xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs_vnops.c (revision 8b80e8cb6855118d46f605e91b5ed4ce83417395)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  *
25  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
26  *	All rights reserved.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/types.h>
31 #include <sys/systm.h>
32 #include <sys/cred.h>
33 #include <sys/time.h>
34 #include <sys/vnode.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/file.h>
38 #include <sys/filio.h>
39 #include <sys/uio.h>
40 #include <sys/buf.h>
41 #include <sys/mman.h>
42 #include <sys/pathname.h>
43 #include <sys/dirent.h>
44 #include <sys/debug.h>
45 #include <sys/vmsystm.h>
46 #include <sys/fcntl.h>
47 #include <sys/flock.h>
48 #include <sys/swap.h>
49 #include <sys/errno.h>
50 #include <sys/strsubr.h>
51 #include <sys/sysmacros.h>
52 #include <sys/kmem.h>
53 #include <sys/cmn_err.h>
54 #include <sys/pathconf.h>
55 #include <sys/utsname.h>
56 #include <sys/dnlc.h>
57 #include <sys/acl.h>
58 #include <sys/atomic.h>
59 #include <sys/policy.h>
60 #include <sys/sdt.h>
61 
62 #include <rpc/types.h>
63 #include <rpc/auth.h>
64 #include <rpc/clnt.h>
65 
66 #include <nfs/nfs.h>
67 #include <nfs/nfs_clnt.h>
68 #include <nfs/rnode.h>
69 #include <nfs/nfs_acl.h>
70 #include <nfs/lm.h>
71 
72 #include <vm/hat.h>
73 #include <vm/as.h>
74 #include <vm/page.h>
75 #include <vm/pvn.h>
76 #include <vm/seg.h>
77 #include <vm/seg_map.h>
78 #include <vm/seg_kpm.h>
79 #include <vm/seg_vn.h>
80 
81 #include <fs/fs_subr.h>
82 
83 #include <sys/ddi.h>
84 
85 static int	nfs_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int,
86 			cred_t *);
87 static int	nfswrite(vnode_t *, caddr_t, uint_t, int, cred_t *);
88 static int	nfsread(vnode_t *, caddr_t, uint_t, int, size_t *, cred_t *);
89 static int	nfssetattr(vnode_t *, struct vattr *, int, cred_t *);
90 static int	nfslookup_dnlc(vnode_t *, char *, vnode_t **, cred_t *);
91 static int	nfslookup_otw(vnode_t *, char *, vnode_t **, cred_t *, int);
92 static int	nfsrename(vnode_t *, char *, vnode_t *, char *, cred_t *,
93 			caller_context_t *);
94 static int	nfsreaddir(vnode_t *, rddir_cache *, cred_t *);
95 static int	nfs_bio(struct buf *, cred_t *);
96 static int	nfs_getapage(vnode_t *, u_offset_t, size_t, uint_t *,
97 			page_t *[], size_t, struct seg *, caddr_t,
98 			enum seg_rw, cred_t *);
99 static void	nfs_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *,
100 			cred_t *);
101 static int	nfs_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t,
102 			int, cred_t *);
103 static int	nfs_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t,
104 			int, cred_t *);
105 static void	nfs_delmap_callback(struct as *, void *, uint_t);
106 
107 /*
108  * Error flags used to pass information about certain special errors
109  * which need to be handled specially.
110  */
111 #define	NFS_EOF			-98
112 
113 /*
114  * These are the vnode ops routines which implement the vnode interface to
115  * the networked file system.  These routines just take their parameters,
116  * make them look networkish by putting the right info into interface structs,
117  * and then calling the appropriate remote routine(s) to do the work.
118  *
119  * Note on directory name lookup cacheing:  If we detect a stale fhandle,
120  * we purge the directory cache relative to that vnode.  This way, the
121  * user won't get burned by the cache repeatedly.  See <nfs/rnode.h> for
122  * more details on rnode locking.
123  */
124 
125 static int	nfs_open(vnode_t **, int, cred_t *, caller_context_t *);
126 static int	nfs_close(vnode_t *, int, int, offset_t, cred_t *,
127 			caller_context_t *);
128 static int	nfs_read(vnode_t *, struct uio *, int, cred_t *,
129 			caller_context_t *);
130 static int	nfs_write(vnode_t *, struct uio *, int, cred_t *,
131 			caller_context_t *);
132 static int	nfs_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
133 			caller_context_t *);
134 static int	nfs_getattr(vnode_t *, struct vattr *, int, cred_t *,
135 			caller_context_t *);
136 static int	nfs_setattr(vnode_t *, struct vattr *, int, cred_t *,
137 			caller_context_t *);
138 static int	nfs_access(vnode_t *, int, int, cred_t *, caller_context_t *);
139 static int	nfs_accessx(void *, int, cred_t *);
140 static int	nfs_readlink(vnode_t *, struct uio *, cred_t *,
141 			caller_context_t *);
142 static int	nfs_fsync(vnode_t *, int, cred_t *, caller_context_t *);
143 static void	nfs_inactive(vnode_t *, cred_t *, caller_context_t *);
144 static int	nfs_lookup(vnode_t *, char *, vnode_t **, struct pathname *,
145 			int, vnode_t *, cred_t *, caller_context_t *,
146 			int *, pathname_t *);
147 static int	nfs_create(vnode_t *, char *, struct vattr *, enum vcexcl,
148 			int, vnode_t **, cred_t *, int, caller_context_t *,
149 			vsecattr_t *);
150 static int	nfs_remove(vnode_t *, char *, cred_t *, caller_context_t *,
151 			int);
152 static int	nfs_link(vnode_t *, vnode_t *, char *, cred_t *,
153 			caller_context_t *, int);
154 static int	nfs_rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
155 			caller_context_t *, int);
156 static int	nfs_mkdir(vnode_t *, char *, struct vattr *, vnode_t **,
157 			cred_t *, caller_context_t *, int, vsecattr_t *);
158 static int	nfs_rmdir(vnode_t *, char *, vnode_t *, cred_t *,
159 			caller_context_t *, int);
160 static int	nfs_symlink(vnode_t *, char *, struct vattr *, char *,
161 			cred_t *, caller_context_t *, int);
162 static int	nfs_readdir(vnode_t *, struct uio *, cred_t *, int *,
163 			caller_context_t *, int);
164 static int	nfs_fid(vnode_t *, fid_t *, caller_context_t *);
165 static int	nfs_rwlock(vnode_t *, int, caller_context_t *);
166 static void	nfs_rwunlock(vnode_t *, int, caller_context_t *);
167 static int	nfs_seek(vnode_t *, offset_t, offset_t *, caller_context_t *);
168 static int	nfs_getpage(vnode_t *, offset_t, size_t, uint_t *,
169 			page_t *[], size_t, struct seg *, caddr_t,
170 			enum seg_rw, cred_t *, caller_context_t *);
171 static int	nfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
172 			caller_context_t *);
173 static int	nfs_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
174 			uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
175 static int	nfs_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
176 			uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
177 static int	nfs_frlock(vnode_t *, int, struct flock64 *, int, offset_t,
178 			struct flk_callback *, cred_t *, caller_context_t *);
179 static int	nfs_space(vnode_t *, int, struct flock64 *, int, offset_t,
180 			cred_t *, caller_context_t *);
181 static int	nfs_realvp(vnode_t *, vnode_t **, caller_context_t *);
182 static int	nfs_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
183 			uint_t, uint_t, uint_t, cred_t *, caller_context_t *);
184 static int	nfs_pathconf(vnode_t *, int, ulong_t *, cred_t *,
185 			caller_context_t *);
186 static int	nfs_pageio(vnode_t *, page_t *, u_offset_t, size_t, int,
187 			cred_t *, caller_context_t *);
188 static int	nfs_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
189 			caller_context_t *);
190 static int	nfs_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
191 			caller_context_t *);
192 static int	nfs_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *,
193 			caller_context_t *);
194 
195 struct vnodeops *nfs_vnodeops;
196 
197 const fs_operation_def_t nfs_vnodeops_template[] = {
198 	VOPNAME_OPEN,		{ .vop_open = nfs_open },
199 	VOPNAME_CLOSE,		{ .vop_close = nfs_close },
200 	VOPNAME_READ,		{ .vop_read = nfs_read },
201 	VOPNAME_WRITE,		{ .vop_write = nfs_write },
202 	VOPNAME_IOCTL,		{ .vop_ioctl = nfs_ioctl },
203 	VOPNAME_GETATTR,	{ .vop_getattr = nfs_getattr },
204 	VOPNAME_SETATTR,	{ .vop_setattr = nfs_setattr },
205 	VOPNAME_ACCESS,		{ .vop_access = nfs_access },
206 	VOPNAME_LOOKUP,		{ .vop_lookup = nfs_lookup },
207 	VOPNAME_CREATE,		{ .vop_create = nfs_create },
208 	VOPNAME_REMOVE,		{ .vop_remove = nfs_remove },
209 	VOPNAME_LINK,		{ .vop_link = nfs_link },
210 	VOPNAME_RENAME,		{ .vop_rename = nfs_rename },
211 	VOPNAME_MKDIR,		{ .vop_mkdir = nfs_mkdir },
212 	VOPNAME_RMDIR,		{ .vop_rmdir = nfs_rmdir },
213 	VOPNAME_READDIR,	{ .vop_readdir = nfs_readdir },
214 	VOPNAME_SYMLINK,	{ .vop_symlink = nfs_symlink },
215 	VOPNAME_READLINK,	{ .vop_readlink = nfs_readlink },
216 	VOPNAME_FSYNC,		{ .vop_fsync = nfs_fsync },
217 	VOPNAME_INACTIVE,	{ .vop_inactive = nfs_inactive },
218 	VOPNAME_FID,		{ .vop_fid = nfs_fid },
219 	VOPNAME_RWLOCK,		{ .vop_rwlock = nfs_rwlock },
220 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = nfs_rwunlock },
221 	VOPNAME_SEEK,		{ .vop_seek = nfs_seek },
222 	VOPNAME_FRLOCK,		{ .vop_frlock = nfs_frlock },
223 	VOPNAME_SPACE,		{ .vop_space = nfs_space },
224 	VOPNAME_REALVP,		{ .vop_realvp = nfs_realvp },
225 	VOPNAME_GETPAGE,	{ .vop_getpage = nfs_getpage },
226 	VOPNAME_PUTPAGE,	{ .vop_putpage = nfs_putpage },
227 	VOPNAME_MAP,		{ .vop_map = nfs_map },
228 	VOPNAME_ADDMAP,		{ .vop_addmap = nfs_addmap },
229 	VOPNAME_DELMAP,		{ .vop_delmap = nfs_delmap },
230 	VOPNAME_DUMP,		{ .vop_dump = nfs_dump },
231 	VOPNAME_PATHCONF,	{ .vop_pathconf = nfs_pathconf },
232 	VOPNAME_PAGEIO,		{ .vop_pageio = nfs_pageio },
233 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = nfs_setsecattr },
234 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = nfs_getsecattr },
235 	VOPNAME_SHRLOCK,	{ .vop_shrlock = nfs_shrlock },
236 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
237 	NULL,			NULL
238 };
239 
240 /*
241  * XXX:  This is referenced in modstubs.s
242  */
243 struct vnodeops *
244 nfs_getvnodeops(void)
245 {
246 	return (nfs_vnodeops);
247 }
248 
249 /* ARGSUSED */
250 static int
251 nfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
252 {
253 	int error;
254 	struct vattr va;
255 	rnode_t *rp;
256 	vnode_t *vp;
257 
258 	vp = *vpp;
259 	rp = VTOR(vp);
260 	if (nfs_zone() != VTOMI(vp)->mi_zone)
261 		return (EIO);
262 	mutex_enter(&rp->r_statelock);
263 	if (rp->r_cred == NULL) {
264 		crhold(cr);
265 		rp->r_cred = cr;
266 	}
267 	mutex_exit(&rp->r_statelock);
268 
269 	/*
270 	 * If there is no cached data or if close-to-open
271 	 * consistency checking is turned off, we can avoid
272 	 * the over the wire getattr.  Otherwise, if the
273 	 * file system is mounted readonly, then just verify
274 	 * the caches are up to date using the normal mechanism.
275 	 * Else, if the file is not mmap'd, then just mark
276 	 * the attributes as timed out.  They will be refreshed
277 	 * and the caches validated prior to being used.
278 	 * Else, the file system is mounted writeable so
279 	 * force an over the wire GETATTR in order to ensure
280 	 * that all cached data is valid.
281 	 */
282 	if (vp->v_count > 1 ||
283 	    ((vn_has_cached_data(vp) || HAVE_RDDIR_CACHE(rp)) &&
284 	    !(VTOMI(vp)->mi_flags & MI_NOCTO))) {
285 		if (vn_is_readonly(vp))
286 			error = nfs_validate_caches(vp, cr);
287 		else if (rp->r_mapcnt == 0 && vp->v_count == 1) {
288 			PURGE_ATTRCACHE(vp);
289 			error = 0;
290 		} else {
291 			va.va_mask = AT_ALL;
292 			error = nfs_getattr_otw(vp, &va, cr);
293 		}
294 	} else
295 		error = 0;
296 
297 	return (error);
298 }
299 
300 /* ARGSUSED */
301 static int
302 nfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
303 	caller_context_t *ct)
304 {
305 	rnode_t *rp;
306 	int error;
307 	struct vattr va;
308 
309 	/*
310 	 * zone_enter(2) prevents processes from changing zones with NFS files
311 	 * open; if we happen to get here from the wrong zone we can't do
312 	 * anything over the wire.
313 	 */
314 	if (VTOMI(vp)->mi_zone != nfs_zone()) {
315 		/*
316 		 * We could attempt to clean up locks, except we're sure
317 		 * that the current process didn't acquire any locks on
318 		 * the file: any attempt to lock a file belong to another zone
319 		 * will fail, and one can't lock an NFS file and then change
320 		 * zones, as that fails too.
321 		 *
322 		 * Returning an error here is the sane thing to do.  A
323 		 * subsequent call to VN_RELE() which translates to a
324 		 * nfs_inactive() will clean up state: if the zone of the
325 		 * vnode's origin is still alive and kicking, an async worker
326 		 * thread will handle the request (from the correct zone), and
327 		 * everything (minus the final nfs_getattr_otw() call) should
328 		 * be OK. If the zone is going away nfs_async_inactive() will
329 		 * throw away cached pages inline.
330 		 */
331 		return (EIO);
332 	}
333 
334 	/*
335 	 * If we are using local locking for this filesystem, then
336 	 * release all of the SYSV style record locks.  Otherwise,
337 	 * we are doing network locking and we need to release all
338 	 * of the network locks.  All of the locks held by this
339 	 * process on this file are released no matter what the
340 	 * incoming reference count is.
341 	 */
342 	if (VTOMI(vp)->mi_flags & MI_LLOCK) {
343 		cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
344 		cleanshares(vp, ttoproc(curthread)->p_pid);
345 	} else
346 		nfs_lockrelease(vp, flag, offset, cr);
347 
348 	if (count > 1)
349 		return (0);
350 
351 	/*
352 	 * If the file has been `unlinked', then purge the
353 	 * DNLC so that this vnode will get reycled quicker
354 	 * and the .nfs* file on the server will get removed.
355 	 */
356 	rp = VTOR(vp);
357 	if (rp->r_unldvp != NULL)
358 		dnlc_purge_vp(vp);
359 
360 	/*
361 	 * If the file was open for write and there are pages,
362 	 * then if the file system was mounted using the "no-close-
363 	 *	to-open" semantics, then start an asynchronous flush
364 	 *	of the all of the pages in the file.
365 	 * else the file system was not mounted using the "no-close-
366 	 *	to-open" semantics, then do a synchronous flush and
367 	 *	commit of all of the dirty and uncommitted pages.
368 	 *
369 	 * The asynchronous flush of the pages in the "nocto" path
370 	 * mostly just associates a cred pointer with the rnode so
371 	 * writes which happen later will have a better chance of
372 	 * working.  It also starts the data being written to the
373 	 * server, but without unnecessarily delaying the application.
374 	 */
375 	if ((flag & FWRITE) && vn_has_cached_data(vp)) {
376 		if ((VTOMI(vp)->mi_flags & MI_NOCTO)) {
377 			error = nfs_putpage(vp, (offset_t)0, 0, B_ASYNC,
378 			    cr, ct);
379 			if (error == EAGAIN)
380 				error = 0;
381 		} else
382 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
383 		if (!error) {
384 			mutex_enter(&rp->r_statelock);
385 			error = rp->r_error;
386 			rp->r_error = 0;
387 			mutex_exit(&rp->r_statelock);
388 		}
389 	} else {
390 		mutex_enter(&rp->r_statelock);
391 		error = rp->r_error;
392 		rp->r_error = 0;
393 		mutex_exit(&rp->r_statelock);
394 	}
395 
396 	/*
397 	 * If RWRITEATTR is set, then issue an over the wire GETATTR to
398 	 * refresh the attribute cache with a set of attributes which
399 	 * weren't returned from a WRITE.  This will enable the close-
400 	 * to-open processing to work.
401 	 */
402 	if (rp->r_flags & RWRITEATTR)
403 		(void) nfs_getattr_otw(vp, &va, cr);
404 
405 	return (error);
406 }
407 
408 /* ARGSUSED */
409 static int
410 nfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
411 	caller_context_t *ct)
412 {
413 	rnode_t *rp;
414 	u_offset_t off;
415 	offset_t diff;
416 	int on;
417 	size_t n;
418 	caddr_t base;
419 	uint_t flags;
420 	int error;
421 	mntinfo_t *mi;
422 
423 	rp = VTOR(vp);
424 	mi = VTOMI(vp);
425 
426 	if (nfs_zone() != mi->mi_zone)
427 		return (EIO);
428 
429 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
430 
431 	if (vp->v_type != VREG)
432 		return (EISDIR);
433 
434 	if (uiop->uio_resid == 0)
435 		return (0);
436 
437 	if (uiop->uio_loffset > MAXOFF32_T)
438 		return (EFBIG);
439 
440 	if (uiop->uio_loffset < 0 ||
441 	    uiop->uio_loffset + uiop->uio_resid > MAXOFF32_T)
442 		return (EINVAL);
443 
444 	/*
445 	 * Bypass VM if caching has been disabled (e.g., locking) or if
446 	 * using client-side direct I/O and the file is not mmap'd and
447 	 * there are no cached pages.
448 	 */
449 	if ((vp->v_flag & VNOCACHE) ||
450 	    (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
451 	    rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
452 	    !vn_has_cached_data(vp))) {
453 		size_t bufsize;
454 		size_t resid = 0;
455 
456 		/*
457 		 * Let's try to do read in as large a chunk as we can
458 		 * (Filesystem (NFS client) bsize if possible/needed).
459 		 * For V3, this is 32K and for V2, this is 8K.
460 		 */
461 		bufsize = MIN(uiop->uio_resid, VTOMI(vp)->mi_curread);
462 		base = kmem_alloc(bufsize, KM_SLEEP);
463 		do {
464 			n = MIN(uiop->uio_resid, bufsize);
465 			error = nfsread(vp, base, uiop->uio_offset, n,
466 			    &resid, cr);
467 			if (!error) {
468 				n -= resid;
469 				error = uiomove(base, n, UIO_READ, uiop);
470 			}
471 		} while (!error && uiop->uio_resid > 0 && n > 0);
472 		kmem_free(base, bufsize);
473 		return (error);
474 	}
475 
476 	error = 0;
477 
478 	do {
479 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
480 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
481 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
482 
483 		error = nfs_validate_caches(vp, cr);
484 		if (error)
485 			break;
486 
487 		mutex_enter(&rp->r_statelock);
488 		while (rp->r_flags & RINCACHEPURGE) {
489 			if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
490 				mutex_exit(&rp->r_statelock);
491 				return (EINTR);
492 			}
493 		}
494 		diff = rp->r_size - uiop->uio_loffset;
495 		mutex_exit(&rp->r_statelock);
496 		if (diff <= 0)
497 			break;
498 		if (diff < n)
499 			n = (size_t)diff;
500 
501 		if (vpm_enable) {
502 			/*
503 			 * Copy data.
504 			 */
505 			error = vpm_data_copy(vp, off + on, n, uiop,
506 			    1, NULL, 0, S_READ);
507 		} else {
508 			base = segmap_getmapflt(segkmap, vp, off + on, n,
509 			    1, S_READ);
510 			error = uiomove(base + on, n, UIO_READ, uiop);
511 		}
512 
513 		if (!error) {
514 			/*
515 			 * If read a whole block or read to eof,
516 			 * won't need this buffer again soon.
517 			 */
518 			mutex_enter(&rp->r_statelock);
519 			if (n + on == MAXBSIZE ||
520 			    uiop->uio_loffset == rp->r_size)
521 				flags = SM_DONTNEED;
522 			else
523 				flags = 0;
524 			mutex_exit(&rp->r_statelock);
525 			if (vpm_enable) {
526 				error = vpm_sync_pages(vp, off, n, flags);
527 			} else {
528 				error = segmap_release(segkmap, base, flags);
529 			}
530 		} else {
531 			if (vpm_enable) {
532 				(void) vpm_sync_pages(vp, off, n, 0);
533 			} else {
534 				(void) segmap_release(segkmap, base, 0);
535 			}
536 		}
537 	} while (!error && uiop->uio_resid > 0);
538 
539 	return (error);
540 }
541 
542 /* ARGSUSED */
543 static int
544 nfs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
545 	caller_context_t *ct)
546 {
547 	rnode_t *rp;
548 	u_offset_t off;
549 	caddr_t base;
550 	uint_t flags;
551 	int remainder;
552 	size_t n;
553 	int on;
554 	int error;
555 	int resid;
556 	offset_t offset;
557 	rlim_t limit;
558 	mntinfo_t *mi;
559 
560 	rp = VTOR(vp);
561 
562 	mi = VTOMI(vp);
563 	if (nfs_zone() != mi->mi_zone)
564 		return (EIO);
565 	if (vp->v_type != VREG)
566 		return (EISDIR);
567 
568 	if (uiop->uio_resid == 0)
569 		return (0);
570 
571 	if (ioflag & FAPPEND) {
572 		struct vattr va;
573 
574 		/*
575 		 * Must serialize if appending.
576 		 */
577 		if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) {
578 			nfs_rw_exit(&rp->r_rwlock);
579 			if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER,
580 			    INTR(vp)))
581 				return (EINTR);
582 		}
583 
584 		va.va_mask = AT_SIZE;
585 		error = nfsgetattr(vp, &va, cr);
586 		if (error)
587 			return (error);
588 		uiop->uio_loffset = va.va_size;
589 	}
590 
591 	if (uiop->uio_loffset > MAXOFF32_T)
592 		return (EFBIG);
593 
594 	offset = uiop->uio_loffset + uiop->uio_resid;
595 
596 	if (uiop->uio_loffset < 0 || offset > MAXOFF32_T)
597 		return (EINVAL);
598 
599 	if (uiop->uio_llimit > (rlim64_t)MAXOFF32_T) {
600 		limit = MAXOFF32_T;
601 	} else {
602 		limit = (rlim_t)uiop->uio_llimit;
603 	}
604 
605 	/*
606 	 * Check to make sure that the process will not exceed
607 	 * its limit on file size.  It is okay to write up to
608 	 * the limit, but not beyond.  Thus, the write which
609 	 * reaches the limit will be short and the next write
610 	 * will return an error.
611 	 */
612 	remainder = 0;
613 	if (offset > limit) {
614 		remainder = offset - limit;
615 		uiop->uio_resid = limit - uiop->uio_offset;
616 		if (uiop->uio_resid <= 0) {
617 			proc_t *p = ttoproc(curthread);
618 
619 			uiop->uio_resid += remainder;
620 			mutex_enter(&p->p_lock);
621 			(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE],
622 			    p->p_rctls, p, RCA_UNSAFE_SIGINFO);
623 			mutex_exit(&p->p_lock);
624 			return (EFBIG);
625 		}
626 	}
627 
628 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp)))
629 		return (EINTR);
630 
631 	/*
632 	 * Bypass VM if caching has been disabled (e.g., locking) or if
633 	 * using client-side direct I/O and the file is not mmap'd and
634 	 * there are no cached pages.
635 	 */
636 	if ((vp->v_flag & VNOCACHE) ||
637 	    (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
638 	    rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
639 	    !vn_has_cached_data(vp))) {
640 		size_t bufsize;
641 		int count;
642 		uint_t org_offset;
643 
644 nfs_fwrite:
645 		if (rp->r_flags & RSTALE) {
646 			resid = uiop->uio_resid;
647 			offset = uiop->uio_loffset;
648 			error = rp->r_error;
649 			goto bottom;
650 		}
651 		bufsize = MIN(uiop->uio_resid, mi->mi_curwrite);
652 		base = kmem_alloc(bufsize, KM_SLEEP);
653 		do {
654 			resid = uiop->uio_resid;
655 			offset = uiop->uio_loffset;
656 			count = MIN(uiop->uio_resid, bufsize);
657 			org_offset = uiop->uio_offset;
658 			error = uiomove(base, count, UIO_WRITE, uiop);
659 			if (!error) {
660 				error = nfswrite(vp, base, org_offset,
661 				    count, cr);
662 			}
663 		} while (!error && uiop->uio_resid > 0);
664 		kmem_free(base, bufsize);
665 		goto bottom;
666 	}
667 
668 	do {
669 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
670 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
671 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
672 
673 		resid = uiop->uio_resid;
674 		offset = uiop->uio_loffset;
675 
676 		if (rp->r_flags & RSTALE) {
677 			error = rp->r_error;
678 			break;
679 		}
680 
681 		/*
682 		 * Don't create dirty pages faster than they
683 		 * can be cleaned so that the system doesn't
684 		 * get imbalanced.  If the async queue is
685 		 * maxed out, then wait for it to drain before
686 		 * creating more dirty pages.  Also, wait for
687 		 * any threads doing pagewalks in the vop_getattr
688 		 * entry points so that they don't block for
689 		 * long periods.
690 		 */
691 		mutex_enter(&rp->r_statelock);
692 		while ((mi->mi_max_threads != 0 &&
693 		    rp->r_awcount > 2 * mi->mi_max_threads) ||
694 		    rp->r_gcount > 0)
695 			cv_wait(&rp->r_cv, &rp->r_statelock);
696 		mutex_exit(&rp->r_statelock);
697 
698 		/*
699 		 * Touch the page and fault it in if it is not in core
700 		 * before segmap_getmapflt or vpm_data_copy can lock it.
701 		 * This is to avoid the deadlock if the buffer is mapped
702 		 * to the same file through mmap which we want to write.
703 		 */
704 		uio_prefaultpages((long)n, uiop);
705 
706 		if (vpm_enable) {
707 			/*
708 			 * It will use kpm mappings, so no need to
709 			 * pass an address.
710 			 */
711 			error = writerp(rp, NULL, n, uiop, 0);
712 		} else  {
713 			if (segmap_kpm) {
714 				int pon = uiop->uio_loffset & PAGEOFFSET;
715 				size_t pn = MIN(PAGESIZE - pon,
716 				    uiop->uio_resid);
717 				int pagecreate;
718 
719 				mutex_enter(&rp->r_statelock);
720 				pagecreate = (pon == 0) && (pn == PAGESIZE ||
721 				    uiop->uio_loffset + pn >= rp->r_size);
722 				mutex_exit(&rp->r_statelock);
723 
724 				base = segmap_getmapflt(segkmap, vp, off + on,
725 				    pn, !pagecreate, S_WRITE);
726 
727 				error = writerp(rp, base + pon, n, uiop,
728 				    pagecreate);
729 
730 			} else {
731 				base = segmap_getmapflt(segkmap, vp, off + on,
732 				    n, 0, S_READ);
733 				error = writerp(rp, base + on, n, uiop, 0);
734 			}
735 		}
736 
737 		if (!error) {
738 			if (mi->mi_flags & MI_NOAC)
739 				flags = SM_WRITE;
740 			else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
741 				/*
742 				 * Have written a whole block.
743 				 * Start an asynchronous write
744 				 * and mark the buffer to
745 				 * indicate that it won't be
746 				 * needed again soon.
747 				 */
748 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
749 			} else
750 				flags = 0;
751 			if ((ioflag & (FSYNC|FDSYNC)) ||
752 			    (rp->r_flags & ROUTOFSPACE)) {
753 				flags &= ~SM_ASYNC;
754 				flags |= SM_WRITE;
755 			}
756 			if (vpm_enable) {
757 				error = vpm_sync_pages(vp, off, n, flags);
758 			} else {
759 				error = segmap_release(segkmap, base, flags);
760 			}
761 		} else {
762 			if (vpm_enable) {
763 				(void) vpm_sync_pages(vp, off, n, 0);
764 			} else {
765 				(void) segmap_release(segkmap, base, 0);
766 			}
767 			/*
768 			 * In the event that we got an access error while
769 			 * faulting in a page for a write-only file just
770 			 * force a write.
771 			 */
772 			if (error == EACCES)
773 				goto nfs_fwrite;
774 		}
775 	} while (!error && uiop->uio_resid > 0);
776 
777 bottom:
778 	if (error) {
779 		uiop->uio_resid = resid + remainder;
780 		uiop->uio_loffset = offset;
781 	} else
782 		uiop->uio_resid += remainder;
783 
784 	nfs_rw_exit(&rp->r_lkserlock);
785 
786 	return (error);
787 }
788 
789 /*
790  * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
791  */
792 static int
793 nfs_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
794 	int flags, cred_t *cr)
795 {
796 	struct buf *bp;
797 	int error;
798 
799 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
800 	bp = pageio_setup(pp, len, vp, flags);
801 	ASSERT(bp != NULL);
802 
803 	/*
804 	 * pageio_setup should have set b_addr to 0.  This
805 	 * is correct since we want to do I/O on a page
806 	 * boundary.  bp_mapin will use this addr to calculate
807 	 * an offset, and then set b_addr to the kernel virtual
808 	 * address it allocated for us.
809 	 */
810 	ASSERT(bp->b_un.b_addr == 0);
811 
812 	bp->b_edev = 0;
813 	bp->b_dev = 0;
814 	bp->b_lblkno = lbtodb(off);
815 	bp->b_file = vp;
816 	bp->b_offset = (offset_t)off;
817 	bp_mapin(bp);
818 
819 	error = nfs_bio(bp, cr);
820 
821 	bp_mapout(bp);
822 	pageio_done(bp);
823 
824 	return (error);
825 }
826 
827 /*
828  * Write to file.  Writes to remote server in largest size
829  * chunks that the server can handle.  Write is synchronous.
830  */
831 static int
832 nfswrite(vnode_t *vp, caddr_t base, uint_t offset, int count, cred_t *cr)
833 {
834 	rnode_t *rp;
835 	mntinfo_t *mi;
836 	struct nfswriteargs wa;
837 	struct nfsattrstat ns;
838 	int error;
839 	int tsize;
840 	int douprintf;
841 
842 	douprintf = 1;
843 
844 	rp = VTOR(vp);
845 	mi = VTOMI(vp);
846 
847 	ASSERT(nfs_zone() == mi->mi_zone);
848 
849 	wa.wa_args = &wa.wa_args_buf;
850 	wa.wa_fhandle = *VTOFH(vp);
851 
852 	do {
853 		tsize = MIN(mi->mi_curwrite, count);
854 		wa.wa_data = base;
855 		wa.wa_begoff = offset;
856 		wa.wa_totcount = tsize;
857 		wa.wa_count = tsize;
858 		wa.wa_offset = offset;
859 
860 		if (mi->mi_io_kstats) {
861 			mutex_enter(&mi->mi_lock);
862 			kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
863 			mutex_exit(&mi->mi_lock);
864 		}
865 		wa.wa_mblk = NULL;
866 		do {
867 			error = rfs2call(mi, RFS_WRITE,
868 			    xdr_writeargs, (caddr_t)&wa,
869 			    xdr_attrstat, (caddr_t)&ns, cr,
870 			    &douprintf, &ns.ns_status, 0, NULL);
871 		} while (error == ENFS_TRYAGAIN);
872 		if (mi->mi_io_kstats) {
873 			mutex_enter(&mi->mi_lock);
874 			kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
875 			mutex_exit(&mi->mi_lock);
876 		}
877 
878 		if (!error) {
879 			error = geterrno(ns.ns_status);
880 			/*
881 			 * Can't check for stale fhandle and purge caches
882 			 * here because pages are held by nfs_getpage.
883 			 * Just mark the attribute cache as timed out
884 			 * and set RWRITEATTR to indicate that the file
885 			 * was modified with a WRITE operation.
886 			 */
887 			if (!error) {
888 				count -= tsize;
889 				base += tsize;
890 				offset += tsize;
891 				if (mi->mi_io_kstats) {
892 					mutex_enter(&mi->mi_lock);
893 					KSTAT_IO_PTR(mi->mi_io_kstats)->
894 					    writes++;
895 					KSTAT_IO_PTR(mi->mi_io_kstats)->
896 					    nwritten += tsize;
897 					mutex_exit(&mi->mi_lock);
898 				}
899 				lwp_stat_update(LWP_STAT_OUBLK, 1);
900 				mutex_enter(&rp->r_statelock);
901 				PURGE_ATTRCACHE_LOCKED(rp);
902 				rp->r_flags |= RWRITEATTR;
903 				mutex_exit(&rp->r_statelock);
904 			}
905 		}
906 	} while (!error && count);
907 
908 	return (error);
909 }
910 
911 /*
912  * Read from a file.  Reads data in largest chunks our interface can handle.
913  */
914 static int
915 nfsread(vnode_t *vp, caddr_t base, uint_t offset,
916     int count, size_t *residp, cred_t *cr)
917 {
918 	mntinfo_t *mi;
919 	struct nfsreadargs ra;
920 	struct nfsrdresult rr;
921 	int tsize;
922 	int error;
923 	int douprintf;
924 	failinfo_t fi;
925 	rnode_t *rp;
926 	struct vattr va;
927 	hrtime_t t;
928 
929 	rp = VTOR(vp);
930 	mi = VTOMI(vp);
931 
932 	ASSERT(nfs_zone() == mi->mi_zone);
933 
934 	douprintf = 1;
935 
936 	ra.ra_fhandle = *VTOFH(vp);
937 
938 	fi.vp = vp;
939 	fi.fhp = (caddr_t)&ra.ra_fhandle;
940 	fi.copyproc = nfscopyfh;
941 	fi.lookupproc = nfslookup;
942 	fi.xattrdirproc = acl_getxattrdir2;
943 
944 	do {
945 		if (mi->mi_io_kstats) {
946 			mutex_enter(&mi->mi_lock);
947 			kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
948 			mutex_exit(&mi->mi_lock);
949 		}
950 
951 		do {
952 			tsize = MIN(mi->mi_curread, count);
953 			rr.rr_data = base;
954 			ra.ra_offset = offset;
955 			ra.ra_totcount = tsize;
956 			ra.ra_count = tsize;
957 			ra.ra_data = base;
958 			t = gethrtime();
959 			error = rfs2call(mi, RFS_READ,
960 			    xdr_readargs, (caddr_t)&ra,
961 			    xdr_rdresult, (caddr_t)&rr, cr,
962 			    &douprintf, &rr.rr_status, 0, &fi);
963 		} while (error == ENFS_TRYAGAIN);
964 
965 		if (mi->mi_io_kstats) {
966 			mutex_enter(&mi->mi_lock);
967 			kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
968 			mutex_exit(&mi->mi_lock);
969 		}
970 
971 		if (!error) {
972 			error = geterrno(rr.rr_status);
973 			if (!error) {
974 				count -= rr.rr_count;
975 				base += rr.rr_count;
976 				offset += rr.rr_count;
977 				if (mi->mi_io_kstats) {
978 					mutex_enter(&mi->mi_lock);
979 					KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
980 					KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
981 					    rr.rr_count;
982 					mutex_exit(&mi->mi_lock);
983 				}
984 				lwp_stat_update(LWP_STAT_INBLK, 1);
985 			}
986 		}
987 	} while (!error && count && rr.rr_count == tsize);
988 
989 	*residp = count;
990 
991 	if (!error) {
992 		/*
993 		 * Since no error occurred, we have the current
994 		 * attributes and we need to do a cache check and then
995 		 * potentially update the cached attributes.  We can't
996 		 * use the normal attribute check and cache mechanisms
997 		 * because they might cause a cache flush which would
998 		 * deadlock.  Instead, we just check the cache to see
999 		 * if the attributes have changed.  If it is, then we
1000 		 * just mark the attributes as out of date.  The next
1001 		 * time that the attributes are checked, they will be
1002 		 * out of date, new attributes will be fetched, and
1003 		 * the page cache will be flushed.  If the attributes
1004 		 * weren't changed, then we just update the cached
1005 		 * attributes with these attributes.
1006 		 */
1007 		/*
1008 		 * If NFS_ACL is supported on the server, then the
1009 		 * attributes returned by server may have minimal
1010 		 * permissions sometimes denying access to users having
1011 		 * proper access.  To get the proper attributes, mark
1012 		 * the attributes as expired so that they will be
1013 		 * regotten via the NFS_ACL GETATTR2 procedure.
1014 		 */
1015 		error = nattr_to_vattr(vp, &rr.rr_attr, &va);
1016 		mutex_enter(&rp->r_statelock);
1017 		if (error || !CACHE_VALID(rp, va.va_mtime, va.va_size) ||
1018 		    (mi->mi_flags & MI_ACL)) {
1019 			mutex_exit(&rp->r_statelock);
1020 			PURGE_ATTRCACHE(vp);
1021 		} else {
1022 			if (rp->r_mtime <= t) {
1023 				nfs_attrcache_va(vp, &va);
1024 			}
1025 			mutex_exit(&rp->r_statelock);
1026 		}
1027 	}
1028 
1029 	return (error);
1030 }
1031 
1032 /* ARGSUSED */
1033 static int
1034 nfs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp,
1035 	caller_context_t *ct)
1036 {
1037 
1038 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1039 		return (EIO);
1040 	switch (cmd) {
1041 		case _FIODIRECTIO:
1042 			return (nfs_directio(vp, (int)arg, cr));
1043 		default:
1044 			return (ENOTTY);
1045 	}
1046 }
1047 
1048 /* ARGSUSED */
1049 static int
1050 nfs_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1051 	caller_context_t *ct)
1052 {
1053 	int error;
1054 	rnode_t *rp;
1055 
1056 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1057 		return (EIO);
1058 	/*
1059 	 * If it has been specified that the return value will
1060 	 * just be used as a hint, and we are only being asked
1061 	 * for size, fsid or rdevid, then return the client's
1062 	 * notion of these values without checking to make sure
1063 	 * that the attribute cache is up to date.
1064 	 * The whole point is to avoid an over the wire GETATTR
1065 	 * call.
1066 	 */
1067 	rp = VTOR(vp);
1068 	if (flags & ATTR_HINT) {
1069 		if (vap->va_mask ==
1070 		    (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) {
1071 			mutex_enter(&rp->r_statelock);
1072 			if (vap->va_mask | AT_SIZE)
1073 				vap->va_size = rp->r_size;
1074 			if (vap->va_mask | AT_FSID)
1075 				vap->va_fsid = rp->r_attr.va_fsid;
1076 			if (vap->va_mask | AT_RDEV)
1077 				vap->va_rdev = rp->r_attr.va_rdev;
1078 			mutex_exit(&rp->r_statelock);
1079 			return (0);
1080 		}
1081 	}
1082 
1083 	/*
1084 	 * Only need to flush pages if asking for the mtime
1085 	 * and if there any dirty pages or any outstanding
1086 	 * asynchronous (write) requests for this file.
1087 	 */
1088 	if (vap->va_mask & AT_MTIME) {
1089 		if (vn_has_cached_data(vp) &&
1090 		    ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) {
1091 			mutex_enter(&rp->r_statelock);
1092 			rp->r_gcount++;
1093 			mutex_exit(&rp->r_statelock);
1094 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1095 			mutex_enter(&rp->r_statelock);
1096 			if (error && (error == ENOSPC || error == EDQUOT)) {
1097 				if (!rp->r_error)
1098 					rp->r_error = error;
1099 			}
1100 			if (--rp->r_gcount == 0)
1101 				cv_broadcast(&rp->r_cv);
1102 			mutex_exit(&rp->r_statelock);
1103 		}
1104 	}
1105 
1106 	return (nfsgetattr(vp, vap, cr));
1107 }
1108 
1109 /*ARGSUSED4*/
1110 static int
1111 nfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1112 		caller_context_t *ct)
1113 {
1114 	int error;
1115 	uint_t mask;
1116 	struct vattr va;
1117 
1118 	mask = vap->va_mask;
1119 
1120 	if (mask & AT_NOSET)
1121 		return (EINVAL);
1122 
1123 	if ((mask & AT_SIZE) &&
1124 	    vap->va_type == VREG &&
1125 	    vap->va_size > MAXOFF32_T)
1126 		return (EFBIG);
1127 
1128 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1129 		return (EIO);
1130 
1131 	va.va_mask = AT_UID | AT_MODE;
1132 
1133 	error = nfsgetattr(vp, &va, cr);
1134 	if (error)
1135 		return (error);
1136 
1137 	error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs_accessx,
1138 	    vp);
1139 
1140 	if (error)
1141 		return (error);
1142 
1143 	return (nfssetattr(vp, vap, flags, cr));
1144 }
1145 
1146 static int
1147 nfssetattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr)
1148 {
1149 	int error;
1150 	uint_t mask;
1151 	struct nfssaargs args;
1152 	struct nfsattrstat ns;
1153 	int douprintf;
1154 	rnode_t *rp;
1155 	struct vattr va;
1156 	mode_t omode;
1157 	mntinfo_t *mi;
1158 	vsecattr_t *vsp;
1159 	hrtime_t t;
1160 
1161 	mask = vap->va_mask;
1162 
1163 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
1164 
1165 	rp = VTOR(vp);
1166 
1167 	/*
1168 	 * Only need to flush pages if there are any pages and
1169 	 * if the file is marked as dirty in some fashion.  The
1170 	 * file must be flushed so that we can accurately
1171 	 * determine the size of the file and the cached data
1172 	 * after the SETATTR returns.  A file is considered to
1173 	 * be dirty if it is either marked with RDIRTY, has
1174 	 * outstanding i/o's active, or is mmap'd.  In this
1175 	 * last case, we can't tell whether there are dirty
1176 	 * pages, so we flush just to be sure.
1177 	 */
1178 	if (vn_has_cached_data(vp) &&
1179 	    ((rp->r_flags & RDIRTY) ||
1180 	    rp->r_count > 0 ||
1181 	    rp->r_mapcnt > 0)) {
1182 		ASSERT(vp->v_type != VCHR);
1183 		error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, NULL);
1184 		if (error && (error == ENOSPC || error == EDQUOT)) {
1185 			mutex_enter(&rp->r_statelock);
1186 			if (!rp->r_error)
1187 				rp->r_error = error;
1188 			mutex_exit(&rp->r_statelock);
1189 		}
1190 	}
1191 
1192 	/*
1193 	 * If the system call was utime(2) or utimes(2) and the
1194 	 * application did not specify the times, then set the
1195 	 * mtime nanosecond field to 1 billion.  This will get
1196 	 * translated from 1 billion nanoseconds to 1 million
1197 	 * microseconds in the over the wire request.  The
1198 	 * server will use 1 million in the microsecond field
1199 	 * to tell whether both the mtime and atime should be
1200 	 * set to the server's current time.
1201 	 *
1202 	 * This is an overload of the protocol and should be
1203 	 * documented in the NFS Version 2 protocol specification.
1204 	 */
1205 	if ((mask & AT_MTIME) && !(flags & ATTR_UTIME)) {
1206 		vap->va_mtime.tv_nsec = 1000000000;
1207 		if (NFS_TIME_T_OK(vap->va_mtime.tv_sec) &&
1208 		    NFS_TIME_T_OK(vap->va_atime.tv_sec)) {
1209 			error = vattr_to_sattr(vap, &args.saa_sa);
1210 		} else {
1211 			/*
1212 			 * Use server times. vap time values will not be used.
1213 			 * To ensure no time overflow, make sure vap has
1214 			 * valid values, but retain the original values.
1215 			 */
1216 			timestruc_t	mtime = vap->va_mtime;
1217 			timestruc_t	atime = vap->va_atime;
1218 			time_t		now;
1219 
1220 			now = gethrestime_sec();
1221 			if (NFS_TIME_T_OK(now)) {
1222 				/* Just in case server does not know of this */
1223 				vap->va_mtime.tv_sec = now;
1224 				vap->va_atime.tv_sec = now;
1225 			} else {
1226 				vap->va_mtime.tv_sec = 0;
1227 				vap->va_atime.tv_sec = 0;
1228 			}
1229 			error = vattr_to_sattr(vap, &args.saa_sa);
1230 			/* set vap times back on */
1231 			vap->va_mtime = mtime;
1232 			vap->va_atime = atime;
1233 		}
1234 	} else {
1235 		/* Either do not set times or use the client specified times */
1236 		error = vattr_to_sattr(vap, &args.saa_sa);
1237 	}
1238 	if (error) {
1239 		/* req time field(s) overflow - return immediately */
1240 		return (error);
1241 	}
1242 	args.saa_fh = *VTOFH(vp);
1243 
1244 	va.va_mask = AT_MODE;
1245 	error = nfsgetattr(vp, &va, cr);
1246 	if (error)
1247 		return (error);
1248 	omode = va.va_mode;
1249 
1250 	mi = VTOMI(vp);
1251 
1252 	douprintf = 1;
1253 
1254 	t = gethrtime();
1255 
1256 	error = rfs2call(mi, RFS_SETATTR,
1257 	    xdr_saargs, (caddr_t)&args,
1258 	    xdr_attrstat, (caddr_t)&ns, cr,
1259 	    &douprintf, &ns.ns_status, 0, NULL);
1260 
1261 	/*
1262 	 * Purge the access cache and ACL cache if changing either the
1263 	 * owner of the file, the group owner, or the mode.  These may
1264 	 * change the access permissions of the file, so purge old
1265 	 * information and start over again.
1266 	 */
1267 	if ((mask & (AT_UID | AT_GID | AT_MODE)) && (mi->mi_flags & MI_ACL)) {
1268 		(void) nfs_access_purge_rp(rp);
1269 		if (rp->r_secattr != NULL) {
1270 			mutex_enter(&rp->r_statelock);
1271 			vsp = rp->r_secattr;
1272 			rp->r_secattr = NULL;
1273 			mutex_exit(&rp->r_statelock);
1274 			if (vsp != NULL)
1275 				nfs_acl_free(vsp);
1276 		}
1277 	}
1278 
1279 	if (!error) {
1280 		error = geterrno(ns.ns_status);
1281 		if (!error) {
1282 			/*
1283 			 * If changing the size of the file, invalidate
1284 			 * any local cached data which is no longer part
1285 			 * of the file.  We also possibly invalidate the
1286 			 * last page in the file.  We could use
1287 			 * pvn_vpzero(), but this would mark the page as
1288 			 * modified and require it to be written back to
1289 			 * the server for no particularly good reason.
1290 			 * This way, if we access it, then we bring it
1291 			 * back in.  A read should be cheaper than a
1292 			 * write.
1293 			 */
1294 			if (mask & AT_SIZE) {
1295 				nfs_invalidate_pages(vp,
1296 				    (vap->va_size & PAGEMASK), cr);
1297 			}
1298 			(void) nfs_cache_fattr(vp, &ns.ns_attr, &va, t, cr);
1299 			/*
1300 			 * If NFS_ACL is supported on the server, then the
1301 			 * attributes returned by server may have minimal
1302 			 * permissions sometimes denying access to users having
1303 			 * proper access.  To get the proper attributes, mark
1304 			 * the attributes as expired so that they will be
1305 			 * regotten via the NFS_ACL GETATTR2 procedure.
1306 			 */
1307 			if (mi->mi_flags & MI_ACL) {
1308 				PURGE_ATTRCACHE(vp);
1309 			}
1310 			/*
1311 			 * This next check attempts to deal with NFS
1312 			 * servers which can not handle increasing
1313 			 * the size of the file via setattr.  Most
1314 			 * of these servers do not return an error,
1315 			 * but do not change the size of the file.
1316 			 * Hence, this check and then attempt to set
1317 			 * the file size by writing 1 byte at the
1318 			 * offset of the end of the file that we need.
1319 			 */
1320 			if ((mask & AT_SIZE) &&
1321 			    ns.ns_attr.na_size < (uint32_t)vap->va_size) {
1322 				char zb = '\0';
1323 
1324 				error = nfswrite(vp, &zb,
1325 				    vap->va_size - sizeof (zb),
1326 				    sizeof (zb), cr);
1327 			}
1328 			/*
1329 			 * Some servers will change the mode to clear the setuid
1330 			 * and setgid bits when changing the uid or gid.  The
1331 			 * client needs to compensate appropriately.
1332 			 */
1333 			if (mask & (AT_UID | AT_GID)) {
1334 				int terror;
1335 
1336 				va.va_mask = AT_MODE;
1337 				terror = nfsgetattr(vp, &va, cr);
1338 				if (!terror &&
1339 				    (((mask & AT_MODE) &&
1340 				    va.va_mode != vap->va_mode) ||
1341 				    (!(mask & AT_MODE) &&
1342 				    va.va_mode != omode))) {
1343 					va.va_mask = AT_MODE;
1344 					if (mask & AT_MODE)
1345 						va.va_mode = vap->va_mode;
1346 					else
1347 						va.va_mode = omode;
1348 					(void) nfssetattr(vp, &va, 0, cr);
1349 				}
1350 			}
1351 		} else {
1352 			PURGE_ATTRCACHE(vp);
1353 			PURGE_STALE_FH(error, vp, cr);
1354 		}
1355 	} else {
1356 		PURGE_ATTRCACHE(vp);
1357 	}
1358 
1359 	return (error);
1360 }
1361 
1362 static int
1363 nfs_accessx(void *vp, int mode, cred_t *cr)
1364 {
1365 	ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone);
1366 	return (nfs_access(vp, mode, 0, cr, NULL));
1367 }
1368 
1369 /* ARGSUSED */
1370 static int
1371 nfs_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
1372 {
1373 	struct vattr va;
1374 	int error;
1375 	mntinfo_t *mi;
1376 	int shift = 0;
1377 
1378 	mi = VTOMI(vp);
1379 
1380 	if (nfs_zone() != mi->mi_zone)
1381 		return (EIO);
1382 	if (mi->mi_flags & MI_ACL) {
1383 		error = acl_access2(vp, mode, flags, cr);
1384 		if (mi->mi_flags & MI_ACL)
1385 			return (error);
1386 	}
1387 
1388 	va.va_mask = AT_MODE | AT_UID | AT_GID;
1389 	error = nfsgetattr(vp, &va, cr);
1390 	if (error)
1391 		return (error);
1392 
1393 	/*
1394 	 * Disallow write attempts on read-only
1395 	 * file systems, unless the file is a
1396 	 * device node.
1397 	 */
1398 	if ((mode & VWRITE) && vn_is_readonly(vp) && !IS_DEVVP(vp))
1399 		return (EROFS);
1400 
1401 	/*
1402 	 * Disallow attempts to access mandatory lock files.
1403 	 */
1404 	if ((mode & (VWRITE | VREAD | VEXEC)) &&
1405 	    MANDLOCK(vp, va.va_mode))
1406 		return (EACCES);
1407 
1408 	/*
1409 	 * Access check is based on only
1410 	 * one of owner, group, public.
1411 	 * If not owner, then check group.
1412 	 * If not a member of the group,
1413 	 * then check public access.
1414 	 */
1415 	if (crgetuid(cr) != va.va_uid) {
1416 		shift += 3;
1417 		if (!groupmember(va.va_gid, cr))
1418 			shift += 3;
1419 	}
1420 found:
1421 	mode &= ~(va.va_mode << shift);
1422 	if (mode == 0)
1423 		return (0);
1424 
1425 	return (secpolicy_vnode_access(cr, vp, va.va_uid, mode));
1426 }
1427 
1428 static int nfs_do_symlink_cache = 1;
1429 
1430 /* ARGSUSED */
1431 static int
1432 nfs_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct)
1433 {
1434 	int error;
1435 	struct nfsrdlnres rl;
1436 	rnode_t *rp;
1437 	int douprintf;
1438 	failinfo_t fi;
1439 
1440 	/*
1441 	 * We want to be consistent with UFS semantics so we will return
1442 	 * EINVAL instead of ENXIO. This violates the XNFS spec and
1443 	 * the RFC 1094, which are wrong any way. BUGID 1138002.
1444 	 */
1445 	if (vp->v_type != VLNK)
1446 		return (EINVAL);
1447 
1448 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1449 		return (EIO);
1450 
1451 	rp = VTOR(vp);
1452 	if (nfs_do_symlink_cache && rp->r_symlink.contents != NULL) {
1453 		error = nfs_validate_caches(vp, cr);
1454 		if (error)
1455 			return (error);
1456 		mutex_enter(&rp->r_statelock);
1457 		if (rp->r_symlink.contents != NULL) {
1458 			error = uiomove(rp->r_symlink.contents,
1459 			    rp->r_symlink.len, UIO_READ, uiop);
1460 			mutex_exit(&rp->r_statelock);
1461 			return (error);
1462 		}
1463 		mutex_exit(&rp->r_statelock);
1464 	}
1465 
1466 
1467 	rl.rl_data = kmem_alloc(NFS_MAXPATHLEN, KM_SLEEP);
1468 
1469 	fi.vp = vp;
1470 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1471 	fi.copyproc = nfscopyfh;
1472 	fi.lookupproc = nfslookup;
1473 	fi.xattrdirproc = acl_getxattrdir2;
1474 
1475 	douprintf = 1;
1476 
1477 	error = rfs2call(VTOMI(vp), RFS_READLINK,
1478 	    xdr_readlink, (caddr_t)VTOFH(vp),
1479 	    xdr_rdlnres, (caddr_t)&rl, cr,
1480 	    &douprintf, &rl.rl_status, 0, &fi);
1481 
1482 	if (error) {
1483 
1484 		kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1485 		return (error);
1486 	}
1487 
1488 	error = geterrno(rl.rl_status);
1489 	if (!error) {
1490 		error = uiomove(rl.rl_data, (int)rl.rl_count, UIO_READ, uiop);
1491 		if (nfs_do_symlink_cache && rp->r_symlink.contents == NULL) {
1492 			mutex_enter(&rp->r_statelock);
1493 			if (rp->r_symlink.contents == NULL) {
1494 				rp->r_symlink.contents = rl.rl_data;
1495 				rp->r_symlink.len = (int)rl.rl_count;
1496 				rp->r_symlink.size = NFS_MAXPATHLEN;
1497 				mutex_exit(&rp->r_statelock);
1498 			} else {
1499 				mutex_exit(&rp->r_statelock);
1500 
1501 				kmem_free((void *)rl.rl_data,
1502 				    NFS_MAXPATHLEN);
1503 			}
1504 		} else {
1505 
1506 			kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1507 		}
1508 	} else {
1509 		PURGE_STALE_FH(error, vp, cr);
1510 
1511 		kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1512 	}
1513 
1514 	/*
1515 	 * Conform to UFS semantics (see comment above)
1516 	 */
1517 	return (error == ENXIO ? EINVAL : error);
1518 }
1519 
1520 /*
1521  * Flush local dirty pages to stable storage on the server.
1522  *
1523  * If FNODSYNC is specified, then there is nothing to do because
1524  * metadata changes are not cached on the client before being
1525  * sent to the server.
1526  */
1527 /* ARGSUSED */
1528 static int
1529 nfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
1530 {
1531 	int error;
1532 
1533 	if ((syncflag & FNODSYNC) || IS_SWAPVP(vp))
1534 		return (0);
1535 
1536 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1537 		return (EIO);
1538 
1539 	error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1540 	if (!error)
1541 		error = VTOR(vp)->r_error;
1542 	return (error);
1543 }
1544 
1545 
1546 /*
1547  * Weirdness: if the file was removed or the target of a rename
1548  * operation while it was open, it got renamed instead.  Here we
1549  * remove the renamed file.
1550  */
1551 /* ARGSUSED */
1552 static void
1553 nfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1554 {
1555 	rnode_t *rp;
1556 
1557 	ASSERT(vp != DNLC_NO_VNODE);
1558 
1559 	/*
1560 	 * If this is coming from the wrong zone, we let someone in the right
1561 	 * zone take care of it asynchronously.  We can get here due to
1562 	 * VN_RELE() being called from pageout() or fsflush().  This call may
1563 	 * potentially turn into an expensive no-op if, for instance, v_count
1564 	 * gets incremented in the meantime, but it's still correct.
1565 	 */
1566 	if (nfs_zone() != VTOMI(vp)->mi_zone) {
1567 		nfs_async_inactive(vp, cr, nfs_inactive);
1568 		return;
1569 	}
1570 
1571 	rp = VTOR(vp);
1572 redo:
1573 	if (rp->r_unldvp != NULL) {
1574 		/*
1575 		 * Save the vnode pointer for the directory where the
1576 		 * unlinked-open file got renamed, then set it to NULL
1577 		 * to prevent another thread from getting here before
1578 		 * we're done with the remove.  While we have the
1579 		 * statelock, make local copies of the pertinent rnode
1580 		 * fields.  If we weren't to do this in an atomic way, the
1581 		 * the unl* fields could become inconsistent with respect
1582 		 * to each other due to a race condition between this
1583 		 * code and nfs_remove().  See bug report 1034328.
1584 		 */
1585 		mutex_enter(&rp->r_statelock);
1586 		if (rp->r_unldvp != NULL) {
1587 			vnode_t *unldvp;
1588 			char *unlname;
1589 			cred_t *unlcred;
1590 			struct nfsdiropargs da;
1591 			enum nfsstat status;
1592 			int douprintf;
1593 			int error;
1594 
1595 			unldvp = rp->r_unldvp;
1596 			rp->r_unldvp = NULL;
1597 			unlname = rp->r_unlname;
1598 			rp->r_unlname = NULL;
1599 			unlcred = rp->r_unlcred;
1600 			rp->r_unlcred = NULL;
1601 			mutex_exit(&rp->r_statelock);
1602 
1603 			/*
1604 			 * If there are any dirty pages left, then flush
1605 			 * them.  This is unfortunate because they just
1606 			 * may get thrown away during the remove operation,
1607 			 * but we have to do this for correctness.
1608 			 */
1609 			if (vn_has_cached_data(vp) &&
1610 			    ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
1611 				ASSERT(vp->v_type != VCHR);
1612 				error = nfs_putpage(vp, (offset_t)0, 0, 0,
1613 				    cr, ct);
1614 				if (error) {
1615 					mutex_enter(&rp->r_statelock);
1616 					if (!rp->r_error)
1617 						rp->r_error = error;
1618 					mutex_exit(&rp->r_statelock);
1619 				}
1620 			}
1621 
1622 			/*
1623 			 * Do the remove operation on the renamed file
1624 			 */
1625 			setdiropargs(&da, unlname, unldvp);
1626 
1627 			douprintf = 1;
1628 
1629 			(void) rfs2call(VTOMI(unldvp), RFS_REMOVE,
1630 			    xdr_diropargs, (caddr_t)&da,
1631 			    xdr_enum, (caddr_t)&status, unlcred,
1632 			    &douprintf, &status, 0, NULL);
1633 
1634 			if (HAVE_RDDIR_CACHE(VTOR(unldvp)))
1635 				nfs_purge_rddir_cache(unldvp);
1636 			PURGE_ATTRCACHE(unldvp);
1637 
1638 			/*
1639 			 * Release stuff held for the remove
1640 			 */
1641 			VN_RELE(unldvp);
1642 			kmem_free(unlname, MAXNAMELEN);
1643 			crfree(unlcred);
1644 			goto redo;
1645 		}
1646 		mutex_exit(&rp->r_statelock);
1647 	}
1648 
1649 	rp_addfree(rp, cr);
1650 }
1651 
1652 /*
1653  * Remote file system operations having to do with directory manipulation.
1654  */
1655 
1656 /* ARGSUSED */
1657 static int
1658 nfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1659 	int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1660 	int *direntflags, pathname_t *realpnp)
1661 {
1662 	int error;
1663 	vnode_t *vp;
1664 	vnode_t *avp = NULL;
1665 	rnode_t *drp;
1666 
1667 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
1668 		return (EPERM);
1669 
1670 	drp = VTOR(dvp);
1671 
1672 	/*
1673 	 * Are we looking up extended attributes?  If so, "dvp" is
1674 	 * the file or directory for which we want attributes, and
1675 	 * we need a lookup of the hidden attribute directory
1676 	 * before we lookup the rest of the path.
1677 	 */
1678 	if (flags & LOOKUP_XATTR) {
1679 		bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0);
1680 		mntinfo_t *mi;
1681 
1682 		mi = VTOMI(dvp);
1683 		if (!(mi->mi_flags & MI_EXTATTR))
1684 			return (EINVAL);
1685 
1686 		if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp)))
1687 			return (EINTR);
1688 
1689 		(void) nfslookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr);
1690 		if (avp == NULL)
1691 			error = acl_getxattrdir2(dvp, &avp, cflag, cr, 0);
1692 		else
1693 			error = 0;
1694 
1695 		nfs_rw_exit(&drp->r_rwlock);
1696 
1697 		if (error) {
1698 			if (mi->mi_flags & MI_EXTATTR)
1699 				return (error);
1700 			return (EINVAL);
1701 		}
1702 		dvp = avp;
1703 		drp = VTOR(dvp);
1704 	}
1705 
1706 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) {
1707 		error = EINTR;
1708 		goto out;
1709 	}
1710 
1711 	error = nfslookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0);
1712 
1713 	nfs_rw_exit(&drp->r_rwlock);
1714 
1715 	/*
1716 	 * If vnode is a device, create special vnode.
1717 	 */
1718 	if (!error && IS_DEVVP(*vpp)) {
1719 		vp = *vpp;
1720 		*vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
1721 		VN_RELE(vp);
1722 	}
1723 
1724 out:
1725 	if (avp != NULL)
1726 		VN_RELE(avp);
1727 
1728 	return (error);
1729 }
1730 
1731 static int nfs_lookup_neg_cache = 1;
1732 
1733 #ifdef DEBUG
1734 static int nfs_lookup_dnlc_hits = 0;
1735 static int nfs_lookup_dnlc_misses = 0;
1736 static int nfs_lookup_dnlc_neg_hits = 0;
1737 static int nfs_lookup_dnlc_disappears = 0;
1738 static int nfs_lookup_dnlc_lookups = 0;
1739 #endif
1740 
1741 /* ARGSUSED */
1742 int
1743 nfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1744 	int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags)
1745 {
1746 	int error;
1747 
1748 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1749 
1750 	/*
1751 	 * If lookup is for "", just return dvp.  Don't need
1752 	 * to send it over the wire, look it up in the dnlc,
1753 	 * or perform any access checks.
1754 	 */
1755 	if (*nm == '\0') {
1756 		VN_HOLD(dvp);
1757 		*vpp = dvp;
1758 		return (0);
1759 	}
1760 
1761 	/*
1762 	 * Can't do lookups in non-directories.
1763 	 */
1764 	if (dvp->v_type != VDIR)
1765 		return (ENOTDIR);
1766 
1767 	/*
1768 	 * If we're called with RFSCALL_SOFT, it's important that
1769 	 * the only rfscall is one we make directly; if we permit
1770 	 * an access call because we're looking up "." or validating
1771 	 * a dnlc hit, we'll deadlock because that rfscall will not
1772 	 * have the RFSCALL_SOFT set.
1773 	 */
1774 	if (rfscall_flags & RFSCALL_SOFT)
1775 		goto callit;
1776 
1777 	/*
1778 	 * If lookup is for ".", just return dvp.  Don't need
1779 	 * to send it over the wire or look it up in the dnlc,
1780 	 * just need to check access.
1781 	 */
1782 	if (strcmp(nm, ".") == 0) {
1783 		error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1784 		if (error)
1785 			return (error);
1786 		VN_HOLD(dvp);
1787 		*vpp = dvp;
1788 		return (0);
1789 	}
1790 
1791 	/*
1792 	 * Lookup this name in the DNLC.  If there was a valid entry,
1793 	 * then return the results of the lookup.
1794 	 */
1795 	error = nfslookup_dnlc(dvp, nm, vpp, cr);
1796 	if (error || *vpp != NULL)
1797 		return (error);
1798 
1799 callit:
1800 	error = nfslookup_otw(dvp, nm, vpp, cr, rfscall_flags);
1801 
1802 	return (error);
1803 }
1804 
1805 static int
1806 nfslookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr)
1807 {
1808 	int error;
1809 	vnode_t *vp;
1810 
1811 	ASSERT(*nm != '\0');
1812 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1813 
1814 	/*
1815 	 * Lookup this name in the DNLC.  If successful, then validate
1816 	 * the caches and then recheck the DNLC.  The DNLC is rechecked
1817 	 * just in case this entry got invalidated during the call
1818 	 * to nfs_validate_caches.
1819 	 *
1820 	 * An assumption is being made that it is safe to say that a
1821 	 * file exists which may not on the server.  Any operations to
1822 	 * the server will fail with ESTALE.
1823 	 */
1824 #ifdef DEBUG
1825 	nfs_lookup_dnlc_lookups++;
1826 #endif
1827 	vp = dnlc_lookup(dvp, nm);
1828 	if (vp != NULL) {
1829 		VN_RELE(vp);
1830 		if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) {
1831 			PURGE_ATTRCACHE(dvp);
1832 		}
1833 		error = nfs_validate_caches(dvp, cr);
1834 		if (error)
1835 			return (error);
1836 		vp = dnlc_lookup(dvp, nm);
1837 		if (vp != NULL) {
1838 			error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1839 			if (error) {
1840 				VN_RELE(vp);
1841 				return (error);
1842 			}
1843 			if (vp == DNLC_NO_VNODE) {
1844 				VN_RELE(vp);
1845 #ifdef DEBUG
1846 				nfs_lookup_dnlc_neg_hits++;
1847 #endif
1848 				return (ENOENT);
1849 			}
1850 			*vpp = vp;
1851 #ifdef DEBUG
1852 			nfs_lookup_dnlc_hits++;
1853 #endif
1854 			return (0);
1855 		}
1856 #ifdef DEBUG
1857 		nfs_lookup_dnlc_disappears++;
1858 #endif
1859 	}
1860 #ifdef DEBUG
1861 	else
1862 		nfs_lookup_dnlc_misses++;
1863 #endif
1864 
1865 	*vpp = NULL;
1866 
1867 	return (0);
1868 }
1869 
1870 static int
1871 nfslookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr,
1872 	int rfscall_flags)
1873 {
1874 	int error;
1875 	struct nfsdiropargs da;
1876 	struct nfsdiropres dr;
1877 	int douprintf;
1878 	failinfo_t fi;
1879 	hrtime_t t;
1880 
1881 	ASSERT(*nm != '\0');
1882 	ASSERT(dvp->v_type == VDIR);
1883 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1884 
1885 	setdiropargs(&da, nm, dvp);
1886 
1887 	fi.vp = dvp;
1888 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1889 	fi.copyproc = nfscopyfh;
1890 	fi.lookupproc = nfslookup;
1891 	fi.xattrdirproc = acl_getxattrdir2;
1892 
1893 	douprintf = 1;
1894 
1895 	t = gethrtime();
1896 
1897 	error = rfs2call(VTOMI(dvp), RFS_LOOKUP,
1898 	    xdr_diropargs, (caddr_t)&da,
1899 	    xdr_diropres, (caddr_t)&dr, cr,
1900 	    &douprintf, &dr.dr_status, rfscall_flags, &fi);
1901 
1902 	if (!error) {
1903 		error = geterrno(dr.dr_status);
1904 		if (!error) {
1905 			*vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
1906 			    dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm);
1907 			/*
1908 			 * If NFS_ACL is supported on the server, then the
1909 			 * attributes returned by server may have minimal
1910 			 * permissions sometimes denying access to users having
1911 			 * proper access.  To get the proper attributes, mark
1912 			 * the attributes as expired so that they will be
1913 			 * regotten via the NFS_ACL GETATTR2 procedure.
1914 			 */
1915 			if (VTOMI(*vpp)->mi_flags & MI_ACL) {
1916 				PURGE_ATTRCACHE(*vpp);
1917 			}
1918 			if (!(rfscall_flags & RFSCALL_SOFT))
1919 				dnlc_update(dvp, nm, *vpp);
1920 		} else {
1921 			PURGE_STALE_FH(error, dvp, cr);
1922 			if (error == ENOENT && nfs_lookup_neg_cache)
1923 				dnlc_enter(dvp, nm, DNLC_NO_VNODE);
1924 		}
1925 	}
1926 
1927 	return (error);
1928 }
1929 
1930 /* ARGSUSED */
1931 static int
1932 nfs_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
1933 	int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct,
1934 	vsecattr_t *vsecp)
1935 {
1936 	int error;
1937 	struct nfscreatargs args;
1938 	struct nfsdiropres dr;
1939 	int douprintf;
1940 	vnode_t *vp;
1941 	rnode_t *rp;
1942 	struct vattr vattr;
1943 	rnode_t *drp;
1944 	vnode_t *tempvp;
1945 	hrtime_t t;
1946 
1947 	drp = VTOR(dvp);
1948 
1949 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
1950 		return (EPERM);
1951 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
1952 		return (EINTR);
1953 
1954 	/*
1955 	 * We make a copy of the attributes because the caller does not
1956 	 * expect us to change what va points to.
1957 	 */
1958 	vattr = *va;
1959 
1960 	/*
1961 	 * If the pathname is "", just use dvp.  Don't need
1962 	 * to send it over the wire, look it up in the dnlc,
1963 	 * or perform any access checks.
1964 	 */
1965 	if (*nm == '\0') {
1966 		error = 0;
1967 		VN_HOLD(dvp);
1968 		vp = dvp;
1969 	/*
1970 	 * If the pathname is ".", just use dvp.  Don't need
1971 	 * to send it over the wire or look it up in the dnlc,
1972 	 * just need to check access.
1973 	 */
1974 	} else if (strcmp(nm, ".") == 0) {
1975 		error = nfs_access(dvp, VEXEC, 0, cr, ct);
1976 		if (error) {
1977 			nfs_rw_exit(&drp->r_rwlock);
1978 			return (error);
1979 		}
1980 		VN_HOLD(dvp);
1981 		vp = dvp;
1982 	/*
1983 	 * We need to go over the wire, just to be sure whether the
1984 	 * file exists or not.  Using the DNLC can be dangerous in
1985 	 * this case when making a decision regarding existence.
1986 	 */
1987 	} else {
1988 		error = nfslookup_otw(dvp, nm, &vp, cr, 0);
1989 	}
1990 	if (!error) {
1991 		if (exclusive == EXCL)
1992 			error = EEXIST;
1993 		else if (vp->v_type == VDIR && (mode & VWRITE))
1994 			error = EISDIR;
1995 		else {
1996 			/*
1997 			 * If vnode is a device, create special vnode.
1998 			 */
1999 			if (IS_DEVVP(vp)) {
2000 				tempvp = vp;
2001 				vp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2002 				VN_RELE(tempvp);
2003 			}
2004 			if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) {
2005 				if ((vattr.va_mask & AT_SIZE) &&
2006 				    vp->v_type == VREG) {
2007 					vattr.va_mask = AT_SIZE;
2008 					error = nfssetattr(vp, &vattr, 0, cr);
2009 				}
2010 			}
2011 		}
2012 		nfs_rw_exit(&drp->r_rwlock);
2013 		if (error) {
2014 			VN_RELE(vp);
2015 		} else {
2016 			/*
2017 			 * existing file got truncated, notify.
2018 			 */
2019 			vnevent_create(vp, ct);
2020 			*vpp = vp;
2021 		}
2022 		return (error);
2023 	}
2024 
2025 	ASSERT(vattr.va_mask & AT_TYPE);
2026 	if (vattr.va_type == VREG) {
2027 		ASSERT(vattr.va_mask & AT_MODE);
2028 		if (MANDMODE(vattr.va_mode)) {
2029 			nfs_rw_exit(&drp->r_rwlock);
2030 			return (EACCES);
2031 		}
2032 	}
2033 
2034 	dnlc_remove(dvp, nm);
2035 
2036 	setdiropargs(&args.ca_da, nm, dvp);
2037 
2038 	/*
2039 	 * Decide what the group-id of the created file should be.
2040 	 * Set it in attribute list as advisory...then do a setattr
2041 	 * if the server didn't get it right the first time.
2042 	 */
2043 	error = setdirgid(dvp, &vattr.va_gid, cr);
2044 	if (error) {
2045 		nfs_rw_exit(&drp->r_rwlock);
2046 		return (error);
2047 	}
2048 	vattr.va_mask |= AT_GID;
2049 
2050 	/*
2051 	 * This is a completely gross hack to make mknod
2052 	 * work over the wire until we can wack the protocol
2053 	 */
2054 #define	IFCHR		0020000		/* character special */
2055 #define	IFBLK		0060000		/* block special */
2056 #define	IFSOCK		0140000		/* socket */
2057 
2058 	/*
2059 	 * dev_t is uint_t in 5.x and short in 4.x. Both 4.x
2060 	 * supports 8 bit majors. 5.x supports 14 bit majors. 5.x supports 18
2061 	 * bits in the minor number where 4.x supports 8 bits.  If the 5.x
2062 	 * minor/major numbers <= 8 bits long, compress the device
2063 	 * number before sending it. Otherwise, the 4.x server will not
2064 	 * create the device with the correct device number and nothing can be
2065 	 * done about this.
2066 	 */
2067 	if (vattr.va_type == VCHR || vattr.va_type == VBLK) {
2068 		dev_t d = vattr.va_rdev;
2069 		dev32_t dev32;
2070 
2071 		if (vattr.va_type == VCHR)
2072 			vattr.va_mode |= IFCHR;
2073 		else
2074 			vattr.va_mode |= IFBLK;
2075 
2076 		(void) cmpldev(&dev32, d);
2077 		if (dev32 & ~((SO4_MAXMAJ << L_BITSMINOR32) | SO4_MAXMIN))
2078 			vattr.va_size = (u_offset_t)dev32;
2079 		else
2080 			vattr.va_size = (u_offset_t)nfsv2_cmpdev(d);
2081 
2082 		vattr.va_mask |= AT_MODE|AT_SIZE;
2083 	} else if (vattr.va_type == VFIFO) {
2084 		vattr.va_mode |= IFCHR;		/* xtra kludge for namedpipe */
2085 		vattr.va_size = (u_offset_t)NFS_FIFO_DEV;	/* blech */
2086 		vattr.va_mask |= AT_MODE|AT_SIZE;
2087 	} else if (vattr.va_type == VSOCK) {
2088 		vattr.va_mode |= IFSOCK;
2089 		/*
2090 		 * To avoid triggering bugs in the servers set AT_SIZE
2091 		 * (all other RFS_CREATE calls set this).
2092 		 */
2093 		vattr.va_size = 0;
2094 		vattr.va_mask |= AT_MODE|AT_SIZE;
2095 	}
2096 
2097 	args.ca_sa = &args.ca_sa_buf;
2098 	error = vattr_to_sattr(&vattr, args.ca_sa);
2099 	if (error) {
2100 		/* req time field(s) overflow - return immediately */
2101 		nfs_rw_exit(&drp->r_rwlock);
2102 		return (error);
2103 	}
2104 
2105 	douprintf = 1;
2106 
2107 	t = gethrtime();
2108 
2109 	error = rfs2call(VTOMI(dvp), RFS_CREATE,
2110 	    xdr_creatargs, (caddr_t)&args,
2111 	    xdr_diropres, (caddr_t)&dr, cr,
2112 	    &douprintf, &dr.dr_status, 0, NULL);
2113 
2114 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2115 
2116 	if (!error) {
2117 		error = geterrno(dr.dr_status);
2118 		if (!error) {
2119 			if (HAVE_RDDIR_CACHE(drp))
2120 				nfs_purge_rddir_cache(dvp);
2121 			vp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2122 			    dvp->v_vfsp, t, cr, NULL, NULL);
2123 			/*
2124 			 * If NFS_ACL is supported on the server, then the
2125 			 * attributes returned by server may have minimal
2126 			 * permissions sometimes denying access to users having
2127 			 * proper access.  To get the proper attributes, mark
2128 			 * the attributes as expired so that they will be
2129 			 * regotten via the NFS_ACL GETATTR2 procedure.
2130 			 */
2131 			if (VTOMI(vp)->mi_flags & MI_ACL) {
2132 				PURGE_ATTRCACHE(vp);
2133 			}
2134 			dnlc_update(dvp, nm, vp);
2135 			rp = VTOR(vp);
2136 			if (vattr.va_size == 0) {
2137 				mutex_enter(&rp->r_statelock);
2138 				rp->r_size = 0;
2139 				mutex_exit(&rp->r_statelock);
2140 				if (vn_has_cached_data(vp)) {
2141 					ASSERT(vp->v_type != VCHR);
2142 					nfs_invalidate_pages(vp,
2143 					    (u_offset_t)0, cr);
2144 				}
2145 			}
2146 
2147 			/*
2148 			 * Make sure the gid was set correctly.
2149 			 * If not, try to set it (but don't lose
2150 			 * any sleep over it).
2151 			 */
2152 			if (vattr.va_gid != rp->r_attr.va_gid) {
2153 				vattr.va_mask = AT_GID;
2154 				(void) nfssetattr(vp, &vattr, 0, cr);
2155 			}
2156 
2157 			/*
2158 			 * If vnode is a device create special vnode
2159 			 */
2160 			if (IS_DEVVP(vp)) {
2161 				*vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2162 				VN_RELE(vp);
2163 			} else
2164 				*vpp = vp;
2165 		} else {
2166 			PURGE_STALE_FH(error, dvp, cr);
2167 		}
2168 	}
2169 
2170 	nfs_rw_exit(&drp->r_rwlock);
2171 
2172 	return (error);
2173 }
2174 
2175 /*
2176  * Weirdness: if the vnode to be removed is open
2177  * we rename it instead of removing it and nfs_inactive
2178  * will remove the new name.
2179  */
2180 /* ARGSUSED */
2181 static int
2182 nfs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags)
2183 {
2184 	int error;
2185 	struct nfsdiropargs da;
2186 	enum nfsstat status;
2187 	vnode_t *vp;
2188 	char *tmpname;
2189 	int douprintf;
2190 	rnode_t *rp;
2191 	rnode_t *drp;
2192 
2193 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2194 		return (EPERM);
2195 	drp = VTOR(dvp);
2196 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2197 		return (EINTR);
2198 
2199 	error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2200 	if (error) {
2201 		nfs_rw_exit(&drp->r_rwlock);
2202 		return (error);
2203 	}
2204 
2205 	if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) {
2206 		VN_RELE(vp);
2207 		nfs_rw_exit(&drp->r_rwlock);
2208 		return (EPERM);
2209 	}
2210 
2211 	/*
2212 	 * First just remove the entry from the name cache, as it
2213 	 * is most likely the only entry for this vp.
2214 	 */
2215 	dnlc_remove(dvp, nm);
2216 
2217 	/*
2218 	 * If the file has a v_count > 1 then there may be more than one
2219 	 * entry in the name cache due multiple links or an open file,
2220 	 * but we don't have the real reference count so flush all
2221 	 * possible entries.
2222 	 */
2223 	if (vp->v_count > 1)
2224 		dnlc_purge_vp(vp);
2225 
2226 	/*
2227 	 * Now we have the real reference count on the vnode
2228 	 */
2229 	rp = VTOR(vp);
2230 	mutex_enter(&rp->r_statelock);
2231 	if (vp->v_count > 1 &&
2232 	    (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) {
2233 		mutex_exit(&rp->r_statelock);
2234 		tmpname = newname();
2235 		error = nfsrename(dvp, nm, dvp, tmpname, cr, ct);
2236 		if (error)
2237 			kmem_free(tmpname, MAXNAMELEN);
2238 		else {
2239 			mutex_enter(&rp->r_statelock);
2240 			if (rp->r_unldvp == NULL) {
2241 				VN_HOLD(dvp);
2242 				rp->r_unldvp = dvp;
2243 				if (rp->r_unlcred != NULL)
2244 					crfree(rp->r_unlcred);
2245 				crhold(cr);
2246 				rp->r_unlcred = cr;
2247 				rp->r_unlname = tmpname;
2248 			} else {
2249 				kmem_free(rp->r_unlname, MAXNAMELEN);
2250 				rp->r_unlname = tmpname;
2251 			}
2252 			mutex_exit(&rp->r_statelock);
2253 		}
2254 	} else {
2255 		mutex_exit(&rp->r_statelock);
2256 		/*
2257 		 * We need to flush any dirty pages which happen to
2258 		 * be hanging around before removing the file.  This
2259 		 * shouldn't happen very often and mostly on file
2260 		 * systems mounted "nocto".
2261 		 */
2262 		if (vn_has_cached_data(vp) &&
2263 		    ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
2264 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
2265 			if (error && (error == ENOSPC || error == EDQUOT)) {
2266 				mutex_enter(&rp->r_statelock);
2267 				if (!rp->r_error)
2268 					rp->r_error = error;
2269 				mutex_exit(&rp->r_statelock);
2270 			}
2271 		}
2272 
2273 		setdiropargs(&da, nm, dvp);
2274 
2275 		douprintf = 1;
2276 
2277 		error = rfs2call(VTOMI(dvp), RFS_REMOVE,
2278 		    xdr_diropargs, (caddr_t)&da,
2279 		    xdr_enum, (caddr_t)&status, cr,
2280 		    &douprintf, &status, 0, NULL);
2281 
2282 		/*
2283 		 * The xattr dir may be gone after last attr is removed,
2284 		 * so flush it from dnlc.
2285 		 */
2286 		if (dvp->v_flag & V_XATTRDIR)
2287 			dnlc_purge_vp(dvp);
2288 
2289 		PURGE_ATTRCACHE(dvp);	/* mod time changed */
2290 		PURGE_ATTRCACHE(vp);	/* link count changed */
2291 
2292 		if (!error) {
2293 			error = geterrno(status);
2294 			if (!error) {
2295 				if (HAVE_RDDIR_CACHE(drp))
2296 					nfs_purge_rddir_cache(dvp);
2297 			} else {
2298 				PURGE_STALE_FH(error, dvp, cr);
2299 			}
2300 		}
2301 	}
2302 
2303 	if (error == 0) {
2304 		vnevent_remove(vp, dvp, nm, ct);
2305 	}
2306 	VN_RELE(vp);
2307 
2308 	nfs_rw_exit(&drp->r_rwlock);
2309 
2310 	return (error);
2311 }
2312 
2313 /* ARGSUSED */
2314 static int
2315 nfs_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr,
2316 	caller_context_t *ct, int flags)
2317 {
2318 	int error;
2319 	struct nfslinkargs args;
2320 	enum nfsstat status;
2321 	vnode_t *realvp;
2322 	int douprintf;
2323 	rnode_t *tdrp;
2324 
2325 	if (nfs_zone() != VTOMI(tdvp)->mi_zone)
2326 		return (EPERM);
2327 	if (VOP_REALVP(svp, &realvp, ct) == 0)
2328 		svp = realvp;
2329 
2330 	args.la_from = VTOFH(svp);
2331 	setdiropargs(&args.la_to, tnm, tdvp);
2332 
2333 	tdrp = VTOR(tdvp);
2334 	if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp)))
2335 		return (EINTR);
2336 
2337 	dnlc_remove(tdvp, tnm);
2338 
2339 	douprintf = 1;
2340 
2341 	error = rfs2call(VTOMI(svp), RFS_LINK,
2342 	    xdr_linkargs, (caddr_t)&args,
2343 	    xdr_enum, (caddr_t)&status, cr,
2344 	    &douprintf, &status, 0, NULL);
2345 
2346 	PURGE_ATTRCACHE(tdvp);	/* mod time changed */
2347 	PURGE_ATTRCACHE(svp);	/* link count changed */
2348 
2349 	if (!error) {
2350 		error = geterrno(status);
2351 		if (!error) {
2352 			if (HAVE_RDDIR_CACHE(tdrp))
2353 				nfs_purge_rddir_cache(tdvp);
2354 		}
2355 	}
2356 
2357 	nfs_rw_exit(&tdrp->r_rwlock);
2358 
2359 	if (!error) {
2360 		/*
2361 		 * Notify the source file of this link operation.
2362 		 */
2363 		vnevent_link(svp, ct);
2364 	}
2365 	return (error);
2366 }
2367 
2368 /* ARGSUSED */
2369 static int
2370 nfs_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2371 	caller_context_t *ct, int flags)
2372 {
2373 	vnode_t *realvp;
2374 
2375 	if (nfs_zone() != VTOMI(odvp)->mi_zone)
2376 		return (EPERM);
2377 	if (VOP_REALVP(ndvp, &realvp, ct) == 0)
2378 		ndvp = realvp;
2379 
2380 	return (nfsrename(odvp, onm, ndvp, nnm, cr, ct));
2381 }
2382 
2383 /*
2384  * nfsrename does the real work of renaming in NFS Version 2.
2385  */
2386 static int
2387 nfsrename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2388     caller_context_t *ct)
2389 {
2390 	int error;
2391 	enum nfsstat status;
2392 	struct nfsrnmargs args;
2393 	int douprintf;
2394 	vnode_t *nvp = NULL;
2395 	vnode_t *ovp = NULL;
2396 	char *tmpname;
2397 	rnode_t *rp;
2398 	rnode_t *odrp;
2399 	rnode_t *ndrp;
2400 
2401 	ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone);
2402 	if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 ||
2403 	    strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0)
2404 		return (EINVAL);
2405 
2406 	odrp = VTOR(odvp);
2407 	ndrp = VTOR(ndvp);
2408 	if ((intptr_t)odrp < (intptr_t)ndrp) {
2409 		if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp)))
2410 			return (EINTR);
2411 		if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) {
2412 			nfs_rw_exit(&odrp->r_rwlock);
2413 			return (EINTR);
2414 		}
2415 	} else {
2416 		if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp)))
2417 			return (EINTR);
2418 		if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) {
2419 			nfs_rw_exit(&ndrp->r_rwlock);
2420 			return (EINTR);
2421 		}
2422 	}
2423 
2424 	/*
2425 	 * Lookup the target file.  If it exists, it needs to be
2426 	 * checked to see whether it is a mount point and whether
2427 	 * it is active (open).
2428 	 */
2429 	error = nfslookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0);
2430 	if (!error) {
2431 		/*
2432 		 * If this file has been mounted on, then just
2433 		 * return busy because renaming to it would remove
2434 		 * the mounted file system from the name space.
2435 		 */
2436 		if (vn_mountedvfs(nvp) != NULL) {
2437 			VN_RELE(nvp);
2438 			nfs_rw_exit(&odrp->r_rwlock);
2439 			nfs_rw_exit(&ndrp->r_rwlock);
2440 			return (EBUSY);
2441 		}
2442 
2443 		/*
2444 		 * Purge the name cache of all references to this vnode
2445 		 * so that we can check the reference count to infer
2446 		 * whether it is active or not.
2447 		 */
2448 		/*
2449 		 * First just remove the entry from the name cache, as it
2450 		 * is most likely the only entry for this vp.
2451 		 */
2452 		dnlc_remove(ndvp, nnm);
2453 		/*
2454 		 * If the file has a v_count > 1 then there may be more
2455 		 * than one entry in the name cache due multiple links
2456 		 * or an open file, but we don't have the real reference
2457 		 * count so flush all possible entries.
2458 		 */
2459 		if (nvp->v_count > 1)
2460 			dnlc_purge_vp(nvp);
2461 
2462 		/*
2463 		 * If the vnode is active and is not a directory,
2464 		 * arrange to rename it to a
2465 		 * temporary file so that it will continue to be
2466 		 * accessible.  This implements the "unlink-open-file"
2467 		 * semantics for the target of a rename operation.
2468 		 * Before doing this though, make sure that the
2469 		 * source and target files are not already the same.
2470 		 */
2471 		if (nvp->v_count > 1 && nvp->v_type != VDIR) {
2472 			/*
2473 			 * Lookup the source name.
2474 			 */
2475 			error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL,
2476 			    cr, 0);
2477 
2478 			/*
2479 			 * The source name *should* already exist.
2480 			 */
2481 			if (error) {
2482 				VN_RELE(nvp);
2483 				nfs_rw_exit(&odrp->r_rwlock);
2484 				nfs_rw_exit(&ndrp->r_rwlock);
2485 				return (error);
2486 			}
2487 
2488 			/*
2489 			 * Compare the two vnodes.  If they are the same,
2490 			 * just release all held vnodes and return success.
2491 			 */
2492 			if (ovp == nvp) {
2493 				VN_RELE(ovp);
2494 				VN_RELE(nvp);
2495 				nfs_rw_exit(&odrp->r_rwlock);
2496 				nfs_rw_exit(&ndrp->r_rwlock);
2497 				return (0);
2498 			}
2499 
2500 			/*
2501 			 * Can't mix and match directories and non-
2502 			 * directories in rename operations.  We already
2503 			 * know that the target is not a directory.  If
2504 			 * the source is a directory, return an error.
2505 			 */
2506 			if (ovp->v_type == VDIR) {
2507 				VN_RELE(ovp);
2508 				VN_RELE(nvp);
2509 				nfs_rw_exit(&odrp->r_rwlock);
2510 				nfs_rw_exit(&ndrp->r_rwlock);
2511 				return (ENOTDIR);
2512 			}
2513 
2514 			/*
2515 			 * The target file exists, is not the same as
2516 			 * the source file, and is active.  Link it
2517 			 * to a temporary filename to avoid having
2518 			 * the server removing the file completely.
2519 			 */
2520 			tmpname = newname();
2521 			error = nfs_link(ndvp, nvp, tmpname, cr, NULL, 0);
2522 			if (error == EOPNOTSUPP) {
2523 				error = nfs_rename(ndvp, nnm, ndvp, tmpname,
2524 				    cr, NULL, 0);
2525 			}
2526 			if (error) {
2527 				kmem_free(tmpname, MAXNAMELEN);
2528 				VN_RELE(ovp);
2529 				VN_RELE(nvp);
2530 				nfs_rw_exit(&odrp->r_rwlock);
2531 				nfs_rw_exit(&ndrp->r_rwlock);
2532 				return (error);
2533 			}
2534 			rp = VTOR(nvp);
2535 			mutex_enter(&rp->r_statelock);
2536 			if (rp->r_unldvp == NULL) {
2537 				VN_HOLD(ndvp);
2538 				rp->r_unldvp = ndvp;
2539 				if (rp->r_unlcred != NULL)
2540 					crfree(rp->r_unlcred);
2541 				crhold(cr);
2542 				rp->r_unlcred = cr;
2543 				rp->r_unlname = tmpname;
2544 			} else {
2545 				kmem_free(rp->r_unlname, MAXNAMELEN);
2546 				rp->r_unlname = tmpname;
2547 			}
2548 			mutex_exit(&rp->r_statelock);
2549 		}
2550 	}
2551 
2552 	if (ovp == NULL) {
2553 		/*
2554 		 * When renaming directories to be a subdirectory of a
2555 		 * different parent, the dnlc entry for ".." will no
2556 		 * longer be valid, so it must be removed.
2557 		 *
2558 		 * We do a lookup here to determine whether we are renaming
2559 		 * a directory and we need to check if we are renaming
2560 		 * an unlinked file.  This might have already been done
2561 		 * in previous code, so we check ovp == NULL to avoid
2562 		 * doing it twice.
2563 		 */
2564 
2565 		error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0);
2566 
2567 		/*
2568 		 * The source name *should* already exist.
2569 		 */
2570 		if (error) {
2571 			nfs_rw_exit(&odrp->r_rwlock);
2572 			nfs_rw_exit(&ndrp->r_rwlock);
2573 			if (nvp) {
2574 				VN_RELE(nvp);
2575 			}
2576 			return (error);
2577 		}
2578 		ASSERT(ovp != NULL);
2579 	}
2580 
2581 	dnlc_remove(odvp, onm);
2582 	dnlc_remove(ndvp, nnm);
2583 
2584 	setdiropargs(&args.rna_from, onm, odvp);
2585 	setdiropargs(&args.rna_to, nnm, ndvp);
2586 
2587 	douprintf = 1;
2588 
2589 	error = rfs2call(VTOMI(odvp), RFS_RENAME,
2590 	    xdr_rnmargs, (caddr_t)&args,
2591 	    xdr_enum, (caddr_t)&status, cr,
2592 	    &douprintf, &status, 0, NULL);
2593 
2594 	PURGE_ATTRCACHE(odvp);	/* mod time changed */
2595 	PURGE_ATTRCACHE(ndvp);	/* mod time changed */
2596 
2597 	if (!error) {
2598 		error = geterrno(status);
2599 		if (!error) {
2600 			if (HAVE_RDDIR_CACHE(odrp))
2601 				nfs_purge_rddir_cache(odvp);
2602 			if (HAVE_RDDIR_CACHE(ndrp))
2603 				nfs_purge_rddir_cache(ndvp);
2604 			/*
2605 			 * when renaming directories to be a subdirectory of a
2606 			 * different parent, the dnlc entry for ".." will no
2607 			 * longer be valid, so it must be removed
2608 			 */
2609 			rp = VTOR(ovp);
2610 			if (ndvp != odvp) {
2611 				if (ovp->v_type == VDIR) {
2612 					dnlc_remove(ovp, "..");
2613 					if (HAVE_RDDIR_CACHE(rp))
2614 						nfs_purge_rddir_cache(ovp);
2615 				}
2616 			}
2617 
2618 			/*
2619 			 * If we are renaming the unlinked file, update the
2620 			 * r_unldvp and r_unlname as needed.
2621 			 */
2622 			mutex_enter(&rp->r_statelock);
2623 			if (rp->r_unldvp != NULL) {
2624 				if (strcmp(rp->r_unlname, onm) == 0) {
2625 					(void) strncpy(rp->r_unlname,
2626 					    nnm, MAXNAMELEN);
2627 					rp->r_unlname[MAXNAMELEN - 1] = '\0';
2628 
2629 					if (ndvp != rp->r_unldvp) {
2630 						VN_RELE(rp->r_unldvp);
2631 						rp->r_unldvp = ndvp;
2632 						VN_HOLD(ndvp);
2633 					}
2634 				}
2635 			}
2636 			mutex_exit(&rp->r_statelock);
2637 		} else {
2638 			/*
2639 			 * System V defines rename to return EEXIST, not
2640 			 * ENOTEMPTY if the target directory is not empty.
2641 			 * Over the wire, the error is NFSERR_ENOTEMPTY
2642 			 * which geterrno maps to ENOTEMPTY.
2643 			 */
2644 			if (error == ENOTEMPTY)
2645 				error = EEXIST;
2646 		}
2647 	}
2648 
2649 	if (error == 0) {
2650 		if (nvp)
2651 			vnevent_rename_dest(nvp, ndvp, nnm, ct);
2652 
2653 		if (odvp != ndvp)
2654 			vnevent_rename_dest_dir(ndvp, ct);
2655 
2656 		ASSERT(ovp != NULL);
2657 		vnevent_rename_src(ovp, odvp, onm, ct);
2658 	}
2659 
2660 	if (nvp) {
2661 		VN_RELE(nvp);
2662 	}
2663 	VN_RELE(ovp);
2664 
2665 	nfs_rw_exit(&odrp->r_rwlock);
2666 	nfs_rw_exit(&ndrp->r_rwlock);
2667 
2668 	return (error);
2669 }
2670 
2671 /* ARGSUSED */
2672 static int
2673 nfs_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr,
2674 	caller_context_t *ct, int flags, vsecattr_t *vsecp)
2675 {
2676 	int error;
2677 	struct nfscreatargs args;
2678 	struct nfsdiropres dr;
2679 	int douprintf;
2680 	rnode_t *drp;
2681 	hrtime_t t;
2682 
2683 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2684 		return (EPERM);
2685 
2686 	setdiropargs(&args.ca_da, nm, dvp);
2687 
2688 	/*
2689 	 * Decide what the group-id and set-gid bit of the created directory
2690 	 * should be.  May have to do a setattr to get the gid right.
2691 	 */
2692 	error = setdirgid(dvp, &va->va_gid, cr);
2693 	if (error)
2694 		return (error);
2695 	error = setdirmode(dvp, &va->va_mode, cr);
2696 	if (error)
2697 		return (error);
2698 	va->va_mask |= AT_MODE|AT_GID;
2699 
2700 	args.ca_sa = &args.ca_sa_buf;
2701 	error = vattr_to_sattr(va, args.ca_sa);
2702 	if (error) {
2703 		/* req time field(s) overflow - return immediately */
2704 		return (error);
2705 	}
2706 
2707 	drp = VTOR(dvp);
2708 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2709 		return (EINTR);
2710 
2711 	dnlc_remove(dvp, nm);
2712 
2713 	douprintf = 1;
2714 
2715 	t = gethrtime();
2716 
2717 	error = rfs2call(VTOMI(dvp), RFS_MKDIR,
2718 	    xdr_creatargs, (caddr_t)&args,
2719 	    xdr_diropres, (caddr_t)&dr, cr,
2720 	    &douprintf, &dr.dr_status, 0, NULL);
2721 
2722 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2723 
2724 	if (!error) {
2725 		error = geterrno(dr.dr_status);
2726 		if (!error) {
2727 			if (HAVE_RDDIR_CACHE(drp))
2728 				nfs_purge_rddir_cache(dvp);
2729 			/*
2730 			 * The attributes returned by RFS_MKDIR can not
2731 			 * be depended upon, so mark the attribute cache
2732 			 * as purged.  A subsequent GETATTR will get the
2733 			 * correct attributes from the server.
2734 			 */
2735 			*vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2736 			    dvp->v_vfsp, t, cr, NULL, NULL);
2737 			PURGE_ATTRCACHE(*vpp);
2738 			dnlc_update(dvp, nm, *vpp);
2739 
2740 			/*
2741 			 * Make sure the gid was set correctly.
2742 			 * If not, try to set it (but don't lose
2743 			 * any sleep over it).
2744 			 */
2745 			if (va->va_gid != VTOR(*vpp)->r_attr.va_gid) {
2746 				va->va_mask = AT_GID;
2747 				(void) nfssetattr(*vpp, va, 0, cr);
2748 			}
2749 		} else {
2750 			PURGE_STALE_FH(error, dvp, cr);
2751 		}
2752 	}
2753 
2754 	nfs_rw_exit(&drp->r_rwlock);
2755 
2756 	return (error);
2757 }
2758 
2759 /* ARGSUSED */
2760 static int
2761 nfs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
2762 	caller_context_t *ct, int flags)
2763 {
2764 	int error;
2765 	enum nfsstat status;
2766 	struct nfsdiropargs da;
2767 	vnode_t *vp;
2768 	int douprintf;
2769 	rnode_t *drp;
2770 
2771 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2772 		return (EPERM);
2773 	drp = VTOR(dvp);
2774 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2775 		return (EINTR);
2776 
2777 	/*
2778 	 * Attempt to prevent a rmdir(".") from succeeding.
2779 	 */
2780 	error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2781 	if (error) {
2782 		nfs_rw_exit(&drp->r_rwlock);
2783 		return (error);
2784 	}
2785 
2786 	if (vp == cdir) {
2787 		VN_RELE(vp);
2788 		nfs_rw_exit(&drp->r_rwlock);
2789 		return (EINVAL);
2790 	}
2791 
2792 	setdiropargs(&da, nm, dvp);
2793 
2794 	/*
2795 	 * First just remove the entry from the name cache, as it
2796 	 * is most likely an entry for this vp.
2797 	 */
2798 	dnlc_remove(dvp, nm);
2799 
2800 	/*
2801 	 * If there vnode reference count is greater than one, then
2802 	 * there may be additional references in the DNLC which will
2803 	 * need to be purged.  First, trying removing the entry for
2804 	 * the parent directory and see if that removes the additional
2805 	 * reference(s).  If that doesn't do it, then use dnlc_purge_vp
2806 	 * to completely remove any references to the directory which
2807 	 * might still exist in the DNLC.
2808 	 */
2809 	if (vp->v_count > 1) {
2810 		dnlc_remove(vp, "..");
2811 		if (vp->v_count > 1)
2812 			dnlc_purge_vp(vp);
2813 	}
2814 
2815 	douprintf = 1;
2816 
2817 	error = rfs2call(VTOMI(dvp), RFS_RMDIR,
2818 	    xdr_diropargs, (caddr_t)&da,
2819 	    xdr_enum, (caddr_t)&status, cr,
2820 	    &douprintf, &status, 0, NULL);
2821 
2822 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2823 
2824 	if (error) {
2825 		VN_RELE(vp);
2826 		nfs_rw_exit(&drp->r_rwlock);
2827 		return (error);
2828 	}
2829 
2830 	error = geterrno(status);
2831 	if (!error) {
2832 		if (HAVE_RDDIR_CACHE(drp))
2833 			nfs_purge_rddir_cache(dvp);
2834 		if (HAVE_RDDIR_CACHE(VTOR(vp)))
2835 			nfs_purge_rddir_cache(vp);
2836 	} else {
2837 		PURGE_STALE_FH(error, dvp, cr);
2838 		/*
2839 		 * System V defines rmdir to return EEXIST, not
2840 		 * ENOTEMPTY if the directory is not empty.  Over
2841 		 * the wire, the error is NFSERR_ENOTEMPTY which
2842 		 * geterrno maps to ENOTEMPTY.
2843 		 */
2844 		if (error == ENOTEMPTY)
2845 			error = EEXIST;
2846 	}
2847 
2848 	if (error == 0) {
2849 		vnevent_rmdir(vp, dvp, nm, ct);
2850 	}
2851 	VN_RELE(vp);
2852 
2853 	nfs_rw_exit(&drp->r_rwlock);
2854 
2855 	return (error);
2856 }
2857 
2858 /* ARGSUSED */
2859 static int
2860 nfs_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr,
2861 	caller_context_t *ct, int flags)
2862 {
2863 	int error;
2864 	struct nfsslargs args;
2865 	enum nfsstat status;
2866 	int douprintf;
2867 	rnode_t *drp;
2868 
2869 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2870 		return (EPERM);
2871 	setdiropargs(&args.sla_from, lnm, dvp);
2872 	args.sla_sa = &args.sla_sa_buf;
2873 	error = vattr_to_sattr(tva, args.sla_sa);
2874 	if (error) {
2875 		/* req time field(s) overflow - return immediately */
2876 		return (error);
2877 	}
2878 	args.sla_tnm = tnm;
2879 
2880 	drp = VTOR(dvp);
2881 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2882 		return (EINTR);
2883 
2884 	dnlc_remove(dvp, lnm);
2885 
2886 	douprintf = 1;
2887 
2888 	error = rfs2call(VTOMI(dvp), RFS_SYMLINK,
2889 	    xdr_slargs, (caddr_t)&args,
2890 	    xdr_enum, (caddr_t)&status, cr,
2891 	    &douprintf, &status, 0, NULL);
2892 
2893 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2894 
2895 	if (!error) {
2896 		error = geterrno(status);
2897 		if (!error) {
2898 			if (HAVE_RDDIR_CACHE(drp))
2899 				nfs_purge_rddir_cache(dvp);
2900 		} else {
2901 			PURGE_STALE_FH(error, dvp, cr);
2902 		}
2903 	}
2904 
2905 	nfs_rw_exit(&drp->r_rwlock);
2906 
2907 	return (error);
2908 }
2909 
2910 #ifdef DEBUG
2911 static int nfs_readdir_cache_hits = 0;
2912 static int nfs_readdir_cache_shorts = 0;
2913 static int nfs_readdir_cache_waits = 0;
2914 static int nfs_readdir_cache_misses = 0;
2915 static int nfs_readdir_readahead = 0;
2916 #endif
2917 
2918 static int nfs_shrinkreaddir = 0;
2919 
2920 /*
2921  * Read directory entries.
2922  * There are some weird things to look out for here.  The uio_offset
2923  * field is either 0 or it is the offset returned from a previous
2924  * readdir.  It is an opaque value used by the server to find the
2925  * correct directory block to read. The count field is the number
2926  * of blocks to read on the server.  This is advisory only, the server
2927  * may return only one block's worth of entries.  Entries may be compressed
2928  * on the server.
2929  */
2930 /* ARGSUSED */
2931 static int
2932 nfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
2933 	caller_context_t *ct, int flags)
2934 {
2935 	int error;
2936 	size_t count;
2937 	rnode_t *rp;
2938 	rddir_cache *rdc;
2939 	rddir_cache *nrdc;
2940 	rddir_cache *rrdc;
2941 #ifdef DEBUG
2942 	int missed;
2943 #endif
2944 	rddir_cache srdc;
2945 	avl_index_t where;
2946 
2947 	rp = VTOR(vp);
2948 
2949 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
2950 	if (nfs_zone() != VTOMI(vp)->mi_zone)
2951 		return (EIO);
2952 	/*
2953 	 * Make sure that the directory cache is valid.
2954 	 */
2955 	if (HAVE_RDDIR_CACHE(rp)) {
2956 		if (nfs_disable_rddir_cache) {
2957 			/*
2958 			 * Setting nfs_disable_rddir_cache in /etc/system
2959 			 * allows interoperability with servers that do not
2960 			 * properly update the attributes of directories.
2961 			 * Any cached information gets purged before an
2962 			 * access is made to it.
2963 			 */
2964 			nfs_purge_rddir_cache(vp);
2965 		} else {
2966 			error = nfs_validate_caches(vp, cr);
2967 			if (error)
2968 				return (error);
2969 		}
2970 	}
2971 
2972 	/*
2973 	 * UGLINESS: SunOS 3.2 servers apparently cannot always handle an
2974 	 * RFS_READDIR request with rda_count set to more than 0x400. So
2975 	 * we reduce the request size here purely for compatibility.
2976 	 *
2977 	 * In general, this is no longer required.  However, if a server
2978 	 * is discovered which can not handle requests larger than 1024,
2979 	 * nfs_shrinkreaddir can be set to 1 to enable this backwards
2980 	 * compatibility.
2981 	 *
2982 	 * In any case, the request size is limited to NFS_MAXDATA bytes.
2983 	 */
2984 	count = MIN(uiop->uio_iov->iov_len,
2985 	    nfs_shrinkreaddir ? 0x400 : NFS_MAXDATA);
2986 
2987 	nrdc = NULL;
2988 #ifdef DEBUG
2989 	missed = 0;
2990 #endif
2991 top:
2992 	/*
2993 	 * Short circuit last readdir which always returns 0 bytes.
2994 	 * This can be done after the directory has been read through
2995 	 * completely at least once.  This will set r_direof which
2996 	 * can be used to find the value of the last cookie.
2997 	 */
2998 	mutex_enter(&rp->r_statelock);
2999 	if (rp->r_direof != NULL &&
3000 	    uiop->uio_offset == rp->r_direof->nfs_ncookie) {
3001 		mutex_exit(&rp->r_statelock);
3002 #ifdef DEBUG
3003 		nfs_readdir_cache_shorts++;
3004 #endif
3005 		if (eofp)
3006 			*eofp = 1;
3007 		if (nrdc != NULL)
3008 			rddir_cache_rele(nrdc);
3009 		return (0);
3010 	}
3011 	/*
3012 	 * Look for a cache entry.  Cache entries are identified
3013 	 * by the NFS cookie value and the byte count requested.
3014 	 */
3015 	srdc.nfs_cookie = uiop->uio_offset;
3016 	srdc.buflen = count;
3017 	rdc = avl_find(&rp->r_dir, &srdc, &where);
3018 	if (rdc != NULL) {
3019 		rddir_cache_hold(rdc);
3020 		/*
3021 		 * If the cache entry is in the process of being
3022 		 * filled in, wait until this completes.  The
3023 		 * RDDIRWAIT bit is set to indicate that someone
3024 		 * is waiting and then the thread currently
3025 		 * filling the entry is done, it should do a
3026 		 * cv_broadcast to wakeup all of the threads
3027 		 * waiting for it to finish.
3028 		 */
3029 		if (rdc->flags & RDDIR) {
3030 			nfs_rw_exit(&rp->r_rwlock);
3031 			rdc->flags |= RDDIRWAIT;
3032 #ifdef DEBUG
3033 			nfs_readdir_cache_waits++;
3034 #endif
3035 			if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) {
3036 				/*
3037 				 * We got interrupted, probably
3038 				 * the user typed ^C or an alarm
3039 				 * fired.  We free the new entry
3040 				 * if we allocated one.
3041 				 */
3042 				mutex_exit(&rp->r_statelock);
3043 				(void) nfs_rw_enter_sig(&rp->r_rwlock,
3044 				    RW_READER, FALSE);
3045 				rddir_cache_rele(rdc);
3046 				if (nrdc != NULL)
3047 					rddir_cache_rele(nrdc);
3048 				return (EINTR);
3049 			}
3050 			mutex_exit(&rp->r_statelock);
3051 			(void) nfs_rw_enter_sig(&rp->r_rwlock,
3052 			    RW_READER, FALSE);
3053 			rddir_cache_rele(rdc);
3054 			goto top;
3055 		}
3056 		/*
3057 		 * Check to see if a readdir is required to
3058 		 * fill the entry.  If so, mark this entry
3059 		 * as being filled, remove our reference,
3060 		 * and branch to the code to fill the entry.
3061 		 */
3062 		if (rdc->flags & RDDIRREQ) {
3063 			rdc->flags &= ~RDDIRREQ;
3064 			rdc->flags |= RDDIR;
3065 			if (nrdc != NULL)
3066 				rddir_cache_rele(nrdc);
3067 			nrdc = rdc;
3068 			mutex_exit(&rp->r_statelock);
3069 			goto bottom;
3070 		}
3071 #ifdef DEBUG
3072 		if (!missed)
3073 			nfs_readdir_cache_hits++;
3074 #endif
3075 		/*
3076 		 * If an error occurred while attempting
3077 		 * to fill the cache entry, just return it.
3078 		 */
3079 		if (rdc->error) {
3080 			error = rdc->error;
3081 			mutex_exit(&rp->r_statelock);
3082 			rddir_cache_rele(rdc);
3083 			if (nrdc != NULL)
3084 				rddir_cache_rele(nrdc);
3085 			return (error);
3086 		}
3087 
3088 		/*
3089 		 * The cache entry is complete and good,
3090 		 * copyout the dirent structs to the calling
3091 		 * thread.
3092 		 */
3093 		error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop);
3094 
3095 		/*
3096 		 * If no error occurred during the copyout,
3097 		 * update the offset in the uio struct to
3098 		 * contain the value of the next cookie
3099 		 * and set the eof value appropriately.
3100 		 */
3101 		if (!error) {
3102 			uiop->uio_offset = rdc->nfs_ncookie;
3103 			if (eofp)
3104 				*eofp = rdc->eof;
3105 		}
3106 
3107 		/*
3108 		 * Decide whether to do readahead.  Don't if
3109 		 * have already read to the end of directory.
3110 		 */
3111 		if (rdc->eof) {
3112 			rp->r_direof = rdc;
3113 			mutex_exit(&rp->r_statelock);
3114 			rddir_cache_rele(rdc);
3115 			if (nrdc != NULL)
3116 				rddir_cache_rele(nrdc);
3117 			return (error);
3118 		}
3119 
3120 		/*
3121 		 * Check to see whether we found an entry
3122 		 * for the readahead.  If so, we don't need
3123 		 * to do anything further, so free the new
3124 		 * entry if one was allocated.  Otherwise,
3125 		 * allocate a new entry, add it to the cache,
3126 		 * and then initiate an asynchronous readdir
3127 		 * operation to fill it.
3128 		 */
3129 		srdc.nfs_cookie = rdc->nfs_ncookie;
3130 		srdc.buflen = count;
3131 		rrdc = avl_find(&rp->r_dir, &srdc, &where);
3132 		if (rrdc != NULL) {
3133 			if (nrdc != NULL)
3134 				rddir_cache_rele(nrdc);
3135 		} else {
3136 			if (nrdc != NULL)
3137 				rrdc = nrdc;
3138 			else {
3139 				rrdc = rddir_cache_alloc(KM_NOSLEEP);
3140 			}
3141 			if (rrdc != NULL) {
3142 				rrdc->nfs_cookie = rdc->nfs_ncookie;
3143 				rrdc->buflen = count;
3144 				avl_insert(&rp->r_dir, rrdc, where);
3145 				rddir_cache_hold(rrdc);
3146 				mutex_exit(&rp->r_statelock);
3147 				rddir_cache_rele(rdc);
3148 #ifdef DEBUG
3149 				nfs_readdir_readahead++;
3150 #endif
3151 				nfs_async_readdir(vp, rrdc, cr, nfsreaddir);
3152 				return (error);
3153 			}
3154 		}
3155 
3156 		mutex_exit(&rp->r_statelock);
3157 		rddir_cache_rele(rdc);
3158 		return (error);
3159 	}
3160 
3161 	/*
3162 	 * Didn't find an entry in the cache.  Construct a new empty
3163 	 * entry and link it into the cache.  Other processes attempting
3164 	 * to access this entry will need to wait until it is filled in.
3165 	 *
3166 	 * Since kmem_alloc may block, another pass through the cache
3167 	 * will need to be taken to make sure that another process
3168 	 * hasn't already added an entry to the cache for this request.
3169 	 */
3170 	if (nrdc == NULL) {
3171 		mutex_exit(&rp->r_statelock);
3172 		nrdc = rddir_cache_alloc(KM_SLEEP);
3173 		nrdc->nfs_cookie = uiop->uio_offset;
3174 		nrdc->buflen = count;
3175 		goto top;
3176 	}
3177 
3178 	/*
3179 	 * Add this entry to the cache.
3180 	 */
3181 	avl_insert(&rp->r_dir, nrdc, where);
3182 	rddir_cache_hold(nrdc);
3183 	mutex_exit(&rp->r_statelock);
3184 
3185 bottom:
3186 #ifdef DEBUG
3187 	missed = 1;
3188 	nfs_readdir_cache_misses++;
3189 #endif
3190 	/*
3191 	 * Do the readdir.
3192 	 */
3193 	error = nfsreaddir(vp, nrdc, cr);
3194 
3195 	/*
3196 	 * If this operation failed, just return the error which occurred.
3197 	 */
3198 	if (error != 0)
3199 		return (error);
3200 
3201 	/*
3202 	 * Since the RPC operation will have taken sometime and blocked
3203 	 * this process, another pass through the cache will need to be
3204 	 * taken to find the correct cache entry.  It is possible that
3205 	 * the correct cache entry will not be there (although one was
3206 	 * added) because the directory changed during the RPC operation
3207 	 * and the readdir cache was flushed.  In this case, just start
3208 	 * over.  It is hoped that this will not happen too often... :-)
3209 	 */
3210 	nrdc = NULL;
3211 	goto top;
3212 	/* NOTREACHED */
3213 }
3214 
3215 static int
3216 nfsreaddir(vnode_t *vp, rddir_cache *rdc, cred_t *cr)
3217 {
3218 	int error;
3219 	struct nfsrddirargs rda;
3220 	struct nfsrddirres rd;
3221 	rnode_t *rp;
3222 	mntinfo_t *mi;
3223 	uint_t count;
3224 	int douprintf;
3225 	failinfo_t fi, *fip;
3226 
3227 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3228 	count = rdc->buflen;
3229 
3230 	rp = VTOR(vp);
3231 	mi = VTOMI(vp);
3232 
3233 	rda.rda_fh = *VTOFH(vp);
3234 	rda.rda_offset = rdc->nfs_cookie;
3235 
3236 	/*
3237 	 * NFS client failover support
3238 	 * suppress failover unless we have a zero cookie
3239 	 */
3240 	if (rdc->nfs_cookie == (off_t)0) {
3241 		fi.vp = vp;
3242 		fi.fhp = (caddr_t)&rda.rda_fh;
3243 		fi.copyproc = nfscopyfh;
3244 		fi.lookupproc = nfslookup;
3245 		fi.xattrdirproc = acl_getxattrdir2;
3246 		fip = &fi;
3247 	} else {
3248 		fip = NULL;
3249 	}
3250 
3251 	rd.rd_entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3252 	rd.rd_size = count;
3253 	rd.rd_offset = rda.rda_offset;
3254 
3255 	douprintf = 1;
3256 
3257 	if (mi->mi_io_kstats) {
3258 		mutex_enter(&mi->mi_lock);
3259 		kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
3260 		mutex_exit(&mi->mi_lock);
3261 	}
3262 
3263 	do {
3264 		rda.rda_count = MIN(count, mi->mi_curread);
3265 		error = rfs2call(mi, RFS_READDIR,
3266 		    xdr_rddirargs, (caddr_t)&rda,
3267 		    xdr_getrddirres, (caddr_t)&rd, cr,
3268 		    &douprintf, &rd.rd_status, 0, fip);
3269 	} while (error == ENFS_TRYAGAIN);
3270 
3271 	if (mi->mi_io_kstats) {
3272 		mutex_enter(&mi->mi_lock);
3273 		kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
3274 		mutex_exit(&mi->mi_lock);
3275 	}
3276 
3277 	/*
3278 	 * Since we are actually doing a READDIR RPC, we must have
3279 	 * exclusive access to the cache entry being filled.  Thus,
3280 	 * it is safe to update all fields except for the flags
3281 	 * field.  The r_statelock in the rnode must be held to
3282 	 * prevent two different threads from simultaneously
3283 	 * attempting to update the flags field.  This can happen
3284 	 * if we are turning off RDDIR and the other thread is
3285 	 * trying to set RDDIRWAIT.
3286 	 */
3287 	ASSERT(rdc->flags & RDDIR);
3288 	if (!error) {
3289 		error = geterrno(rd.rd_status);
3290 		if (!error) {
3291 			rdc->nfs_ncookie = rd.rd_offset;
3292 			rdc->eof = rd.rd_eof ? 1 : 0;
3293 			rdc->entlen = rd.rd_size;
3294 			ASSERT(rdc->entlen <= rdc->buflen);
3295 #ifdef DEBUG
3296 			rdc->entries = rddir_cache_buf_alloc(rdc->buflen,
3297 			    KM_SLEEP);
3298 #else
3299 			rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3300 #endif
3301 			bcopy(rd.rd_entries, rdc->entries, rdc->entlen);
3302 			rdc->error = 0;
3303 			if (mi->mi_io_kstats) {
3304 				mutex_enter(&mi->mi_lock);
3305 				KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
3306 				KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
3307 				    rd.rd_size;
3308 				mutex_exit(&mi->mi_lock);
3309 			}
3310 		} else {
3311 			PURGE_STALE_FH(error, vp, cr);
3312 		}
3313 	}
3314 	if (error) {
3315 		rdc->entries = NULL;
3316 		rdc->error = error;
3317 	}
3318 	kmem_free(rd.rd_entries, rdc->buflen);
3319 
3320 	mutex_enter(&rp->r_statelock);
3321 	rdc->flags &= ~RDDIR;
3322 	if (rdc->flags & RDDIRWAIT) {
3323 		rdc->flags &= ~RDDIRWAIT;
3324 		cv_broadcast(&rdc->cv);
3325 	}
3326 	if (error)
3327 		rdc->flags |= RDDIRREQ;
3328 	mutex_exit(&rp->r_statelock);
3329 
3330 	rddir_cache_rele(rdc);
3331 
3332 	return (error);
3333 }
3334 
3335 #ifdef DEBUG
3336 static int nfs_bio_do_stop = 0;
3337 #endif
3338 
3339 static int
3340 nfs_bio(struct buf *bp, cred_t *cr)
3341 {
3342 	rnode_t *rp = VTOR(bp->b_vp);
3343 	int count;
3344 	int error;
3345 	cred_t *cred;
3346 	uint_t offset;
3347 
3348 	DTRACE_IO1(start, struct buf *, bp);
3349 
3350 	ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone);
3351 	offset = dbtob(bp->b_blkno);
3352 
3353 	if (bp->b_flags & B_READ) {
3354 		mutex_enter(&rp->r_statelock);
3355 		if (rp->r_cred != NULL) {
3356 			cred = rp->r_cred;
3357 			crhold(cred);
3358 		} else {
3359 			rp->r_cred = cr;
3360 			crhold(cr);
3361 			cred = cr;
3362 			crhold(cred);
3363 		}
3364 		mutex_exit(&rp->r_statelock);
3365 	read_again:
3366 		error = bp->b_error = nfsread(bp->b_vp, bp->b_un.b_addr,
3367 		    offset, bp->b_bcount, &bp->b_resid, cred);
3368 
3369 		crfree(cred);
3370 		if (!error) {
3371 			if (bp->b_resid) {
3372 				/*
3373 				 * Didn't get it all because we hit EOF,
3374 				 * zero all the memory beyond the EOF.
3375 				 */
3376 				/* bzero(rdaddr + */
3377 				bzero(bp->b_un.b_addr +
3378 				    bp->b_bcount - bp->b_resid, bp->b_resid);
3379 			}
3380 			mutex_enter(&rp->r_statelock);
3381 			if (bp->b_resid == bp->b_bcount &&
3382 			    offset >= rp->r_size) {
3383 				/*
3384 				 * We didn't read anything at all as we are
3385 				 * past EOF.  Return an error indicator back
3386 				 * but don't destroy the pages (yet).
3387 				 */
3388 				error = NFS_EOF;
3389 			}
3390 			mutex_exit(&rp->r_statelock);
3391 		} else if (error == EACCES) {
3392 			mutex_enter(&rp->r_statelock);
3393 			if (cred != cr) {
3394 				if (rp->r_cred != NULL)
3395 					crfree(rp->r_cred);
3396 				rp->r_cred = cr;
3397 				crhold(cr);
3398 				cred = cr;
3399 				crhold(cred);
3400 				mutex_exit(&rp->r_statelock);
3401 				goto read_again;
3402 			}
3403 			mutex_exit(&rp->r_statelock);
3404 		}
3405 	} else {
3406 		if (!(rp->r_flags & RSTALE)) {
3407 			mutex_enter(&rp->r_statelock);
3408 			if (rp->r_cred != NULL) {
3409 				cred = rp->r_cred;
3410 				crhold(cred);
3411 			} else {
3412 				rp->r_cred = cr;
3413 				crhold(cr);
3414 				cred = cr;
3415 				crhold(cred);
3416 			}
3417 			mutex_exit(&rp->r_statelock);
3418 		write_again:
3419 			mutex_enter(&rp->r_statelock);
3420 			count = MIN(bp->b_bcount, rp->r_size - offset);
3421 			mutex_exit(&rp->r_statelock);
3422 			if (count < 0)
3423 				cmn_err(CE_PANIC, "nfs_bio: write count < 0");
3424 #ifdef DEBUG
3425 			if (count == 0) {
3426 				zcmn_err(getzoneid(), CE_WARN,
3427 				    "nfs_bio: zero length write at %d",
3428 				    offset);
3429 				nfs_printfhandle(&rp->r_fh);
3430 				if (nfs_bio_do_stop)
3431 					debug_enter("nfs_bio");
3432 			}
3433 #endif
3434 			error = nfswrite(bp->b_vp, bp->b_un.b_addr, offset,
3435 			    count, cred);
3436 			if (error == EACCES) {
3437 				mutex_enter(&rp->r_statelock);
3438 				if (cred != cr) {
3439 					if (rp->r_cred != NULL)
3440 						crfree(rp->r_cred);
3441 					rp->r_cred = cr;
3442 					crhold(cr);
3443 					crfree(cred);
3444 					cred = cr;
3445 					crhold(cred);
3446 					mutex_exit(&rp->r_statelock);
3447 					goto write_again;
3448 				}
3449 				mutex_exit(&rp->r_statelock);
3450 			}
3451 			bp->b_error = error;
3452 			if (error && error != EINTR) {
3453 				/*
3454 				 * Don't print EDQUOT errors on the console.
3455 				 * Don't print asynchronous EACCES errors.
3456 				 * Don't print EFBIG errors.
3457 				 * Print all other write errors.
3458 				 */
3459 				if (error != EDQUOT && error != EFBIG &&
3460 				    (error != EACCES ||
3461 				    !(bp->b_flags & B_ASYNC)))
3462 					nfs_write_error(bp->b_vp, error, cred);
3463 				/*
3464 				 * Update r_error and r_flags as appropriate.
3465 				 * If the error was ESTALE, then mark the
3466 				 * rnode as not being writeable and save
3467 				 * the error status.  Otherwise, save any
3468 				 * errors which occur from asynchronous
3469 				 * page invalidations.  Any errors occurring
3470 				 * from other operations should be saved
3471 				 * by the caller.
3472 				 */
3473 				mutex_enter(&rp->r_statelock);
3474 				if (error == ESTALE) {
3475 					rp->r_flags |= RSTALE;
3476 					if (!rp->r_error)
3477 						rp->r_error = error;
3478 				} else if (!rp->r_error &&
3479 				    (bp->b_flags &
3480 				    (B_INVAL|B_FORCE|B_ASYNC)) ==
3481 				    (B_INVAL|B_FORCE|B_ASYNC)) {
3482 					rp->r_error = error;
3483 				}
3484 				mutex_exit(&rp->r_statelock);
3485 			}
3486 			crfree(cred);
3487 		} else
3488 			error = rp->r_error;
3489 	}
3490 
3491 	if (error != 0 && error != NFS_EOF)
3492 		bp->b_flags |= B_ERROR;
3493 
3494 	DTRACE_IO1(done, struct buf *, bp);
3495 
3496 	return (error);
3497 }
3498 
3499 /* ARGSUSED */
3500 static int
3501 nfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3502 {
3503 	struct nfs_fid *fp;
3504 	rnode_t *rp;
3505 
3506 	rp = VTOR(vp);
3507 
3508 	if (fidp->fid_len < (sizeof (struct nfs_fid) - sizeof (short))) {
3509 		fidp->fid_len = sizeof (struct nfs_fid) - sizeof (short);
3510 		return (ENOSPC);
3511 	}
3512 	fp = (struct nfs_fid *)fidp;
3513 	fp->nf_pad = 0;
3514 	fp->nf_len = sizeof (struct nfs_fid) - sizeof (short);
3515 	bcopy(rp->r_fh.fh_buf, fp->nf_data, NFS_FHSIZE);
3516 	return (0);
3517 }
3518 
3519 /* ARGSUSED2 */
3520 static int
3521 nfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3522 {
3523 	rnode_t *rp = VTOR(vp);
3524 
3525 	if (!write_lock) {
3526 		(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3527 		return (V_WRITELOCK_FALSE);
3528 	}
3529 
3530 	if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) {
3531 		(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3532 		if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp))
3533 			return (V_WRITELOCK_FALSE);
3534 		nfs_rw_exit(&rp->r_rwlock);
3535 	}
3536 
3537 	(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE);
3538 	return (V_WRITELOCK_TRUE);
3539 }
3540 
3541 /* ARGSUSED */
3542 static void
3543 nfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3544 {
3545 	rnode_t *rp = VTOR(vp);
3546 
3547 	nfs_rw_exit(&rp->r_rwlock);
3548 }
3549 
3550 /* ARGSUSED */
3551 static int
3552 nfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
3553 {
3554 
3555 	/*
3556 	 * Because we stuff the readdir cookie into the offset field
3557 	 * someone may attempt to do an lseek with the cookie which
3558 	 * we want to succeed.
3559 	 */
3560 	if (vp->v_type == VDIR)
3561 		return (0);
3562 	if (*noffp < 0 || *noffp > MAXOFF32_T)
3563 		return (EINVAL);
3564 	return (0);
3565 }
3566 
3567 /*
3568  * number of NFS_MAXDATA blocks to read ahead
3569  * optimized for 100 base-T.
3570  */
3571 static int nfs_nra = 4;
3572 
3573 #ifdef DEBUG
3574 static int nfs_lostpage = 0;	/* number of times we lost original page */
3575 #endif
3576 
3577 /*
3578  * Return all the pages from [off..off+len) in file
3579  */
3580 /* ARGSUSED */
3581 static int
3582 nfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3583 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3584 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3585 {
3586 	rnode_t *rp;
3587 	int error;
3588 	mntinfo_t *mi;
3589 
3590 	if (vp->v_flag & VNOMAP)
3591 		return (ENOSYS);
3592 
3593 	ASSERT(off <= MAXOFF32_T);
3594 	if (nfs_zone() != VTOMI(vp)->mi_zone)
3595 		return (EIO);
3596 	if (protp != NULL)
3597 		*protp = PROT_ALL;
3598 
3599 	/*
3600 	 * Now valididate that the caches are up to date.
3601 	 */
3602 	error = nfs_validate_caches(vp, cr);
3603 	if (error)
3604 		return (error);
3605 
3606 	rp = VTOR(vp);
3607 	mi = VTOMI(vp);
3608 retry:
3609 	mutex_enter(&rp->r_statelock);
3610 
3611 	/*
3612 	 * Don't create dirty pages faster than they
3613 	 * can be cleaned so that the system doesn't
3614 	 * get imbalanced.  If the async queue is
3615 	 * maxed out, then wait for it to drain before
3616 	 * creating more dirty pages.  Also, wait for
3617 	 * any threads doing pagewalks in the vop_getattr
3618 	 * entry points so that they don't block for
3619 	 * long periods.
3620 	 */
3621 	if (rw == S_CREATE) {
3622 		while ((mi->mi_max_threads != 0 &&
3623 		    rp->r_awcount > 2 * mi->mi_max_threads) ||
3624 		    rp->r_gcount > 0)
3625 			cv_wait(&rp->r_cv, &rp->r_statelock);
3626 	}
3627 
3628 	/*
3629 	 * If we are getting called as a side effect of an nfs_write()
3630 	 * operation the local file size might not be extended yet.
3631 	 * In this case we want to be able to return pages of zeroes.
3632 	 */
3633 	if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) {
3634 		mutex_exit(&rp->r_statelock);
3635 		return (EFAULT);		/* beyond EOF */
3636 	}
3637 
3638 	mutex_exit(&rp->r_statelock);
3639 
3640 	if (len <= PAGESIZE) {
3641 		error = nfs_getapage(vp, off, len, protp, pl, plsz,
3642 		    seg, addr, rw, cr);
3643 	} else {
3644 		error = pvn_getpages(nfs_getapage, vp, off, len, protp,
3645 		    pl, plsz, seg, addr, rw, cr);
3646 	}
3647 
3648 	switch (error) {
3649 	case NFS_EOF:
3650 		nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr);
3651 		goto retry;
3652 	case ESTALE:
3653 		PURGE_STALE_FH(error, vp, cr);
3654 	}
3655 
3656 	return (error);
3657 }
3658 
3659 /*
3660  * Called from pvn_getpages or nfs_getpage to get a particular page.
3661  */
3662 /* ARGSUSED */
3663 static int
3664 nfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
3665 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3666 	enum seg_rw rw, cred_t *cr)
3667 {
3668 	rnode_t *rp;
3669 	uint_t bsize;
3670 	struct buf *bp;
3671 	page_t *pp;
3672 	u_offset_t lbn;
3673 	u_offset_t io_off;
3674 	u_offset_t blkoff;
3675 	u_offset_t rablkoff;
3676 	size_t io_len;
3677 	uint_t blksize;
3678 	int error;
3679 	int readahead;
3680 	int readahead_issued = 0;
3681 	int ra_window; /* readahead window */
3682 	page_t *pagefound;
3683 
3684 	if (nfs_zone() != VTOMI(vp)->mi_zone)
3685 		return (EIO);
3686 	rp = VTOR(vp);
3687 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3688 
3689 reread:
3690 	bp = NULL;
3691 	pp = NULL;
3692 	pagefound = NULL;
3693 
3694 	if (pl != NULL)
3695 		pl[0] = NULL;
3696 
3697 	error = 0;
3698 	lbn = off / bsize;
3699 	blkoff = lbn * bsize;
3700 
3701 	/*
3702 	 * Queueing up the readahead before doing the synchronous read
3703 	 * results in a significant increase in read throughput because
3704 	 * of the increased parallelism between the async threads and
3705 	 * the process context.
3706 	 */
3707 	if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 &&
3708 	    rw != S_CREATE &&
3709 	    !(vp->v_flag & VNOCACHE)) {
3710 		mutex_enter(&rp->r_statelock);
3711 
3712 		/*
3713 		 * Calculate the number of readaheads to do.
3714 		 * a) No readaheads at offset = 0.
3715 		 * b) Do maximum(nfs_nra) readaheads when the readahead
3716 		 *    window is closed.
3717 		 * c) Do readaheads between 1 to (nfs_nra - 1) depending
3718 		 *    upon how far the readahead window is open or close.
3719 		 * d) No readaheads if rp->r_nextr is not within the scope
3720 		 *    of the readahead window (random i/o).
3721 		 */
3722 
3723 		if (off == 0)
3724 			readahead = 0;
3725 		else if (blkoff == rp->r_nextr)
3726 			readahead = nfs_nra;
3727 		else if (rp->r_nextr > blkoff &&
3728 		    ((ra_window = (rp->r_nextr - blkoff) / bsize)
3729 		    <= (nfs_nra - 1)))
3730 			readahead = nfs_nra - ra_window;
3731 		else
3732 			readahead = 0;
3733 
3734 		rablkoff = rp->r_nextr;
3735 		while (readahead > 0 && rablkoff + bsize < rp->r_size) {
3736 			mutex_exit(&rp->r_statelock);
3737 			if (nfs_async_readahead(vp, rablkoff + bsize,
3738 			    addr + (rablkoff + bsize - off), seg, cr,
3739 			    nfs_readahead) < 0) {
3740 				mutex_enter(&rp->r_statelock);
3741 				break;
3742 			}
3743 			readahead--;
3744 			rablkoff += bsize;
3745 			/*
3746 			 * Indicate that we did a readahead so
3747 			 * readahead offset is not updated
3748 			 * by the synchronous read below.
3749 			 */
3750 			readahead_issued = 1;
3751 			mutex_enter(&rp->r_statelock);
3752 			/*
3753 			 * set readahead offset to
3754 			 * offset of last async readahead
3755 			 * request.
3756 			 */
3757 			rp->r_nextr = rablkoff;
3758 		}
3759 		mutex_exit(&rp->r_statelock);
3760 	}
3761 
3762 again:
3763 	if ((pagefound = page_exists(vp, off)) == NULL) {
3764 		if (pl == NULL) {
3765 			(void) nfs_async_readahead(vp, blkoff, addr, seg, cr,
3766 			    nfs_readahead);
3767 		} else if (rw == S_CREATE) {
3768 			/*
3769 			 * Block for this page is not allocated, or the offset
3770 			 * is beyond the current allocation size, or we're
3771 			 * allocating a swap slot and the page was not found,
3772 			 * so allocate it and return a zero page.
3773 			 */
3774 			if ((pp = page_create_va(vp, off,
3775 			    PAGESIZE, PG_WAIT, seg, addr)) == NULL)
3776 				cmn_err(CE_PANIC, "nfs_getapage: page_create");
3777 			io_len = PAGESIZE;
3778 			mutex_enter(&rp->r_statelock);
3779 			rp->r_nextr = off + PAGESIZE;
3780 			mutex_exit(&rp->r_statelock);
3781 		} else {
3782 			/*
3783 			 * Need to go to server to get a BLOCK, exception to
3784 			 * that being while reading at offset = 0 or doing
3785 			 * random i/o, in that case read only a PAGE.
3786 			 */
3787 			mutex_enter(&rp->r_statelock);
3788 			if (blkoff < rp->r_size &&
3789 			    blkoff + bsize >= rp->r_size) {
3790 				/*
3791 				 * If only a block or less is left in
3792 				 * the file, read all that is remaining.
3793 				 */
3794 				if (rp->r_size <= off) {
3795 					/*
3796 					 * Trying to access beyond EOF,
3797 					 * set up to get at least one page.
3798 					 */
3799 					blksize = off + PAGESIZE - blkoff;
3800 				} else
3801 					blksize = rp->r_size - blkoff;
3802 			} else if ((off == 0) ||
3803 			    (off != rp->r_nextr && !readahead_issued)) {
3804 				blksize = PAGESIZE;
3805 				blkoff = off; /* block = page here */
3806 			} else
3807 				blksize = bsize;
3808 			mutex_exit(&rp->r_statelock);
3809 
3810 			pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3811 			    &io_len, blkoff, blksize, 0);
3812 
3813 			/*
3814 			 * Some other thread has entered the page,
3815 			 * so just use it.
3816 			 */
3817 			if (pp == NULL)
3818 				goto again;
3819 
3820 			/*
3821 			 * Now round the request size up to page boundaries.
3822 			 * This ensures that the entire page will be
3823 			 * initialized to zeroes if EOF is encountered.
3824 			 */
3825 			io_len = ptob(btopr(io_len));
3826 
3827 			bp = pageio_setup(pp, io_len, vp, B_READ);
3828 			ASSERT(bp != NULL);
3829 
3830 			/*
3831 			 * pageio_setup should have set b_addr to 0.  This
3832 			 * is correct since we want to do I/O on a page
3833 			 * boundary.  bp_mapin will use this addr to calculate
3834 			 * an offset, and then set b_addr to the kernel virtual
3835 			 * address it allocated for us.
3836 			 */
3837 			ASSERT(bp->b_un.b_addr == 0);
3838 
3839 			bp->b_edev = 0;
3840 			bp->b_dev = 0;
3841 			bp->b_lblkno = lbtodb(io_off);
3842 			bp->b_file = vp;
3843 			bp->b_offset = (offset_t)off;
3844 			bp_mapin(bp);
3845 
3846 			/*
3847 			 * If doing a write beyond what we believe is EOF,
3848 			 * don't bother trying to read the pages from the
3849 			 * server, we'll just zero the pages here.  We
3850 			 * don't check that the rw flag is S_WRITE here
3851 			 * because some implementations may attempt a
3852 			 * read access to the buffer before copying data.
3853 			 */
3854 			mutex_enter(&rp->r_statelock);
3855 			if (io_off >= rp->r_size && seg == segkmap) {
3856 				mutex_exit(&rp->r_statelock);
3857 				bzero(bp->b_un.b_addr, io_len);
3858 			} else {
3859 				mutex_exit(&rp->r_statelock);
3860 				error = nfs_bio(bp, cr);
3861 			}
3862 
3863 			/*
3864 			 * Unmap the buffer before freeing it.
3865 			 */
3866 			bp_mapout(bp);
3867 			pageio_done(bp);
3868 
3869 			if (error == NFS_EOF) {
3870 				/*
3871 				 * If doing a write system call just return
3872 				 * zeroed pages, else user tried to get pages
3873 				 * beyond EOF, return error.  We don't check
3874 				 * that the rw flag is S_WRITE here because
3875 				 * some implementations may attempt a read
3876 				 * access to the buffer before copying data.
3877 				 */
3878 				if (seg == segkmap)
3879 					error = 0;
3880 				else
3881 					error = EFAULT;
3882 			}
3883 
3884 			if (!readahead_issued && !error) {
3885 				mutex_enter(&rp->r_statelock);
3886 				rp->r_nextr = io_off + io_len;
3887 				mutex_exit(&rp->r_statelock);
3888 			}
3889 		}
3890 	}
3891 
3892 out:
3893 	if (pl == NULL)
3894 		return (error);
3895 
3896 	if (error) {
3897 		if (pp != NULL)
3898 			pvn_read_done(pp, B_ERROR);
3899 		return (error);
3900 	}
3901 
3902 	if (pagefound) {
3903 		se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED);
3904 
3905 		/*
3906 		 * Page exists in the cache, acquire the appropriate lock.
3907 		 * If this fails, start all over again.
3908 		 */
3909 		if ((pp = page_lookup(vp, off, se)) == NULL) {
3910 #ifdef DEBUG
3911 			nfs_lostpage++;
3912 #endif
3913 			goto reread;
3914 		}
3915 		pl[0] = pp;
3916 		pl[1] = NULL;
3917 		return (0);
3918 	}
3919 
3920 	if (pp != NULL)
3921 		pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3922 
3923 	return (error);
3924 }
3925 
3926 static void
3927 nfs_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg,
3928 	cred_t *cr)
3929 {
3930 	int error;
3931 	page_t *pp;
3932 	u_offset_t io_off;
3933 	size_t io_len;
3934 	struct buf *bp;
3935 	uint_t bsize, blksize;
3936 	rnode_t *rp = VTOR(vp);
3937 
3938 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3939 
3940 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3941 
3942 	mutex_enter(&rp->r_statelock);
3943 	if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) {
3944 		/*
3945 		 * If less than a block left in file read less
3946 		 * than a block.
3947 		 */
3948 		blksize = rp->r_size - blkoff;
3949 	} else
3950 		blksize = bsize;
3951 	mutex_exit(&rp->r_statelock);
3952 
3953 	pp = pvn_read_kluster(vp, blkoff, segkmap, addr,
3954 	    &io_off, &io_len, blkoff, blksize, 1);
3955 	/*
3956 	 * The isra flag passed to the kluster function is 1, we may have
3957 	 * gotten a return value of NULL for a variety of reasons (# of free
3958 	 * pages < minfree, someone entered the page on the vnode etc). In all
3959 	 * cases, we want to punt on the readahead.
3960 	 */
3961 	if (pp == NULL)
3962 		return;
3963 
3964 	/*
3965 	 * Now round the request size up to page boundaries.
3966 	 * This ensures that the entire page will be
3967 	 * initialized to zeroes if EOF is encountered.
3968 	 */
3969 	io_len = ptob(btopr(io_len));
3970 
3971 	bp = pageio_setup(pp, io_len, vp, B_READ);
3972 	ASSERT(bp != NULL);
3973 
3974 	/*
3975 	 * pageio_setup should have set b_addr to 0.  This is correct since
3976 	 * we want to do I/O on a page boundary. bp_mapin() will use this addr
3977 	 * to calculate an offset, and then set b_addr to the kernel virtual
3978 	 * address it allocated for us.
3979 	 */
3980 	ASSERT(bp->b_un.b_addr == 0);
3981 
3982 	bp->b_edev = 0;
3983 	bp->b_dev = 0;
3984 	bp->b_lblkno = lbtodb(io_off);
3985 	bp->b_file = vp;
3986 	bp->b_offset = (offset_t)blkoff;
3987 	bp_mapin(bp);
3988 
3989 	/*
3990 	 * If doing a write beyond what we believe is EOF, don't bother trying
3991 	 * to read the pages from the server, we'll just zero the pages here.
3992 	 * We don't check that the rw flag is S_WRITE here because some
3993 	 * implementations may attempt a read access to the buffer before
3994 	 * copying data.
3995 	 */
3996 	mutex_enter(&rp->r_statelock);
3997 	if (io_off >= rp->r_size && seg == segkmap) {
3998 		mutex_exit(&rp->r_statelock);
3999 		bzero(bp->b_un.b_addr, io_len);
4000 		error = 0;
4001 	} else {
4002 		mutex_exit(&rp->r_statelock);
4003 		error = nfs_bio(bp, cr);
4004 		if (error == NFS_EOF)
4005 			error = 0;
4006 	}
4007 
4008 	/*
4009 	 * Unmap the buffer before freeing it.
4010 	 */
4011 	bp_mapout(bp);
4012 	pageio_done(bp);
4013 
4014 	pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ);
4015 
4016 	/*
4017 	 * In case of error set readahead offset
4018 	 * to the lowest offset.
4019 	 * pvn_read_done() calls VN_DISPOSE to destroy the pages
4020 	 */
4021 	if (error && rp->r_nextr > io_off) {
4022 		mutex_enter(&rp->r_statelock);
4023 		if (rp->r_nextr > io_off)
4024 			rp->r_nextr = io_off;
4025 		mutex_exit(&rp->r_statelock);
4026 	}
4027 }
4028 
4029 /*
4030  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
4031  * If len == 0, do from off to EOF.
4032  *
4033  * The normal cases should be len == 0 && off == 0 (entire vp list),
4034  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
4035  * (from pageout).
4036  */
4037 /* ARGSUSED */
4038 static int
4039 nfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4040 	caller_context_t *ct)
4041 {
4042 	int error;
4043 	rnode_t *rp;
4044 
4045 	ASSERT(cr != NULL);
4046 
4047 	/*
4048 	 * XXX - Why should this check be made here?
4049 	 */
4050 	if (vp->v_flag & VNOMAP)
4051 		return (ENOSYS);
4052 
4053 	if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp))
4054 		return (0);
4055 
4056 	if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone)
4057 		return (EIO);
4058 	ASSERT(off <= MAXOFF32_T);
4059 
4060 	rp = VTOR(vp);
4061 	mutex_enter(&rp->r_statelock);
4062 	rp->r_count++;
4063 	mutex_exit(&rp->r_statelock);
4064 	error = nfs_putpages(vp, off, len, flags, cr);
4065 	mutex_enter(&rp->r_statelock);
4066 	rp->r_count--;
4067 	cv_broadcast(&rp->r_cv);
4068 	mutex_exit(&rp->r_statelock);
4069 
4070 	return (error);
4071 }
4072 
4073 /*
4074  * Write out a single page, possibly klustering adjacent dirty pages.
4075  */
4076 int
4077 nfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
4078 	int flags, cred_t *cr)
4079 {
4080 	u_offset_t io_off;
4081 	u_offset_t lbn_off;
4082 	u_offset_t lbn;
4083 	size_t io_len;
4084 	uint_t bsize;
4085 	int error;
4086 	rnode_t *rp;
4087 
4088 	ASSERT(!vn_is_readonly(vp));
4089 	ASSERT(pp != NULL);
4090 	ASSERT(cr != NULL);
4091 	ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone);
4092 
4093 	rp = VTOR(vp);
4094 	ASSERT(rp->r_count > 0);
4095 
4096 	ASSERT(pp->p_offset <= MAXOFF32_T);
4097 
4098 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
4099 	lbn = pp->p_offset / bsize;
4100 	lbn_off = lbn * bsize;
4101 
4102 	/*
4103 	 * Find a kluster that fits in one block, or in
4104 	 * one page if pages are bigger than blocks.  If
4105 	 * there is less file space allocated than a whole
4106 	 * page, we'll shorten the i/o request below.
4107 	 */
4108 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off,
4109 	    roundup(bsize, PAGESIZE), flags);
4110 
4111 	/*
4112 	 * pvn_write_kluster shouldn't have returned a page with offset
4113 	 * behind the original page we were given.  Verify that.
4114 	 */
4115 	ASSERT((pp->p_offset / bsize) >= lbn);
4116 
4117 	/*
4118 	 * Now pp will have the list of kept dirty pages marked for
4119 	 * write back.  It will also handle invalidation and freeing
4120 	 * of pages that are not dirty.  Check for page length rounding
4121 	 * problems.
4122 	 */
4123 	if (io_off + io_len > lbn_off + bsize) {
4124 		ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE);
4125 		io_len = lbn_off + bsize - io_off;
4126 	}
4127 	/*
4128 	 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4129 	 * consistent value of r_size. RMODINPROGRESS is set in writerp().
4130 	 * When RMODINPROGRESS is set it indicates that a uiomove() is in
4131 	 * progress and the r_size has not been made consistent with the
4132 	 * new size of the file. When the uiomove() completes the r_size is
4133 	 * updated and the RMODINPROGRESS flag is cleared.
4134 	 *
4135 	 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4136 	 * consistent value of r_size. Without this handshaking, it is
4137 	 * possible that nfs(3)_bio() picks  up the old value of r_size
4138 	 * before the uiomove() in writerp() completes. This will result
4139 	 * in the write through nfs(3)_bio() being dropped.
4140 	 *
4141 	 * More precisely, there is a window between the time the uiomove()
4142 	 * completes and the time the r_size is updated. If a VOP_PUTPAGE()
4143 	 * operation intervenes in this window, the page will be picked up,
4144 	 * because it is dirty (it will be unlocked, unless it was
4145 	 * pagecreate'd). When the page is picked up as dirty, the dirty
4146 	 * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is
4147 	 * checked. This will still be the old size. Therefore the page will
4148 	 * not be written out. When segmap_release() calls VOP_PUTPAGE(),
4149 	 * the page will be found to be clean and the write will be dropped.
4150 	 */
4151 	if (rp->r_flags & RMODINPROGRESS) {
4152 		mutex_enter(&rp->r_statelock);
4153 		if ((rp->r_flags & RMODINPROGRESS) &&
4154 		    rp->r_modaddr + MAXBSIZE > io_off &&
4155 		    rp->r_modaddr < io_off + io_len) {
4156 			page_t *plist;
4157 			/*
4158 			 * A write is in progress for this region of the file.
4159 			 * If we did not detect RMODINPROGRESS here then this
4160 			 * path through nfs_putapage() would eventually go to
4161 			 * nfs(3)_bio() and may not write out all of the data
4162 			 * in the pages. We end up losing data. So we decide
4163 			 * to set the modified bit on each page in the page
4164 			 * list and mark the rnode with RDIRTY. This write
4165 			 * will be restarted at some later time.
4166 			 */
4167 			plist = pp;
4168 			while (plist != NULL) {
4169 				pp = plist;
4170 				page_sub(&plist, pp);
4171 				hat_setmod(pp);
4172 				page_io_unlock(pp);
4173 				page_unlock(pp);
4174 			}
4175 			rp->r_flags |= RDIRTY;
4176 			mutex_exit(&rp->r_statelock);
4177 			if (offp)
4178 				*offp = io_off;
4179 			if (lenp)
4180 				*lenp = io_len;
4181 			return (0);
4182 		}
4183 		mutex_exit(&rp->r_statelock);
4184 	}
4185 
4186 	if (flags & B_ASYNC) {
4187 		error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr,
4188 		    nfs_sync_putapage);
4189 	} else
4190 		error = nfs_sync_putapage(vp, pp, io_off, io_len, flags, cr);
4191 
4192 	if (offp)
4193 		*offp = io_off;
4194 	if (lenp)
4195 		*lenp = io_len;
4196 	return (error);
4197 }
4198 
4199 static int
4200 nfs_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4201 	int flags, cred_t *cr)
4202 {
4203 	int error;
4204 	rnode_t *rp;
4205 
4206 	flags |= B_WRITE;
4207 
4208 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4209 	error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4210 
4211 	rp = VTOR(vp);
4212 
4213 	if ((error == ENOSPC || error == EDQUOT || error == EACCES) &&
4214 	    (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) {
4215 		if (!(rp->r_flags & ROUTOFSPACE)) {
4216 			mutex_enter(&rp->r_statelock);
4217 			rp->r_flags |= ROUTOFSPACE;
4218 			mutex_exit(&rp->r_statelock);
4219 		}
4220 		flags |= B_ERROR;
4221 		pvn_write_done(pp, flags);
4222 		/*
4223 		 * If this was not an async thread, then try again to
4224 		 * write out the pages, but this time, also destroy
4225 		 * them whether or not the write is successful.  This
4226 		 * will prevent memory from filling up with these
4227 		 * pages and destroying them is the only alternative
4228 		 * if they can't be written out.
4229 		 *
4230 		 * Don't do this if this is an async thread because
4231 		 * when the pages are unlocked in pvn_write_done,
4232 		 * some other thread could have come along, locked
4233 		 * them, and queued for an async thread.  It would be
4234 		 * possible for all of the async threads to be tied
4235 		 * up waiting to lock the pages again and they would
4236 		 * all already be locked and waiting for an async
4237 		 * thread to handle them.  Deadlock.
4238 		 */
4239 		if (!(flags & B_ASYNC)) {
4240 			error = nfs_putpage(vp, io_off, io_len,
4241 			    B_INVAL | B_FORCE, cr, NULL);
4242 		}
4243 	} else {
4244 		if (error)
4245 			flags |= B_ERROR;
4246 		else if (rp->r_flags & ROUTOFSPACE) {
4247 			mutex_enter(&rp->r_statelock);
4248 			rp->r_flags &= ~ROUTOFSPACE;
4249 			mutex_exit(&rp->r_statelock);
4250 		}
4251 		pvn_write_done(pp, flags);
4252 	}
4253 
4254 	return (error);
4255 }
4256 
4257 /* ARGSUSED */
4258 static int
4259 nfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4260 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4261 	caller_context_t *ct)
4262 {
4263 	struct segvn_crargs vn_a;
4264 	int error;
4265 	rnode_t *rp;
4266 	struct vattr va;
4267 
4268 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4269 		return (EIO);
4270 
4271 	if (vp->v_flag & VNOMAP)
4272 		return (ENOSYS);
4273 
4274 	if (off > MAXOFF32_T)
4275 		return (EFBIG);
4276 
4277 	if (off < 0 || off + len < 0)
4278 		return (ENXIO);
4279 
4280 	if (vp->v_type != VREG)
4281 		return (ENODEV);
4282 
4283 	/*
4284 	 * If there is cached data and if close-to-open consistency
4285 	 * checking is not turned off and if the file system is not
4286 	 * mounted readonly, then force an over the wire getattr.
4287 	 * Otherwise, just invoke nfsgetattr to get a copy of the
4288 	 * attributes.  The attribute cache will be used unless it
4289 	 * is timed out and if it is, then an over the wire getattr
4290 	 * will be issued.
4291 	 */
4292 	va.va_mask = AT_ALL;
4293 	if (vn_has_cached_data(vp) &&
4294 	    !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp))
4295 		error = nfs_getattr_otw(vp, &va, cr);
4296 	else
4297 		error = nfsgetattr(vp, &va, cr);
4298 	if (error)
4299 		return (error);
4300 
4301 	/*
4302 	 * Check to see if the vnode is currently marked as not cachable.
4303 	 * This means portions of the file are locked (through VOP_FRLOCK).
4304 	 * In this case the map request must be refused.  We use
4305 	 * rp->r_lkserlock to avoid a race with concurrent lock requests.
4306 	 */
4307 	rp = VTOR(vp);
4308 
4309 	/*
4310 	 * Atomically increment r_inmap after acquiring r_rwlock. The
4311 	 * idea here is to acquire r_rwlock to block read/write and
4312 	 * not to protect r_inmap. r_inmap will inform nfs_read/write()
4313 	 * that we are in nfs_map(). Now, r_rwlock is acquired in order
4314 	 * and we can prevent the deadlock that would have occurred
4315 	 * when nfs_addmap() would have acquired it out of order.
4316 	 *
4317 	 * Since we are not protecting r_inmap by any lock, we do not
4318 	 * hold any lock when we decrement it. We atomically decrement
4319 	 * r_inmap after we release r_lkserlock.
4320 	 */
4321 
4322 	if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp)))
4323 		return (EINTR);
4324 	atomic_add_int(&rp->r_inmap, 1);
4325 	nfs_rw_exit(&rp->r_rwlock);
4326 
4327 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) {
4328 		atomic_add_int(&rp->r_inmap, -1);
4329 		return (EINTR);
4330 	}
4331 	if (vp->v_flag & VNOCACHE) {
4332 		error = EAGAIN;
4333 		goto done;
4334 	}
4335 
4336 	/*
4337 	 * Don't allow concurrent locks and mapping if mandatory locking is
4338 	 * enabled.
4339 	 */
4340 	if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) &&
4341 	    MANDLOCK(vp, va.va_mode)) {
4342 		error = EAGAIN;
4343 		goto done;
4344 	}
4345 
4346 	as_rangelock(as);
4347 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4348 	if (error != 0) {
4349 		as_rangeunlock(as);
4350 		goto done;
4351 	}
4352 
4353 	vn_a.vp = vp;
4354 	vn_a.offset = off;
4355 	vn_a.type = (flags & MAP_TYPE);
4356 	vn_a.prot = (uchar_t)prot;
4357 	vn_a.maxprot = (uchar_t)maxprot;
4358 	vn_a.flags = (flags & ~MAP_TYPE);
4359 	vn_a.cred = cr;
4360 	vn_a.amp = NULL;
4361 	vn_a.szc = 0;
4362 	vn_a.lgrp_mem_policy_flags = 0;
4363 
4364 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4365 	as_rangeunlock(as);
4366 
4367 done:
4368 	nfs_rw_exit(&rp->r_lkserlock);
4369 	atomic_add_int(&rp->r_inmap, -1);
4370 	return (error);
4371 }
4372 
4373 /* ARGSUSED */
4374 static int
4375 nfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4376 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4377 	caller_context_t *ct)
4378 {
4379 	rnode_t *rp;
4380 
4381 	if (vp->v_flag & VNOMAP)
4382 		return (ENOSYS);
4383 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4384 		return (EIO);
4385 
4386 	rp = VTOR(vp);
4387 	atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len));
4388 
4389 	return (0);
4390 }
4391 
4392 /* ARGSUSED */
4393 static int
4394 nfs_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, offset_t offset,
4395 	struct flk_callback *flk_cbp, cred_t *cr, caller_context_t *ct)
4396 {
4397 	netobj lm_fh;
4398 	int rc;
4399 	u_offset_t start, end;
4400 	rnode_t *rp;
4401 	int error = 0, intr = INTR(vp);
4402 
4403 	/* check for valid cmd parameter */
4404 	if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW)
4405 		return (EINVAL);
4406 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4407 		return (EIO);
4408 
4409 	/* Verify l_type. */
4410 	switch (bfp->l_type) {
4411 	case F_RDLCK:
4412 		if (cmd != F_GETLK && !(flag & FREAD))
4413 			return (EBADF);
4414 		break;
4415 	case F_WRLCK:
4416 		if (cmd != F_GETLK && !(flag & FWRITE))
4417 			return (EBADF);
4418 		break;
4419 	case F_UNLCK:
4420 		intr = 0;
4421 		break;
4422 
4423 	default:
4424 		return (EINVAL);
4425 	}
4426 
4427 	/* check the validity of the lock range */
4428 	if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset))
4429 		return (rc);
4430 	if (rc = flk_check_lock_data(start, end, MAXOFF32_T))
4431 		return (rc);
4432 
4433 	/*
4434 	 * If the filesystem is mounted using local locking, pass the
4435 	 * request off to the local locking code.
4436 	 */
4437 	if (VTOMI(vp)->mi_flags & MI_LLOCK) {
4438 		if (offset > MAXOFF32_T)
4439 			return (EFBIG);
4440 		if (cmd == F_SETLK || cmd == F_SETLKW) {
4441 			/*
4442 			 * For complete safety, we should be holding
4443 			 * r_lkserlock.  However, we can't call
4444 			 * lm_safelock and then fs_frlock while
4445 			 * holding r_lkserlock, so just invoke
4446 			 * lm_safelock and expect that this will
4447 			 * catch enough of the cases.
4448 			 */
4449 			if (!lm_safelock(vp, bfp, cr))
4450 				return (EAGAIN);
4451 		}
4452 		return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4453 	}
4454 
4455 	rp = VTOR(vp);
4456 
4457 	/*
4458 	 * Check whether the given lock request can proceed, given the
4459 	 * current file mappings.
4460 	 */
4461 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr))
4462 		return (EINTR);
4463 	if (cmd == F_SETLK || cmd == F_SETLKW) {
4464 		if (!lm_safelock(vp, bfp, cr)) {
4465 			rc = EAGAIN;
4466 			goto done;
4467 		}
4468 	}
4469 
4470 	/*
4471 	 * Flush the cache after waiting for async I/O to finish.  For new
4472 	 * locks, this is so that the process gets the latest bits from the
4473 	 * server.  For unlocks, this is so that other clients see the
4474 	 * latest bits once the file has been unlocked.  If currently dirty
4475 	 * pages can't be flushed, then don't allow a lock to be set.  But
4476 	 * allow unlocks to succeed, to avoid having orphan locks on the
4477 	 * server.
4478 	 */
4479 	if (cmd != F_GETLK) {
4480 		mutex_enter(&rp->r_statelock);
4481 		while (rp->r_count > 0) {
4482 			if (intr) {
4483 				klwp_t *lwp = ttolwp(curthread);
4484 
4485 				if (lwp != NULL)
4486 					lwp->lwp_nostop++;
4487 				if (cv_wait_sig(&rp->r_cv, &rp->r_statelock)
4488 				    == 0) {
4489 					if (lwp != NULL)
4490 						lwp->lwp_nostop--;
4491 					rc = EINTR;
4492 					break;
4493 				}
4494 				if (lwp != NULL)
4495 					lwp->lwp_nostop--;
4496 			} else
4497 			cv_wait(&rp->r_cv, &rp->r_statelock);
4498 		}
4499 		mutex_exit(&rp->r_statelock);
4500 		if (rc != 0)
4501 			goto done;
4502 		error = nfs_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct);
4503 		if (error) {
4504 			if (error == ENOSPC || error == EDQUOT) {
4505 				mutex_enter(&rp->r_statelock);
4506 				if (!rp->r_error)
4507 					rp->r_error = error;
4508 				mutex_exit(&rp->r_statelock);
4509 			}
4510 			if (bfp->l_type != F_UNLCK) {
4511 				rc = ENOLCK;
4512 				goto done;
4513 			}
4514 		}
4515 	}
4516 
4517 	lm_fh.n_len = sizeof (fhandle_t);
4518 	lm_fh.n_bytes = (char *)VTOFH(vp);
4519 
4520 	/*
4521 	 * Call the lock manager to do the real work of contacting
4522 	 * the server and obtaining the lock.
4523 	 */
4524 	rc = lm_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh, flk_cbp);
4525 
4526 	if (rc == 0)
4527 		nfs_lockcompletion(vp, cmd);
4528 
4529 done:
4530 	nfs_rw_exit(&rp->r_lkserlock);
4531 	return (rc);
4532 }
4533 
4534 /*
4535  * Free storage space associated with the specified vnode.  The portion
4536  * to be freed is specified by bfp->l_start and bfp->l_len (already
4537  * normalized to a "whence" of 0).
4538  *
4539  * This is an experimental facility whose continued existence is not
4540  * guaranteed.  Currently, we only support the special case
4541  * of l_len == 0, meaning free to end of file.
4542  */
4543 /* ARGSUSED */
4544 static int
4545 nfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
4546 	offset_t offset, cred_t *cr, caller_context_t *ct)
4547 {
4548 	int error;
4549 
4550 	ASSERT(vp->v_type == VREG);
4551 	if (cmd != F_FREESP)
4552 		return (EINVAL);
4553 
4554 	if (offset > MAXOFF32_T)
4555 		return (EFBIG);
4556 
4557 	if ((bfp->l_start > MAXOFF32_T) || (bfp->l_end > MAXOFF32_T) ||
4558 	    (bfp->l_len > MAXOFF32_T))
4559 		return (EFBIG);
4560 
4561 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4562 		return (EIO);
4563 
4564 	error = convoff(vp, bfp, 0, offset);
4565 	if (!error) {
4566 		ASSERT(bfp->l_start >= 0);
4567 		if (bfp->l_len == 0) {
4568 			struct vattr va;
4569 
4570 			/*
4571 			 * ftruncate should not change the ctime and
4572 			 * mtime if we truncate the file to its
4573 			 * previous size.
4574 			 */
4575 			va.va_mask = AT_SIZE;
4576 			error = nfsgetattr(vp, &va, cr);
4577 			if (error || va.va_size == bfp->l_start)
4578 				return (error);
4579 			va.va_mask = AT_SIZE;
4580 			va.va_size = bfp->l_start;
4581 			error = nfssetattr(vp, &va, 0, cr);
4582 		} else
4583 			error = EINVAL;
4584 	}
4585 
4586 	return (error);
4587 }
4588 
4589 /* ARGSUSED */
4590 static int
4591 nfs_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
4592 {
4593 
4594 	return (EINVAL);
4595 }
4596 
4597 /*
4598  * Setup and add an address space callback to do the work of the delmap call.
4599  * The callback will (and must be) deleted in the actual callback function.
4600  *
4601  * This is done in order to take care of the problem that we have with holding
4602  * the address space's a_lock for a long period of time (e.g. if the NFS server
4603  * is down).  Callbacks will be executed in the address space code while the
4604  * a_lock is not held.	Holding the address space's a_lock causes things such
4605  * as ps and fork to hang because they are trying to acquire this lock as well.
4606  */
4607 /* ARGSUSED */
4608 static int
4609 nfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4610 	size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4611 	caller_context_t *ct)
4612 {
4613 	int			caller_found;
4614 	int			error;
4615 	rnode_t			*rp;
4616 	nfs_delmap_args_t	*dmapp;
4617 	nfs_delmapcall_t	*delmap_call;
4618 
4619 	if (vp->v_flag & VNOMAP)
4620 		return (ENOSYS);
4621 	/*
4622 	 * A process may not change zones if it has NFS pages mmap'ed
4623 	 * in, so we can't legitimately get here from the wrong zone.
4624 	 */
4625 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4626 
4627 	rp = VTOR(vp);
4628 
4629 	/*
4630 	 * The way that the address space of this process deletes its mapping
4631 	 * of this file is via the following call chains:
4632 	 * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4633 	 * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4634 	 *
4635 	 * With the use of address space callbacks we are allowed to drop the
4636 	 * address space lock, a_lock, while executing the NFS operations that
4637 	 * need to go over the wire.  Returning EAGAIN to the caller of this
4638 	 * function is what drives the execution of the callback that we add
4639 	 * below.  The callback will be executed by the address space code
4640 	 * after dropping the a_lock.  When the callback is finished, since
4641 	 * we dropped the a_lock, it must be re-acquired and segvn_unmap()
4642 	 * is called again on the same segment to finish the rest of the work
4643 	 * that needs to happen during unmapping.
4644 	 *
4645 	 * This action of calling back into the segment driver causes
4646 	 * nfs_delmap() to get called again, but since the callback was
4647 	 * already executed at this point, it already did the work and there
4648 	 * is nothing left for us to do.
4649 	 *
4650 	 * To Summarize:
4651 	 * - The first time nfs_delmap is called by the current thread is when
4652 	 * we add the caller associated with this delmap to the delmap caller
4653 	 * list, add the callback, and return EAGAIN.
4654 	 * - The second time in this call chain when nfs_delmap is called we
4655 	 * will find this caller in the delmap caller list and realize there
4656 	 * is no more work to do thus removing this caller from the list and
4657 	 * returning the error that was set in the callback execution.
4658 	 */
4659 	caller_found = nfs_find_and_delete_delmapcall(rp, &error);
4660 	if (caller_found) {
4661 		/*
4662 		 * 'error' is from the actual delmap operations.  To avoid
4663 		 * hangs, we need to handle the return of EAGAIN differently
4664 		 * since this is what drives the callback execution.
4665 		 * In this case, we don't want to return EAGAIN and do the
4666 		 * callback execution because there are none to execute.
4667 		 */
4668 		if (error == EAGAIN)
4669 			return (0);
4670 		else
4671 			return (error);
4672 	}
4673 
4674 	/* current caller was not in the list */
4675 	delmap_call = nfs_init_delmapcall();
4676 
4677 	mutex_enter(&rp->r_statelock);
4678 	list_insert_tail(&rp->r_indelmap, delmap_call);
4679 	mutex_exit(&rp->r_statelock);
4680 
4681 	dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP);
4682 
4683 	dmapp->vp = vp;
4684 	dmapp->off = off;
4685 	dmapp->addr = addr;
4686 	dmapp->len = len;
4687 	dmapp->prot = prot;
4688 	dmapp->maxprot = maxprot;
4689 	dmapp->flags = flags;
4690 	dmapp->cr = cr;
4691 	dmapp->caller = delmap_call;
4692 
4693 	error = as_add_callback(as, nfs_delmap_callback, dmapp,
4694 	    AS_UNMAP_EVENT, addr, len, KM_SLEEP);
4695 
4696 	return (error ? error : EAGAIN);
4697 }
4698 
4699 /*
4700  * Remove some pages from an mmap'd vnode.  Just update the
4701  * count of pages.  If doing close-to-open, then flush all
4702  * of the pages associated with this file.  Otherwise, start
4703  * an asynchronous page flush to write out any dirty pages.
4704  * This will also associate a credential with the rnode which
4705  * can be used to write the pages.
4706  */
4707 /* ARGSUSED */
4708 static void
4709 nfs_delmap_callback(struct as *as, void *arg, uint_t event)
4710 {
4711 	int			error;
4712 	rnode_t			*rp;
4713 	mntinfo_t		*mi;
4714 	nfs_delmap_args_t	*dmapp = (nfs_delmap_args_t *)arg;
4715 
4716 	rp = VTOR(dmapp->vp);
4717 	mi = VTOMI(dmapp->vp);
4718 
4719 	atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len));
4720 	ASSERT(rp->r_mapcnt >= 0);
4721 
4722 	/*
4723 	 * Initiate a page flush if there are pages, the file system
4724 	 * was not mounted readonly, the segment was mapped shared, and
4725 	 * the pages themselves were writeable.
4726 	 */
4727 	if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) &&
4728 	    dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) {
4729 		mutex_enter(&rp->r_statelock);
4730 		rp->r_flags |= RDIRTY;
4731 		mutex_exit(&rp->r_statelock);
4732 		/*
4733 		 * If this is a cross-zone access a sync putpage won't work, so
4734 		 * the best we can do is try an async putpage.  That seems
4735 		 * better than something more draconian such as discarding the
4736 		 * dirty pages.
4737 		 */
4738 		if ((mi->mi_flags & MI_NOCTO) ||
4739 		    nfs_zone() != mi->mi_zone)
4740 			error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4741 			    B_ASYNC, dmapp->cr, NULL);
4742 		else
4743 			error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4744 			    0, dmapp->cr, NULL);
4745 		if (!error) {
4746 			mutex_enter(&rp->r_statelock);
4747 			error = rp->r_error;
4748 			rp->r_error = 0;
4749 			mutex_exit(&rp->r_statelock);
4750 		}
4751 	} else
4752 		error = 0;
4753 
4754 	if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO))
4755 		(void) nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4756 		    B_INVAL, dmapp->cr, NULL);
4757 
4758 	dmapp->caller->error = error;
4759 	(void) as_delete_callback(as, arg);
4760 	kmem_free(dmapp, sizeof (nfs_delmap_args_t));
4761 }
4762 
4763 /* ARGSUSED */
4764 static int
4765 nfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4766 	caller_context_t *ct)
4767 {
4768 	int error = 0;
4769 
4770 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4771 		return (EIO);
4772 	/*
4773 	 * This looks a little weird because it's written in a general
4774 	 * manner but we make little use of cases.  If cntl() ever gets
4775 	 * widely used, the outer switch will make more sense.
4776 	 */
4777 
4778 	switch (cmd) {
4779 
4780 	/*
4781 	 * Large file spec - need to base answer new query with
4782 	 * hardcoded constant based on the protocol.
4783 	 */
4784 	case _PC_FILESIZEBITS:
4785 		*valp = 32;
4786 		return (0);
4787 
4788 	case _PC_LINK_MAX:
4789 	case _PC_NAME_MAX:
4790 	case _PC_PATH_MAX:
4791 	case _PC_SYMLINK_MAX:
4792 	case _PC_CHOWN_RESTRICTED:
4793 	case _PC_NO_TRUNC: {
4794 		mntinfo_t *mi;
4795 		struct pathcnf *pc;
4796 
4797 		if ((mi = VTOMI(vp)) == NULL || (pc = mi->mi_pathconf) == NULL)
4798 			return (EINVAL);
4799 		error = _PC_ISSET(cmd, pc->pc_mask);    /* error or bool */
4800 		switch (cmd) {
4801 		case _PC_LINK_MAX:
4802 			*valp = pc->pc_link_max;
4803 			break;
4804 		case _PC_NAME_MAX:
4805 			*valp = pc->pc_name_max;
4806 			break;
4807 		case _PC_PATH_MAX:
4808 		case _PC_SYMLINK_MAX:
4809 			*valp = pc->pc_path_max;
4810 			break;
4811 		case _PC_CHOWN_RESTRICTED:
4812 			/*
4813 			 * if we got here, error is really a boolean which
4814 			 * indicates whether cmd is set or not.
4815 			 */
4816 			*valp = error ? 1 : 0;	/* see above */
4817 			error = 0;
4818 			break;
4819 		case _PC_NO_TRUNC:
4820 			/*
4821 			 * if we got here, error is really a boolean which
4822 			 * indicates whether cmd is set or not.
4823 			 */
4824 			*valp = error ? 1 : 0;	/* see above */
4825 			error = 0;
4826 			break;
4827 		}
4828 		return (error ? EINVAL : 0);
4829 		}
4830 
4831 	case _PC_XATTR_EXISTS:
4832 		*valp = 0;
4833 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
4834 			vnode_t *avp;
4835 			rnode_t *rp;
4836 			mntinfo_t *mi = VTOMI(vp);
4837 
4838 			if (!(mi->mi_flags & MI_EXTATTR))
4839 				return (0);
4840 
4841 			rp = VTOR(vp);
4842 			if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER,
4843 			    INTR(vp)))
4844 				return (EINTR);
4845 
4846 			error = nfslookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr);
4847 			if (error || avp == NULL)
4848 				error = acl_getxattrdir2(vp, &avp, 0, cr, 0);
4849 
4850 			nfs_rw_exit(&rp->r_rwlock);
4851 
4852 			if (error == 0 && avp != NULL) {
4853 				error = do_xattr_exists_check(avp, valp, cr);
4854 				VN_RELE(avp);
4855 			}
4856 		}
4857 		return (error ? EINVAL : 0);
4858 
4859 	case _PC_ACL_ENABLED:
4860 		*valp = _ACL_ACLENT_ENABLED;
4861 		return (0);
4862 
4863 	default:
4864 		return (EINVAL);
4865 	}
4866 }
4867 
4868 /*
4869  * Called by async thread to do synchronous pageio. Do the i/o, wait
4870  * for it to complete, and cleanup the page list when done.
4871  */
4872 static int
4873 nfs_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4874 	int flags, cred_t *cr)
4875 {
4876 	int error;
4877 
4878 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4879 	error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4880 	if (flags & B_READ)
4881 		pvn_read_done(pp, (error ? B_ERROR : 0) | flags);
4882 	else
4883 		pvn_write_done(pp, (error ? B_ERROR : 0) | flags);
4884 	return (error);
4885 }
4886 
4887 /* ARGSUSED */
4888 static int
4889 nfs_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4890 	int flags, cred_t *cr, caller_context_t *ct)
4891 {
4892 	int error;
4893 	rnode_t *rp;
4894 
4895 	if (pp == NULL)
4896 		return (EINVAL);
4897 
4898 	if (io_off > MAXOFF32_T)
4899 		return (EFBIG);
4900 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4901 		return (EIO);
4902 	rp = VTOR(vp);
4903 	mutex_enter(&rp->r_statelock);
4904 	rp->r_count++;
4905 	mutex_exit(&rp->r_statelock);
4906 
4907 	if (flags & B_ASYNC) {
4908 		error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr,
4909 		    nfs_sync_pageio);
4910 	} else
4911 		error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4912 	mutex_enter(&rp->r_statelock);
4913 	rp->r_count--;
4914 	cv_broadcast(&rp->r_cv);
4915 	mutex_exit(&rp->r_statelock);
4916 	return (error);
4917 }
4918 
4919 /* ARGSUSED */
4920 static int
4921 nfs_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4922 	caller_context_t *ct)
4923 {
4924 	int error;
4925 	mntinfo_t *mi;
4926 
4927 	mi = VTOMI(vp);
4928 
4929 	if (nfs_zone() != mi->mi_zone)
4930 		return (EIO);
4931 	if (mi->mi_flags & MI_ACL) {
4932 		error = acl_setacl2(vp, vsecattr, flag, cr);
4933 		if (mi->mi_flags & MI_ACL)
4934 			return (error);
4935 	}
4936 
4937 	return (ENOSYS);
4938 }
4939 
4940 /* ARGSUSED */
4941 static int
4942 nfs_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4943 	caller_context_t *ct)
4944 {
4945 	int error;
4946 	mntinfo_t *mi;
4947 
4948 	mi = VTOMI(vp);
4949 
4950 	if (nfs_zone() != mi->mi_zone)
4951 		return (EIO);
4952 	if (mi->mi_flags & MI_ACL) {
4953 		error = acl_getacl2(vp, vsecattr, flag, cr);
4954 		if (mi->mi_flags & MI_ACL)
4955 			return (error);
4956 	}
4957 
4958 	return (fs_fab_acl(vp, vsecattr, flag, cr, ct));
4959 }
4960 
4961 /* ARGSUSED */
4962 static int
4963 nfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
4964 	caller_context_t *ct)
4965 {
4966 	int error;
4967 	struct shrlock nshr;
4968 	struct nfs_owner nfs_owner;
4969 	netobj lm_fh;
4970 
4971 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4972 		return (EIO);
4973 
4974 	/*
4975 	 * check for valid cmd parameter
4976 	 */
4977 	if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS)
4978 		return (EINVAL);
4979 
4980 	/*
4981 	 * Check access permissions
4982 	 */
4983 	if (cmd == F_SHARE &&
4984 	    (((shr->s_access & F_RDACC) && !(flag & FREAD)) ||
4985 	    ((shr->s_access & F_WRACC) && !(flag & FWRITE))))
4986 		return (EBADF);
4987 
4988 	/*
4989 	 * If the filesystem is mounted using local locking, pass the
4990 	 * request off to the local share code.
4991 	 */
4992 	if (VTOMI(vp)->mi_flags & MI_LLOCK)
4993 		return (fs_shrlock(vp, cmd, shr, flag, cr, ct));
4994 
4995 	switch (cmd) {
4996 	case F_SHARE:
4997 	case F_UNSHARE:
4998 		lm_fh.n_len = sizeof (fhandle_t);
4999 		lm_fh.n_bytes = (char *)VTOFH(vp);
5000 
5001 		/*
5002 		 * If passed an owner that is too large to fit in an
5003 		 * nfs_owner it is likely a recursive call from the
5004 		 * lock manager client and pass it straight through.  If
5005 		 * it is not a nfs_owner then simply return an error.
5006 		 */
5007 		if (shr->s_own_len > sizeof (nfs_owner.lowner)) {
5008 			if (((struct nfs_owner *)shr->s_owner)->magic !=
5009 			    NFS_OWNER_MAGIC)
5010 				return (EINVAL);
5011 
5012 			if (error = lm_shrlock(vp, cmd, shr, flag, &lm_fh)) {
5013 				error = set_errno(error);
5014 			}
5015 			return (error);
5016 		}
5017 		/*
5018 		 * Remote share reservations owner is a combination of
5019 		 * a magic number, hostname, and the local owner
5020 		 */
5021 		bzero(&nfs_owner, sizeof (nfs_owner));
5022 		nfs_owner.magic = NFS_OWNER_MAGIC;
5023 		(void) strncpy(nfs_owner.hname, uts_nodename(),
5024 		    sizeof (nfs_owner.hname));
5025 		bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len);
5026 		nshr.s_access = shr->s_access;
5027 		nshr.s_deny = shr->s_deny;
5028 		nshr.s_sysid = 0;
5029 		nshr.s_pid = ttoproc(curthread)->p_pid;
5030 		nshr.s_own_len = sizeof (nfs_owner);
5031 		nshr.s_owner = (caddr_t)&nfs_owner;
5032 
5033 		if (error = lm_shrlock(vp, cmd, &nshr, flag, &lm_fh)) {
5034 			error = set_errno(error);
5035 		}
5036 
5037 		break;
5038 
5039 	case F_HASREMOTELOCKS:
5040 		/*
5041 		 * NFS client can't store remote locks itself
5042 		 */
5043 		shr->s_access = 0;
5044 		error = 0;
5045 		break;
5046 
5047 	default:
5048 		error = EINVAL;
5049 		break;
5050 	}
5051 
5052 	return (error);
5053 }
5054