. 28207c478bd9Sstevel@tonic-gate */ 28217c478bd9Sstevel@tonic-gate /* ARGSUSED */ 28227c478bd9Sstevel@tonic-gate static void 28237c478bd9Sstevel@tonic-gate nfs4_mi_shutdown(zoneid_t zoneid, void *data) 28247c478bd9Sstevel@tonic-gate { 28257c478bd9Sstevel@tonic-gate struct mi4_globals *mig = data; 28267c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 28277c478bd9Sstevel@tonic-gate nfs4_server_t *np; 28287c478bd9Sstevel@tonic-gate 28297c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE, 28307c478bd9Sstevel@tonic-gate "nfs4_mi_shutdown zone %d\n", zoneid)); 28317c478bd9Sstevel@tonic-gate ASSERT(mig != NULL); 283250a83466Sjwahlig for (;;) { 283350a83466Sjwahlig mutex_enter(&mig->mig_lock); 283450a83466Sjwahlig mi = list_head(&mig->mig_list); 283550a83466Sjwahlig if (mi == NULL) { 283650a83466Sjwahlig mutex_exit(&mig->mig_lock); 283750a83466Sjwahlig break; 283850a83466Sjwahlig } 28393fd6cc29Sthurlow 28407c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE, 28417c478bd9Sstevel@tonic-gate "nfs4_mi_shutdown stopping vfs %p\n", (void *)mi->mi_vfsp)); 28427c478bd9Sstevel@tonic-gate /* 28437c478bd9Sstevel@tonic-gate * purge the DNLC for this filesystem 28447c478bd9Sstevel@tonic-gate */ 28457c478bd9Sstevel@tonic-gate (void) dnlc_purge_vfsp(mi->mi_vfsp, 0); 28467c478bd9Sstevel@tonic-gate /* 28477c478bd9Sstevel@tonic-gate * Tell existing async worker threads to exit. 28487c478bd9Sstevel@tonic-gate */ 284950a83466Sjwahlig mutex_enter(&mi->mi_async_lock); 28507c478bd9Sstevel@tonic-gate mi->mi_max_threads = 0; 28517c478bd9Sstevel@tonic-gate cv_broadcast(&mi->mi_async_work_cv); 28527c478bd9Sstevel@tonic-gate /* 285350a83466Sjwahlig * Set the appropriate flags, signal and wait for both the 285450a83466Sjwahlig * async manager and the inactive thread to exit when they're 285550a83466Sjwahlig * done with their current work. 28567c478bd9Sstevel@tonic-gate */ 28577c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 28587c478bd9Sstevel@tonic-gate mi->mi_flags |= (MI4_ASYNC_MGR_STOP|MI4_DEAD); 28597c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 286050a83466Sjwahlig mutex_exit(&mi->mi_async_lock); 286150a83466Sjwahlig if (mi->mi_manager_thread) { 286250a83466Sjwahlig nfs4_async_manager_stop(mi->mi_vfsp); 286350a83466Sjwahlig } 286450a83466Sjwahlig if (mi->mi_inactive_thread) { 286550a83466Sjwahlig mutex_enter(&mi->mi_async_lock); 286650a83466Sjwahlig cv_signal(&mi->mi_inact_req_cv); 286750a83466Sjwahlig /* 286850a83466Sjwahlig * Wait for the inactive thread to exit. 286950a83466Sjwahlig */ 287050a83466Sjwahlig while (mi->mi_inactive_thread != NULL) { 287150a83466Sjwahlig cv_wait(&mi->mi_async_cv, &mi->mi_async_lock); 287250a83466Sjwahlig } 287350a83466Sjwahlig mutex_exit(&mi->mi_async_lock); 287450a83466Sjwahlig } 28757c478bd9Sstevel@tonic-gate /* 287650a83466Sjwahlig * Wait for the recovery thread to complete, that is, it will 287750a83466Sjwahlig * signal when it is done using the "mi" structure and about 287850a83466Sjwahlig * to exit 28797c478bd9Sstevel@tonic-gate */ 288050a83466Sjwahlig mutex_enter(&mi->mi_lock); 288150a83466Sjwahlig while (mi->mi_in_recovery > 0) 288250a83466Sjwahlig cv_wait(&mi->mi_cv_in_recov, &mi->mi_lock); 288350a83466Sjwahlig mutex_exit(&mi->mi_lock); 28843fd6cc29Sthurlow /* 28853fd6cc29Sthurlow * We're done when every mi has been done or the list is empty. 288650a83466Sjwahlig * This one is done, remove it from the list. 28873fd6cc29Sthurlow */ 288850a83466Sjwahlig list_remove(&mig->mig_list, mi); 28893fd6cc29Sthurlow mutex_exit(&mig->mig_lock); 289050a83466Sjwahlig zone_rele(mi->mi_zone); 289150a83466Sjwahlig /* 289250a83466Sjwahlig * Release hold on vfs and mi done to prevent race with zone 289350a83466Sjwahlig * shutdown. This releases the hold in nfs4_mi_zonelist_add. 289450a83466Sjwahlig */ 28953fd6cc29Sthurlow VFS_RELE(mi->mi_vfsp); 289650a83466Sjwahlig MI4_RELE(mi); 28977c478bd9Sstevel@tonic-gate } 28987c478bd9Sstevel@tonic-gate /* 28997c478bd9Sstevel@tonic-gate * Tell each renew thread in the zone to exit 29007c478bd9Sstevel@tonic-gate */ 29017c478bd9Sstevel@tonic-gate mutex_enter(&nfs4_server_lst_lock); 29027c478bd9Sstevel@tonic-gate for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) { 29037c478bd9Sstevel@tonic-gate mutex_enter(&np->s_lock); 29047c478bd9Sstevel@tonic-gate if (np->zoneid == zoneid) { 29057c478bd9Sstevel@tonic-gate /* 29067c478bd9Sstevel@tonic-gate * We add another hold onto the nfs4_server_t 29077c478bd9Sstevel@tonic-gate * because this will make sure tha the nfs4_server_t 29087c478bd9Sstevel@tonic-gate * stays around until nfs4_callback_fini_zone destroys 29097c478bd9Sstevel@tonic-gate * the zone. This way, the renew thread can 29107c478bd9Sstevel@tonic-gate * unconditionally release its holds on the 29117c478bd9Sstevel@tonic-gate * nfs4_server_t. 29127c478bd9Sstevel@tonic-gate */ 29137c478bd9Sstevel@tonic-gate np->s_refcnt++; 29147c478bd9Sstevel@tonic-gate nfs4_mark_srv_dead(np); 29157c478bd9Sstevel@tonic-gate } 29167c478bd9Sstevel@tonic-gate mutex_exit(&np->s_lock); 29177c478bd9Sstevel@tonic-gate } 29187c478bd9Sstevel@tonic-gate mutex_exit(&nfs4_server_lst_lock); 29197c478bd9Sstevel@tonic-gate } 29207c478bd9Sstevel@tonic-gate 29217c478bd9Sstevel@tonic-gate static void 29227c478bd9Sstevel@tonic-gate nfs4_mi_free_globals(struct mi4_globals *mig) 29237c478bd9Sstevel@tonic-gate { 29247c478bd9Sstevel@tonic-gate list_destroy(&mig->mig_list); /* makes sure the list is empty */ 29257c478bd9Sstevel@tonic-gate mutex_destroy(&mig->mig_lock); 29267c478bd9Sstevel@tonic-gate kmem_free(mig, sizeof (*mig)); 29277c478bd9Sstevel@tonic-gate } 29287c478bd9Sstevel@tonic-gate 29297c478bd9Sstevel@tonic-gate /* ARGSUSED */ 29307c478bd9Sstevel@tonic-gate static void 29317c478bd9Sstevel@tonic-gate nfs4_mi_destroy(zoneid_t zoneid, void *data) 29327c478bd9Sstevel@tonic-gate { 29337c478bd9Sstevel@tonic-gate struct mi4_globals *mig = data; 29347c478bd9Sstevel@tonic-gate 29357c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE, 29367c478bd9Sstevel@tonic-gate "nfs4_mi_destroy zone %d\n", zoneid)); 29377c478bd9Sstevel@tonic-gate ASSERT(mig != NULL); 29387c478bd9Sstevel@tonic-gate mutex_enter(&mig->mig_lock); 29397c478bd9Sstevel@tonic-gate if (list_head(&mig->mig_list) != NULL) { 29407c478bd9Sstevel@tonic-gate /* Still waiting for VFS_FREEVFS() */ 29417c478bd9Sstevel@tonic-gate mig->mig_destructor_called = B_TRUE; 29427c478bd9Sstevel@tonic-gate mutex_exit(&mig->mig_lock); 29437c478bd9Sstevel@tonic-gate return; 29447c478bd9Sstevel@tonic-gate } 29457c478bd9Sstevel@tonic-gate nfs4_mi_free_globals(mig); 29467c478bd9Sstevel@tonic-gate } 29477c478bd9Sstevel@tonic-gate 29487c478bd9Sstevel@tonic-gate /* 29497c478bd9Sstevel@tonic-gate * Add an NFS mount to the per-zone list of NFS mounts. 29507c478bd9Sstevel@tonic-gate */ 29517c478bd9Sstevel@tonic-gate void 29527c478bd9Sstevel@tonic-gate nfs4_mi_zonelist_add(mntinfo4_t *mi) 29537c478bd9Sstevel@tonic-gate { 29547c478bd9Sstevel@tonic-gate struct mi4_globals *mig; 29557c478bd9Sstevel@tonic-gate 29567c478bd9Sstevel@tonic-gate mig = zone_getspecific(mi4_list_key, mi->mi_zone); 29577c478bd9Sstevel@tonic-gate mutex_enter(&mig->mig_lock); 29587c478bd9Sstevel@tonic-gate list_insert_head(&mig->mig_list, mi); 295950a83466Sjwahlig /* 296050a83466Sjwahlig * hold added to eliminate race with zone shutdown -this will be 296150a83466Sjwahlig * released in mi_shutdown 296250a83466Sjwahlig */ 296350a83466Sjwahlig MI4_HOLD(mi); 296450a83466Sjwahlig VFS_HOLD(mi->mi_vfsp); 29657c478bd9Sstevel@tonic-gate mutex_exit(&mig->mig_lock); 29667c478bd9Sstevel@tonic-gate } 29677c478bd9Sstevel@tonic-gate 29687c478bd9Sstevel@tonic-gate /* 29697c478bd9Sstevel@tonic-gate * Remove an NFS mount from the per-zone list of NFS mounts. 29707c478bd9Sstevel@tonic-gate */ 297150a83466Sjwahlig int 29727c478bd9Sstevel@tonic-gate nfs4_mi_zonelist_remove(mntinfo4_t *mi) 29737c478bd9Sstevel@tonic-gate { 29747c478bd9Sstevel@tonic-gate struct mi4_globals *mig; 297550a83466Sjwahlig int ret = 0; 29767c478bd9Sstevel@tonic-gate 29777c478bd9Sstevel@tonic-gate mig = zone_getspecific(mi4_list_key, mi->mi_zone); 29787c478bd9Sstevel@tonic-gate mutex_enter(&mig->mig_lock); 297950a83466Sjwahlig mutex_enter(&mi->mi_lock); 298050a83466Sjwahlig /* if this mi is marked dead, then the zone already released it */ 298150a83466Sjwahlig if (!(mi->mi_flags & MI4_DEAD)) { 298250a83466Sjwahlig list_remove(&mig->mig_list, mi); 298350a83466Sjwahlig 298450a83466Sjwahlig /* release the holds put on in zonelist_add(). */ 298550a83466Sjwahlig VFS_RELE(mi->mi_vfsp); 298650a83466Sjwahlig MI4_RELE(mi); 298750a83466Sjwahlig ret = 1; 298850a83466Sjwahlig } 298950a83466Sjwahlig mutex_exit(&mi->mi_lock); 299050a83466Sjwahlig 29917c478bd9Sstevel@tonic-gate /* 29927c478bd9Sstevel@tonic-gate * We can be called asynchronously by VFS_FREEVFS() after the zone 29937c478bd9Sstevel@tonic-gate * shutdown/destroy callbacks have executed; if so, clean up the zone's 29947c478bd9Sstevel@tonic-gate * mi globals. 29957c478bd9Sstevel@tonic-gate */ 29967c478bd9Sstevel@tonic-gate if (list_head(&mig->mig_list) == NULL && 29977c478bd9Sstevel@tonic-gate mig->mig_destructor_called == B_TRUE) { 29987c478bd9Sstevel@tonic-gate nfs4_mi_free_globals(mig); 299950a83466Sjwahlig return (ret); 30007c478bd9Sstevel@tonic-gate } 30017c478bd9Sstevel@tonic-gate mutex_exit(&mig->mig_lock); 300250a83466Sjwahlig return (ret); 30037c478bd9Sstevel@tonic-gate } 30047c478bd9Sstevel@tonic-gate 30057c478bd9Sstevel@tonic-gate void 30067c478bd9Sstevel@tonic-gate nfs_free_mi4(mntinfo4_t *mi) 30077c478bd9Sstevel@tonic-gate { 30087c478bd9Sstevel@tonic-gate nfs4_open_owner_t *foop; 300950a83466Sjwahlig nfs4_oo_hash_bucket_t *bucketp; 30107c478bd9Sstevel@tonic-gate nfs4_debug_msg_t *msgp; 30117c478bd9Sstevel@tonic-gate int i; 301250a83466Sjwahlig servinfo4_t *svp; 30137c478bd9Sstevel@tonic-gate 30147c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_lock); 30157c478bd9Sstevel@tonic-gate ASSERT(mi->mi_recovthread == NULL); 30167c478bd9Sstevel@tonic-gate ASSERT(mi->mi_flags & MI4_ASYNC_MGR_STOP); 30177c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_lock); 30187c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_async_lock); 30197c478bd9Sstevel@tonic-gate ASSERT(mi->mi_threads == 0); 30207c478bd9Sstevel@tonic-gate ASSERT(mi->mi_manager_thread == NULL); 302150a83466Sjwahlig mutex_exit(&mi->mi_async_lock); 302250a83466Sjwahlig svp = mi->mi_servers; 302350a83466Sjwahlig sv4_free(svp); 302450a83466Sjwahlig if (mi->mi_io_kstats) { 302550a83466Sjwahlig kstat_delete(mi->mi_io_kstats); 302650a83466Sjwahlig mi->mi_io_kstats = NULL; 30277c478bd9Sstevel@tonic-gate } 302850a83466Sjwahlig if (mi->mi_ro_kstats) { 302950a83466Sjwahlig kstat_delete(mi->mi_ro_kstats); 303050a83466Sjwahlig mi->mi_ro_kstats = NULL; 303150a83466Sjwahlig } 303250a83466Sjwahlig if (mi->mi_recov_ksp) { 303350a83466Sjwahlig kstat_delete(mi->mi_recov_ksp); 303450a83466Sjwahlig mi->mi_recov_ksp = NULL; 30357c478bd9Sstevel@tonic-gate } 30367c478bd9Sstevel@tonic-gate mutex_enter(&mi->mi_msg_list_lock); 30377c478bd9Sstevel@tonic-gate while (msgp = list_head(&mi->mi_msg_list)) { 30387c478bd9Sstevel@tonic-gate list_remove(&mi->mi_msg_list, msgp); 30397c478bd9Sstevel@tonic-gate nfs4_free_msg(msgp); 30407c478bd9Sstevel@tonic-gate } 30417c478bd9Sstevel@tonic-gate mutex_exit(&mi->mi_msg_list_lock); 30427c478bd9Sstevel@tonic-gate list_destroy(&mi->mi_msg_list); 30437c478bd9Sstevel@tonic-gate if (mi->mi_rootfh != NULL) 30447c478bd9Sstevel@tonic-gate sfh4_rele(&mi->mi_rootfh); 30457c478bd9Sstevel@tonic-gate if (mi->mi_srvparentfh != NULL) 30467c478bd9Sstevel@tonic-gate sfh4_rele(&mi->mi_srvparentfh); 30477c478bd9Sstevel@tonic-gate mutex_destroy(&mi->mi_lock); 30487c478bd9Sstevel@tonic-gate mutex_destroy(&mi->mi_async_lock); 30497c478bd9Sstevel@tonic-gate mutex_destroy(&mi->mi_msg_list_lock); 30507c478bd9Sstevel@tonic-gate nfs_rw_destroy(&mi->mi_recovlock); 30517c478bd9Sstevel@tonic-gate nfs_rw_destroy(&mi->mi_rename_lock); 30527c478bd9Sstevel@tonic-gate nfs_rw_destroy(&mi->mi_fh_lock); 30537c478bd9Sstevel@tonic-gate cv_destroy(&mi->mi_failover_cv); 30547c478bd9Sstevel@tonic-gate cv_destroy(&mi->mi_async_reqs_cv); 30557c478bd9Sstevel@tonic-gate cv_destroy(&mi->mi_async_work_cv); 30567c478bd9Sstevel@tonic-gate cv_destroy(&mi->mi_async_cv); 30577c478bd9Sstevel@tonic-gate cv_destroy(&mi->mi_inact_req_cv); 30587c478bd9Sstevel@tonic-gate /* 30597c478bd9Sstevel@tonic-gate * Destroy the oo hash lists and mutexes for the cred hash table. 30607c478bd9Sstevel@tonic-gate */ 30617c478bd9Sstevel@tonic-gate for (i = 0; i < NFS4_NUM_OO_BUCKETS; i++) { 30627c478bd9Sstevel@tonic-gate bucketp = &(mi->mi_oo_list[i]); 30637c478bd9Sstevel@tonic-gate /* Destroy any remaining open owners on the list */ 30647c478bd9Sstevel@tonic-gate foop = list_head(&bucketp->b_oo_hash_list); 30657c478bd9Sstevel@tonic-gate while (foop != NULL) { 30667c478bd9Sstevel@tonic-gate list_remove(&bucketp->b_oo_hash_list, foop); 30677c478bd9Sstevel@tonic-gate nfs4_destroy_open_owner(foop); 30687c478bd9Sstevel@tonic-gate foop = list_head(&bucketp->b_oo_hash_list); 30697c478bd9Sstevel@tonic-gate } 30707c478bd9Sstevel@tonic-gate list_destroy(&bucketp->b_oo_hash_list); 30717c478bd9Sstevel@tonic-gate mutex_destroy(&bucketp->b_lock); 30727c478bd9Sstevel@tonic-gate } 30737c478bd9Sstevel@tonic-gate /* 30747c478bd9Sstevel@tonic-gate * Empty and destroy the freed open owner list. 30757c478bd9Sstevel@tonic-gate */ 30767c478bd9Sstevel@tonic-gate foop = list_head(&mi->mi_foo_list); 30777c478bd9Sstevel@tonic-gate while (foop != NULL) { 30787c478bd9Sstevel@tonic-gate list_remove(&mi->mi_foo_list, foop); 30797c478bd9Sstevel@tonic-gate nfs4_destroy_open_owner(foop); 30807c478bd9Sstevel@tonic-gate foop = list_head(&mi->mi_foo_list); 30817c478bd9Sstevel@tonic-gate } 30827c478bd9Sstevel@tonic-gate list_destroy(&mi->mi_foo_list); 30837c478bd9Sstevel@tonic-gate list_destroy(&mi->mi_bseqid_list); 30847c478bd9Sstevel@tonic-gate list_destroy(&mi->mi_lost_state); 30857c478bd9Sstevel@tonic-gate avl_destroy(&mi->mi_filehandles); 30867c478bd9Sstevel@tonic-gate fn_rele(&mi->mi_fname); 30877c478bd9Sstevel@tonic-gate kmem_free(mi, sizeof (*mi)); 30887c478bd9Sstevel@tonic-gate } 308950a83466Sjwahlig void 309050a83466Sjwahlig mi_hold(mntinfo4_t *mi) 309150a83466Sjwahlig { 309250a83466Sjwahlig atomic_add_32(&mi->mi_count, 1); 309350a83466Sjwahlig ASSERT(mi->mi_count != 0); 309450a83466Sjwahlig } 309550a83466Sjwahlig 309650a83466Sjwahlig void 309750a83466Sjwahlig mi_rele(mntinfo4_t *mi) 309850a83466Sjwahlig { 309950a83466Sjwahlig ASSERT(mi->mi_count != 0); 310050a83466Sjwahlig if (atomic_add_32_nv(&mi->mi_count, -1) == 0) { 310150a83466Sjwahlig nfs_free_mi4(mi); 310250a83466Sjwahlig } 310350a83466Sjwahlig } 31047c478bd9Sstevel@tonic-gate 31057c478bd9Sstevel@tonic-gate vnode_t nfs4_xattr_notsupp_vnode; 31067c478bd9Sstevel@tonic-gate 31077c478bd9Sstevel@tonic-gate void 31087c478bd9Sstevel@tonic-gate nfs4_clnt_init(void) 31097c478bd9Sstevel@tonic-gate { 31107c478bd9Sstevel@tonic-gate nfs4_vnops_init(); 31117c478bd9Sstevel@tonic-gate (void) nfs4_rnode_init(); 31127c478bd9Sstevel@tonic-gate (void) nfs4_shadow_init(); 31137c478bd9Sstevel@tonic-gate (void) nfs4_acache_init(); 31147c478bd9Sstevel@tonic-gate (void) nfs4_subr_init(); 31157c478bd9Sstevel@tonic-gate nfs4_acl_init(); 31167c478bd9Sstevel@tonic-gate nfs_idmap_init(); 31177c478bd9Sstevel@tonic-gate nfs4_callback_init(); 31187c478bd9Sstevel@tonic-gate nfs4_secinfo_init(); 31197c478bd9Sstevel@tonic-gate #ifdef DEBUG 31207c478bd9Sstevel@tonic-gate tsd_create(&nfs4_tsd_key, NULL); 31217c478bd9Sstevel@tonic-gate #endif 31227c478bd9Sstevel@tonic-gate 31237c478bd9Sstevel@tonic-gate /* 31247c478bd9Sstevel@tonic-gate * Add a CPR callback so that we can update client 31257c478bd9Sstevel@tonic-gate * lease after a suspend and resume. 31267c478bd9Sstevel@tonic-gate */ 31277c478bd9Sstevel@tonic-gate cid = callb_add(nfs4_client_cpr_callb, 0, CB_CL_CPR_RPC, "nfs4"); 31287c478bd9Sstevel@tonic-gate 31297c478bd9Sstevel@tonic-gate zone_key_create(&mi4_list_key, nfs4_mi_init, nfs4_mi_shutdown, 31307c478bd9Sstevel@tonic-gate nfs4_mi_destroy); 31317c478bd9Sstevel@tonic-gate 31327c478bd9Sstevel@tonic-gate /* 31337c478bd9Sstevel@tonic-gate * Initialise the reference count of the notsupp xattr cache vnode to 1 31347c478bd9Sstevel@tonic-gate * so that it never goes away (VOP_INACTIVE isn't called on it). 31357c478bd9Sstevel@tonic-gate */ 31367c478bd9Sstevel@tonic-gate nfs4_xattr_notsupp_vnode.v_count = 1; 31377c478bd9Sstevel@tonic-gate } 31387c478bd9Sstevel@tonic-gate 31397c478bd9Sstevel@tonic-gate void 31407c478bd9Sstevel@tonic-gate nfs4_clnt_fini(void) 31417c478bd9Sstevel@tonic-gate { 31427c478bd9Sstevel@tonic-gate (void) zone_key_delete(mi4_list_key); 31437c478bd9Sstevel@tonic-gate nfs4_vnops_fini(); 31447c478bd9Sstevel@tonic-gate (void) nfs4_rnode_fini(); 31457c478bd9Sstevel@tonic-gate (void) nfs4_shadow_fini(); 31467c478bd9Sstevel@tonic-gate (void) nfs4_acache_fini(); 31477c478bd9Sstevel@tonic-gate (void) nfs4_subr_fini(); 31487c478bd9Sstevel@tonic-gate nfs_idmap_fini(); 31497c478bd9Sstevel@tonic-gate nfs4_callback_fini(); 31507c478bd9Sstevel@tonic-gate nfs4_secinfo_fini(); 31517c478bd9Sstevel@tonic-gate #ifdef DEBUG 31527c478bd9Sstevel@tonic-gate tsd_destroy(&nfs4_tsd_key); 31537c478bd9Sstevel@tonic-gate #endif 31547c478bd9Sstevel@tonic-gate if (cid) 31557c478bd9Sstevel@tonic-gate (void) callb_delete(cid); 31567c478bd9Sstevel@tonic-gate } 31577c478bd9Sstevel@tonic-gate 31587c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31597c478bd9Sstevel@tonic-gate static boolean_t 31607c478bd9Sstevel@tonic-gate nfs4_client_cpr_callb(void *arg, int code) 31617c478bd9Sstevel@tonic-gate { 31627c478bd9Sstevel@tonic-gate /* 31637c478bd9Sstevel@tonic-gate * We get called for Suspend and Resume events. 31647c478bd9Sstevel@tonic-gate * For the suspend case we simply don't care! 31657c478bd9Sstevel@tonic-gate */ 31667c478bd9Sstevel@tonic-gate if (code == CB_CODE_CPR_CHKPT) { 31677c478bd9Sstevel@tonic-gate return (B_TRUE); 31687c478bd9Sstevel@tonic-gate } 31697c478bd9Sstevel@tonic-gate 31707c478bd9Sstevel@tonic-gate /* 31717c478bd9Sstevel@tonic-gate * When we get to here we are in the process of 31727c478bd9Sstevel@tonic-gate * resuming the system from a previous suspend. 31737c478bd9Sstevel@tonic-gate */ 31747c478bd9Sstevel@tonic-gate nfs4_client_resumed = gethrestime_sec(); 31757c478bd9Sstevel@tonic-gate return (B_TRUE); 31767c478bd9Sstevel@tonic-gate } 31777c478bd9Sstevel@tonic-gate 31787c478bd9Sstevel@tonic-gate void 31797c478bd9Sstevel@tonic-gate nfs4_renew_lease_thread(nfs4_server_t *sp) 31807c478bd9Sstevel@tonic-gate { 31817c478bd9Sstevel@tonic-gate int error = 0; 31827c478bd9Sstevel@tonic-gate time_t tmp_last_renewal_time, tmp_time, tmp_now_time, kip_secs; 31837c478bd9Sstevel@tonic-gate clock_t tick_delay = 0; 31847c478bd9Sstevel@tonic-gate clock_t time_left = 0; 31857c478bd9Sstevel@tonic-gate callb_cpr_t cpr_info; 31867c478bd9Sstevel@tonic-gate kmutex_t cpr_lock; 31877c478bd9Sstevel@tonic-gate 31887c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3189b9238976Sth "nfs4_renew_lease_thread: acting on sp 0x%p", (void*)sp)); 31907c478bd9Sstevel@tonic-gate mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL); 31917c478bd9Sstevel@tonic-gate CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr, "nfsv4Lease"); 31927c478bd9Sstevel@tonic-gate 31937c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 31947c478bd9Sstevel@tonic-gate /* sp->s_lease_time is set via a GETATTR */ 31957c478bd9Sstevel@tonic-gate sp->last_renewal_time = gethrestime_sec(); 31967c478bd9Sstevel@tonic-gate sp->lease_valid = NFS4_LEASE_UNINITIALIZED; 31977c478bd9Sstevel@tonic-gate ASSERT(sp->s_refcnt >= 1); 31987c478bd9Sstevel@tonic-gate 31997c478bd9Sstevel@tonic-gate for (;;) { 32007c478bd9Sstevel@tonic-gate if (!sp->state_ref_count || 3201b9238976Sth sp->lease_valid != NFS4_LEASE_VALID) { 32027c478bd9Sstevel@tonic-gate 32037c478bd9Sstevel@tonic-gate kip_secs = MAX((sp->s_lease_time >> 1) - 3204b9238976Sth (3 * sp->propagation_delay.tv_sec), 1); 32057c478bd9Sstevel@tonic-gate 32067c478bd9Sstevel@tonic-gate tick_delay = SEC_TO_TICK(kip_secs); 32077c478bd9Sstevel@tonic-gate 32087c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3209b9238976Sth "nfs4_renew_lease_thread: no renew : thread " 3210b9238976Sth "wait %ld secs", kip_secs)); 32117c478bd9Sstevel@tonic-gate 32127c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3213b9238976Sth "nfs4_renew_lease_thread: no renew : " 3214b9238976Sth "state_ref_count %d, lease_valid %d", 3215b9238976Sth sp->state_ref_count, sp->lease_valid)); 32167c478bd9Sstevel@tonic-gate 32177c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 32187c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_BEGIN(&cpr_info); 32197c478bd9Sstevel@tonic-gate mutex_exit(&cpr_lock); 32207c478bd9Sstevel@tonic-gate time_left = cv_timedwait(&sp->cv_thread_exit, 3221b9238976Sth &sp->s_lock, tick_delay + lbolt); 32227c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 32237c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock); 32247c478bd9Sstevel@tonic-gate mutex_exit(&cpr_lock); 32257c478bd9Sstevel@tonic-gate 32267c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3227b9238976Sth "nfs4_renew_lease_thread: no renew: " 3228b9238976Sth "time left %ld", time_left)); 32297c478bd9Sstevel@tonic-gate 32307c478bd9Sstevel@tonic-gate if (sp->s_thread_exit == NFS4_THREAD_EXIT) 32317c478bd9Sstevel@tonic-gate goto die; 32327c478bd9Sstevel@tonic-gate continue; 32337c478bd9Sstevel@tonic-gate } 32347c478bd9Sstevel@tonic-gate 32357c478bd9Sstevel@tonic-gate tmp_last_renewal_time = sp->last_renewal_time; 32367c478bd9Sstevel@tonic-gate 32377c478bd9Sstevel@tonic-gate tmp_time = gethrestime_sec() - sp->last_renewal_time + 3238b9238976Sth (3 * sp->propagation_delay.tv_sec); 32397c478bd9Sstevel@tonic-gate 32407c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3241b9238976Sth "nfs4_renew_lease_thread: tmp_time %ld, " 3242b9238976Sth "sp->last_renewal_time %ld", tmp_time, 3243b9238976Sth sp->last_renewal_time)); 32447c478bd9Sstevel@tonic-gate 32457c478bd9Sstevel@tonic-gate kip_secs = MAX((sp->s_lease_time >> 1) - tmp_time, 1); 32467c478bd9Sstevel@tonic-gate 32477c478bd9Sstevel@tonic-gate tick_delay = SEC_TO_TICK(kip_secs); 32487c478bd9Sstevel@tonic-gate 32497c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3250b9238976Sth "nfs4_renew_lease_thread: valid lease: sleep for %ld " 3251b9238976Sth "secs", kip_secs)); 32527c478bd9Sstevel@tonic-gate 32537c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 32547c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_BEGIN(&cpr_info); 32557c478bd9Sstevel@tonic-gate mutex_exit(&cpr_lock); 32567c478bd9Sstevel@tonic-gate time_left = cv_timedwait(&sp->cv_thread_exit, &sp->s_lock, 3257b9238976Sth tick_delay + lbolt); 32587c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 32597c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock); 32607c478bd9Sstevel@tonic-gate mutex_exit(&cpr_lock); 32617c478bd9Sstevel@tonic-gate 32627c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3263b9238976Sth "nfs4_renew_lease_thread: valid lease: time left %ld :" 3264b9238976Sth "sp last_renewal_time %ld, nfs4_client_resumed %ld, " 3265b9238976Sth "tmp_last_renewal_time %ld", time_left, 3266b9238976Sth sp->last_renewal_time, nfs4_client_resumed, 3267b9238976Sth tmp_last_renewal_time)); 32687c478bd9Sstevel@tonic-gate 32697c478bd9Sstevel@tonic-gate if (sp->s_thread_exit == NFS4_THREAD_EXIT) 32707c478bd9Sstevel@tonic-gate goto die; 32717c478bd9Sstevel@tonic-gate 32727c478bd9Sstevel@tonic-gate if (tmp_last_renewal_time == sp->last_renewal_time || 3273b9238976Sth (nfs4_client_resumed != 0 && 3274b9238976Sth nfs4_client_resumed > sp->last_renewal_time)) { 32757c478bd9Sstevel@tonic-gate /* 32767c478bd9Sstevel@tonic-gate * Issue RENEW op since we haven't renewed the lease 32777c478bd9Sstevel@tonic-gate * since we slept. 32787c478bd9Sstevel@tonic-gate */ 32797c478bd9Sstevel@tonic-gate tmp_now_time = gethrestime_sec(); 32807c478bd9Sstevel@tonic-gate error = nfs4renew(sp); 32817c478bd9Sstevel@tonic-gate /* 32827c478bd9Sstevel@tonic-gate * Need to re-acquire sp's lock, nfs4renew() 32837c478bd9Sstevel@tonic-gate * relinqueshes it. 32847c478bd9Sstevel@tonic-gate */ 32857c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 32867c478bd9Sstevel@tonic-gate 32877c478bd9Sstevel@tonic-gate /* 32887c478bd9Sstevel@tonic-gate * See if someone changed s_thread_exit while we gave 32897c478bd9Sstevel@tonic-gate * up s_lock. 32907c478bd9Sstevel@tonic-gate */ 32917c478bd9Sstevel@tonic-gate if (sp->s_thread_exit == NFS4_THREAD_EXIT) 32927c478bd9Sstevel@tonic-gate goto die; 32937c478bd9Sstevel@tonic-gate 32947c478bd9Sstevel@tonic-gate if (!error) { 32957c478bd9Sstevel@tonic-gate /* 32967c478bd9Sstevel@tonic-gate * check to see if we implicitly renewed while 32977c478bd9Sstevel@tonic-gate * we waited for a reply for our RENEW call. 32987c478bd9Sstevel@tonic-gate */ 32997c478bd9Sstevel@tonic-gate if (tmp_last_renewal_time == 3300b9238976Sth sp->last_renewal_time) { 33017c478bd9Sstevel@tonic-gate /* no implicit renew came */ 33027c478bd9Sstevel@tonic-gate sp->last_renewal_time = tmp_now_time; 33037c478bd9Sstevel@tonic-gate } else { 33047c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, 3305b9238976Sth (CE_NOTE, "renew_thread: did " 3306b9238976Sth "implicit renewal before reply " 3307b9238976Sth "from server for RENEW")); 33087c478bd9Sstevel@tonic-gate } 33097c478bd9Sstevel@tonic-gate } else { 33107c478bd9Sstevel@tonic-gate /* figure out error */ 33117c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3312b9238976Sth "renew_thread: nfs4renew returned error" 3313b9238976Sth " %d", error)); 33147c478bd9Sstevel@tonic-gate } 33157c478bd9Sstevel@tonic-gate 33167c478bd9Sstevel@tonic-gate } 33177c478bd9Sstevel@tonic-gate } 33187c478bd9Sstevel@tonic-gate 33197c478bd9Sstevel@tonic-gate die: 33207c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3321b9238976Sth "nfs4_renew_lease_thread: thread exiting")); 33227c478bd9Sstevel@tonic-gate 33237c478bd9Sstevel@tonic-gate while (sp->s_otw_call_count != 0) { 33247c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3325b9238976Sth "nfs4_renew_lease_thread: waiting for outstanding " 3326b9238976Sth "otw calls to finish for sp 0x%p, current " 3327b9238976Sth "s_otw_call_count %d", (void *)sp, 3328b9238976Sth sp->s_otw_call_count)); 33297c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 33307c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_BEGIN(&cpr_info); 33317c478bd9Sstevel@tonic-gate mutex_exit(&cpr_lock); 33327c478bd9Sstevel@tonic-gate cv_wait(&sp->s_cv_otw_count, &sp->s_lock); 33337c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 33347c478bd9Sstevel@tonic-gate CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock); 33357c478bd9Sstevel@tonic-gate mutex_exit(&cpr_lock); 33367c478bd9Sstevel@tonic-gate } 33377c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 33387c478bd9Sstevel@tonic-gate 33397c478bd9Sstevel@tonic-gate nfs4_server_rele(sp); /* free the thread's reference */ 33407c478bd9Sstevel@tonic-gate nfs4_server_rele(sp); /* free the list's reference */ 33417c478bd9Sstevel@tonic-gate sp = NULL; 33427c478bd9Sstevel@tonic-gate 33437c478bd9Sstevel@tonic-gate done: 33447c478bd9Sstevel@tonic-gate mutex_enter(&cpr_lock); 33457c478bd9Sstevel@tonic-gate CALLB_CPR_EXIT(&cpr_info); /* drops cpr_lock */ 33467c478bd9Sstevel@tonic-gate mutex_destroy(&cpr_lock); 33477c478bd9Sstevel@tonic-gate 33487c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3349b9238976Sth "nfs4_renew_lease_thread: renew thread exit officially")); 33507c478bd9Sstevel@tonic-gate 33517c478bd9Sstevel@tonic-gate zthread_exit(); 33527c478bd9Sstevel@tonic-gate /* NOT REACHED */ 33537c478bd9Sstevel@tonic-gate } 33547c478bd9Sstevel@tonic-gate 33557c478bd9Sstevel@tonic-gate /* 33567c478bd9Sstevel@tonic-gate * Send out a RENEW op to the server. 33577c478bd9Sstevel@tonic-gate * Assumes sp is locked down. 33587c478bd9Sstevel@tonic-gate */ 33597c478bd9Sstevel@tonic-gate static int 33607c478bd9Sstevel@tonic-gate nfs4renew(nfs4_server_t *sp) 33617c478bd9Sstevel@tonic-gate { 33627c478bd9Sstevel@tonic-gate COMPOUND4args_clnt args; 33637c478bd9Sstevel@tonic-gate COMPOUND4res_clnt res; 33647c478bd9Sstevel@tonic-gate nfs_argop4 argop[1]; 33657c478bd9Sstevel@tonic-gate int doqueue = 1; 33667c478bd9Sstevel@tonic-gate int rpc_error; 33677c478bd9Sstevel@tonic-gate cred_t *cr; 33687c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 33697c478bd9Sstevel@tonic-gate timespec_t prop_time, after_time; 33707c478bd9Sstevel@tonic-gate int needrecov = FALSE; 33717c478bd9Sstevel@tonic-gate nfs4_recov_state_t recov_state; 33727c478bd9Sstevel@tonic-gate nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS }; 33737c478bd9Sstevel@tonic-gate 33747c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4renew")); 33757c478bd9Sstevel@tonic-gate 33767c478bd9Sstevel@tonic-gate recov_state.rs_flags = 0; 33777c478bd9Sstevel@tonic-gate recov_state.rs_num_retry_despite_err = 0; 33787c478bd9Sstevel@tonic-gate 33797c478bd9Sstevel@tonic-gate recov_retry: 33807c478bd9Sstevel@tonic-gate mi = sp->mntinfo4_list; 33817c478bd9Sstevel@tonic-gate VFS_HOLD(mi->mi_vfsp); 33827c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 33837c478bd9Sstevel@tonic-gate ASSERT(mi != NULL); 33847c478bd9Sstevel@tonic-gate 33857c478bd9Sstevel@tonic-gate e.error = nfs4_start_op(mi, NULL, NULL, &recov_state); 33867c478bd9Sstevel@tonic-gate if (e.error) { 33877c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 33887c478bd9Sstevel@tonic-gate return (e.error); 33897c478bd9Sstevel@tonic-gate } 33907c478bd9Sstevel@tonic-gate 33917c478bd9Sstevel@tonic-gate /* Check to see if we're dealing with a marked-dead sp */ 33927c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 33937c478bd9Sstevel@tonic-gate if (sp->s_thread_exit == NFS4_THREAD_EXIT) { 33947c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 33957c478bd9Sstevel@tonic-gate nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 33967c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 33977c478bd9Sstevel@tonic-gate return (0); 33987c478bd9Sstevel@tonic-gate } 33997c478bd9Sstevel@tonic-gate 34007c478bd9Sstevel@tonic-gate /* Make sure mi hasn't changed on us */ 34017c478bd9Sstevel@tonic-gate if (mi != sp->mntinfo4_list) { 34027c478bd9Sstevel@tonic-gate /* Must drop sp's lock to avoid a recursive mutex enter */ 34037c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 34047c478bd9Sstevel@tonic-gate nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 34057c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 34067c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 34077c478bd9Sstevel@tonic-gate goto recov_retry; 34087c478bd9Sstevel@tonic-gate } 34097c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 34107c478bd9Sstevel@tonic-gate 34117c478bd9Sstevel@tonic-gate args.ctag = TAG_RENEW; 34127c478bd9Sstevel@tonic-gate 34137c478bd9Sstevel@tonic-gate args.array_len = 1; 34147c478bd9Sstevel@tonic-gate args.array = argop; 34157c478bd9Sstevel@tonic-gate 34167c478bd9Sstevel@tonic-gate argop[0].argop = OP_RENEW; 34177c478bd9Sstevel@tonic-gate 34187c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 34197c478bd9Sstevel@tonic-gate argop[0].nfs_argop4_u.oprenew.clientid = sp->clientid; 34207c478bd9Sstevel@tonic-gate cr = sp->s_cred; 34217c478bd9Sstevel@tonic-gate crhold(cr); 34227c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 34237c478bd9Sstevel@tonic-gate 34247c478bd9Sstevel@tonic-gate ASSERT(cr != NULL); 34257c478bd9Sstevel@tonic-gate 34267c478bd9Sstevel@tonic-gate /* used to figure out RTT for sp */ 34277c478bd9Sstevel@tonic-gate gethrestime(&prop_time); 34287c478bd9Sstevel@tonic-gate 34297c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE, 34307c478bd9Sstevel@tonic-gate "nfs4renew: %s call, sp 0x%p", needrecov ? "recov" : "first", 34317c478bd9Sstevel@tonic-gate (void*)sp)); 34327c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "before: %ld s %ld ns ", 3433b9238976Sth prop_time.tv_sec, prop_time.tv_nsec)); 34347c478bd9Sstevel@tonic-gate 34357c478bd9Sstevel@tonic-gate DTRACE_PROBE2(nfs4__renew__start, nfs4_server_t *, sp, 3436b9238976Sth mntinfo4_t *, mi); 34377c478bd9Sstevel@tonic-gate 34387c478bd9Sstevel@tonic-gate rfs4call(mi, &args, &res, cr, &doqueue, 0, &e); 34397c478bd9Sstevel@tonic-gate crfree(cr); 34407c478bd9Sstevel@tonic-gate 34417c478bd9Sstevel@tonic-gate DTRACE_PROBE2(nfs4__renew__end, nfs4_server_t *, sp, 3442b9238976Sth mntinfo4_t *, mi); 34437c478bd9Sstevel@tonic-gate 34447c478bd9Sstevel@tonic-gate gethrestime(&after_time); 34457c478bd9Sstevel@tonic-gate 34467c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 34477c478bd9Sstevel@tonic-gate sp->propagation_delay.tv_sec = 3448b9238976Sth MAX(1, after_time.tv_sec - prop_time.tv_sec); 34497c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 34507c478bd9Sstevel@tonic-gate 34517c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "after : %ld s %ld ns ", 3452b9238976Sth after_time.tv_sec, after_time.tv_nsec)); 34537c478bd9Sstevel@tonic-gate 34547c478bd9Sstevel@tonic-gate if (e.error == 0 && res.status == NFS4ERR_CB_PATH_DOWN) { 3455d1ea0e82Swebaker (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 34567c478bd9Sstevel@tonic-gate nfs4_delegreturn_all(sp); 34577c478bd9Sstevel@tonic-gate nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 34587c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 34597c478bd9Sstevel@tonic-gate /* 34607c478bd9Sstevel@tonic-gate * If the server returns CB_PATH_DOWN, it has renewed 34617c478bd9Sstevel@tonic-gate * the lease and informed us that the callback path is 34627c478bd9Sstevel@tonic-gate * down. Since the lease is renewed, just return 0 and 34637c478bd9Sstevel@tonic-gate * let the renew thread proceed as normal. 34647c478bd9Sstevel@tonic-gate */ 34657c478bd9Sstevel@tonic-gate return (0); 34667c478bd9Sstevel@tonic-gate } 34677c478bd9Sstevel@tonic-gate 34687c478bd9Sstevel@tonic-gate needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp); 34697c478bd9Sstevel@tonic-gate if (!needrecov && e.error) { 34707c478bd9Sstevel@tonic-gate nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 34717c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 34727c478bd9Sstevel@tonic-gate return (e.error); 34737c478bd9Sstevel@tonic-gate } 34747c478bd9Sstevel@tonic-gate 34757c478bd9Sstevel@tonic-gate rpc_error = e.error; 34767c478bd9Sstevel@tonic-gate 34777c478bd9Sstevel@tonic-gate if (needrecov) { 34787c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE, 34797c478bd9Sstevel@tonic-gate "nfs4renew: initiating recovery\n")); 34807c478bd9Sstevel@tonic-gate 34817c478bd9Sstevel@tonic-gate if (nfs4_start_recovery(&e, mi, NULL, NULL, NULL, NULL, 34827c478bd9Sstevel@tonic-gate OP_RENEW, NULL) == FALSE) { 34837c478bd9Sstevel@tonic-gate nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 34847c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 34857c478bd9Sstevel@tonic-gate if (!e.error) 34867c478bd9Sstevel@tonic-gate (void) xdr_free(xdr_COMPOUND4res_clnt, 3487b9238976Sth (caddr_t)&res); 34887c478bd9Sstevel@tonic-gate mutex_enter(&sp->s_lock); 34897c478bd9Sstevel@tonic-gate goto recov_retry; 34907c478bd9Sstevel@tonic-gate } 34917c478bd9Sstevel@tonic-gate /* fall through for res.status case */ 34927c478bd9Sstevel@tonic-gate } 34937c478bd9Sstevel@tonic-gate 34947c478bd9Sstevel@tonic-gate if (res.status) { 34957c478bd9Sstevel@tonic-gate if (res.status == NFS4ERR_LEASE_MOVED) { 34967c478bd9Sstevel@tonic-gate /*EMPTY*/ 34977c478bd9Sstevel@tonic-gate /* 34987c478bd9Sstevel@tonic-gate * XXX need to try every mntinfo4 in sp->mntinfo4_list 34997c478bd9Sstevel@tonic-gate * to renew the lease on that server 35007c478bd9Sstevel@tonic-gate */ 35017c478bd9Sstevel@tonic-gate } 35027c478bd9Sstevel@tonic-gate e.error = geterrno4(res.status); 35037c478bd9Sstevel@tonic-gate } 35047c478bd9Sstevel@tonic-gate 35057c478bd9Sstevel@tonic-gate if (!rpc_error) 35067c478bd9Sstevel@tonic-gate (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 35077c478bd9Sstevel@tonic-gate 35087c478bd9Sstevel@tonic-gate nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 35097c478bd9Sstevel@tonic-gate 35107c478bd9Sstevel@tonic-gate VFS_RELE(mi->mi_vfsp); 35117c478bd9Sstevel@tonic-gate 35127c478bd9Sstevel@tonic-gate return (e.error); 35137c478bd9Sstevel@tonic-gate } 35147c478bd9Sstevel@tonic-gate 35157c478bd9Sstevel@tonic-gate void 35167c478bd9Sstevel@tonic-gate nfs4_inc_state_ref_count(mntinfo4_t *mi) 35177c478bd9Sstevel@tonic-gate { 35187c478bd9Sstevel@tonic-gate nfs4_server_t *sp; 35197c478bd9Sstevel@tonic-gate 35207c478bd9Sstevel@tonic-gate /* this locks down sp if it is found */ 35217c478bd9Sstevel@tonic-gate sp = find_nfs4_server(mi); 35227c478bd9Sstevel@tonic-gate 35237c478bd9Sstevel@tonic-gate if (sp != NULL) { 35247c478bd9Sstevel@tonic-gate nfs4_inc_state_ref_count_nolock(sp, mi); 35257c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 35267c478bd9Sstevel@tonic-gate nfs4_server_rele(sp); 35277c478bd9Sstevel@tonic-gate } 35287c478bd9Sstevel@tonic-gate } 35297c478bd9Sstevel@tonic-gate 35307c478bd9Sstevel@tonic-gate /* 35317c478bd9Sstevel@tonic-gate * Bump the number of OPEN files (ie: those with state) so we know if this 35327c478bd9Sstevel@tonic-gate * nfs4_server has any state to maintain a lease for or not. 35337c478bd9Sstevel@tonic-gate * 35347c478bd9Sstevel@tonic-gate * Also, marks the nfs4_server's lease valid if it hasn't been done so already. 35357c478bd9Sstevel@tonic-gate */ 35367c478bd9Sstevel@tonic-gate void 35377c478bd9Sstevel@tonic-gate nfs4_inc_state_ref_count_nolock(nfs4_server_t *sp, mntinfo4_t *mi) 35387c478bd9Sstevel@tonic-gate { 35397c478bd9Sstevel@tonic-gate ASSERT(mutex_owned(&sp->s_lock)); 35407c478bd9Sstevel@tonic-gate 35417c478bd9Sstevel@tonic-gate sp->state_ref_count++; 35427c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3543b9238976Sth "nfs4_inc_state_ref_count: state_ref_count now %d", 3544b9238976Sth sp->state_ref_count)); 35457c478bd9Sstevel@tonic-gate 35467c478bd9Sstevel@tonic-gate if (sp->lease_valid == NFS4_LEASE_UNINITIALIZED) 35477c478bd9Sstevel@tonic-gate sp->lease_valid = NFS4_LEASE_VALID; 35487c478bd9Sstevel@tonic-gate 35497c478bd9Sstevel@tonic-gate /* 35507c478bd9Sstevel@tonic-gate * If this call caused the lease to be marked valid and/or 35517c478bd9Sstevel@tonic-gate * took the state_ref_count from 0 to 1, then start the time 35527c478bd9Sstevel@tonic-gate * on lease renewal. 35537c478bd9Sstevel@tonic-gate */ 35547c478bd9Sstevel@tonic-gate if (sp->lease_valid == NFS4_LEASE_VALID && sp->state_ref_count == 1) 35557c478bd9Sstevel@tonic-gate sp->last_renewal_time = gethrestime_sec(); 35567c478bd9Sstevel@tonic-gate 35577c478bd9Sstevel@tonic-gate /* update the number of open files for mi */ 35587c478bd9Sstevel@tonic-gate mi->mi_open_files++; 35597c478bd9Sstevel@tonic-gate } 35607c478bd9Sstevel@tonic-gate 35617c478bd9Sstevel@tonic-gate void 35627c478bd9Sstevel@tonic-gate nfs4_dec_state_ref_count(mntinfo4_t *mi) 35637c478bd9Sstevel@tonic-gate { 35647c478bd9Sstevel@tonic-gate nfs4_server_t *sp; 35657c478bd9Sstevel@tonic-gate 35667c478bd9Sstevel@tonic-gate /* this locks down sp if it is found */ 35677c478bd9Sstevel@tonic-gate sp = find_nfs4_server_all(mi, 1); 35687c478bd9Sstevel@tonic-gate 35697c478bd9Sstevel@tonic-gate if (sp != NULL) { 35707c478bd9Sstevel@tonic-gate nfs4_dec_state_ref_count_nolock(sp, mi); 35717c478bd9Sstevel@tonic-gate mutex_exit(&sp->s_lock); 35727c478bd9Sstevel@tonic-gate nfs4_server_rele(sp); 35737c478bd9Sstevel@tonic-gate } 35747c478bd9Sstevel@tonic-gate } 35757c478bd9Sstevel@tonic-gate 35767c478bd9Sstevel@tonic-gate /* 35777c478bd9Sstevel@tonic-gate * Decrement the number of OPEN files (ie: those with state) so we know if 35787c478bd9Sstevel@tonic-gate * this nfs4_server has any state to maintain a lease for or not. 35797c478bd9Sstevel@tonic-gate */ 35807c478bd9Sstevel@tonic-gate void 35817c478bd9Sstevel@tonic-gate nfs4_dec_state_ref_count_nolock(nfs4_server_t *sp, mntinfo4_t *mi) 35827c478bd9Sstevel@tonic-gate { 35837c478bd9Sstevel@tonic-gate ASSERT(mutex_owned(&sp->s_lock)); 35847c478bd9Sstevel@tonic-gate ASSERT(sp->state_ref_count != 0); 35857c478bd9Sstevel@tonic-gate sp->state_ref_count--; 35867c478bd9Sstevel@tonic-gate 35877c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3588b9238976Sth "nfs4_dec_state_ref_count: state ref count now %d", 3589b9238976Sth sp->state_ref_count)); 35907c478bd9Sstevel@tonic-gate 35917c478bd9Sstevel@tonic-gate mi->mi_open_files--; 35927c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3593b9238976Sth "nfs4_dec_state_ref_count: mi open files %d, v4 flags 0x%x", 3594b9238976Sth mi->mi_open_files, mi->mi_flags)); 35957c478bd9Sstevel@tonic-gate 35967c478bd9Sstevel@tonic-gate /* We don't have to hold the mi_lock to test mi_flags */ 35977c478bd9Sstevel@tonic-gate if (mi->mi_open_files == 0 && 35987c478bd9Sstevel@tonic-gate (mi->mi_flags & MI4_REMOVE_ON_LAST_CLOSE)) { 35997c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3600b9238976Sth "nfs4_dec_state_ref_count: remove mntinfo4 %p since " 3601b9238976Sth "we have closed the last open file", (void*)mi)); 36027c478bd9Sstevel@tonic-gate nfs4_remove_mi_from_server(mi, sp); 36037c478bd9Sstevel@tonic-gate } 36047c478bd9Sstevel@tonic-gate } 36057c478bd9Sstevel@tonic-gate 36067c478bd9Sstevel@tonic-gate bool_t 36077c478bd9Sstevel@tonic-gate inlease(nfs4_server_t *sp) 36087c478bd9Sstevel@tonic-gate { 36097c478bd9Sstevel@tonic-gate bool_t result; 36107c478bd9Sstevel@tonic-gate 36117c478bd9Sstevel@tonic-gate ASSERT(mutex_owned(&sp->s_lock)); 36127c478bd9Sstevel@tonic-gate 36137c478bd9Sstevel@tonic-gate if (sp->lease_valid == NFS4_LEASE_VALID && 36147c478bd9Sstevel@tonic-gate gethrestime_sec() < sp->last_renewal_time + sp->s_lease_time) 36157c478bd9Sstevel@tonic-gate result = TRUE; 36167c478bd9Sstevel@tonic-gate else 36177c478bd9Sstevel@tonic-gate result = FALSE; 36187c478bd9Sstevel@tonic-gate 36197c478bd9Sstevel@tonic-gate return (result); 36207c478bd9Sstevel@tonic-gate } 36217c478bd9Sstevel@tonic-gate 36227c478bd9Sstevel@tonic-gate 36237c478bd9Sstevel@tonic-gate /* 36247c478bd9Sstevel@tonic-gate * Return non-zero if the given nfs4_server_t is going through recovery. 36257c478bd9Sstevel@tonic-gate */ 36267c478bd9Sstevel@tonic-gate 36277c478bd9Sstevel@tonic-gate int 36287c478bd9Sstevel@tonic-gate nfs4_server_in_recovery(nfs4_server_t *sp) 36297c478bd9Sstevel@tonic-gate { 36307c478bd9Sstevel@tonic-gate return (nfs_rw_lock_held(&sp->s_recovlock, RW_WRITER)); 36317c478bd9Sstevel@tonic-gate } 36327c478bd9Sstevel@tonic-gate 36337c478bd9Sstevel@tonic-gate /* 36347c478bd9Sstevel@tonic-gate * Compare two shared filehandle objects. Returns -1, 0, or +1, if the 36357c478bd9Sstevel@tonic-gate * first is less than, equal to, or greater than the second. 36367c478bd9Sstevel@tonic-gate */ 36377c478bd9Sstevel@tonic-gate 36387c478bd9Sstevel@tonic-gate int 36397c478bd9Sstevel@tonic-gate sfh4cmp(const void *p1, const void *p2) 36407c478bd9Sstevel@tonic-gate { 36417c478bd9Sstevel@tonic-gate const nfs4_sharedfh_t *sfh1 = (const nfs4_sharedfh_t *)p1; 36427c478bd9Sstevel@tonic-gate const nfs4_sharedfh_t *sfh2 = (const nfs4_sharedfh_t *)p2; 36437c478bd9Sstevel@tonic-gate 36447c478bd9Sstevel@tonic-gate return (nfs4cmpfh(&sfh1->sfh_fh, &sfh2->sfh_fh)); 36457c478bd9Sstevel@tonic-gate } 36467c478bd9Sstevel@tonic-gate 36477c478bd9Sstevel@tonic-gate /* 36487c478bd9Sstevel@tonic-gate * Create a table for shared filehandle objects. 36497c478bd9Sstevel@tonic-gate */ 36507c478bd9Sstevel@tonic-gate 36517c478bd9Sstevel@tonic-gate void 36527c478bd9Sstevel@tonic-gate sfh4_createtab(avl_tree_t *tab) 36537c478bd9Sstevel@tonic-gate { 36547c478bd9Sstevel@tonic-gate avl_create(tab, sfh4cmp, sizeof (nfs4_sharedfh_t), 3655b9238976Sth offsetof(nfs4_sharedfh_t, sfh_tree)); 36567c478bd9Sstevel@tonic-gate } 36577c478bd9Sstevel@tonic-gate 36587c478bd9Sstevel@tonic-gate /* 36597c478bd9Sstevel@tonic-gate * Return a shared filehandle object for the given filehandle. The caller 36607c478bd9Sstevel@tonic-gate * is responsible for eventually calling sfh4_rele(). 36617c478bd9Sstevel@tonic-gate */ 36627c478bd9Sstevel@tonic-gate 36637c478bd9Sstevel@tonic-gate nfs4_sharedfh_t * 36647c478bd9Sstevel@tonic-gate sfh4_put(const nfs_fh4 *fh, mntinfo4_t *mi, nfs4_sharedfh_t *key) 36657c478bd9Sstevel@tonic-gate { 36667c478bd9Sstevel@tonic-gate nfs4_sharedfh_t *sfh, *nsfh; 36677c478bd9Sstevel@tonic-gate avl_index_t where; 36687c478bd9Sstevel@tonic-gate nfs4_sharedfh_t skey; 36697c478bd9Sstevel@tonic-gate 36707c478bd9Sstevel@tonic-gate if (!key) { 36717c478bd9Sstevel@tonic-gate skey.sfh_fh = *fh; 36727c478bd9Sstevel@tonic-gate key = &skey; 36737c478bd9Sstevel@tonic-gate } 36747c478bd9Sstevel@tonic-gate 36757c478bd9Sstevel@tonic-gate nsfh = kmem_alloc(sizeof (nfs4_sharedfh_t), KM_SLEEP); 36767c478bd9Sstevel@tonic-gate nsfh->sfh_fh.nfs_fh4_len = fh->nfs_fh4_len; 36777c478bd9Sstevel@tonic-gate /* 36787c478bd9Sstevel@tonic-gate * We allocate the largest possible filehandle size because it's 36797c478bd9Sstevel@tonic-gate * not that big, and it saves us from possibly having to resize the 36807c478bd9Sstevel@tonic-gate * buffer later. 36817c478bd9Sstevel@tonic-gate */ 36827c478bd9Sstevel@tonic-gate nsfh->sfh_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP); 36837c478bd9Sstevel@tonic-gate bcopy(fh->nfs_fh4_val, nsfh->sfh_fh.nfs_fh4_val, fh->nfs_fh4_len); 36847c478bd9Sstevel@tonic-gate mutex_init(&nsfh->sfh_lock, NULL, MUTEX_DEFAULT, NULL); 36857c478bd9Sstevel@tonic-gate nsfh->sfh_refcnt = 1; 36867c478bd9Sstevel@tonic-gate nsfh->sfh_flags = SFH4_IN_TREE; 36877c478bd9Sstevel@tonic-gate nsfh->sfh_mi = mi; 36887c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE, "sfh4_get: new object (%p)", 3689b9238976Sth (void *)nsfh)); 36907c478bd9Sstevel@tonic-gate 36917c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_WRITER, 0); 36927c478bd9Sstevel@tonic-gate sfh = avl_find(&mi->mi_filehandles, key, &where); 36937c478bd9Sstevel@tonic-gate if (sfh != NULL) { 36947c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 36957c478bd9Sstevel@tonic-gate sfh->sfh_refcnt++; 36967c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 36977c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 36987c478bd9Sstevel@tonic-gate /* free our speculative allocs */ 36997c478bd9Sstevel@tonic-gate kmem_free(nsfh->sfh_fh.nfs_fh4_val, NFS4_FHSIZE); 37007c478bd9Sstevel@tonic-gate kmem_free(nsfh, sizeof (nfs4_sharedfh_t)); 37017c478bd9Sstevel@tonic-gate return (sfh); 37027c478bd9Sstevel@tonic-gate } 37037c478bd9Sstevel@tonic-gate 37047c478bd9Sstevel@tonic-gate avl_insert(&mi->mi_filehandles, nsfh, where); 37057c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 37067c478bd9Sstevel@tonic-gate 37077c478bd9Sstevel@tonic-gate return (nsfh); 37087c478bd9Sstevel@tonic-gate } 37097c478bd9Sstevel@tonic-gate 37107c478bd9Sstevel@tonic-gate /* 37117c478bd9Sstevel@tonic-gate * Return a shared filehandle object for the given filehandle. The caller 37127c478bd9Sstevel@tonic-gate * is responsible for eventually calling sfh4_rele(). 37137c478bd9Sstevel@tonic-gate */ 37147c478bd9Sstevel@tonic-gate 37157c478bd9Sstevel@tonic-gate nfs4_sharedfh_t * 37167c478bd9Sstevel@tonic-gate sfh4_get(const nfs_fh4 *fh, mntinfo4_t *mi) 37177c478bd9Sstevel@tonic-gate { 37187c478bd9Sstevel@tonic-gate nfs4_sharedfh_t *sfh; 37197c478bd9Sstevel@tonic-gate nfs4_sharedfh_t key; 37207c478bd9Sstevel@tonic-gate 37217c478bd9Sstevel@tonic-gate ASSERT(fh->nfs_fh4_len <= NFS4_FHSIZE); 37227c478bd9Sstevel@tonic-gate 37237c478bd9Sstevel@tonic-gate #ifdef DEBUG 37247c478bd9Sstevel@tonic-gate if (nfs4_sharedfh_debug) { 37257c478bd9Sstevel@tonic-gate nfs4_fhandle_t fhandle; 37267c478bd9Sstevel@tonic-gate 37277c478bd9Sstevel@tonic-gate fhandle.fh_len = fh->nfs_fh4_len; 37287c478bd9Sstevel@tonic-gate bcopy(fh->nfs_fh4_val, fhandle.fh_buf, fhandle.fh_len); 37297c478bd9Sstevel@tonic-gate zcmn_err(mi->mi_zone->zone_id, CE_NOTE, "sfh4_get:"); 37307c478bd9Sstevel@tonic-gate nfs4_printfhandle(&fhandle); 37317c478bd9Sstevel@tonic-gate } 37327c478bd9Sstevel@tonic-gate #endif 37337c478bd9Sstevel@tonic-gate 37347c478bd9Sstevel@tonic-gate /* 37357c478bd9Sstevel@tonic-gate * If there's already an object for the given filehandle, bump the 37367c478bd9Sstevel@tonic-gate * reference count and return it. Otherwise, create a new object 37377c478bd9Sstevel@tonic-gate * and add it to the AVL tree. 37387c478bd9Sstevel@tonic-gate */ 37397c478bd9Sstevel@tonic-gate 37407c478bd9Sstevel@tonic-gate key.sfh_fh = *fh; 37417c478bd9Sstevel@tonic-gate 37427c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_READER, 0); 37437c478bd9Sstevel@tonic-gate sfh = avl_find(&mi->mi_filehandles, &key, NULL); 37447c478bd9Sstevel@tonic-gate if (sfh != NULL) { 37457c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 37467c478bd9Sstevel@tonic-gate sfh->sfh_refcnt++; 37477c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE, 3748b9238976Sth "sfh4_get: found existing %p, new refcnt=%d", 3749b9238976Sth (void *)sfh, sfh->sfh_refcnt)); 37507c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 37517c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 37527c478bd9Sstevel@tonic-gate return (sfh); 37537c478bd9Sstevel@tonic-gate } 37547c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 37557c478bd9Sstevel@tonic-gate 37567c478bd9Sstevel@tonic-gate return (sfh4_put(fh, mi, &key)); 37577c478bd9Sstevel@tonic-gate } 37587c478bd9Sstevel@tonic-gate 37597c478bd9Sstevel@tonic-gate /* 37607c478bd9Sstevel@tonic-gate * Get a reference to the given shared filehandle object. 37617c478bd9Sstevel@tonic-gate */ 37627c478bd9Sstevel@tonic-gate 37637c478bd9Sstevel@tonic-gate void 37647c478bd9Sstevel@tonic-gate sfh4_hold(nfs4_sharedfh_t *sfh) 37657c478bd9Sstevel@tonic-gate { 37667c478bd9Sstevel@tonic-gate ASSERT(sfh->sfh_refcnt > 0); 37677c478bd9Sstevel@tonic-gate 37687c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 37697c478bd9Sstevel@tonic-gate sfh->sfh_refcnt++; 37707c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_sharedfh_debug, 3771b9238976Sth (CE_NOTE, "sfh4_hold %p, new refcnt=%d", 3772b9238976Sth (void *)sfh, sfh->sfh_refcnt)); 37737c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 37747c478bd9Sstevel@tonic-gate } 37757c478bd9Sstevel@tonic-gate 37767c478bd9Sstevel@tonic-gate /* 37777c478bd9Sstevel@tonic-gate * Release a reference to the given shared filehandle object and null out 37787c478bd9Sstevel@tonic-gate * the given pointer. 37797c478bd9Sstevel@tonic-gate */ 37807c478bd9Sstevel@tonic-gate 37817c478bd9Sstevel@tonic-gate void 37827c478bd9Sstevel@tonic-gate sfh4_rele(nfs4_sharedfh_t **sfhpp) 37837c478bd9Sstevel@tonic-gate { 37847c478bd9Sstevel@tonic-gate mntinfo4_t *mi; 37857c478bd9Sstevel@tonic-gate nfs4_sharedfh_t *sfh = *sfhpp; 37867c478bd9Sstevel@tonic-gate 37877c478bd9Sstevel@tonic-gate ASSERT(sfh->sfh_refcnt > 0); 37887c478bd9Sstevel@tonic-gate 37897c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 37907c478bd9Sstevel@tonic-gate if (sfh->sfh_refcnt > 1) { 37917c478bd9Sstevel@tonic-gate sfh->sfh_refcnt--; 37927c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE, 37937c478bd9Sstevel@tonic-gate "sfh4_rele %p, new refcnt=%d", 37947c478bd9Sstevel@tonic-gate (void *)sfh, sfh->sfh_refcnt)); 37957c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 37967c478bd9Sstevel@tonic-gate goto finish; 37977c478bd9Sstevel@tonic-gate } 37987c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 37997c478bd9Sstevel@tonic-gate 38007c478bd9Sstevel@tonic-gate /* 38017c478bd9Sstevel@tonic-gate * Possibly the last reference, so get the lock for the table in 38027c478bd9Sstevel@tonic-gate * case it's time to remove the object from the table. 38037c478bd9Sstevel@tonic-gate */ 38047c478bd9Sstevel@tonic-gate mi = sfh->sfh_mi; 38057c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_WRITER, 0); 38067c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 38077c478bd9Sstevel@tonic-gate sfh->sfh_refcnt--; 38087c478bd9Sstevel@tonic-gate if (sfh->sfh_refcnt > 0) { 38097c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE, 38107c478bd9Sstevel@tonic-gate "sfh4_rele %p, new refcnt=%d", 38117c478bd9Sstevel@tonic-gate (void *)sfh, sfh->sfh_refcnt)); 38127c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 38137c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 38147c478bd9Sstevel@tonic-gate goto finish; 38157c478bd9Sstevel@tonic-gate } 38167c478bd9Sstevel@tonic-gate 38177c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE, 3818b9238976Sth "sfh4_rele %p, last ref", (void *)sfh)); 38197c478bd9Sstevel@tonic-gate if (sfh->sfh_flags & SFH4_IN_TREE) { 38207c478bd9Sstevel@tonic-gate avl_remove(&mi->mi_filehandles, sfh); 38217c478bd9Sstevel@tonic-gate sfh->sfh_flags &= ~SFH4_IN_TREE; 38227c478bd9Sstevel@tonic-gate } 38237c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 38247c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 38257c478bd9Sstevel@tonic-gate mutex_destroy(&sfh->sfh_lock); 38267c478bd9Sstevel@tonic-gate kmem_free(sfh->sfh_fh.nfs_fh4_val, NFS4_FHSIZE); 38277c478bd9Sstevel@tonic-gate kmem_free(sfh, sizeof (nfs4_sharedfh_t)); 38287c478bd9Sstevel@tonic-gate 38297c478bd9Sstevel@tonic-gate finish: 38307c478bd9Sstevel@tonic-gate *sfhpp = NULL; 38317c478bd9Sstevel@tonic-gate } 38327c478bd9Sstevel@tonic-gate 38337c478bd9Sstevel@tonic-gate /* 38347c478bd9Sstevel@tonic-gate * Update the filehandle for the given shared filehandle object. 38357c478bd9Sstevel@tonic-gate */ 38367c478bd9Sstevel@tonic-gate 38377c478bd9Sstevel@tonic-gate int nfs4_warn_dupfh = 0; /* if set, always warn about dup fhs below */ 38387c478bd9Sstevel@tonic-gate 38397c478bd9Sstevel@tonic-gate void 38407c478bd9Sstevel@tonic-gate sfh4_update(nfs4_sharedfh_t *sfh, const nfs_fh4 *newfh) 38417c478bd9Sstevel@tonic-gate { 38427c478bd9Sstevel@tonic-gate mntinfo4_t *mi = sfh->sfh_mi; 38437c478bd9Sstevel@tonic-gate nfs4_sharedfh_t *dupsfh; 38447c478bd9Sstevel@tonic-gate avl_index_t where; 38457c478bd9Sstevel@tonic-gate nfs4_sharedfh_t key; 38467c478bd9Sstevel@tonic-gate 38477c478bd9Sstevel@tonic-gate #ifdef DEBUG 38487c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 38497c478bd9Sstevel@tonic-gate ASSERT(sfh->sfh_refcnt > 0); 38507c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 38517c478bd9Sstevel@tonic-gate #endif 38527c478bd9Sstevel@tonic-gate ASSERT(newfh->nfs_fh4_len <= NFS4_FHSIZE); 38537c478bd9Sstevel@tonic-gate 38547c478bd9Sstevel@tonic-gate /* 38557c478bd9Sstevel@tonic-gate * The basic plan is to remove the shared filehandle object from 38567c478bd9Sstevel@tonic-gate * the table, update it to have the new filehandle, then reinsert 38577c478bd9Sstevel@tonic-gate * it. 38587c478bd9Sstevel@tonic-gate */ 38597c478bd9Sstevel@tonic-gate 38607c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_WRITER, 0); 38617c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 38627c478bd9Sstevel@tonic-gate if (sfh->sfh_flags & SFH4_IN_TREE) { 38637c478bd9Sstevel@tonic-gate avl_remove(&mi->mi_filehandles, sfh); 38647c478bd9Sstevel@tonic-gate sfh->sfh_flags &= ~SFH4_IN_TREE; 38657c478bd9Sstevel@tonic-gate } 38667c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 38677c478bd9Sstevel@tonic-gate sfh->sfh_fh.nfs_fh4_len = newfh->nfs_fh4_len; 38687c478bd9Sstevel@tonic-gate bcopy(newfh->nfs_fh4_val, sfh->sfh_fh.nfs_fh4_val, 38697c478bd9Sstevel@tonic-gate sfh->sfh_fh.nfs_fh4_len); 38707c478bd9Sstevel@tonic-gate 38717c478bd9Sstevel@tonic-gate /* 38727c478bd9Sstevel@tonic-gate * XXX If there is already a shared filehandle object with the new 38737c478bd9Sstevel@tonic-gate * filehandle, we're in trouble, because the rnode code assumes 38747c478bd9Sstevel@tonic-gate * that there is only one shared filehandle object for a given 38757c478bd9Sstevel@tonic-gate * filehandle. So issue a warning (for read-write mounts only) 38767c478bd9Sstevel@tonic-gate * and don't try to re-insert the given object into the table. 38777c478bd9Sstevel@tonic-gate * Hopefully the given object will quickly go away and everyone 38787c478bd9Sstevel@tonic-gate * will use the new object. 38797c478bd9Sstevel@tonic-gate */ 38807c478bd9Sstevel@tonic-gate key.sfh_fh = *newfh; 38817c478bd9Sstevel@tonic-gate dupsfh = avl_find(&mi->mi_filehandles, &key, &where); 38827c478bd9Sstevel@tonic-gate if (dupsfh != NULL) { 38837c478bd9Sstevel@tonic-gate if (!(mi->mi_vfsp->vfs_flag & VFS_RDONLY) || nfs4_warn_dupfh) { 38847c478bd9Sstevel@tonic-gate zcmn_err(mi->mi_zone->zone_id, CE_WARN, "sfh4_update: " 38857c478bd9Sstevel@tonic-gate "duplicate filehandle detected"); 38867c478bd9Sstevel@tonic-gate sfh4_printfhandle(dupsfh); 38877c478bd9Sstevel@tonic-gate } 38887c478bd9Sstevel@tonic-gate } else { 38897c478bd9Sstevel@tonic-gate avl_insert(&mi->mi_filehandles, sfh, where); 38907c478bd9Sstevel@tonic-gate mutex_enter(&sfh->sfh_lock); 38917c478bd9Sstevel@tonic-gate sfh->sfh_flags |= SFH4_IN_TREE; 38927c478bd9Sstevel@tonic-gate mutex_exit(&sfh->sfh_lock); 38937c478bd9Sstevel@tonic-gate } 38947c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 38957c478bd9Sstevel@tonic-gate } 38967c478bd9Sstevel@tonic-gate 38977c478bd9Sstevel@tonic-gate /* 38987c478bd9Sstevel@tonic-gate * Copy out the current filehandle for the given shared filehandle object. 38997c478bd9Sstevel@tonic-gate */ 39007c478bd9Sstevel@tonic-gate 39017c478bd9Sstevel@tonic-gate void 39027c478bd9Sstevel@tonic-gate sfh4_copyval(const nfs4_sharedfh_t *sfh, nfs4_fhandle_t *fhp) 39037c478bd9Sstevel@tonic-gate { 39047c478bd9Sstevel@tonic-gate mntinfo4_t *mi = sfh->sfh_mi; 39057c478bd9Sstevel@tonic-gate 39067c478bd9Sstevel@tonic-gate ASSERT(sfh->sfh_refcnt > 0); 39077c478bd9Sstevel@tonic-gate 39087c478bd9Sstevel@tonic-gate (void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_READER, 0); 39097c478bd9Sstevel@tonic-gate fhp->fh_len = sfh->sfh_fh.nfs_fh4_len; 39107c478bd9Sstevel@tonic-gate ASSERT(fhp->fh_len <= NFS4_FHSIZE); 39117c478bd9Sstevel@tonic-gate bcopy(sfh->sfh_fh.nfs_fh4_val, fhp->fh_buf, fhp->fh_len); 39127c478bd9Sstevel@tonic-gate nfs_rw_exit(&mi->mi_fh_lock); 39137c478bd9Sstevel@tonic-gate } 39147c478bd9Sstevel@tonic-gate 39157c478bd9Sstevel@tonic-gate /* 39167c478bd9Sstevel@tonic-gate * Print out the filehandle for the given shared filehandle object. 39177c478bd9Sstevel@tonic-gate */ 39187c478bd9Sstevel@tonic-gate 39197c478bd9Sstevel@tonic-gate void 39207c478bd9Sstevel@tonic-gate sfh4_printfhandle(const nfs4_sharedfh_t *sfh) 39217c478bd9Sstevel@tonic-gate { 39227c478bd9Sstevel@tonic-gate nfs4_fhandle_t fhandle; 39237c478bd9Sstevel@tonic-gate 39247c478bd9Sstevel@tonic-gate sfh4_copyval(sfh, &fhandle); 39257c478bd9Sstevel@tonic-gate nfs4_printfhandle(&fhandle); 39267c478bd9Sstevel@tonic-gate } 39277c478bd9Sstevel@tonic-gate 39287c478bd9Sstevel@tonic-gate /* 39297c478bd9Sstevel@tonic-gate * Compare 2 fnames. Returns -1 if the first is "less" than the second, 0 39307c478bd9Sstevel@tonic-gate * if they're the same, +1 if the first is "greater" than the second. The 39317c478bd9Sstevel@tonic-gate * caller (or whoever's calling the AVL package) is responsible for 39327c478bd9Sstevel@tonic-gate * handling locking issues. 39337c478bd9Sstevel@tonic-gate */ 39347c478bd9Sstevel@tonic-gate 39357c478bd9Sstevel@tonic-gate static int 39367c478bd9Sstevel@tonic-gate fncmp(const void *p1, const void *p2) 39377c478bd9Sstevel@tonic-gate { 39387c478bd9Sstevel@tonic-gate const nfs4_fname_t *f1 = p1; 39397c478bd9Sstevel@tonic-gate const nfs4_fname_t *f2 = p2; 39407c478bd9Sstevel@tonic-gate int res; 39417c478bd9Sstevel@tonic-gate 39427c478bd9Sstevel@tonic-gate res = strcmp(f1->fn_name, f2->fn_name); 39437c478bd9Sstevel@tonic-gate /* 39447c478bd9Sstevel@tonic-gate * The AVL package wants +/-1, not arbitrary positive or negative 39457c478bd9Sstevel@tonic-gate * integers. 39467c478bd9Sstevel@tonic-gate */ 39477c478bd9Sstevel@tonic-gate if (res > 0) 39487c478bd9Sstevel@tonic-gate res = 1; 39497c478bd9Sstevel@tonic-gate else if (res < 0) 39507c478bd9Sstevel@tonic-gate res = -1; 39517c478bd9Sstevel@tonic-gate return (res); 39527c478bd9Sstevel@tonic-gate } 39537c478bd9Sstevel@tonic-gate 39547c478bd9Sstevel@tonic-gate /* 39557c478bd9Sstevel@tonic-gate * Get or create an fname with the given name, as a child of the given 39567c478bd9Sstevel@tonic-gate * fname. The caller is responsible for eventually releasing the reference 39577c478bd9Sstevel@tonic-gate * (fn_rele()). parent may be NULL. 39587c478bd9Sstevel@tonic-gate */ 39597c478bd9Sstevel@tonic-gate 39607c478bd9Sstevel@tonic-gate nfs4_fname_t * 39617c478bd9Sstevel@tonic-gate fn_get(nfs4_fname_t *parent, char *name) 39627c478bd9Sstevel@tonic-gate { 39637c478bd9Sstevel@tonic-gate nfs4_fname_t key; 39647c478bd9Sstevel@tonic-gate nfs4_fname_t *fnp; 39657c478bd9Sstevel@tonic-gate avl_index_t where; 39667c478bd9Sstevel@tonic-gate 39677c478bd9Sstevel@tonic-gate key.fn_name = name; 39687c478bd9Sstevel@tonic-gate 39697c478bd9Sstevel@tonic-gate /* 39707c478bd9Sstevel@tonic-gate * If there's already an fname registered with the given name, bump 39717c478bd9Sstevel@tonic-gate * its reference count and return it. Otherwise, create a new one 39727c478bd9Sstevel@tonic-gate * and add it to the parent's AVL tree. 39737c478bd9Sstevel@tonic-gate */ 39747c478bd9Sstevel@tonic-gate 39757c478bd9Sstevel@tonic-gate if (parent != NULL) { 39767c478bd9Sstevel@tonic-gate mutex_enter(&parent->fn_lock); 39777c478bd9Sstevel@tonic-gate fnp = avl_find(&parent->fn_children, &key, &where); 39787c478bd9Sstevel@tonic-gate if (fnp != NULL) { 39797c478bd9Sstevel@tonic-gate fn_hold(fnp); 39807c478bd9Sstevel@tonic-gate mutex_exit(&parent->fn_lock); 39817c478bd9Sstevel@tonic-gate return (fnp); 39827c478bd9Sstevel@tonic-gate } 39837c478bd9Sstevel@tonic-gate } 39847c478bd9Sstevel@tonic-gate 39857c478bd9Sstevel@tonic-gate fnp = kmem_alloc(sizeof (nfs4_fname_t), KM_SLEEP); 39867c478bd9Sstevel@tonic-gate mutex_init(&fnp->fn_lock, NULL, MUTEX_DEFAULT, NULL); 39877c478bd9Sstevel@tonic-gate fnp->fn_parent = parent; 39887c478bd9Sstevel@tonic-gate if (parent != NULL) 39897c478bd9Sstevel@tonic-gate fn_hold(parent); 39907c478bd9Sstevel@tonic-gate fnp->fn_len = strlen(name); 39917c478bd9Sstevel@tonic-gate ASSERT(fnp->fn_len < MAXNAMELEN); 39927c478bd9Sstevel@tonic-gate fnp->fn_name = kmem_alloc(fnp->fn_len + 1, KM_SLEEP); 39937c478bd9Sstevel@tonic-gate (void) strcpy(fnp->fn_name, name); 39947c478bd9Sstevel@tonic-gate fnp->fn_refcnt = 1; 39957c478bd9Sstevel@tonic-gate avl_create(&fnp->fn_children, fncmp, sizeof (nfs4_fname_t), 39967c478bd9Sstevel@tonic-gate offsetof(nfs4_fname_t, fn_tree)); 39977c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE, 3998b9238976Sth "fn_get %p:%s, a new nfs4_fname_t!", 3999b9238976Sth (void *)fnp, fnp->fn_name)); 40007c478bd9Sstevel@tonic-gate if (parent != NULL) { 40017c478bd9Sstevel@tonic-gate avl_insert(&parent->fn_children, fnp, where); 40027c478bd9Sstevel@tonic-gate mutex_exit(&parent->fn_lock); 40037c478bd9Sstevel@tonic-gate } 40047c478bd9Sstevel@tonic-gate 40057c478bd9Sstevel@tonic-gate return (fnp); 40067c478bd9Sstevel@tonic-gate } 40077c478bd9Sstevel@tonic-gate 40087c478bd9Sstevel@tonic-gate void 40097c478bd9Sstevel@tonic-gate fn_hold(nfs4_fname_t *fnp) 40107c478bd9Sstevel@tonic-gate { 40117c478bd9Sstevel@tonic-gate atomic_add_32(&fnp->fn_refcnt, 1); 40127c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE, 4013b9238976Sth "fn_hold %p:%s, new refcnt=%d", 4014b9238976Sth (void *)fnp, fnp->fn_name, fnp->fn_refcnt)); 40157c478bd9Sstevel@tonic-gate } 40167c478bd9Sstevel@tonic-gate 40177c478bd9Sstevel@tonic-gate /* 40187c478bd9Sstevel@tonic-gate * Decrement the reference count of the given fname, and destroy it if its 40197c478bd9Sstevel@tonic-gate * reference count goes to zero. Nulls out the given pointer. 40207c478bd9Sstevel@tonic-gate */ 40217c478bd9Sstevel@tonic-gate 40227c478bd9Sstevel@tonic-gate void 40237c478bd9Sstevel@tonic-gate fn_rele(nfs4_fname_t **fnpp) 40247c478bd9Sstevel@tonic-gate { 40257c478bd9Sstevel@tonic-gate nfs4_fname_t *parent; 40267c478bd9Sstevel@tonic-gate uint32_t newref; 40277c478bd9Sstevel@tonic-gate nfs4_fname_t *fnp; 40287c478bd9Sstevel@tonic-gate 40297c478bd9Sstevel@tonic-gate recur: 40307c478bd9Sstevel@tonic-gate fnp = *fnpp; 40317c478bd9Sstevel@tonic-gate *fnpp = NULL; 40327c478bd9Sstevel@tonic-gate 40337c478bd9Sstevel@tonic-gate mutex_enter(&fnp->fn_lock); 40347c478bd9Sstevel@tonic-gate parent = fnp->fn_parent; 40357c478bd9Sstevel@tonic-gate if (parent != NULL) 40367c478bd9Sstevel@tonic-gate mutex_enter(&parent->fn_lock); /* prevent new references */ 40377c478bd9Sstevel@tonic-gate newref = atomic_add_32_nv(&fnp->fn_refcnt, -1); 40387c478bd9Sstevel@tonic-gate if (newref > 0) { 40397c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE, 4040b9238976Sth "fn_rele %p:%s, new refcnt=%d", 4041b9238976Sth (void *)fnp, fnp->fn_name, fnp->fn_refcnt)); 40427c478bd9Sstevel@tonic-gate if (parent != NULL) 40437c478bd9Sstevel@tonic-gate mutex_exit(&parent->fn_lock); 40447c478bd9Sstevel@tonic-gate mutex_exit(&fnp->fn_lock); 40457c478bd9Sstevel@tonic-gate return; 40467c478bd9Sstevel@tonic-gate } 40477c478bd9Sstevel@tonic-gate 40487c478bd9Sstevel@tonic-gate NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE, 4049b9238976Sth "fn_rele %p:%s, last reference, deleting...", 4050b9238976Sth (void *)fnp, fnp->fn_name)); 40517c478bd9Sstevel@tonic-gate if (parent != NULL) { 40527c478bd9Sstevel@tonic-gate avl_remove(&parent->fn_children, fnp); 40537c478bd9Sstevel@tonic-gate mutex_exit(&parent->fn_lock); 40547c478bd9Sstevel@tonic-gate } 40557c478bd9Sstevel@tonic-gate kmem_free(fnp->fn_name, fnp->fn_len + 1); 40567c478bd9Sstevel@tonic-gate mutex_destroy(&fnp->fn_lock); 40577c478bd9Sstevel@tonic-gate avl_destroy(&fnp->fn_children); 40587c478bd9Sstevel@tonic-gate kmem_free(fnp, sizeof (nfs4_fname_t)); 40597c478bd9Sstevel@tonic-gate /* 40607c478bd9Sstevel@tonic-gate * Recursivly fn_rele the parent. 40617c478bd9Sstevel@tonic-gate * Use goto instead of a recursive call to avoid stack overflow. 40627c478bd9Sstevel@tonic-gate */ 40637c478bd9Sstevel@tonic-gate if (parent != NULL) { 40647c478bd9Sstevel@tonic-gate fnpp = &parent; 40657c478bd9Sstevel@tonic-gate goto recur; 40667c478bd9Sstevel@tonic-gate } 40677c478bd9Sstevel@tonic-gate } 40687c478bd9Sstevel@tonic-gate 40697c478bd9Sstevel@tonic-gate /* 40707c478bd9Sstevel@tonic-gate * Returns the single component name of the given fname, in a MAXNAMELEN 40717c478bd9Sstevel@tonic-gate * string buffer, which the caller is responsible for freeing. Note that 40727c478bd9Sstevel@tonic-gate * the name may become invalid as a result of fn_move(). 40737c478bd9Sstevel@tonic-gate */ 40747c478bd9Sstevel@tonic-gate 40757c478bd9Sstevel@tonic-gate char * 40767c478bd9Sstevel@tonic-gate fn_name(nfs4_fname_t *fnp) 40777c478bd9Sstevel@tonic-gate { 40787c478bd9Sstevel@tonic-gate char *name; 40797c478bd9Sstevel@tonic-gate 40807c478bd9Sstevel@tonic-gate ASSERT(fnp->fn_len < MAXNAMELEN); 40817c478bd9Sstevel@tonic-gate name = kmem_alloc(MAXNAMELEN, KM_SLEEP); 40827c478bd9Sstevel@tonic-gate mutex_enter(&fnp->fn_lock); 40837c478bd9Sstevel@tonic-gate (void) strcpy(name, fnp->fn_name); 40847c478bd9Sstevel@tonic-gate mutex_exit(&fnp->fn_lock); 40857c478bd9Sstevel@tonic-gate 40867c478bd9Sstevel@tonic-gate return (name); 40877c478bd9Sstevel@tonic-gate } 40887c478bd9Sstevel@tonic-gate 40897c478bd9Sstevel@tonic-gate 40907c478bd9Sstevel@tonic-gate /* 40917c478bd9Sstevel@tonic-gate * fn_path_realloc 40927c478bd9Sstevel@tonic-gate * 40937c478bd9Sstevel@tonic-gate * This function, used only by fn_path, constructs 40947c478bd9Sstevel@tonic-gate * a new string which looks like "prepend" + "/" + "current". 40957c478bd9Sstevel@tonic-gate * by allocating a new string and freeing the old one. 40967c478bd9Sstevel@tonic-gate */ 40977c478bd9Sstevel@tonic-gate static void 40987c478bd9Sstevel@tonic-gate fn_path_realloc(char **curses, char *prepend) 40997c478bd9Sstevel@tonic-gate { 41007c478bd9Sstevel@tonic-gate int len, curlen = 0; 41017c478bd9Sstevel@tonic-gate char *news; 41027c478bd9Sstevel@tonic-gate 41037c478bd9Sstevel@tonic-gate if (*curses == NULL) { 41047c478bd9Sstevel@tonic-gate /* 41057c478bd9Sstevel@tonic-gate * Prime the pump, allocate just the 41067c478bd9Sstevel@tonic-gate * space for prepend and return that. 41077c478bd9Sstevel@tonic-gate */ 41087c478bd9Sstevel@tonic-gate len = strlen(prepend) + 1; 41097c478bd9Sstevel@tonic-gate news = kmem_alloc(len, KM_SLEEP); 41107c478bd9Sstevel@tonic-gate (void) strncpy(news, prepend, len); 41117c478bd9Sstevel@tonic-gate } else { 41127c478bd9Sstevel@tonic-gate /* 41137c478bd9Sstevel@tonic-gate * Allocate the space for a new string 41147c478bd9Sstevel@tonic-gate * +1 +1 is for the "/" and the NULL 41157c478bd9Sstevel@tonic-gate * byte at the end of it all. 41167c478bd9Sstevel@tonic-gate */ 41177c478bd9Sstevel@tonic-gate curlen = strlen(*curses); 41187c478bd9Sstevel@tonic-gate len = curlen + strlen(prepend) + 1 + 1; 41197c478bd9Sstevel@tonic-gate news = kmem_alloc(len, KM_SLEEP); 41207c478bd9Sstevel@tonic-gate (void) strncpy(news, prepend, len); 41217c478bd9Sstevel@tonic-gate (void) strcat(news, "/"); 41227c478bd9Sstevel@tonic-gate (void) strcat(news, *curses); 41237c478bd9Sstevel@tonic-gate kmem_free(*curses, curlen + 1); 41247c478bd9Sstevel@tonic-gate } 41257c478bd9Sstevel@tonic-gate *curses = news; 41267c478bd9Sstevel@tonic-gate } 41277c478bd9Sstevel@tonic-gate 41287c478bd9Sstevel@tonic-gate /* 41297c478bd9Sstevel@tonic-gate * Returns the path name (starting from the fs root) for the given fname. 41307c478bd9Sstevel@tonic-gate * The caller is responsible for freeing. Note that the path may be or 41317c478bd9Sstevel@tonic-gate * become invalid as a result of fn_move(). 41327c478bd9Sstevel@tonic-gate */ 41337c478bd9Sstevel@tonic-gate 41347c478bd9Sstevel@tonic-gate char * 41357c478bd9Sstevel@tonic-gate fn_path(nfs4_fname_t *fnp) 41367c478bd9Sstevel@tonic-gate { 41377c478bd9Sstevel@tonic-gate char *path; 41387c478bd9Sstevel@tonic-gate nfs4_fname_t *nextfnp; 41397c478bd9Sstevel@tonic-gate 41407c478bd9Sstevel@tonic-gate if (fnp == NULL) 41417c478bd9Sstevel@tonic-gate return (NULL); 41427c478bd9Sstevel@tonic-gate 41437c478bd9Sstevel@tonic-gate path = NULL; 41447c478bd9Sstevel@tonic-gate 41457c478bd9Sstevel@tonic-gate /* walk up the tree constructing the pathname. */ 41467c478bd9Sstevel@tonic-gate 41477c478bd9Sstevel@tonic-gate fn_hold(fnp); /* adjust for later rele */ 41487c478bd9Sstevel@tonic-gate do { 41497c478bd9Sstevel@tonic-gate mutex_enter(&fnp->fn_lock); 41507c478bd9Sstevel@tonic-gate /* 41517c478bd9Sstevel@tonic-gate * Add fn_name in front of the current path 41527c478bd9Sstevel@tonic-gate */ 41537c478bd9Sstevel@tonic-gate fn_path_realloc(&path, fnp->fn_name); 41547c478bd9Sstevel@tonic-gate nextfnp = fnp->fn_parent; 41557c478bd9Sstevel@tonic-gate if (nextfnp != NULL) 41567c478bd9Sstevel@tonic-gate fn_hold(nextfnp); 41577c478bd9Sstevel@tonic-gate mutex_exit(&fnp->fn_lock); 41587c478bd9Sstevel@tonic-gate fn_rele(&fnp); 41597c478bd9Sstevel@tonic-gate fnp = nextfnp; 41607c478bd9Sstevel@tonic-gate } while (fnp != NULL); 41617c478bd9Sstevel@tonic-gate 41627c478bd9Sstevel@tonic-gate return (path); 41637c478bd9Sstevel@tonic-gate } 41647c478bd9Sstevel@tonic-gate 41657c478bd9Sstevel@tonic-gate /* 41667c478bd9Sstevel@tonic-gate * Return a reference to the parent of the given fname, which the caller is 41677c478bd9Sstevel@tonic-gate * responsible for eventually releasing. 41687c478bd9Sstevel@tonic-gate */ 41697c478bd9Sstevel@tonic-gate 41707c478bd9Sstevel@tonic-gate nfs4_fname_t * 41717c478bd9Sstevel@tonic-gate fn_parent(nfs4_fname_t *fnp) 41727c478bd9Sstevel@tonic-gate { 41737c478bd9Sstevel@tonic-gate nfs4_fname_t *parent; 41747c478bd9Sstevel@tonic-gate 41757c478bd9Sstevel@tonic-gate mutex_enter(&fnp->fn_lock); 41767c478bd9Sstevel@tonic-gate parent = fnp->fn_parent; 41777c478bd9Sstevel@tonic-gate if (parent != NULL) 41787c478bd9Sstevel@tonic-gate fn_hold(parent); 41797c478bd9Sstevel@tonic-gate mutex_exit(&fnp->fn_lock); 41807c478bd9Sstevel@tonic-gate 41817c478bd9Sstevel@tonic-gate return (parent); 41827c478bd9Sstevel@tonic-gate } 41837c478bd9Sstevel@tonic-gate 41847c478bd9Sstevel@tonic-gate /* 41857c478bd9Sstevel@tonic-gate * Update fnp so that its parent is newparent and its name is newname. 41867c478bd9Sstevel@tonic-gate */ 41877c478bd9Sstevel@tonic-gate 41887c478bd9Sstevel@tonic-gate void 41897c478bd9Sstevel@tonic-gate fn_move(nfs4_fname_t *fnp, nfs4_fname_t *newparent, char *newname) 41907c478bd9Sstevel@tonic-gate { 41917c478bd9Sstevel@tonic-gate nfs4_fname_t *parent, *tmpfnp; 41927c478bd9Sstevel@tonic-gate ssize_t newlen; 41937c478bd9Sstevel@tonic-gate nfs4_fname_t key; 41947c478bd9Sstevel@tonic-gate avl_index_t where; 41957c478bd9Sstevel@tonic-gate 41967c478bd9Sstevel@tonic-gate /* 41977c478bd9Sstevel@tonic-gate * This assert exists to catch the client trying to rename 41987c478bd9Sstevel@tonic-gate * a dir to be a child of itself. This happened at a recent 41997c478bd9Sstevel@tonic-gate * bakeoff against a 3rd party (broken) server which allowed 42007c478bd9Sstevel@tonic-gate * the rename to succeed. If it trips it means that: 42017c478bd9Sstevel@tonic-gate * a) the code in nfs4rename that detects this case is broken 42027c478bd9Sstevel@tonic-gate * b) the server is broken (since it allowed the bogus rename) 42037c478bd9Sstevel@tonic-gate * 42047c478bd9Sstevel@tonic-gate * For non-DEBUG kernels, prepare for a recursive mutex_enter 42057c478bd9Sstevel@tonic-gate * panic below from: mutex_enter(&newparent->fn_lock); 42067c478bd9Sstevel@tonic-gate */ 42077c478bd9Sstevel@tonic-gate ASSERT(fnp != newparent); 42087c478bd9Sstevel@tonic-gate 42097c478bd9Sstevel@tonic-gate /* 42107c478bd9Sstevel@tonic-gate * Remove fnp from its current parent, change its name, then add it 42117c478bd9Sstevel@tonic-gate * to newparent. 42127c478bd9Sstevel@tonic-gate */ 42137c478bd9Sstevel@tonic-gate mutex_enter(&fnp->fn_lock); 42147c478bd9Sstevel@tonic-gate parent = fnp->fn_parent; 42157c478bd9Sstevel@tonic-gate mutex_enter(&parent->fn_lock); 42167c478bd9Sstevel@tonic-gate avl_remove(&parent->fn_children, fnp); 42177c478bd9Sstevel@tonic-gate mutex_exit(&parent->fn_lock); 42187c478bd9Sstevel@tonic-gate fn_rele(&fnp->fn_parent); 42197c478bd9Sstevel@tonic-gate 42207c478bd9Sstevel@tonic-gate newlen = strlen(newname); 42217c478bd9Sstevel@tonic-gate if (newlen != fnp->fn_len) { 42227c478bd9Sstevel@tonic-gate ASSERT(newlen < MAXNAMELEN); 42237c478bd9Sstevel@tonic-gate kmem_free(fnp->fn_name, fnp->fn_len + 1); 42247c478bd9Sstevel@tonic-gate fnp->fn_name = kmem_alloc(newlen + 1, KM_SLEEP); 42257c478bd9Sstevel@tonic-gate fnp->fn_len = newlen; 42267c478bd9Sstevel@tonic-gate } 42277c478bd9Sstevel@tonic-gate (void) strcpy(fnp->fn_name, newname); 42287c478bd9Sstevel@tonic-gate 42297c478bd9Sstevel@tonic-gate again: 42307c478bd9Sstevel@tonic-gate mutex_enter(&newparent->fn_lock); 42317c478bd9Sstevel@tonic-gate key.fn_name = fnp->fn_name; 42327c478bd9Sstevel@tonic-gate tmpfnp = avl_find(&newparent->fn_children, &key, &where); 42337c478bd9Sstevel@tonic-gate if (tmpfnp != NULL) { 42347c478bd9Sstevel@tonic-gate /* 42357c478bd9Sstevel@tonic-gate * This could be due to a file that was unlinked while 42367c478bd9Sstevel@tonic-gate * open, or perhaps the rnode is in the free list. Remove 42377c478bd9Sstevel@tonic-gate * it from newparent and let it go away on its own. The 42387c478bd9Sstevel@tonic-gate * contorted code is to deal with lock order issues and 42397c478bd9Sstevel@tonic-gate * race conditions. 42407c478bd9Sstevel@tonic-gate */ 42417c478bd9Sstevel@tonic-gate fn_hold(tmpfnp); 42427c478bd9Sstevel@tonic-gate mutex_exit(&newparent->fn_lock); 42437c478bd9Sstevel@tonic-gate mutex_enter(&tmpfnp->fn_lock); 42447c478bd9Sstevel@tonic-gate if (tmpfnp->fn_parent == newparent) { 42457c478bd9Sstevel@tonic-gate mutex_enter(&newparent->fn_lock); 42467c478bd9Sstevel@tonic-gate avl_remove(&newparent->fn_children, tmpfnp); 42477c478bd9Sstevel@tonic-gate mutex_exit(&newparent->fn_lock); 42487c478bd9Sstevel@tonic-gate fn_rele(&tmpfnp->fn_parent); 42497c478bd9Sstevel@tonic-gate } 42507c478bd9Sstevel@tonic-gate mutex_exit(&tmpfnp->fn_lock); 42517c478bd9Sstevel@tonic-gate fn_rele(&tmpfnp); 42527c478bd9Sstevel@tonic-gate goto again; 42537c478bd9Sstevel@tonic-gate } 42547c478bd9Sstevel@tonic-gate fnp->fn_parent = newparent; 42557c478bd9Sstevel@tonic-gate fn_hold(newparent); 42567c478bd9Sstevel@tonic-gate avl_insert(&newparent->fn_children, fnp, where); 42577c478bd9Sstevel@tonic-gate mutex_exit(&newparent->fn_lock); 42587c478bd9Sstevel@tonic-gate mutex_exit(&fnp->fn_lock); 42597c478bd9Sstevel@tonic-gate } 42607c478bd9Sstevel@tonic-gate 42617c478bd9Sstevel@tonic-gate #ifdef DEBUG 42627c478bd9Sstevel@tonic-gate /* 42637c478bd9Sstevel@tonic-gate * Return non-zero if the type information makes sense for the given vnode. 42647c478bd9Sstevel@tonic-gate * Otherwise panic. 42657c478bd9Sstevel@tonic-gate */ 42667c478bd9Sstevel@tonic-gate int 42677c478bd9Sstevel@tonic-gate nfs4_consistent_type(vnode_t *vp) 42687c478bd9Sstevel@tonic-gate { 42697c478bd9Sstevel@tonic-gate rnode4_t *rp = VTOR4(vp); 42707c478bd9Sstevel@tonic-gate 42717c478bd9Sstevel@tonic-gate if (nfs4_vtype_debug && vp->v_type != VNON && 42727c478bd9Sstevel@tonic-gate rp->r_attr.va_type != VNON && vp->v_type != rp->r_attr.va_type) { 42737c478bd9Sstevel@tonic-gate cmn_err(CE_PANIC, "vnode %p type mismatch; v_type=%d, " 4274b9238976Sth "rnode attr type=%d", (void *)vp, vp->v_type, 4275b9238976Sth rp->r_attr.va_type); 42767c478bd9Sstevel@tonic-gate } 42777c478bd9Sstevel@tonic-gate 42787c478bd9Sstevel@tonic-gate return (1); 42797c478bd9Sstevel@tonic-gate } 42807c478bd9Sstevel@tonic-gate #endif /* DEBUG */ 4281