xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs_vnops.c (revision 1022fd2a9aa2c967697116c2ca51a238a3c550ac)
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 (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
23  *
24  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
25  *	All rights reserved.
26  */
27 
28 #include <sys/param.h>
29 #include <sys/types.h>
30 #include <sys/systm.h>
31 #include <sys/cred.h>
32 #include <sys/time.h>
33 #include <sys/vnode.h>
34 #include <sys/vfs.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/file.h>
37 #include <sys/filio.h>
38 #include <sys/uio.h>
39 #include <sys/buf.h>
40 #include <sys/mman.h>
41 #include <sys/pathname.h>
42 #include <sys/dirent.h>
43 #include <sys/debug.h>
44 #include <sys/vmsystm.h>
45 #include <sys/fcntl.h>
46 #include <sys/flock.h>
47 #include <sys/swap.h>
48 #include <sys/errno.h>
49 #include <sys/strsubr.h>
50 #include <sys/sysmacros.h>
51 #include <sys/kmem.h>
52 #include <sys/cmn_err.h>
53 #include <sys/pathconf.h>
54 #include <sys/utsname.h>
55 #include <sys/dnlc.h>
56 #include <sys/acl.h>
57 #include <sys/atomic.h>
58 #include <sys/policy.h>
59 #include <sys/sdt.h>
60 
61 #include <rpc/types.h>
62 #include <rpc/auth.h>
63 #include <rpc/clnt.h>
64 
65 #include <nfs/nfs.h>
66 #include <nfs/nfs_clnt.h>
67 #include <nfs/rnode.h>
68 #include <nfs/nfs_acl.h>
69 #include <nfs/lm.h>
70 
71 #include <vm/hat.h>
72 #include <vm/as.h>
73 #include <vm/page.h>
74 #include <vm/pvn.h>
75 #include <vm/seg.h>
76 #include <vm/seg_map.h>
77 #include <vm/seg_kpm.h>
78 #include <vm/seg_vn.h>
79 
80 #include <fs/fs_subr.h>
81 
82 #include <sys/ddi.h>
83 
84 static int	nfs_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int,
85 			cred_t *);
86 static int	nfswrite(vnode_t *, caddr_t, uint_t, int, cred_t *);
87 static int	nfsread(vnode_t *, caddr_t, uint_t, int, size_t *, cred_t *);
88 static int	nfssetattr(vnode_t *, struct vattr *, int, cred_t *);
89 static int	nfslookup_dnlc(vnode_t *, char *, vnode_t **, cred_t *);
90 static int	nfslookup_otw(vnode_t *, char *, vnode_t **, cred_t *, int);
91 static int	nfsrename(vnode_t *, char *, vnode_t *, char *, cred_t *,
92 			caller_context_t *);
93 static int	nfsreaddir(vnode_t *, rddir_cache *, cred_t *);
94 static int	nfs_bio(struct buf *, cred_t *);
95 static int	nfs_getapage(vnode_t *, u_offset_t, size_t, uint_t *,
96 			page_t *[], size_t, struct seg *, caddr_t,
97 			enum seg_rw, cred_t *);
98 static void	nfs_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *,
99 			cred_t *);
100 static int	nfs_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t,
101 			int, cred_t *);
102 static int	nfs_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t,
103 			int, cred_t *);
104 static void	nfs_delmap_callback(struct as *, void *, uint_t);
105 
106 /*
107  * Error flags used to pass information about certain special errors
108  * which need to be handled specially.
109  */
110 #define	NFS_EOF			-98
111 
112 /*
113  * These are the vnode ops routines which implement the vnode interface to
114  * the networked file system.  These routines just take their parameters,
115  * make them look networkish by putting the right info into interface structs,
116  * and then calling the appropriate remote routine(s) to do the work.
117  *
118  * Note on directory name lookup cacheing:  If we detect a stale fhandle,
119  * we purge the directory cache relative to that vnode.  This way, the
120  * user won't get burned by the cache repeatedly.  See <nfs/rnode.h> for
121  * more details on rnode locking.
122  */
123 
124 static int	nfs_open(vnode_t **, int, cred_t *, caller_context_t *);
125 static int	nfs_close(vnode_t *, int, int, offset_t, cred_t *,
126 			caller_context_t *);
127 static int	nfs_read(vnode_t *, struct uio *, int, cred_t *,
128 			caller_context_t *);
129 static int	nfs_write(vnode_t *, struct uio *, int, cred_t *,
130 			caller_context_t *);
131 static int	nfs_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
132 			caller_context_t *);
133 static int	nfs_getattr(vnode_t *, struct vattr *, int, cred_t *,
134 			caller_context_t *);
135 static int	nfs_setattr(vnode_t *, struct vattr *, int, cred_t *,
136 			caller_context_t *);
137 static int	nfs_access(vnode_t *, int, int, cred_t *, caller_context_t *);
138 static int	nfs_accessx(void *, int, cred_t *);
139 static int	nfs_readlink(vnode_t *, struct uio *, cred_t *,
140 			caller_context_t *);
141 static int	nfs_fsync(vnode_t *, int, cred_t *, caller_context_t *);
142 static void	nfs_inactive(vnode_t *, cred_t *, caller_context_t *);
143 static int	nfs_lookup(vnode_t *, char *, vnode_t **, struct pathname *,
144 			int, vnode_t *, cred_t *, caller_context_t *,
145 			int *, pathname_t *);
146 static int	nfs_create(vnode_t *, char *, struct vattr *, enum vcexcl,
147 			int, vnode_t **, cred_t *, int, caller_context_t *,
148 			vsecattr_t *);
149 static int	nfs_remove(vnode_t *, char *, cred_t *, caller_context_t *,
150 			int);
151 static int	nfs_link(vnode_t *, vnode_t *, char *, cred_t *,
152 			caller_context_t *, int);
153 static int	nfs_rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
154 			caller_context_t *, int);
155 static int	nfs_mkdir(vnode_t *, char *, struct vattr *, vnode_t **,
156 			cred_t *, caller_context_t *, int, vsecattr_t *);
157 static int	nfs_rmdir(vnode_t *, char *, vnode_t *, cred_t *,
158 			caller_context_t *, int);
159 static int	nfs_symlink(vnode_t *, char *, struct vattr *, char *,
160 			cred_t *, caller_context_t *, int);
161 static int	nfs_readdir(vnode_t *, struct uio *, cred_t *, int *,
162 			caller_context_t *, int);
163 static int	nfs_fid(vnode_t *, fid_t *, caller_context_t *);
164 static int	nfs_rwlock(vnode_t *, int, caller_context_t *);
165 static void	nfs_rwunlock(vnode_t *, int, caller_context_t *);
166 static int	nfs_seek(vnode_t *, offset_t, offset_t *, caller_context_t *);
167 static int	nfs_getpage(vnode_t *, offset_t, size_t, uint_t *,
168 			page_t *[], size_t, struct seg *, caddr_t,
169 			enum seg_rw, cred_t *, caller_context_t *);
170 static int	nfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
171 			caller_context_t *);
172 static int	nfs_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
173 			uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
174 static int	nfs_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
175 			uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
176 static int	nfs_frlock(vnode_t *, int, struct flock64 *, int, offset_t,
177 			struct flk_callback *, cred_t *, caller_context_t *);
178 static int	nfs_space(vnode_t *, int, struct flock64 *, int, offset_t,
179 			cred_t *, caller_context_t *);
180 static int	nfs_realvp(vnode_t *, vnode_t **, caller_context_t *);
181 static int	nfs_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
182 			uint_t, uint_t, uint_t, cred_t *, caller_context_t *);
183 static int	nfs_pathconf(vnode_t *, int, ulong_t *, cred_t *,
184 			caller_context_t *);
185 static int	nfs_pageio(vnode_t *, page_t *, u_offset_t, size_t, int,
186 			cred_t *, caller_context_t *);
187 static int	nfs_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
188 			caller_context_t *);
189 static int	nfs_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
190 			caller_context_t *);
191 static int	nfs_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *,
192 			caller_context_t *);
193 
194 struct vnodeops *nfs_vnodeops;
195 
196 const fs_operation_def_t nfs_vnodeops_template[] = {
197 	VOPNAME_OPEN,		{ .vop_open = nfs_open },
198 	VOPNAME_CLOSE,		{ .vop_close = nfs_close },
199 	VOPNAME_READ,		{ .vop_read = nfs_read },
200 	VOPNAME_WRITE,		{ .vop_write = nfs_write },
201 	VOPNAME_IOCTL,		{ .vop_ioctl = nfs_ioctl },
202 	VOPNAME_GETATTR,	{ .vop_getattr = nfs_getattr },
203 	VOPNAME_SETATTR,	{ .vop_setattr = nfs_setattr },
204 	VOPNAME_ACCESS,		{ .vop_access = nfs_access },
205 	VOPNAME_LOOKUP,		{ .vop_lookup = nfs_lookup },
206 	VOPNAME_CREATE,		{ .vop_create = nfs_create },
207 	VOPNAME_REMOVE,		{ .vop_remove = nfs_remove },
208 	VOPNAME_LINK,		{ .vop_link = nfs_link },
209 	VOPNAME_RENAME,		{ .vop_rename = nfs_rename },
210 	VOPNAME_MKDIR,		{ .vop_mkdir = nfs_mkdir },
211 	VOPNAME_RMDIR,		{ .vop_rmdir = nfs_rmdir },
212 	VOPNAME_READDIR,	{ .vop_readdir = nfs_readdir },
213 	VOPNAME_SYMLINK,	{ .vop_symlink = nfs_symlink },
214 	VOPNAME_READLINK,	{ .vop_readlink = nfs_readlink },
215 	VOPNAME_FSYNC,		{ .vop_fsync = nfs_fsync },
216 	VOPNAME_INACTIVE,	{ .vop_inactive = nfs_inactive },
217 	VOPNAME_FID,		{ .vop_fid = nfs_fid },
218 	VOPNAME_RWLOCK,		{ .vop_rwlock = nfs_rwlock },
219 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = nfs_rwunlock },
220 	VOPNAME_SEEK,		{ .vop_seek = nfs_seek },
221 	VOPNAME_FRLOCK,		{ .vop_frlock = nfs_frlock },
222 	VOPNAME_SPACE,		{ .vop_space = nfs_space },
223 	VOPNAME_REALVP,		{ .vop_realvp = nfs_realvp },
224 	VOPNAME_GETPAGE,	{ .vop_getpage = nfs_getpage },
225 	VOPNAME_PUTPAGE,	{ .vop_putpage = nfs_putpage },
226 	VOPNAME_MAP,		{ .vop_map = nfs_map },
227 	VOPNAME_ADDMAP,		{ .vop_addmap = nfs_addmap },
228 	VOPNAME_DELMAP,		{ .vop_delmap = nfs_delmap },
229 	VOPNAME_DUMP,		{ .vop_dump = nfs_dump },
230 	VOPNAME_PATHCONF,	{ .vop_pathconf = nfs_pathconf },
231 	VOPNAME_PAGEIO,		{ .vop_pageio = nfs_pageio },
232 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = nfs_setsecattr },
233 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = nfs_getsecattr },
234 	VOPNAME_SHRLOCK,	{ .vop_shrlock = nfs_shrlock },
235 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
236 	NULL,			NULL
237 };
238 
239 /*
240  * XXX:  This is referenced in modstubs.s
241  */
242 struct vnodeops *
243 nfs_getvnodeops(void)
244 {
245 	return (nfs_vnodeops);
246 }
247 
248 /* ARGSUSED */
249 static int
250 nfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
251 {
252 	int error;
253 	struct vattr va;
254 	rnode_t *rp;
255 	vnode_t *vp;
256 
257 	vp = *vpp;
258 	rp = VTOR(vp);
259 	if (nfs_zone() != VTOMI(vp)->mi_zone)
260 		return (EIO);
261 	mutex_enter(&rp->r_statelock);
262 	if (rp->r_cred == NULL) {
263 		crhold(cr);
264 		rp->r_cred = cr;
265 	}
266 	mutex_exit(&rp->r_statelock);
267 
268 	/*
269 	 * If there is no cached data or if close-to-open
270 	 * consistency checking is turned off, we can avoid
271 	 * the over the wire getattr.  Otherwise, if the
272 	 * file system is mounted readonly, then just verify
273 	 * the caches are up to date using the normal mechanism.
274 	 * Else, if the file is not mmap'd, then just mark
275 	 * the attributes as timed out.  They will be refreshed
276 	 * and the caches validated prior to being used.
277 	 * Else, the file system is mounted writeable so
278 	 * force an over the wire GETATTR in order to ensure
279 	 * that all cached data is valid.
280 	 */
281 	if (vp->v_count > 1 ||
282 	    ((vn_has_cached_data(vp) || HAVE_RDDIR_CACHE(rp)) &&
283 	    !(VTOMI(vp)->mi_flags & MI_NOCTO))) {
284 		if (vn_is_readonly(vp))
285 			error = nfs_validate_caches(vp, cr);
286 		else if (rp->r_mapcnt == 0 && vp->v_count == 1) {
287 			PURGE_ATTRCACHE(vp);
288 			error = 0;
289 		} else {
290 			va.va_mask = AT_ALL;
291 			error = nfs_getattr_otw(vp, &va, cr);
292 		}
293 	} else
294 		error = 0;
295 
296 	return (error);
297 }
298 
299 /* ARGSUSED */
300 static int
301 nfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
302 	caller_context_t *ct)
303 {
304 	rnode_t *rp;
305 	int error;
306 	struct vattr va;
307 
308 	/*
309 	 * zone_enter(2) prevents processes from changing zones with NFS files
310 	 * open; if we happen to get here from the wrong zone we can't do
311 	 * anything over the wire.
312 	 */
313 	if (VTOMI(vp)->mi_zone != nfs_zone()) {
314 		/*
315 		 * We could attempt to clean up locks, except we're sure
316 		 * that the current process didn't acquire any locks on
317 		 * the file: any attempt to lock a file belong to another zone
318 		 * will fail, and one can't lock an NFS file and then change
319 		 * zones, as that fails too.
320 		 *
321 		 * Returning an error here is the sane thing to do.  A
322 		 * subsequent call to VN_RELE() which translates to a
323 		 * nfs_inactive() will clean up state: if the zone of the
324 		 * vnode's origin is still alive and kicking, an async worker
325 		 * thread will handle the request (from the correct zone), and
326 		 * everything (minus the final nfs_getattr_otw() call) should
327 		 * be OK. If the zone is going away nfs_async_inactive() will
328 		 * throw away cached pages inline.
329 		 */
330 		return (EIO);
331 	}
332 
333 	/*
334 	 * If we are using local locking for this filesystem, then
335 	 * release all of the SYSV style record locks.  Otherwise,
336 	 * we are doing network locking and we need to release all
337 	 * of the network locks.  All of the locks held by this
338 	 * process on this file are released no matter what the
339 	 * incoming reference count is.
340 	 */
341 	if (VTOMI(vp)->mi_flags & MI_LLOCK) {
342 		cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
343 		cleanshares(vp, ttoproc(curthread)->p_pid);
344 	} else
345 		nfs_lockrelease(vp, flag, offset, cr);
346 
347 	if (count > 1)
348 		return (0);
349 
350 	/*
351 	 * If the file has been `unlinked', then purge the
352 	 * DNLC so that this vnode will get reycled quicker
353 	 * and the .nfs* file on the server will get removed.
354 	 */
355 	rp = VTOR(vp);
356 	if (rp->r_unldvp != NULL)
357 		dnlc_purge_vp(vp);
358 
359 	/*
360 	 * If the file was open for write and there are pages,
361 	 * then if the file system was mounted using the "no-close-
362 	 *	to-open" semantics, then start an asynchronous flush
363 	 *	of the all of the pages in the file.
364 	 * else the file system was not mounted using the "no-close-
365 	 *	to-open" semantics, then do a synchronous flush and
366 	 *	commit of all of the dirty and uncommitted pages.
367 	 *
368 	 * The asynchronous flush of the pages in the "nocto" path
369 	 * mostly just associates a cred pointer with the rnode so
370 	 * writes which happen later will have a better chance of
371 	 * working.  It also starts the data being written to the
372 	 * server, but without unnecessarily delaying the application.
373 	 */
374 	if ((flag & FWRITE) && vn_has_cached_data(vp)) {
375 		if ((VTOMI(vp)->mi_flags & MI_NOCTO)) {
376 			error = nfs_putpage(vp, (offset_t)0, 0, B_ASYNC,
377 			    cr, ct);
378 			if (error == EAGAIN)
379 				error = 0;
380 		} else
381 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
382 		if (!error) {
383 			mutex_enter(&rp->r_statelock);
384 			error = rp->r_error;
385 			rp->r_error = 0;
386 			mutex_exit(&rp->r_statelock);
387 		}
388 	} else {
389 		mutex_enter(&rp->r_statelock);
390 		error = rp->r_error;
391 		rp->r_error = 0;
392 		mutex_exit(&rp->r_statelock);
393 	}
394 
395 	/*
396 	 * If RWRITEATTR is set, then issue an over the wire GETATTR to
397 	 * refresh the attribute cache with a set of attributes which
398 	 * weren't returned from a WRITE.  This will enable the close-
399 	 * to-open processing to work.
400 	 */
401 	if (rp->r_flags & RWRITEATTR)
402 		(void) nfs_getattr_otw(vp, &va, cr);
403 
404 	return (error);
405 }
406 
407 /* ARGSUSED */
408 static int
409 nfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
410 	caller_context_t *ct)
411 {
412 	rnode_t *rp;
413 	u_offset_t off;
414 	offset_t diff;
415 	int on;
416 	size_t n;
417 	caddr_t base;
418 	uint_t flags;
419 	int error;
420 	mntinfo_t *mi;
421 
422 	rp = VTOR(vp);
423 	mi = VTOMI(vp);
424 
425 	if (nfs_zone() != mi->mi_zone)
426 		return (EIO);
427 
428 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
429 
430 	if (vp->v_type != VREG)
431 		return (EISDIR);
432 
433 	if (uiop->uio_resid == 0)
434 		return (0);
435 
436 	if (uiop->uio_loffset > MAXOFF32_T)
437 		return (EFBIG);
438 
439 	if (uiop->uio_loffset < 0 ||
440 	    uiop->uio_loffset + uiop->uio_resid > MAXOFF32_T)
441 		return (EINVAL);
442 
443 	/*
444 	 * Bypass VM if caching has been disabled (e.g., locking) or if
445 	 * using client-side direct I/O and the file is not mmap'd and
446 	 * there are no cached pages.
447 	 */
448 	if ((vp->v_flag & VNOCACHE) ||
449 	    (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
450 	    rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
451 	    !vn_has_cached_data(vp))) {
452 		size_t bufsize;
453 		size_t resid = 0;
454 
455 		/*
456 		 * Let's try to do read in as large a chunk as we can
457 		 * (Filesystem (NFS client) bsize if possible/needed).
458 		 * For V3, this is 32K and for V2, this is 8K.
459 		 */
460 		bufsize = MIN(uiop->uio_resid, VTOMI(vp)->mi_curread);
461 		base = kmem_alloc(bufsize, KM_SLEEP);
462 		do {
463 			n = MIN(uiop->uio_resid, bufsize);
464 			error = nfsread(vp, base, uiop->uio_offset, n,
465 			    &resid, cr);
466 			if (!error) {
467 				n -= resid;
468 				error = uiomove(base, n, UIO_READ, uiop);
469 			}
470 		} while (!error && uiop->uio_resid > 0 && n > 0);
471 		kmem_free(base, bufsize);
472 		return (error);
473 	}
474 
475 	error = 0;
476 
477 	do {
478 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
479 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
480 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
481 
482 		error = nfs_validate_caches(vp, cr);
483 		if (error)
484 			break;
485 
486 		mutex_enter(&rp->r_statelock);
487 		while (rp->r_flags & RINCACHEPURGE) {
488 			if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
489 				mutex_exit(&rp->r_statelock);
490 				return (EINTR);
491 			}
492 		}
493 		diff = rp->r_size - uiop->uio_loffset;
494 		mutex_exit(&rp->r_statelock);
495 		if (diff <= 0)
496 			break;
497 		if (diff < n)
498 			n = (size_t)diff;
499 
500 		if (vpm_enable) {
501 			/*
502 			 * Copy data.
503 			 */
504 			error = vpm_data_copy(vp, off + on, n, uiop,
505 			    1, NULL, 0, S_READ);
506 		} else {
507 			base = segmap_getmapflt(segkmap, vp, off + on, n,
508 			    1, S_READ);
509 			error = uiomove(base + on, n, UIO_READ, uiop);
510 		}
511 
512 		if (!error) {
513 			/*
514 			 * If read a whole block or read to eof,
515 			 * won't need this buffer again soon.
516 			 */
517 			mutex_enter(&rp->r_statelock);
518 			if (n + on == MAXBSIZE ||
519 			    uiop->uio_loffset == rp->r_size)
520 				flags = SM_DONTNEED;
521 			else
522 				flags = 0;
523 			mutex_exit(&rp->r_statelock);
524 			if (vpm_enable) {
525 				error = vpm_sync_pages(vp, off, n, flags);
526 			} else {
527 				error = segmap_release(segkmap, base, flags);
528 			}
529 		} else {
530 			if (vpm_enable) {
531 				(void) vpm_sync_pages(vp, off, n, 0);
532 			} else {
533 				(void) segmap_release(segkmap, base, 0);
534 			}
535 		}
536 	} while (!error && uiop->uio_resid > 0);
537 
538 	return (error);
539 }
540 
541 /* ARGSUSED */
542 static int
543 nfs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
544 	caller_context_t *ct)
545 {
546 	rnode_t *rp;
547 	u_offset_t off;
548 	caddr_t base;
549 	uint_t flags;
550 	int remainder;
551 	size_t n;
552 	int on;
553 	int error;
554 	int resid;
555 	offset_t offset;
556 	rlim_t limit;
557 	mntinfo_t *mi;
558 
559 	rp = VTOR(vp);
560 
561 	mi = VTOMI(vp);
562 	if (nfs_zone() != mi->mi_zone)
563 		return (EIO);
564 	if (vp->v_type != VREG)
565 		return (EISDIR);
566 
567 	if (uiop->uio_resid == 0)
568 		return (0);
569 
570 	if (ioflag & FAPPEND) {
571 		struct vattr va;
572 
573 		/*
574 		 * Must serialize if appending.
575 		 */
576 		if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) {
577 			nfs_rw_exit(&rp->r_rwlock);
578 			if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER,
579 			    INTR(vp)))
580 				return (EINTR);
581 		}
582 
583 		va.va_mask = AT_SIZE;
584 		error = nfsgetattr(vp, &va, cr);
585 		if (error)
586 			return (error);
587 		uiop->uio_loffset = va.va_size;
588 	}
589 
590 	if (uiop->uio_loffset > MAXOFF32_T)
591 		return (EFBIG);
592 
593 	offset = uiop->uio_loffset + uiop->uio_resid;
594 
595 	if (uiop->uio_loffset < 0 || offset > MAXOFF32_T)
596 		return (EINVAL);
597 
598 	if (uiop->uio_llimit > (rlim64_t)MAXOFF32_T) {
599 		limit = MAXOFF32_T;
600 	} else {
601 		limit = (rlim_t)uiop->uio_llimit;
602 	}
603 
604 	/*
605 	 * Check to make sure that the process will not exceed
606 	 * its limit on file size.  It is okay to write up to
607 	 * the limit, but not beyond.  Thus, the write which
608 	 * reaches the limit will be short and the next write
609 	 * will return an error.
610 	 */
611 	remainder = 0;
612 	if (offset > limit) {
613 		remainder = offset - limit;
614 		uiop->uio_resid = limit - uiop->uio_offset;
615 		if (uiop->uio_resid <= 0) {
616 			proc_t *p = ttoproc(curthread);
617 
618 			uiop->uio_resid += remainder;
619 			mutex_enter(&p->p_lock);
620 			(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE],
621 			    p->p_rctls, p, RCA_UNSAFE_SIGINFO);
622 			mutex_exit(&p->p_lock);
623 			return (EFBIG);
624 		}
625 	}
626 
627 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp)))
628 		return (EINTR);
629 
630 	/*
631 	 * Bypass VM if caching has been disabled (e.g., locking) or if
632 	 * using client-side direct I/O and the file is not mmap'd and
633 	 * there are no cached pages.
634 	 */
635 	if ((vp->v_flag & VNOCACHE) ||
636 	    (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
637 	    rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
638 	    !vn_has_cached_data(vp))) {
639 		size_t bufsize;
640 		int count;
641 		uint_t org_offset;
642 
643 nfs_fwrite:
644 		if (rp->r_flags & RSTALE) {
645 			resid = uiop->uio_resid;
646 			offset = uiop->uio_loffset;
647 			error = rp->r_error;
648 			/*
649 			 * A close may have cleared r_error, if so,
650 			 * propagate ESTALE error return properly
651 			 */
652 			if (error == 0)
653 				error = ESTALE;
654 			goto bottom;
655 		}
656 		bufsize = MIN(uiop->uio_resid, mi->mi_curwrite);
657 		base = kmem_alloc(bufsize, KM_SLEEP);
658 		do {
659 			resid = uiop->uio_resid;
660 			offset = uiop->uio_loffset;
661 			count = MIN(uiop->uio_resid, bufsize);
662 			org_offset = uiop->uio_offset;
663 			error = uiomove(base, count, UIO_WRITE, uiop);
664 			if (!error) {
665 				error = nfswrite(vp, base, org_offset,
666 				    count, cr);
667 			}
668 		} while (!error && uiop->uio_resid > 0);
669 		kmem_free(base, bufsize);
670 		goto bottom;
671 	}
672 
673 	do {
674 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
675 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
676 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
677 
678 		resid = uiop->uio_resid;
679 		offset = uiop->uio_loffset;
680 
681 		if (rp->r_flags & RSTALE) {
682 			error = rp->r_error;
683 			/*
684 			 * A close may have cleared r_error, if so,
685 			 * propagate ESTALE error return properly
686 			 */
687 			if (error == 0)
688 				error = ESTALE;
689 			break;
690 		}
691 
692 		/*
693 		 * Don't create dirty pages faster than they
694 		 * can be cleaned so that the system doesn't
695 		 * get imbalanced.  If the async queue is
696 		 * maxed out, then wait for it to drain before
697 		 * creating more dirty pages.  Also, wait for
698 		 * any threads doing pagewalks in the vop_getattr
699 		 * entry points so that they don't block for
700 		 * long periods.
701 		 */
702 		mutex_enter(&rp->r_statelock);
703 		while ((mi->mi_max_threads != 0 &&
704 		    rp->r_awcount > 2 * mi->mi_max_threads) ||
705 		    rp->r_gcount > 0) {
706 			if (INTR(vp)) {
707 				klwp_t *lwp = ttolwp(curthread);
708 
709 				if (lwp != NULL)
710 					lwp->lwp_nostop++;
711 				if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
712 					mutex_exit(&rp->r_statelock);
713 					if (lwp != NULL)
714 						lwp->lwp_nostop--;
715 					error = EINTR;
716 					goto bottom;
717 				}
718 				if (lwp != NULL)
719 					lwp->lwp_nostop--;
720 			} else
721 				cv_wait(&rp->r_cv, &rp->r_statelock);
722 		}
723 		mutex_exit(&rp->r_statelock);
724 
725 		/*
726 		 * Touch the page and fault it in if it is not in core
727 		 * before segmap_getmapflt or vpm_data_copy can lock it.
728 		 * This is to avoid the deadlock if the buffer is mapped
729 		 * to the same file through mmap which we want to write.
730 		 */
731 		uio_prefaultpages((long)n, uiop);
732 
733 		if (vpm_enable) {
734 			/*
735 			 * It will use kpm mappings, so no need to
736 			 * pass an address.
737 			 */
738 			error = writerp(rp, NULL, n, uiop, 0);
739 		} else  {
740 			if (segmap_kpm) {
741 				int pon = uiop->uio_loffset & PAGEOFFSET;
742 				size_t pn = MIN(PAGESIZE - pon,
743 				    uiop->uio_resid);
744 				int pagecreate;
745 
746 				mutex_enter(&rp->r_statelock);
747 				pagecreate = (pon == 0) && (pn == PAGESIZE ||
748 				    uiop->uio_loffset + pn >= rp->r_size);
749 				mutex_exit(&rp->r_statelock);
750 
751 				base = segmap_getmapflt(segkmap, vp, off + on,
752 				    pn, !pagecreate, S_WRITE);
753 
754 				error = writerp(rp, base + pon, n, uiop,
755 				    pagecreate);
756 
757 			} else {
758 				base = segmap_getmapflt(segkmap, vp, off + on,
759 				    n, 0, S_READ);
760 				error = writerp(rp, base + on, n, uiop, 0);
761 			}
762 		}
763 
764 		if (!error) {
765 			if (mi->mi_flags & MI_NOAC)
766 				flags = SM_WRITE;
767 			else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
768 				/*
769 				 * Have written a whole block.
770 				 * Start an asynchronous write
771 				 * and mark the buffer to
772 				 * indicate that it won't be
773 				 * needed again soon.
774 				 */
775 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
776 			} else
777 				flags = 0;
778 			if ((ioflag & (FSYNC|FDSYNC)) ||
779 			    (rp->r_flags & ROUTOFSPACE)) {
780 				flags &= ~SM_ASYNC;
781 				flags |= SM_WRITE;
782 			}
783 			if (vpm_enable) {
784 				error = vpm_sync_pages(vp, off, n, flags);
785 			} else {
786 				error = segmap_release(segkmap, base, flags);
787 			}
788 		} else {
789 			if (vpm_enable) {
790 				(void) vpm_sync_pages(vp, off, n, 0);
791 			} else {
792 				(void) segmap_release(segkmap, base, 0);
793 			}
794 			/*
795 			 * In the event that we got an access error while
796 			 * faulting in a page for a write-only file just
797 			 * force a write.
798 			 */
799 			if (error == EACCES)
800 				goto nfs_fwrite;
801 		}
802 	} while (!error && uiop->uio_resid > 0);
803 
804 bottom:
805 	if (error) {
806 		uiop->uio_resid = resid + remainder;
807 		uiop->uio_loffset = offset;
808 	} else
809 		uiop->uio_resid += remainder;
810 
811 	nfs_rw_exit(&rp->r_lkserlock);
812 
813 	return (error);
814 }
815 
816 /*
817  * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
818  */
819 static int
820 nfs_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
821 	int flags, cred_t *cr)
822 {
823 	struct buf *bp;
824 	int error;
825 
826 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
827 	bp = pageio_setup(pp, len, vp, flags);
828 	ASSERT(bp != NULL);
829 
830 	/*
831 	 * pageio_setup should have set b_addr to 0.  This
832 	 * is correct since we want to do I/O on a page
833 	 * boundary.  bp_mapin will use this addr to calculate
834 	 * an offset, and then set b_addr to the kernel virtual
835 	 * address it allocated for us.
836 	 */
837 	ASSERT(bp->b_un.b_addr == 0);
838 
839 	bp->b_edev = 0;
840 	bp->b_dev = 0;
841 	bp->b_lblkno = lbtodb(off);
842 	bp->b_file = vp;
843 	bp->b_offset = (offset_t)off;
844 	bp_mapin(bp);
845 
846 	error = nfs_bio(bp, cr);
847 
848 	bp_mapout(bp);
849 	pageio_done(bp);
850 
851 	return (error);
852 }
853 
854 /*
855  * Write to file.  Writes to remote server in largest size
856  * chunks that the server can handle.  Write is synchronous.
857  */
858 static int
859 nfswrite(vnode_t *vp, caddr_t base, uint_t offset, int count, cred_t *cr)
860 {
861 	rnode_t *rp;
862 	mntinfo_t *mi;
863 	struct nfswriteargs wa;
864 	struct nfsattrstat ns;
865 	int error;
866 	int tsize;
867 	int douprintf;
868 
869 	douprintf = 1;
870 
871 	rp = VTOR(vp);
872 	mi = VTOMI(vp);
873 
874 	ASSERT(nfs_zone() == mi->mi_zone);
875 
876 	wa.wa_args = &wa.wa_args_buf;
877 	wa.wa_fhandle = *VTOFH(vp);
878 
879 	do {
880 		tsize = MIN(mi->mi_curwrite, count);
881 		wa.wa_data = base;
882 		wa.wa_begoff = offset;
883 		wa.wa_totcount = tsize;
884 		wa.wa_count = tsize;
885 		wa.wa_offset = offset;
886 
887 		if (mi->mi_io_kstats) {
888 			mutex_enter(&mi->mi_lock);
889 			kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
890 			mutex_exit(&mi->mi_lock);
891 		}
892 		wa.wa_mblk = NULL;
893 		do {
894 			error = rfs2call(mi, RFS_WRITE,
895 			    xdr_writeargs, (caddr_t)&wa,
896 			    xdr_attrstat, (caddr_t)&ns, cr,
897 			    &douprintf, &ns.ns_status, 0, NULL);
898 		} while (error == ENFS_TRYAGAIN);
899 		if (mi->mi_io_kstats) {
900 			mutex_enter(&mi->mi_lock);
901 			kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
902 			mutex_exit(&mi->mi_lock);
903 		}
904 
905 		if (!error) {
906 			error = geterrno(ns.ns_status);
907 			/*
908 			 * Can't check for stale fhandle and purge caches
909 			 * here because pages are held by nfs_getpage.
910 			 * Just mark the attribute cache as timed out
911 			 * and set RWRITEATTR to indicate that the file
912 			 * was modified with a WRITE operation.
913 			 */
914 			if (!error) {
915 				count -= tsize;
916 				base += tsize;
917 				offset += tsize;
918 				if (mi->mi_io_kstats) {
919 					mutex_enter(&mi->mi_lock);
920 					KSTAT_IO_PTR(mi->mi_io_kstats)->
921 					    writes++;
922 					KSTAT_IO_PTR(mi->mi_io_kstats)->
923 					    nwritten += tsize;
924 					mutex_exit(&mi->mi_lock);
925 				}
926 				lwp_stat_update(LWP_STAT_OUBLK, 1);
927 				mutex_enter(&rp->r_statelock);
928 				PURGE_ATTRCACHE_LOCKED(rp);
929 				rp->r_flags |= RWRITEATTR;
930 				mutex_exit(&rp->r_statelock);
931 			}
932 		}
933 	} while (!error && count);
934 
935 	return (error);
936 }
937 
938 /*
939  * Read from a file.  Reads data in largest chunks our interface can handle.
940  */
941 static int
942 nfsread(vnode_t *vp, caddr_t base, uint_t offset,
943     int count, size_t *residp, cred_t *cr)
944 {
945 	mntinfo_t *mi;
946 	struct nfsreadargs ra;
947 	struct nfsrdresult rr;
948 	int tsize;
949 	int error;
950 	int douprintf;
951 	failinfo_t fi;
952 	rnode_t *rp;
953 	struct vattr va;
954 	hrtime_t t;
955 
956 	rp = VTOR(vp);
957 	mi = VTOMI(vp);
958 
959 	ASSERT(nfs_zone() == mi->mi_zone);
960 
961 	douprintf = 1;
962 
963 	ra.ra_fhandle = *VTOFH(vp);
964 
965 	fi.vp = vp;
966 	fi.fhp = (caddr_t)&ra.ra_fhandle;
967 	fi.copyproc = nfscopyfh;
968 	fi.lookupproc = nfslookup;
969 	fi.xattrdirproc = acl_getxattrdir2;
970 
971 	do {
972 		if (mi->mi_io_kstats) {
973 			mutex_enter(&mi->mi_lock);
974 			kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
975 			mutex_exit(&mi->mi_lock);
976 		}
977 
978 		do {
979 			tsize = MIN(mi->mi_curread, count);
980 			rr.rr_data = base;
981 			ra.ra_offset = offset;
982 			ra.ra_totcount = tsize;
983 			ra.ra_count = tsize;
984 			ra.ra_data = base;
985 			t = gethrtime();
986 			error = rfs2call(mi, RFS_READ,
987 			    xdr_readargs, (caddr_t)&ra,
988 			    xdr_rdresult, (caddr_t)&rr, cr,
989 			    &douprintf, &rr.rr_status, 0, &fi);
990 		} while (error == ENFS_TRYAGAIN);
991 
992 		if (mi->mi_io_kstats) {
993 			mutex_enter(&mi->mi_lock);
994 			kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
995 			mutex_exit(&mi->mi_lock);
996 		}
997 
998 		if (!error) {
999 			error = geterrno(rr.rr_status);
1000 			if (!error) {
1001 				count -= rr.rr_count;
1002 				base += rr.rr_count;
1003 				offset += rr.rr_count;
1004 				if (mi->mi_io_kstats) {
1005 					mutex_enter(&mi->mi_lock);
1006 					KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
1007 					KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
1008 					    rr.rr_count;
1009 					mutex_exit(&mi->mi_lock);
1010 				}
1011 				lwp_stat_update(LWP_STAT_INBLK, 1);
1012 			}
1013 		}
1014 	} while (!error && count && rr.rr_count == tsize);
1015 
1016 	*residp = count;
1017 
1018 	if (!error) {
1019 		/*
1020 		 * Since no error occurred, we have the current
1021 		 * attributes and we need to do a cache check and then
1022 		 * potentially update the cached attributes.  We can't
1023 		 * use the normal attribute check and cache mechanisms
1024 		 * because they might cause a cache flush which would
1025 		 * deadlock.  Instead, we just check the cache to see
1026 		 * if the attributes have changed.  If it is, then we
1027 		 * just mark the attributes as out of date.  The next
1028 		 * time that the attributes are checked, they will be
1029 		 * out of date, new attributes will be fetched, and
1030 		 * the page cache will be flushed.  If the attributes
1031 		 * weren't changed, then we just update the cached
1032 		 * attributes with these attributes.
1033 		 */
1034 		/*
1035 		 * If NFS_ACL is supported on the server, then the
1036 		 * attributes returned by server may have minimal
1037 		 * permissions sometimes denying access to users having
1038 		 * proper access.  To get the proper attributes, mark
1039 		 * the attributes as expired so that they will be
1040 		 * regotten via the NFS_ACL GETATTR2 procedure.
1041 		 */
1042 		error = nattr_to_vattr(vp, &rr.rr_attr, &va);
1043 		mutex_enter(&rp->r_statelock);
1044 		if (error || !CACHE_VALID(rp, va.va_mtime, va.va_size) ||
1045 		    (mi->mi_flags & MI_ACL)) {
1046 			mutex_exit(&rp->r_statelock);
1047 			PURGE_ATTRCACHE(vp);
1048 		} else {
1049 			if (rp->r_mtime <= t) {
1050 				nfs_attrcache_va(vp, &va);
1051 			}
1052 			mutex_exit(&rp->r_statelock);
1053 		}
1054 	}
1055 
1056 	return (error);
1057 }
1058 
1059 /* ARGSUSED */
1060 static int
1061 nfs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp,
1062 	caller_context_t *ct)
1063 {
1064 
1065 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1066 		return (EIO);
1067 	switch (cmd) {
1068 		case _FIODIRECTIO:
1069 			return (nfs_directio(vp, (int)arg, cr));
1070 		default:
1071 			return (ENOTTY);
1072 	}
1073 }
1074 
1075 /* ARGSUSED */
1076 static int
1077 nfs_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1078 	caller_context_t *ct)
1079 {
1080 	int error;
1081 	rnode_t *rp;
1082 
1083 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1084 		return (EIO);
1085 	/*
1086 	 * If it has been specified that the return value will
1087 	 * just be used as a hint, and we are only being asked
1088 	 * for size, fsid or rdevid, then return the client's
1089 	 * notion of these values without checking to make sure
1090 	 * that the attribute cache is up to date.
1091 	 * The whole point is to avoid an over the wire GETATTR
1092 	 * call.
1093 	 */
1094 	rp = VTOR(vp);
1095 	if (flags & ATTR_HINT) {
1096 		if (vap->va_mask ==
1097 		    (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) {
1098 			mutex_enter(&rp->r_statelock);
1099 			if (vap->va_mask | AT_SIZE)
1100 				vap->va_size = rp->r_size;
1101 			if (vap->va_mask | AT_FSID)
1102 				vap->va_fsid = rp->r_attr.va_fsid;
1103 			if (vap->va_mask | AT_RDEV)
1104 				vap->va_rdev = rp->r_attr.va_rdev;
1105 			mutex_exit(&rp->r_statelock);
1106 			return (0);
1107 		}
1108 	}
1109 
1110 	/*
1111 	 * Only need to flush pages if asking for the mtime
1112 	 * and if there any dirty pages or any outstanding
1113 	 * asynchronous (write) requests for this file.
1114 	 */
1115 	if (vap->va_mask & AT_MTIME) {
1116 		if (vn_has_cached_data(vp) &&
1117 		    ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) {
1118 			mutex_enter(&rp->r_statelock);
1119 			rp->r_gcount++;
1120 			mutex_exit(&rp->r_statelock);
1121 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1122 			mutex_enter(&rp->r_statelock);
1123 			if (error && (error == ENOSPC || error == EDQUOT)) {
1124 				if (!rp->r_error)
1125 					rp->r_error = error;
1126 			}
1127 			if (--rp->r_gcount == 0)
1128 				cv_broadcast(&rp->r_cv);
1129 			mutex_exit(&rp->r_statelock);
1130 		}
1131 	}
1132 
1133 	return (nfsgetattr(vp, vap, cr));
1134 }
1135 
1136 /*ARGSUSED4*/
1137 static int
1138 nfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1139 		caller_context_t *ct)
1140 {
1141 	int error;
1142 	uint_t mask;
1143 	struct vattr va;
1144 
1145 	mask = vap->va_mask;
1146 
1147 	if (mask & AT_NOSET)
1148 		return (EINVAL);
1149 
1150 	if ((mask & AT_SIZE) &&
1151 	    vap->va_type == VREG &&
1152 	    vap->va_size > MAXOFF32_T)
1153 		return (EFBIG);
1154 
1155 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1156 		return (EIO);
1157 
1158 	va.va_mask = AT_UID | AT_MODE;
1159 
1160 	error = nfsgetattr(vp, &va, cr);
1161 	if (error)
1162 		return (error);
1163 
1164 	error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs_accessx,
1165 	    vp);
1166 
1167 	if (error)
1168 		return (error);
1169 
1170 	return (nfssetattr(vp, vap, flags, cr));
1171 }
1172 
1173 static int
1174 nfssetattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr)
1175 {
1176 	int error;
1177 	uint_t mask;
1178 	struct nfssaargs args;
1179 	struct nfsattrstat ns;
1180 	int douprintf;
1181 	rnode_t *rp;
1182 	struct vattr va;
1183 	mode_t omode;
1184 	mntinfo_t *mi;
1185 	vsecattr_t *vsp;
1186 	hrtime_t t;
1187 
1188 	mask = vap->va_mask;
1189 
1190 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
1191 
1192 	rp = VTOR(vp);
1193 
1194 	/*
1195 	 * Only need to flush pages if there are any pages and
1196 	 * if the file is marked as dirty in some fashion.  The
1197 	 * file must be flushed so that we can accurately
1198 	 * determine the size of the file and the cached data
1199 	 * after the SETATTR returns.  A file is considered to
1200 	 * be dirty if it is either marked with RDIRTY, has
1201 	 * outstanding i/o's active, or is mmap'd.  In this
1202 	 * last case, we can't tell whether there are dirty
1203 	 * pages, so we flush just to be sure.
1204 	 */
1205 	if (vn_has_cached_data(vp) &&
1206 	    ((rp->r_flags & RDIRTY) ||
1207 	    rp->r_count > 0 ||
1208 	    rp->r_mapcnt > 0)) {
1209 		ASSERT(vp->v_type != VCHR);
1210 		error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, NULL);
1211 		if (error && (error == ENOSPC || error == EDQUOT)) {
1212 			mutex_enter(&rp->r_statelock);
1213 			if (!rp->r_error)
1214 				rp->r_error = error;
1215 			mutex_exit(&rp->r_statelock);
1216 		}
1217 	}
1218 
1219 	/*
1220 	 * If the system call was utime(2) or utimes(2) and the
1221 	 * application did not specify the times, then set the
1222 	 * mtime nanosecond field to 1 billion.  This will get
1223 	 * translated from 1 billion nanoseconds to 1 million
1224 	 * microseconds in the over the wire request.  The
1225 	 * server will use 1 million in the microsecond field
1226 	 * to tell whether both the mtime and atime should be
1227 	 * set to the server's current time.
1228 	 *
1229 	 * This is an overload of the protocol and should be
1230 	 * documented in the NFS Version 2 protocol specification.
1231 	 */
1232 	if ((mask & AT_MTIME) && !(flags & ATTR_UTIME)) {
1233 		vap->va_mtime.tv_nsec = 1000000000;
1234 		if (NFS_TIME_T_OK(vap->va_mtime.tv_sec) &&
1235 		    NFS_TIME_T_OK(vap->va_atime.tv_sec)) {
1236 			error = vattr_to_sattr(vap, &args.saa_sa);
1237 		} else {
1238 			/*
1239 			 * Use server times. vap time values will not be used.
1240 			 * To ensure no time overflow, make sure vap has
1241 			 * valid values, but retain the original values.
1242 			 */
1243 			timestruc_t	mtime = vap->va_mtime;
1244 			timestruc_t	atime = vap->va_atime;
1245 			time_t		now;
1246 
1247 			now = gethrestime_sec();
1248 			if (NFS_TIME_T_OK(now)) {
1249 				/* Just in case server does not know of this */
1250 				vap->va_mtime.tv_sec = now;
1251 				vap->va_atime.tv_sec = now;
1252 			} else {
1253 				vap->va_mtime.tv_sec = 0;
1254 				vap->va_atime.tv_sec = 0;
1255 			}
1256 			error = vattr_to_sattr(vap, &args.saa_sa);
1257 			/* set vap times back on */
1258 			vap->va_mtime = mtime;
1259 			vap->va_atime = atime;
1260 		}
1261 	} else {
1262 		/* Either do not set times or use the client specified times */
1263 		error = vattr_to_sattr(vap, &args.saa_sa);
1264 	}
1265 	if (error) {
1266 		/* req time field(s) overflow - return immediately */
1267 		return (error);
1268 	}
1269 	args.saa_fh = *VTOFH(vp);
1270 
1271 	va.va_mask = AT_MODE;
1272 	error = nfsgetattr(vp, &va, cr);
1273 	if (error)
1274 		return (error);
1275 	omode = va.va_mode;
1276 
1277 	mi = VTOMI(vp);
1278 
1279 	douprintf = 1;
1280 
1281 	t = gethrtime();
1282 
1283 	error = rfs2call(mi, RFS_SETATTR,
1284 	    xdr_saargs, (caddr_t)&args,
1285 	    xdr_attrstat, (caddr_t)&ns, cr,
1286 	    &douprintf, &ns.ns_status, 0, NULL);
1287 
1288 	/*
1289 	 * Purge the access cache and ACL cache if changing either the
1290 	 * owner of the file, the group owner, or the mode.  These may
1291 	 * change the access permissions of the file, so purge old
1292 	 * information and start over again.
1293 	 */
1294 	if ((mask & (AT_UID | AT_GID | AT_MODE)) && (mi->mi_flags & MI_ACL)) {
1295 		(void) nfs_access_purge_rp(rp);
1296 		if (rp->r_secattr != NULL) {
1297 			mutex_enter(&rp->r_statelock);
1298 			vsp = rp->r_secattr;
1299 			rp->r_secattr = NULL;
1300 			mutex_exit(&rp->r_statelock);
1301 			if (vsp != NULL)
1302 				nfs_acl_free(vsp);
1303 		}
1304 	}
1305 
1306 	if (!error) {
1307 		error = geterrno(ns.ns_status);
1308 		if (!error) {
1309 			/*
1310 			 * If changing the size of the file, invalidate
1311 			 * any local cached data which is no longer part
1312 			 * of the file.  We also possibly invalidate the
1313 			 * last page in the file.  We could use
1314 			 * pvn_vpzero(), but this would mark the page as
1315 			 * modified and require it to be written back to
1316 			 * the server for no particularly good reason.
1317 			 * This way, if we access it, then we bring it
1318 			 * back in.  A read should be cheaper than a
1319 			 * write.
1320 			 */
1321 			if (mask & AT_SIZE) {
1322 				nfs_invalidate_pages(vp,
1323 				    (vap->va_size & PAGEMASK), cr);
1324 			}
1325 			(void) nfs_cache_fattr(vp, &ns.ns_attr, &va, t, cr);
1326 			/*
1327 			 * If NFS_ACL is supported on the server, then the
1328 			 * attributes returned by server may have minimal
1329 			 * permissions sometimes denying access to users having
1330 			 * proper access.  To get the proper attributes, mark
1331 			 * the attributes as expired so that they will be
1332 			 * regotten via the NFS_ACL GETATTR2 procedure.
1333 			 */
1334 			if (mi->mi_flags & MI_ACL) {
1335 				PURGE_ATTRCACHE(vp);
1336 			}
1337 			/*
1338 			 * This next check attempts to deal with NFS
1339 			 * servers which can not handle increasing
1340 			 * the size of the file via setattr.  Most
1341 			 * of these servers do not return an error,
1342 			 * but do not change the size of the file.
1343 			 * Hence, this check and then attempt to set
1344 			 * the file size by writing 1 byte at the
1345 			 * offset of the end of the file that we need.
1346 			 */
1347 			if ((mask & AT_SIZE) &&
1348 			    ns.ns_attr.na_size < (uint32_t)vap->va_size) {
1349 				char zb = '\0';
1350 
1351 				error = nfswrite(vp, &zb,
1352 				    vap->va_size - sizeof (zb),
1353 				    sizeof (zb), cr);
1354 			}
1355 			/*
1356 			 * Some servers will change the mode to clear the setuid
1357 			 * and setgid bits when changing the uid or gid.  The
1358 			 * client needs to compensate appropriately.
1359 			 */
1360 			if (mask & (AT_UID | AT_GID)) {
1361 				int terror;
1362 
1363 				va.va_mask = AT_MODE;
1364 				terror = nfsgetattr(vp, &va, cr);
1365 				if (!terror &&
1366 				    (((mask & AT_MODE) &&
1367 				    va.va_mode != vap->va_mode) ||
1368 				    (!(mask & AT_MODE) &&
1369 				    va.va_mode != omode))) {
1370 					va.va_mask = AT_MODE;
1371 					if (mask & AT_MODE)
1372 						va.va_mode = vap->va_mode;
1373 					else
1374 						va.va_mode = omode;
1375 					(void) nfssetattr(vp, &va, 0, cr);
1376 				}
1377 			}
1378 		} else {
1379 			PURGE_ATTRCACHE(vp);
1380 			PURGE_STALE_FH(error, vp, cr);
1381 		}
1382 	} else {
1383 		PURGE_ATTRCACHE(vp);
1384 	}
1385 
1386 	return (error);
1387 }
1388 
1389 static int
1390 nfs_accessx(void *vp, int mode, cred_t *cr)
1391 {
1392 	ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone);
1393 	return (nfs_access(vp, mode, 0, cr, NULL));
1394 }
1395 
1396 /* ARGSUSED */
1397 static int
1398 nfs_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
1399 {
1400 	struct vattr va;
1401 	int error;
1402 	mntinfo_t *mi;
1403 	int shift = 0;
1404 
1405 	mi = VTOMI(vp);
1406 
1407 	if (nfs_zone() != mi->mi_zone)
1408 		return (EIO);
1409 	if (mi->mi_flags & MI_ACL) {
1410 		error = acl_access2(vp, mode, flags, cr);
1411 		if (mi->mi_flags & MI_ACL)
1412 			return (error);
1413 	}
1414 
1415 	va.va_mask = AT_MODE | AT_UID | AT_GID;
1416 	error = nfsgetattr(vp, &va, cr);
1417 	if (error)
1418 		return (error);
1419 
1420 	/*
1421 	 * Disallow write attempts on read-only
1422 	 * file systems, unless the file is a
1423 	 * device node.
1424 	 */
1425 	if ((mode & VWRITE) && vn_is_readonly(vp) && !IS_DEVVP(vp))
1426 		return (EROFS);
1427 
1428 	/*
1429 	 * Disallow attempts to access mandatory lock files.
1430 	 */
1431 	if ((mode & (VWRITE | VREAD | VEXEC)) &&
1432 	    MANDLOCK(vp, va.va_mode))
1433 		return (EACCES);
1434 
1435 	/*
1436 	 * Access check is based on only
1437 	 * one of owner, group, public.
1438 	 * If not owner, then check group.
1439 	 * If not a member of the group,
1440 	 * then check public access.
1441 	 */
1442 	if (crgetuid(cr) != va.va_uid) {
1443 		shift += 3;
1444 		if (!groupmember(va.va_gid, cr))
1445 			shift += 3;
1446 	}
1447 
1448 	return (secpolicy_vnode_access2(cr, vp, va.va_uid,
1449 	    va.va_mode << shift, mode));
1450 }
1451 
1452 static int nfs_do_symlink_cache = 1;
1453 
1454 /* ARGSUSED */
1455 static int
1456 nfs_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct)
1457 {
1458 	int error;
1459 	struct nfsrdlnres rl;
1460 	rnode_t *rp;
1461 	int douprintf;
1462 	failinfo_t fi;
1463 
1464 	/*
1465 	 * We want to be consistent with UFS semantics so we will return
1466 	 * EINVAL instead of ENXIO. This violates the XNFS spec and
1467 	 * the RFC 1094, which are wrong any way. BUGID 1138002.
1468 	 */
1469 	if (vp->v_type != VLNK)
1470 		return (EINVAL);
1471 
1472 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1473 		return (EIO);
1474 
1475 	rp = VTOR(vp);
1476 	if (nfs_do_symlink_cache && rp->r_symlink.contents != NULL) {
1477 		error = nfs_validate_caches(vp, cr);
1478 		if (error)
1479 			return (error);
1480 		mutex_enter(&rp->r_statelock);
1481 		if (rp->r_symlink.contents != NULL) {
1482 			error = uiomove(rp->r_symlink.contents,
1483 			    rp->r_symlink.len, UIO_READ, uiop);
1484 			mutex_exit(&rp->r_statelock);
1485 			return (error);
1486 		}
1487 		mutex_exit(&rp->r_statelock);
1488 	}
1489 
1490 
1491 	rl.rl_data = kmem_alloc(NFS_MAXPATHLEN, KM_SLEEP);
1492 
1493 	fi.vp = vp;
1494 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1495 	fi.copyproc = nfscopyfh;
1496 	fi.lookupproc = nfslookup;
1497 	fi.xattrdirproc = acl_getxattrdir2;
1498 
1499 	douprintf = 1;
1500 
1501 	error = rfs2call(VTOMI(vp), RFS_READLINK,
1502 	    xdr_readlink, (caddr_t)VTOFH(vp),
1503 	    xdr_rdlnres, (caddr_t)&rl, cr,
1504 	    &douprintf, &rl.rl_status, 0, &fi);
1505 
1506 	if (error) {
1507 
1508 		kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1509 		return (error);
1510 	}
1511 
1512 	error = geterrno(rl.rl_status);
1513 	if (!error) {
1514 		error = uiomove(rl.rl_data, (int)rl.rl_count, UIO_READ, uiop);
1515 		if (nfs_do_symlink_cache && rp->r_symlink.contents == NULL) {
1516 			mutex_enter(&rp->r_statelock);
1517 			if (rp->r_symlink.contents == NULL) {
1518 				rp->r_symlink.contents = rl.rl_data;
1519 				rp->r_symlink.len = (int)rl.rl_count;
1520 				rp->r_symlink.size = NFS_MAXPATHLEN;
1521 				mutex_exit(&rp->r_statelock);
1522 			} else {
1523 				mutex_exit(&rp->r_statelock);
1524 
1525 				kmem_free((void *)rl.rl_data,
1526 				    NFS_MAXPATHLEN);
1527 			}
1528 		} else {
1529 
1530 			kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1531 		}
1532 	} else {
1533 		PURGE_STALE_FH(error, vp, cr);
1534 
1535 		kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1536 	}
1537 
1538 	/*
1539 	 * Conform to UFS semantics (see comment above)
1540 	 */
1541 	return (error == ENXIO ? EINVAL : error);
1542 }
1543 
1544 /*
1545  * Flush local dirty pages to stable storage on the server.
1546  *
1547  * If FNODSYNC is specified, then there is nothing to do because
1548  * metadata changes are not cached on the client before being
1549  * sent to the server.
1550  */
1551 /* ARGSUSED */
1552 static int
1553 nfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
1554 {
1555 	int error;
1556 
1557 	if ((syncflag & FNODSYNC) || IS_SWAPVP(vp))
1558 		return (0);
1559 
1560 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1561 		return (EIO);
1562 
1563 	error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1564 	if (!error)
1565 		error = VTOR(vp)->r_error;
1566 	return (error);
1567 }
1568 
1569 
1570 /*
1571  * Weirdness: if the file was removed or the target of a rename
1572  * operation while it was open, it got renamed instead.  Here we
1573  * remove the renamed file.
1574  */
1575 /* ARGSUSED */
1576 static void
1577 nfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1578 {
1579 	rnode_t *rp;
1580 
1581 	ASSERT(vp != DNLC_NO_VNODE);
1582 
1583 	/*
1584 	 * If this is coming from the wrong zone, we let someone in the right
1585 	 * zone take care of it asynchronously.  We can get here due to
1586 	 * VN_RELE() being called from pageout() or fsflush().  This call may
1587 	 * potentially turn into an expensive no-op if, for instance, v_count
1588 	 * gets incremented in the meantime, but it's still correct.
1589 	 */
1590 	if (nfs_zone() != VTOMI(vp)->mi_zone) {
1591 		nfs_async_inactive(vp, cr, nfs_inactive);
1592 		return;
1593 	}
1594 
1595 	rp = VTOR(vp);
1596 redo:
1597 	if (rp->r_unldvp != NULL) {
1598 		/*
1599 		 * Save the vnode pointer for the directory where the
1600 		 * unlinked-open file got renamed, then set it to NULL
1601 		 * to prevent another thread from getting here before
1602 		 * we're done with the remove.  While we have the
1603 		 * statelock, make local copies of the pertinent rnode
1604 		 * fields.  If we weren't to do this in an atomic way, the
1605 		 * the unl* fields could become inconsistent with respect
1606 		 * to each other due to a race condition between this
1607 		 * code and nfs_remove().  See bug report 1034328.
1608 		 */
1609 		mutex_enter(&rp->r_statelock);
1610 		if (rp->r_unldvp != NULL) {
1611 			vnode_t *unldvp;
1612 			char *unlname;
1613 			cred_t *unlcred;
1614 			struct nfsdiropargs da;
1615 			enum nfsstat status;
1616 			int douprintf;
1617 			int error;
1618 
1619 			unldvp = rp->r_unldvp;
1620 			rp->r_unldvp = NULL;
1621 			unlname = rp->r_unlname;
1622 			rp->r_unlname = NULL;
1623 			unlcred = rp->r_unlcred;
1624 			rp->r_unlcred = NULL;
1625 			mutex_exit(&rp->r_statelock);
1626 
1627 			/*
1628 			 * If there are any dirty pages left, then flush
1629 			 * them.  This is unfortunate because they just
1630 			 * may get thrown away during the remove operation,
1631 			 * but we have to do this for correctness.
1632 			 */
1633 			if (vn_has_cached_data(vp) &&
1634 			    ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
1635 				ASSERT(vp->v_type != VCHR);
1636 				error = nfs_putpage(vp, (offset_t)0, 0, 0,
1637 				    cr, ct);
1638 				if (error) {
1639 					mutex_enter(&rp->r_statelock);
1640 					if (!rp->r_error)
1641 						rp->r_error = error;
1642 					mutex_exit(&rp->r_statelock);
1643 				}
1644 			}
1645 
1646 			/*
1647 			 * Do the remove operation on the renamed file
1648 			 */
1649 			setdiropargs(&da, unlname, unldvp);
1650 
1651 			douprintf = 1;
1652 
1653 			(void) rfs2call(VTOMI(unldvp), RFS_REMOVE,
1654 			    xdr_diropargs, (caddr_t)&da,
1655 			    xdr_enum, (caddr_t)&status, unlcred,
1656 			    &douprintf, &status, 0, NULL);
1657 
1658 			if (HAVE_RDDIR_CACHE(VTOR(unldvp)))
1659 				nfs_purge_rddir_cache(unldvp);
1660 			PURGE_ATTRCACHE(unldvp);
1661 
1662 			/*
1663 			 * Release stuff held for the remove
1664 			 */
1665 			VN_RELE(unldvp);
1666 			kmem_free(unlname, MAXNAMELEN);
1667 			crfree(unlcred);
1668 			goto redo;
1669 		}
1670 		mutex_exit(&rp->r_statelock);
1671 	}
1672 
1673 	rp_addfree(rp, cr);
1674 }
1675 
1676 /*
1677  * Remote file system operations having to do with directory manipulation.
1678  */
1679 
1680 /* ARGSUSED */
1681 static int
1682 nfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1683 	int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1684 	int *direntflags, pathname_t *realpnp)
1685 {
1686 	int error;
1687 	vnode_t *vp;
1688 	vnode_t *avp = NULL;
1689 	rnode_t *drp;
1690 
1691 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
1692 		return (EPERM);
1693 
1694 	drp = VTOR(dvp);
1695 
1696 	/*
1697 	 * Are we looking up extended attributes?  If so, "dvp" is
1698 	 * the file or directory for which we want attributes, and
1699 	 * we need a lookup of the hidden attribute directory
1700 	 * before we lookup the rest of the path.
1701 	 */
1702 	if (flags & LOOKUP_XATTR) {
1703 		bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0);
1704 		mntinfo_t *mi;
1705 
1706 		mi = VTOMI(dvp);
1707 		if (!(mi->mi_flags & MI_EXTATTR))
1708 			return (EINVAL);
1709 
1710 		if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp)))
1711 			return (EINTR);
1712 
1713 		(void) nfslookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr);
1714 		if (avp == NULL)
1715 			error = acl_getxattrdir2(dvp, &avp, cflag, cr, 0);
1716 		else
1717 			error = 0;
1718 
1719 		nfs_rw_exit(&drp->r_rwlock);
1720 
1721 		if (error) {
1722 			if (mi->mi_flags & MI_EXTATTR)
1723 				return (error);
1724 			return (EINVAL);
1725 		}
1726 		dvp = avp;
1727 		drp = VTOR(dvp);
1728 	}
1729 
1730 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) {
1731 		error = EINTR;
1732 		goto out;
1733 	}
1734 
1735 	error = nfslookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0);
1736 
1737 	nfs_rw_exit(&drp->r_rwlock);
1738 
1739 	/*
1740 	 * If vnode is a device, create special vnode.
1741 	 */
1742 	if (!error && IS_DEVVP(*vpp)) {
1743 		vp = *vpp;
1744 		*vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
1745 		VN_RELE(vp);
1746 	}
1747 
1748 out:
1749 	if (avp != NULL)
1750 		VN_RELE(avp);
1751 
1752 	return (error);
1753 }
1754 
1755 static int nfs_lookup_neg_cache = 1;
1756 
1757 #ifdef DEBUG
1758 static int nfs_lookup_dnlc_hits = 0;
1759 static int nfs_lookup_dnlc_misses = 0;
1760 static int nfs_lookup_dnlc_neg_hits = 0;
1761 static int nfs_lookup_dnlc_disappears = 0;
1762 static int nfs_lookup_dnlc_lookups = 0;
1763 #endif
1764 
1765 /* ARGSUSED */
1766 int
1767 nfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1768 	int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags)
1769 {
1770 	int error;
1771 
1772 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1773 
1774 	/*
1775 	 * If lookup is for "", just return dvp.  Don't need
1776 	 * to send it over the wire, look it up in the dnlc,
1777 	 * or perform any access checks.
1778 	 */
1779 	if (*nm == '\0') {
1780 		VN_HOLD(dvp);
1781 		*vpp = dvp;
1782 		return (0);
1783 	}
1784 
1785 	/*
1786 	 * Can't do lookups in non-directories.
1787 	 */
1788 	if (dvp->v_type != VDIR)
1789 		return (ENOTDIR);
1790 
1791 	/*
1792 	 * If we're called with RFSCALL_SOFT, it's important that
1793 	 * the only rfscall is one we make directly; if we permit
1794 	 * an access call because we're looking up "." or validating
1795 	 * a dnlc hit, we'll deadlock because that rfscall will not
1796 	 * have the RFSCALL_SOFT set.
1797 	 */
1798 	if (rfscall_flags & RFSCALL_SOFT)
1799 		goto callit;
1800 
1801 	/*
1802 	 * If lookup is for ".", just return dvp.  Don't need
1803 	 * to send it over the wire or look it up in the dnlc,
1804 	 * just need to check access.
1805 	 */
1806 	if (strcmp(nm, ".") == 0) {
1807 		error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1808 		if (error)
1809 			return (error);
1810 		VN_HOLD(dvp);
1811 		*vpp = dvp;
1812 		return (0);
1813 	}
1814 
1815 	/*
1816 	 * Lookup this name in the DNLC.  If there was a valid entry,
1817 	 * then return the results of the lookup.
1818 	 */
1819 	error = nfslookup_dnlc(dvp, nm, vpp, cr);
1820 	if (error || *vpp != NULL)
1821 		return (error);
1822 
1823 callit:
1824 	error = nfslookup_otw(dvp, nm, vpp, cr, rfscall_flags);
1825 
1826 	return (error);
1827 }
1828 
1829 static int
1830 nfslookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr)
1831 {
1832 	int error;
1833 	vnode_t *vp;
1834 
1835 	ASSERT(*nm != '\0');
1836 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1837 
1838 	/*
1839 	 * Lookup this name in the DNLC.  If successful, then validate
1840 	 * the caches and then recheck the DNLC.  The DNLC is rechecked
1841 	 * just in case this entry got invalidated during the call
1842 	 * to nfs_validate_caches.
1843 	 *
1844 	 * An assumption is being made that it is safe to say that a
1845 	 * file exists which may not on the server.  Any operations to
1846 	 * the server will fail with ESTALE.
1847 	 */
1848 #ifdef DEBUG
1849 	nfs_lookup_dnlc_lookups++;
1850 #endif
1851 	vp = dnlc_lookup(dvp, nm);
1852 	if (vp != NULL) {
1853 		VN_RELE(vp);
1854 		if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) {
1855 			PURGE_ATTRCACHE(dvp);
1856 		}
1857 		error = nfs_validate_caches(dvp, cr);
1858 		if (error)
1859 			return (error);
1860 		vp = dnlc_lookup(dvp, nm);
1861 		if (vp != NULL) {
1862 			error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1863 			if (error) {
1864 				VN_RELE(vp);
1865 				return (error);
1866 			}
1867 			if (vp == DNLC_NO_VNODE) {
1868 				VN_RELE(vp);
1869 #ifdef DEBUG
1870 				nfs_lookup_dnlc_neg_hits++;
1871 #endif
1872 				return (ENOENT);
1873 			}
1874 			*vpp = vp;
1875 #ifdef DEBUG
1876 			nfs_lookup_dnlc_hits++;
1877 #endif
1878 			return (0);
1879 		}
1880 #ifdef DEBUG
1881 		nfs_lookup_dnlc_disappears++;
1882 #endif
1883 	}
1884 #ifdef DEBUG
1885 	else
1886 		nfs_lookup_dnlc_misses++;
1887 #endif
1888 
1889 	*vpp = NULL;
1890 
1891 	return (0);
1892 }
1893 
1894 static int
1895 nfslookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr,
1896 	int rfscall_flags)
1897 {
1898 	int error;
1899 	struct nfsdiropargs da;
1900 	struct nfsdiropres dr;
1901 	int douprintf;
1902 	failinfo_t fi;
1903 	hrtime_t t;
1904 
1905 	ASSERT(*nm != '\0');
1906 	ASSERT(dvp->v_type == VDIR);
1907 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1908 
1909 	setdiropargs(&da, nm, dvp);
1910 
1911 	fi.vp = dvp;
1912 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1913 	fi.copyproc = nfscopyfh;
1914 	fi.lookupproc = nfslookup;
1915 	fi.xattrdirproc = acl_getxattrdir2;
1916 
1917 	douprintf = 1;
1918 
1919 	t = gethrtime();
1920 
1921 	error = rfs2call(VTOMI(dvp), RFS_LOOKUP,
1922 	    xdr_diropargs, (caddr_t)&da,
1923 	    xdr_diropres, (caddr_t)&dr, cr,
1924 	    &douprintf, &dr.dr_status, rfscall_flags, &fi);
1925 
1926 	if (!error) {
1927 		error = geterrno(dr.dr_status);
1928 		if (!error) {
1929 			*vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
1930 			    dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm);
1931 			/*
1932 			 * If NFS_ACL is supported on the server, then the
1933 			 * attributes returned by server may have minimal
1934 			 * permissions sometimes denying access to users having
1935 			 * proper access.  To get the proper attributes, mark
1936 			 * the attributes as expired so that they will be
1937 			 * regotten via the NFS_ACL GETATTR2 procedure.
1938 			 */
1939 			if (VTOMI(*vpp)->mi_flags & MI_ACL) {
1940 				PURGE_ATTRCACHE(*vpp);
1941 			}
1942 			if (!(rfscall_flags & RFSCALL_SOFT))
1943 				dnlc_update(dvp, nm, *vpp);
1944 		} else {
1945 			PURGE_STALE_FH(error, dvp, cr);
1946 			if (error == ENOENT && nfs_lookup_neg_cache)
1947 				dnlc_enter(dvp, nm, DNLC_NO_VNODE);
1948 		}
1949 	}
1950 
1951 	return (error);
1952 }
1953 
1954 /* ARGSUSED */
1955 static int
1956 nfs_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
1957 	int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct,
1958 	vsecattr_t *vsecp)
1959 {
1960 	int error;
1961 	struct nfscreatargs args;
1962 	struct nfsdiropres dr;
1963 	int douprintf;
1964 	vnode_t *vp;
1965 	rnode_t *rp;
1966 	struct vattr vattr;
1967 	rnode_t *drp;
1968 	vnode_t *tempvp;
1969 	hrtime_t t;
1970 
1971 	drp = VTOR(dvp);
1972 
1973 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
1974 		return (EPERM);
1975 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
1976 		return (EINTR);
1977 
1978 	/*
1979 	 * We make a copy of the attributes because the caller does not
1980 	 * expect us to change what va points to.
1981 	 */
1982 	vattr = *va;
1983 
1984 	/*
1985 	 * If the pathname is "", just use dvp.  Don't need
1986 	 * to send it over the wire, look it up in the dnlc,
1987 	 * or perform any access checks.
1988 	 */
1989 	if (*nm == '\0') {
1990 		error = 0;
1991 		VN_HOLD(dvp);
1992 		vp = dvp;
1993 	/*
1994 	 * If the pathname is ".", just use dvp.  Don't need
1995 	 * to send it over the wire or look it up in the dnlc,
1996 	 * just need to check access.
1997 	 */
1998 	} else if (strcmp(nm, ".") == 0) {
1999 		error = nfs_access(dvp, VEXEC, 0, cr, ct);
2000 		if (error) {
2001 			nfs_rw_exit(&drp->r_rwlock);
2002 			return (error);
2003 		}
2004 		VN_HOLD(dvp);
2005 		vp = dvp;
2006 	/*
2007 	 * We need to go over the wire, just to be sure whether the
2008 	 * file exists or not.  Using the DNLC can be dangerous in
2009 	 * this case when making a decision regarding existence.
2010 	 */
2011 	} else {
2012 		error = nfslookup_otw(dvp, nm, &vp, cr, 0);
2013 	}
2014 	if (!error) {
2015 		if (exclusive == EXCL)
2016 			error = EEXIST;
2017 		else if (vp->v_type == VDIR && (mode & VWRITE))
2018 			error = EISDIR;
2019 		else {
2020 			/*
2021 			 * If vnode is a device, create special vnode.
2022 			 */
2023 			if (IS_DEVVP(vp)) {
2024 				tempvp = vp;
2025 				vp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2026 				VN_RELE(tempvp);
2027 			}
2028 			if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) {
2029 				if ((vattr.va_mask & AT_SIZE) &&
2030 				    vp->v_type == VREG) {
2031 					vattr.va_mask = AT_SIZE;
2032 					error = nfssetattr(vp, &vattr, 0, cr);
2033 				}
2034 			}
2035 		}
2036 		nfs_rw_exit(&drp->r_rwlock);
2037 		if (error) {
2038 			VN_RELE(vp);
2039 		} else {
2040 			/*
2041 			 * existing file got truncated, notify.
2042 			 */
2043 			vnevent_create(vp, ct);
2044 			*vpp = vp;
2045 		}
2046 		return (error);
2047 	}
2048 
2049 	ASSERT(vattr.va_mask & AT_TYPE);
2050 	if (vattr.va_type == VREG) {
2051 		ASSERT(vattr.va_mask & AT_MODE);
2052 		if (MANDMODE(vattr.va_mode)) {
2053 			nfs_rw_exit(&drp->r_rwlock);
2054 			return (EACCES);
2055 		}
2056 	}
2057 
2058 	dnlc_remove(dvp, nm);
2059 
2060 	setdiropargs(&args.ca_da, nm, dvp);
2061 
2062 	/*
2063 	 * Decide what the group-id of the created file should be.
2064 	 * Set it in attribute list as advisory...then do a setattr
2065 	 * if the server didn't get it right the first time.
2066 	 */
2067 	error = setdirgid(dvp, &vattr.va_gid, cr);
2068 	if (error) {
2069 		nfs_rw_exit(&drp->r_rwlock);
2070 		return (error);
2071 	}
2072 	vattr.va_mask |= AT_GID;
2073 
2074 	/*
2075 	 * This is a completely gross hack to make mknod
2076 	 * work over the wire until we can wack the protocol
2077 	 */
2078 #define	IFCHR		0020000		/* character special */
2079 #define	IFBLK		0060000		/* block special */
2080 #define	IFSOCK		0140000		/* socket */
2081 
2082 	/*
2083 	 * dev_t is uint_t in 5.x and short in 4.x. Both 4.x
2084 	 * supports 8 bit majors. 5.x supports 14 bit majors. 5.x supports 18
2085 	 * bits in the minor number where 4.x supports 8 bits.  If the 5.x
2086 	 * minor/major numbers <= 8 bits long, compress the device
2087 	 * number before sending it. Otherwise, the 4.x server will not
2088 	 * create the device with the correct device number and nothing can be
2089 	 * done about this.
2090 	 */
2091 	if (vattr.va_type == VCHR || vattr.va_type == VBLK) {
2092 		dev_t d = vattr.va_rdev;
2093 		dev32_t dev32;
2094 
2095 		if (vattr.va_type == VCHR)
2096 			vattr.va_mode |= IFCHR;
2097 		else
2098 			vattr.va_mode |= IFBLK;
2099 
2100 		(void) cmpldev(&dev32, d);
2101 		if (dev32 & ~((SO4_MAXMAJ << L_BITSMINOR32) | SO4_MAXMIN))
2102 			vattr.va_size = (u_offset_t)dev32;
2103 		else
2104 			vattr.va_size = (u_offset_t)nfsv2_cmpdev(d);
2105 
2106 		vattr.va_mask |= AT_MODE|AT_SIZE;
2107 	} else if (vattr.va_type == VFIFO) {
2108 		vattr.va_mode |= IFCHR;		/* xtra kludge for namedpipe */
2109 		vattr.va_size = (u_offset_t)NFS_FIFO_DEV;	/* blech */
2110 		vattr.va_mask |= AT_MODE|AT_SIZE;
2111 	} else if (vattr.va_type == VSOCK) {
2112 		vattr.va_mode |= IFSOCK;
2113 		/*
2114 		 * To avoid triggering bugs in the servers set AT_SIZE
2115 		 * (all other RFS_CREATE calls set this).
2116 		 */
2117 		vattr.va_size = 0;
2118 		vattr.va_mask |= AT_MODE|AT_SIZE;
2119 	}
2120 
2121 	args.ca_sa = &args.ca_sa_buf;
2122 	error = vattr_to_sattr(&vattr, args.ca_sa);
2123 	if (error) {
2124 		/* req time field(s) overflow - return immediately */
2125 		nfs_rw_exit(&drp->r_rwlock);
2126 		return (error);
2127 	}
2128 
2129 	douprintf = 1;
2130 
2131 	t = gethrtime();
2132 
2133 	error = rfs2call(VTOMI(dvp), RFS_CREATE,
2134 	    xdr_creatargs, (caddr_t)&args,
2135 	    xdr_diropres, (caddr_t)&dr, cr,
2136 	    &douprintf, &dr.dr_status, 0, NULL);
2137 
2138 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2139 
2140 	if (!error) {
2141 		error = geterrno(dr.dr_status);
2142 		if (!error) {
2143 			if (HAVE_RDDIR_CACHE(drp))
2144 				nfs_purge_rddir_cache(dvp);
2145 			vp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2146 			    dvp->v_vfsp, t, cr, NULL, NULL);
2147 			/*
2148 			 * If NFS_ACL is supported on the server, then the
2149 			 * attributes returned by server may have minimal
2150 			 * permissions sometimes denying access to users having
2151 			 * proper access.  To get the proper attributes, mark
2152 			 * the attributes as expired so that they will be
2153 			 * regotten via the NFS_ACL GETATTR2 procedure.
2154 			 */
2155 			if (VTOMI(vp)->mi_flags & MI_ACL) {
2156 				PURGE_ATTRCACHE(vp);
2157 			}
2158 			dnlc_update(dvp, nm, vp);
2159 			rp = VTOR(vp);
2160 			if (vattr.va_size == 0) {
2161 				mutex_enter(&rp->r_statelock);
2162 				rp->r_size = 0;
2163 				mutex_exit(&rp->r_statelock);
2164 				if (vn_has_cached_data(vp)) {
2165 					ASSERT(vp->v_type != VCHR);
2166 					nfs_invalidate_pages(vp,
2167 					    (u_offset_t)0, cr);
2168 				}
2169 			}
2170 
2171 			/*
2172 			 * Make sure the gid was set correctly.
2173 			 * If not, try to set it (but don't lose
2174 			 * any sleep over it).
2175 			 */
2176 			if (vattr.va_gid != rp->r_attr.va_gid) {
2177 				vattr.va_mask = AT_GID;
2178 				(void) nfssetattr(vp, &vattr, 0, cr);
2179 			}
2180 
2181 			/*
2182 			 * If vnode is a device create special vnode
2183 			 */
2184 			if (IS_DEVVP(vp)) {
2185 				*vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2186 				VN_RELE(vp);
2187 			} else
2188 				*vpp = vp;
2189 		} else {
2190 			PURGE_STALE_FH(error, dvp, cr);
2191 		}
2192 	}
2193 
2194 	nfs_rw_exit(&drp->r_rwlock);
2195 
2196 	return (error);
2197 }
2198 
2199 /*
2200  * Weirdness: if the vnode to be removed is open
2201  * we rename it instead of removing it and nfs_inactive
2202  * will remove the new name.
2203  */
2204 /* ARGSUSED */
2205 static int
2206 nfs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags)
2207 {
2208 	int error;
2209 	struct nfsdiropargs da;
2210 	enum nfsstat status;
2211 	vnode_t *vp;
2212 	char *tmpname;
2213 	int douprintf;
2214 	rnode_t *rp;
2215 	rnode_t *drp;
2216 
2217 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2218 		return (EPERM);
2219 	drp = VTOR(dvp);
2220 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2221 		return (EINTR);
2222 
2223 	error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2224 	if (error) {
2225 		nfs_rw_exit(&drp->r_rwlock);
2226 		return (error);
2227 	}
2228 
2229 	if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) {
2230 		VN_RELE(vp);
2231 		nfs_rw_exit(&drp->r_rwlock);
2232 		return (EPERM);
2233 	}
2234 
2235 	/*
2236 	 * First just remove the entry from the name cache, as it
2237 	 * is most likely the only entry for this vp.
2238 	 */
2239 	dnlc_remove(dvp, nm);
2240 
2241 	/*
2242 	 * If the file has a v_count > 1 then there may be more than one
2243 	 * entry in the name cache due multiple links or an open file,
2244 	 * but we don't have the real reference count so flush all
2245 	 * possible entries.
2246 	 */
2247 	if (vp->v_count > 1)
2248 		dnlc_purge_vp(vp);
2249 
2250 	/*
2251 	 * Now we have the real reference count on the vnode
2252 	 */
2253 	rp = VTOR(vp);
2254 	mutex_enter(&rp->r_statelock);
2255 	if (vp->v_count > 1 &&
2256 	    (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) {
2257 		mutex_exit(&rp->r_statelock);
2258 		tmpname = newname();
2259 		error = nfsrename(dvp, nm, dvp, tmpname, cr, ct);
2260 		if (error)
2261 			kmem_free(tmpname, MAXNAMELEN);
2262 		else {
2263 			mutex_enter(&rp->r_statelock);
2264 			if (rp->r_unldvp == NULL) {
2265 				VN_HOLD(dvp);
2266 				rp->r_unldvp = dvp;
2267 				if (rp->r_unlcred != NULL)
2268 					crfree(rp->r_unlcred);
2269 				crhold(cr);
2270 				rp->r_unlcred = cr;
2271 				rp->r_unlname = tmpname;
2272 			} else {
2273 				kmem_free(rp->r_unlname, MAXNAMELEN);
2274 				rp->r_unlname = tmpname;
2275 			}
2276 			mutex_exit(&rp->r_statelock);
2277 		}
2278 	} else {
2279 		mutex_exit(&rp->r_statelock);
2280 		/*
2281 		 * We need to flush any dirty pages which happen to
2282 		 * be hanging around before removing the file.  This
2283 		 * shouldn't happen very often and mostly on file
2284 		 * systems mounted "nocto".
2285 		 */
2286 		if (vn_has_cached_data(vp) &&
2287 		    ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
2288 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
2289 			if (error && (error == ENOSPC || error == EDQUOT)) {
2290 				mutex_enter(&rp->r_statelock);
2291 				if (!rp->r_error)
2292 					rp->r_error = error;
2293 				mutex_exit(&rp->r_statelock);
2294 			}
2295 		}
2296 
2297 		setdiropargs(&da, nm, dvp);
2298 
2299 		douprintf = 1;
2300 
2301 		error = rfs2call(VTOMI(dvp), RFS_REMOVE,
2302 		    xdr_diropargs, (caddr_t)&da,
2303 		    xdr_enum, (caddr_t)&status, cr,
2304 		    &douprintf, &status, 0, NULL);
2305 
2306 		/*
2307 		 * The xattr dir may be gone after last attr is removed,
2308 		 * so flush it from dnlc.
2309 		 */
2310 		if (dvp->v_flag & V_XATTRDIR)
2311 			dnlc_purge_vp(dvp);
2312 
2313 		PURGE_ATTRCACHE(dvp);	/* mod time changed */
2314 		PURGE_ATTRCACHE(vp);	/* link count changed */
2315 
2316 		if (!error) {
2317 			error = geterrno(status);
2318 			if (!error) {
2319 				if (HAVE_RDDIR_CACHE(drp))
2320 					nfs_purge_rddir_cache(dvp);
2321 			} else {
2322 				PURGE_STALE_FH(error, dvp, cr);
2323 			}
2324 		}
2325 	}
2326 
2327 	if (error == 0) {
2328 		vnevent_remove(vp, dvp, nm, ct);
2329 	}
2330 	VN_RELE(vp);
2331 
2332 	nfs_rw_exit(&drp->r_rwlock);
2333 
2334 	return (error);
2335 }
2336 
2337 /* ARGSUSED */
2338 static int
2339 nfs_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr,
2340 	caller_context_t *ct, int flags)
2341 {
2342 	int error;
2343 	struct nfslinkargs args;
2344 	enum nfsstat status;
2345 	vnode_t *realvp;
2346 	int douprintf;
2347 	rnode_t *tdrp;
2348 
2349 	if (nfs_zone() != VTOMI(tdvp)->mi_zone)
2350 		return (EPERM);
2351 	if (VOP_REALVP(svp, &realvp, ct) == 0)
2352 		svp = realvp;
2353 
2354 	args.la_from = VTOFH(svp);
2355 	setdiropargs(&args.la_to, tnm, tdvp);
2356 
2357 	tdrp = VTOR(tdvp);
2358 	if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp)))
2359 		return (EINTR);
2360 
2361 	dnlc_remove(tdvp, tnm);
2362 
2363 	douprintf = 1;
2364 
2365 	error = rfs2call(VTOMI(svp), RFS_LINK,
2366 	    xdr_linkargs, (caddr_t)&args,
2367 	    xdr_enum, (caddr_t)&status, cr,
2368 	    &douprintf, &status, 0, NULL);
2369 
2370 	PURGE_ATTRCACHE(tdvp);	/* mod time changed */
2371 	PURGE_ATTRCACHE(svp);	/* link count changed */
2372 
2373 	if (!error) {
2374 		error = geterrno(status);
2375 		if (!error) {
2376 			if (HAVE_RDDIR_CACHE(tdrp))
2377 				nfs_purge_rddir_cache(tdvp);
2378 		}
2379 	}
2380 
2381 	nfs_rw_exit(&tdrp->r_rwlock);
2382 
2383 	if (!error) {
2384 		/*
2385 		 * Notify the source file of this link operation.
2386 		 */
2387 		vnevent_link(svp, ct);
2388 	}
2389 	return (error);
2390 }
2391 
2392 /* ARGSUSED */
2393 static int
2394 nfs_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2395 	caller_context_t *ct, int flags)
2396 {
2397 	vnode_t *realvp;
2398 
2399 	if (nfs_zone() != VTOMI(odvp)->mi_zone)
2400 		return (EPERM);
2401 	if (VOP_REALVP(ndvp, &realvp, ct) == 0)
2402 		ndvp = realvp;
2403 
2404 	return (nfsrename(odvp, onm, ndvp, nnm, cr, ct));
2405 }
2406 
2407 /*
2408  * nfsrename does the real work of renaming in NFS Version 2.
2409  */
2410 static int
2411 nfsrename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2412     caller_context_t *ct)
2413 {
2414 	int error;
2415 	enum nfsstat status;
2416 	struct nfsrnmargs args;
2417 	int douprintf;
2418 	vnode_t *nvp = NULL;
2419 	vnode_t *ovp = NULL;
2420 	char *tmpname;
2421 	rnode_t *rp;
2422 	rnode_t *odrp;
2423 	rnode_t *ndrp;
2424 
2425 	ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone);
2426 	if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 ||
2427 	    strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0)
2428 		return (EINVAL);
2429 
2430 	odrp = VTOR(odvp);
2431 	ndrp = VTOR(ndvp);
2432 	if ((intptr_t)odrp < (intptr_t)ndrp) {
2433 		if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp)))
2434 			return (EINTR);
2435 		if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) {
2436 			nfs_rw_exit(&odrp->r_rwlock);
2437 			return (EINTR);
2438 		}
2439 	} else {
2440 		if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp)))
2441 			return (EINTR);
2442 		if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) {
2443 			nfs_rw_exit(&ndrp->r_rwlock);
2444 			return (EINTR);
2445 		}
2446 	}
2447 
2448 	/*
2449 	 * Lookup the target file.  If it exists, it needs to be
2450 	 * checked to see whether it is a mount point and whether
2451 	 * it is active (open).
2452 	 */
2453 	error = nfslookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0);
2454 	if (!error) {
2455 		/*
2456 		 * If this file has been mounted on, then just
2457 		 * return busy because renaming to it would remove
2458 		 * the mounted file system from the name space.
2459 		 */
2460 		if (vn_mountedvfs(nvp) != NULL) {
2461 			VN_RELE(nvp);
2462 			nfs_rw_exit(&odrp->r_rwlock);
2463 			nfs_rw_exit(&ndrp->r_rwlock);
2464 			return (EBUSY);
2465 		}
2466 
2467 		/*
2468 		 * Purge the name cache of all references to this vnode
2469 		 * so that we can check the reference count to infer
2470 		 * whether it is active or not.
2471 		 */
2472 		/*
2473 		 * First just remove the entry from the name cache, as it
2474 		 * is most likely the only entry for this vp.
2475 		 */
2476 		dnlc_remove(ndvp, nnm);
2477 		/*
2478 		 * If the file has a v_count > 1 then there may be more
2479 		 * than one entry in the name cache due multiple links
2480 		 * or an open file, but we don't have the real reference
2481 		 * count so flush all possible entries.
2482 		 */
2483 		if (nvp->v_count > 1)
2484 			dnlc_purge_vp(nvp);
2485 
2486 		/*
2487 		 * If the vnode is active and is not a directory,
2488 		 * arrange to rename it to a
2489 		 * temporary file so that it will continue to be
2490 		 * accessible.  This implements the "unlink-open-file"
2491 		 * semantics for the target of a rename operation.
2492 		 * Before doing this though, make sure that the
2493 		 * source and target files are not already the same.
2494 		 */
2495 		if (nvp->v_count > 1 && nvp->v_type != VDIR) {
2496 			/*
2497 			 * Lookup the source name.
2498 			 */
2499 			error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL,
2500 			    cr, 0);
2501 
2502 			/*
2503 			 * The source name *should* already exist.
2504 			 */
2505 			if (error) {
2506 				VN_RELE(nvp);
2507 				nfs_rw_exit(&odrp->r_rwlock);
2508 				nfs_rw_exit(&ndrp->r_rwlock);
2509 				return (error);
2510 			}
2511 
2512 			/*
2513 			 * Compare the two vnodes.  If they are the same,
2514 			 * just release all held vnodes and return success.
2515 			 */
2516 			if (ovp == nvp) {
2517 				VN_RELE(ovp);
2518 				VN_RELE(nvp);
2519 				nfs_rw_exit(&odrp->r_rwlock);
2520 				nfs_rw_exit(&ndrp->r_rwlock);
2521 				return (0);
2522 			}
2523 
2524 			/*
2525 			 * Can't mix and match directories and non-
2526 			 * directories in rename operations.  We already
2527 			 * know that the target is not a directory.  If
2528 			 * the source is a directory, return an error.
2529 			 */
2530 			if (ovp->v_type == VDIR) {
2531 				VN_RELE(ovp);
2532 				VN_RELE(nvp);
2533 				nfs_rw_exit(&odrp->r_rwlock);
2534 				nfs_rw_exit(&ndrp->r_rwlock);
2535 				return (ENOTDIR);
2536 			}
2537 
2538 			/*
2539 			 * The target file exists, is not the same as
2540 			 * the source file, and is active.  Link it
2541 			 * to a temporary filename to avoid having
2542 			 * the server removing the file completely.
2543 			 */
2544 			tmpname = newname();
2545 			error = nfs_link(ndvp, nvp, tmpname, cr, NULL, 0);
2546 			if (error == EOPNOTSUPP) {
2547 				error = nfs_rename(ndvp, nnm, ndvp, tmpname,
2548 				    cr, NULL, 0);
2549 			}
2550 			if (error) {
2551 				kmem_free(tmpname, MAXNAMELEN);
2552 				VN_RELE(ovp);
2553 				VN_RELE(nvp);
2554 				nfs_rw_exit(&odrp->r_rwlock);
2555 				nfs_rw_exit(&ndrp->r_rwlock);
2556 				return (error);
2557 			}
2558 			rp = VTOR(nvp);
2559 			mutex_enter(&rp->r_statelock);
2560 			if (rp->r_unldvp == NULL) {
2561 				VN_HOLD(ndvp);
2562 				rp->r_unldvp = ndvp;
2563 				if (rp->r_unlcred != NULL)
2564 					crfree(rp->r_unlcred);
2565 				crhold(cr);
2566 				rp->r_unlcred = cr;
2567 				rp->r_unlname = tmpname;
2568 			} else {
2569 				kmem_free(rp->r_unlname, MAXNAMELEN);
2570 				rp->r_unlname = tmpname;
2571 			}
2572 			mutex_exit(&rp->r_statelock);
2573 		}
2574 	}
2575 
2576 	if (ovp == NULL) {
2577 		/*
2578 		 * When renaming directories to be a subdirectory of a
2579 		 * different parent, the dnlc entry for ".." will no
2580 		 * longer be valid, so it must be removed.
2581 		 *
2582 		 * We do a lookup here to determine whether we are renaming
2583 		 * a directory and we need to check if we are renaming
2584 		 * an unlinked file.  This might have already been done
2585 		 * in previous code, so we check ovp == NULL to avoid
2586 		 * doing it twice.
2587 		 */
2588 
2589 		error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0);
2590 
2591 		/*
2592 		 * The source name *should* already exist.
2593 		 */
2594 		if (error) {
2595 			nfs_rw_exit(&odrp->r_rwlock);
2596 			nfs_rw_exit(&ndrp->r_rwlock);
2597 			if (nvp) {
2598 				VN_RELE(nvp);
2599 			}
2600 			return (error);
2601 		}
2602 		ASSERT(ovp != NULL);
2603 	}
2604 
2605 	dnlc_remove(odvp, onm);
2606 	dnlc_remove(ndvp, nnm);
2607 
2608 	setdiropargs(&args.rna_from, onm, odvp);
2609 	setdiropargs(&args.rna_to, nnm, ndvp);
2610 
2611 	douprintf = 1;
2612 
2613 	error = rfs2call(VTOMI(odvp), RFS_RENAME,
2614 	    xdr_rnmargs, (caddr_t)&args,
2615 	    xdr_enum, (caddr_t)&status, cr,
2616 	    &douprintf, &status, 0, NULL);
2617 
2618 	PURGE_ATTRCACHE(odvp);	/* mod time changed */
2619 	PURGE_ATTRCACHE(ndvp);	/* mod time changed */
2620 
2621 	if (!error) {
2622 		error = geterrno(status);
2623 		if (!error) {
2624 			if (HAVE_RDDIR_CACHE(odrp))
2625 				nfs_purge_rddir_cache(odvp);
2626 			if (HAVE_RDDIR_CACHE(ndrp))
2627 				nfs_purge_rddir_cache(ndvp);
2628 			/*
2629 			 * when renaming directories to be a subdirectory of a
2630 			 * different parent, the dnlc entry for ".." will no
2631 			 * longer be valid, so it must be removed
2632 			 */
2633 			rp = VTOR(ovp);
2634 			if (ndvp != odvp) {
2635 				if (ovp->v_type == VDIR) {
2636 					dnlc_remove(ovp, "..");
2637 					if (HAVE_RDDIR_CACHE(rp))
2638 						nfs_purge_rddir_cache(ovp);
2639 				}
2640 			}
2641 
2642 			/*
2643 			 * If we are renaming the unlinked file, update the
2644 			 * r_unldvp and r_unlname as needed.
2645 			 */
2646 			mutex_enter(&rp->r_statelock);
2647 			if (rp->r_unldvp != NULL) {
2648 				if (strcmp(rp->r_unlname, onm) == 0) {
2649 					(void) strncpy(rp->r_unlname,
2650 					    nnm, MAXNAMELEN);
2651 					rp->r_unlname[MAXNAMELEN - 1] = '\0';
2652 
2653 					if (ndvp != rp->r_unldvp) {
2654 						VN_RELE(rp->r_unldvp);
2655 						rp->r_unldvp = ndvp;
2656 						VN_HOLD(ndvp);
2657 					}
2658 				}
2659 			}
2660 			mutex_exit(&rp->r_statelock);
2661 		} else {
2662 			/*
2663 			 * System V defines rename to return EEXIST, not
2664 			 * ENOTEMPTY if the target directory is not empty.
2665 			 * Over the wire, the error is NFSERR_ENOTEMPTY
2666 			 * which geterrno maps to ENOTEMPTY.
2667 			 */
2668 			if (error == ENOTEMPTY)
2669 				error = EEXIST;
2670 		}
2671 	}
2672 
2673 	if (error == 0) {
2674 		if (nvp)
2675 			vnevent_rename_dest(nvp, ndvp, nnm, ct);
2676 
2677 		if (odvp != ndvp)
2678 			vnevent_rename_dest_dir(ndvp, ct);
2679 
2680 		ASSERT(ovp != NULL);
2681 		vnevent_rename_src(ovp, odvp, onm, ct);
2682 	}
2683 
2684 	if (nvp) {
2685 		VN_RELE(nvp);
2686 	}
2687 	VN_RELE(ovp);
2688 
2689 	nfs_rw_exit(&odrp->r_rwlock);
2690 	nfs_rw_exit(&ndrp->r_rwlock);
2691 
2692 	return (error);
2693 }
2694 
2695 /* ARGSUSED */
2696 static int
2697 nfs_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr,
2698 	caller_context_t *ct, int flags, vsecattr_t *vsecp)
2699 {
2700 	int error;
2701 	struct nfscreatargs args;
2702 	struct nfsdiropres dr;
2703 	int douprintf;
2704 	rnode_t *drp;
2705 	hrtime_t t;
2706 
2707 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2708 		return (EPERM);
2709 
2710 	setdiropargs(&args.ca_da, nm, dvp);
2711 
2712 	/*
2713 	 * Decide what the group-id and set-gid bit of the created directory
2714 	 * should be.  May have to do a setattr to get the gid right.
2715 	 */
2716 	error = setdirgid(dvp, &va->va_gid, cr);
2717 	if (error)
2718 		return (error);
2719 	error = setdirmode(dvp, &va->va_mode, cr);
2720 	if (error)
2721 		return (error);
2722 	va->va_mask |= AT_MODE|AT_GID;
2723 
2724 	args.ca_sa = &args.ca_sa_buf;
2725 	error = vattr_to_sattr(va, args.ca_sa);
2726 	if (error) {
2727 		/* req time field(s) overflow - return immediately */
2728 		return (error);
2729 	}
2730 
2731 	drp = VTOR(dvp);
2732 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2733 		return (EINTR);
2734 
2735 	dnlc_remove(dvp, nm);
2736 
2737 	douprintf = 1;
2738 
2739 	t = gethrtime();
2740 
2741 	error = rfs2call(VTOMI(dvp), RFS_MKDIR,
2742 	    xdr_creatargs, (caddr_t)&args,
2743 	    xdr_diropres, (caddr_t)&dr, cr,
2744 	    &douprintf, &dr.dr_status, 0, NULL);
2745 
2746 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2747 
2748 	if (!error) {
2749 		error = geterrno(dr.dr_status);
2750 		if (!error) {
2751 			if (HAVE_RDDIR_CACHE(drp))
2752 				nfs_purge_rddir_cache(dvp);
2753 			/*
2754 			 * The attributes returned by RFS_MKDIR can not
2755 			 * be depended upon, so mark the attribute cache
2756 			 * as purged.  A subsequent GETATTR will get the
2757 			 * correct attributes from the server.
2758 			 */
2759 			*vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2760 			    dvp->v_vfsp, t, cr, NULL, NULL);
2761 			PURGE_ATTRCACHE(*vpp);
2762 			dnlc_update(dvp, nm, *vpp);
2763 
2764 			/*
2765 			 * Make sure the gid was set correctly.
2766 			 * If not, try to set it (but don't lose
2767 			 * any sleep over it).
2768 			 */
2769 			if (va->va_gid != VTOR(*vpp)->r_attr.va_gid) {
2770 				va->va_mask = AT_GID;
2771 				(void) nfssetattr(*vpp, va, 0, cr);
2772 			}
2773 		} else {
2774 			PURGE_STALE_FH(error, dvp, cr);
2775 		}
2776 	}
2777 
2778 	nfs_rw_exit(&drp->r_rwlock);
2779 
2780 	return (error);
2781 }
2782 
2783 /* ARGSUSED */
2784 static int
2785 nfs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
2786 	caller_context_t *ct, int flags)
2787 {
2788 	int error;
2789 	enum nfsstat status;
2790 	struct nfsdiropargs da;
2791 	vnode_t *vp;
2792 	int douprintf;
2793 	rnode_t *drp;
2794 
2795 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2796 		return (EPERM);
2797 	drp = VTOR(dvp);
2798 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2799 		return (EINTR);
2800 
2801 	/*
2802 	 * Attempt to prevent a rmdir(".") from succeeding.
2803 	 */
2804 	error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2805 	if (error) {
2806 		nfs_rw_exit(&drp->r_rwlock);
2807 		return (error);
2808 	}
2809 
2810 	if (vp == cdir) {
2811 		VN_RELE(vp);
2812 		nfs_rw_exit(&drp->r_rwlock);
2813 		return (EINVAL);
2814 	}
2815 
2816 	setdiropargs(&da, nm, dvp);
2817 
2818 	/*
2819 	 * First just remove the entry from the name cache, as it
2820 	 * is most likely an entry for this vp.
2821 	 */
2822 	dnlc_remove(dvp, nm);
2823 
2824 	/*
2825 	 * If there vnode reference count is greater than one, then
2826 	 * there may be additional references in the DNLC which will
2827 	 * need to be purged.  First, trying removing the entry for
2828 	 * the parent directory and see if that removes the additional
2829 	 * reference(s).  If that doesn't do it, then use dnlc_purge_vp
2830 	 * to completely remove any references to the directory which
2831 	 * might still exist in the DNLC.
2832 	 */
2833 	if (vp->v_count > 1) {
2834 		dnlc_remove(vp, "..");
2835 		if (vp->v_count > 1)
2836 			dnlc_purge_vp(vp);
2837 	}
2838 
2839 	douprintf = 1;
2840 
2841 	error = rfs2call(VTOMI(dvp), RFS_RMDIR,
2842 	    xdr_diropargs, (caddr_t)&da,
2843 	    xdr_enum, (caddr_t)&status, cr,
2844 	    &douprintf, &status, 0, NULL);
2845 
2846 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2847 
2848 	if (error) {
2849 		VN_RELE(vp);
2850 		nfs_rw_exit(&drp->r_rwlock);
2851 		return (error);
2852 	}
2853 
2854 	error = geterrno(status);
2855 	if (!error) {
2856 		if (HAVE_RDDIR_CACHE(drp))
2857 			nfs_purge_rddir_cache(dvp);
2858 		if (HAVE_RDDIR_CACHE(VTOR(vp)))
2859 			nfs_purge_rddir_cache(vp);
2860 	} else {
2861 		PURGE_STALE_FH(error, dvp, cr);
2862 		/*
2863 		 * System V defines rmdir to return EEXIST, not
2864 		 * ENOTEMPTY if the directory is not empty.  Over
2865 		 * the wire, the error is NFSERR_ENOTEMPTY which
2866 		 * geterrno maps to ENOTEMPTY.
2867 		 */
2868 		if (error == ENOTEMPTY)
2869 			error = EEXIST;
2870 	}
2871 
2872 	if (error == 0) {
2873 		vnevent_rmdir(vp, dvp, nm, ct);
2874 	}
2875 	VN_RELE(vp);
2876 
2877 	nfs_rw_exit(&drp->r_rwlock);
2878 
2879 	return (error);
2880 }
2881 
2882 /* ARGSUSED */
2883 static int
2884 nfs_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr,
2885 	caller_context_t *ct, int flags)
2886 {
2887 	int error;
2888 	struct nfsslargs args;
2889 	enum nfsstat status;
2890 	int douprintf;
2891 	rnode_t *drp;
2892 
2893 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2894 		return (EPERM);
2895 	setdiropargs(&args.sla_from, lnm, dvp);
2896 	args.sla_sa = &args.sla_sa_buf;
2897 	error = vattr_to_sattr(tva, args.sla_sa);
2898 	if (error) {
2899 		/* req time field(s) overflow - return immediately */
2900 		return (error);
2901 	}
2902 	args.sla_tnm = tnm;
2903 
2904 	drp = VTOR(dvp);
2905 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2906 		return (EINTR);
2907 
2908 	dnlc_remove(dvp, lnm);
2909 
2910 	douprintf = 1;
2911 
2912 	error = rfs2call(VTOMI(dvp), RFS_SYMLINK,
2913 	    xdr_slargs, (caddr_t)&args,
2914 	    xdr_enum, (caddr_t)&status, cr,
2915 	    &douprintf, &status, 0, NULL);
2916 
2917 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2918 
2919 	if (!error) {
2920 		error = geterrno(status);
2921 		if (!error) {
2922 			if (HAVE_RDDIR_CACHE(drp))
2923 				nfs_purge_rddir_cache(dvp);
2924 		} else {
2925 			PURGE_STALE_FH(error, dvp, cr);
2926 		}
2927 	}
2928 
2929 	nfs_rw_exit(&drp->r_rwlock);
2930 
2931 	return (error);
2932 }
2933 
2934 #ifdef DEBUG
2935 static int nfs_readdir_cache_hits = 0;
2936 static int nfs_readdir_cache_shorts = 0;
2937 static int nfs_readdir_cache_waits = 0;
2938 static int nfs_readdir_cache_misses = 0;
2939 static int nfs_readdir_readahead = 0;
2940 #endif
2941 
2942 static int nfs_shrinkreaddir = 0;
2943 
2944 /*
2945  * Read directory entries.
2946  * There are some weird things to look out for here.  The uio_offset
2947  * field is either 0 or it is the offset returned from a previous
2948  * readdir.  It is an opaque value used by the server to find the
2949  * correct directory block to read. The count field is the number
2950  * of blocks to read on the server.  This is advisory only, the server
2951  * may return only one block's worth of entries.  Entries may be compressed
2952  * on the server.
2953  */
2954 /* ARGSUSED */
2955 static int
2956 nfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
2957 	caller_context_t *ct, int flags)
2958 {
2959 	int error;
2960 	size_t count;
2961 	rnode_t *rp;
2962 	rddir_cache *rdc;
2963 	rddir_cache *nrdc;
2964 	rddir_cache *rrdc;
2965 #ifdef DEBUG
2966 	int missed;
2967 #endif
2968 	rddir_cache srdc;
2969 	avl_index_t where;
2970 
2971 	rp = VTOR(vp);
2972 
2973 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
2974 	if (nfs_zone() != VTOMI(vp)->mi_zone)
2975 		return (EIO);
2976 	/*
2977 	 * Make sure that the directory cache is valid.
2978 	 */
2979 	if (HAVE_RDDIR_CACHE(rp)) {
2980 		if (nfs_disable_rddir_cache) {
2981 			/*
2982 			 * Setting nfs_disable_rddir_cache in /etc/system
2983 			 * allows interoperability with servers that do not
2984 			 * properly update the attributes of directories.
2985 			 * Any cached information gets purged before an
2986 			 * access is made to it.
2987 			 */
2988 			nfs_purge_rddir_cache(vp);
2989 		} else {
2990 			error = nfs_validate_caches(vp, cr);
2991 			if (error)
2992 				return (error);
2993 		}
2994 	}
2995 
2996 	/*
2997 	 * UGLINESS: SunOS 3.2 servers apparently cannot always handle an
2998 	 * RFS_READDIR request with rda_count set to more than 0x400. So
2999 	 * we reduce the request size here purely for compatibility.
3000 	 *
3001 	 * In general, this is no longer required.  However, if a server
3002 	 * is discovered which can not handle requests larger than 1024,
3003 	 * nfs_shrinkreaddir can be set to 1 to enable this backwards
3004 	 * compatibility.
3005 	 *
3006 	 * In any case, the request size is limited to NFS_MAXDATA bytes.
3007 	 */
3008 	count = MIN(uiop->uio_iov->iov_len,
3009 	    nfs_shrinkreaddir ? 0x400 : NFS_MAXDATA);
3010 
3011 	nrdc = NULL;
3012 #ifdef DEBUG
3013 	missed = 0;
3014 #endif
3015 top:
3016 	/*
3017 	 * Short circuit last readdir which always returns 0 bytes.
3018 	 * This can be done after the directory has been read through
3019 	 * completely at least once.  This will set r_direof which
3020 	 * can be used to find the value of the last cookie.
3021 	 */
3022 	mutex_enter(&rp->r_statelock);
3023 	if (rp->r_direof != NULL &&
3024 	    uiop->uio_offset == rp->r_direof->nfs_ncookie) {
3025 		mutex_exit(&rp->r_statelock);
3026 #ifdef DEBUG
3027 		nfs_readdir_cache_shorts++;
3028 #endif
3029 		if (eofp)
3030 			*eofp = 1;
3031 		if (nrdc != NULL)
3032 			rddir_cache_rele(nrdc);
3033 		return (0);
3034 	}
3035 	/*
3036 	 * Look for a cache entry.  Cache entries are identified
3037 	 * by the NFS cookie value and the byte count requested.
3038 	 */
3039 	srdc.nfs_cookie = uiop->uio_offset;
3040 	srdc.buflen = count;
3041 	rdc = avl_find(&rp->r_dir, &srdc, &where);
3042 	if (rdc != NULL) {
3043 		rddir_cache_hold(rdc);
3044 		/*
3045 		 * If the cache entry is in the process of being
3046 		 * filled in, wait until this completes.  The
3047 		 * RDDIRWAIT bit is set to indicate that someone
3048 		 * is waiting and then the thread currently
3049 		 * filling the entry is done, it should do a
3050 		 * cv_broadcast to wakeup all of the threads
3051 		 * waiting for it to finish.
3052 		 */
3053 		if (rdc->flags & RDDIR) {
3054 			nfs_rw_exit(&rp->r_rwlock);
3055 			rdc->flags |= RDDIRWAIT;
3056 #ifdef DEBUG
3057 			nfs_readdir_cache_waits++;
3058 #endif
3059 			if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) {
3060 				/*
3061 				 * We got interrupted, probably
3062 				 * the user typed ^C or an alarm
3063 				 * fired.  We free the new entry
3064 				 * if we allocated one.
3065 				 */
3066 				mutex_exit(&rp->r_statelock);
3067 				(void) nfs_rw_enter_sig(&rp->r_rwlock,
3068 				    RW_READER, FALSE);
3069 				rddir_cache_rele(rdc);
3070 				if (nrdc != NULL)
3071 					rddir_cache_rele(nrdc);
3072 				return (EINTR);
3073 			}
3074 			mutex_exit(&rp->r_statelock);
3075 			(void) nfs_rw_enter_sig(&rp->r_rwlock,
3076 			    RW_READER, FALSE);
3077 			rddir_cache_rele(rdc);
3078 			goto top;
3079 		}
3080 		/*
3081 		 * Check to see if a readdir is required to
3082 		 * fill the entry.  If so, mark this entry
3083 		 * as being filled, remove our reference,
3084 		 * and branch to the code to fill the entry.
3085 		 */
3086 		if (rdc->flags & RDDIRREQ) {
3087 			rdc->flags &= ~RDDIRREQ;
3088 			rdc->flags |= RDDIR;
3089 			if (nrdc != NULL)
3090 				rddir_cache_rele(nrdc);
3091 			nrdc = rdc;
3092 			mutex_exit(&rp->r_statelock);
3093 			goto bottom;
3094 		}
3095 #ifdef DEBUG
3096 		if (!missed)
3097 			nfs_readdir_cache_hits++;
3098 #endif
3099 		/*
3100 		 * If an error occurred while attempting
3101 		 * to fill the cache entry, just return it.
3102 		 */
3103 		if (rdc->error) {
3104 			error = rdc->error;
3105 			mutex_exit(&rp->r_statelock);
3106 			rddir_cache_rele(rdc);
3107 			if (nrdc != NULL)
3108 				rddir_cache_rele(nrdc);
3109 			return (error);
3110 		}
3111 
3112 		/*
3113 		 * The cache entry is complete and good,
3114 		 * copyout the dirent structs to the calling
3115 		 * thread.
3116 		 */
3117 		error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop);
3118 
3119 		/*
3120 		 * If no error occurred during the copyout,
3121 		 * update the offset in the uio struct to
3122 		 * contain the value of the next cookie
3123 		 * and set the eof value appropriately.
3124 		 */
3125 		if (!error) {
3126 			uiop->uio_offset = rdc->nfs_ncookie;
3127 			if (eofp)
3128 				*eofp = rdc->eof;
3129 		}
3130 
3131 		/*
3132 		 * Decide whether to do readahead.  Don't if
3133 		 * have already read to the end of directory.
3134 		 */
3135 		if (rdc->eof) {
3136 			rp->r_direof = rdc;
3137 			mutex_exit(&rp->r_statelock);
3138 			rddir_cache_rele(rdc);
3139 			if (nrdc != NULL)
3140 				rddir_cache_rele(nrdc);
3141 			return (error);
3142 		}
3143 
3144 		/*
3145 		 * Check to see whether we found an entry
3146 		 * for the readahead.  If so, we don't need
3147 		 * to do anything further, so free the new
3148 		 * entry if one was allocated.  Otherwise,
3149 		 * allocate a new entry, add it to the cache,
3150 		 * and then initiate an asynchronous readdir
3151 		 * operation to fill it.
3152 		 */
3153 		srdc.nfs_cookie = rdc->nfs_ncookie;
3154 		srdc.buflen = count;
3155 		rrdc = avl_find(&rp->r_dir, &srdc, &where);
3156 		if (rrdc != NULL) {
3157 			if (nrdc != NULL)
3158 				rddir_cache_rele(nrdc);
3159 		} else {
3160 			if (nrdc != NULL)
3161 				rrdc = nrdc;
3162 			else {
3163 				rrdc = rddir_cache_alloc(KM_NOSLEEP);
3164 			}
3165 			if (rrdc != NULL) {
3166 				rrdc->nfs_cookie = rdc->nfs_ncookie;
3167 				rrdc->buflen = count;
3168 				avl_insert(&rp->r_dir, rrdc, where);
3169 				rddir_cache_hold(rrdc);
3170 				mutex_exit(&rp->r_statelock);
3171 				rddir_cache_rele(rdc);
3172 #ifdef DEBUG
3173 				nfs_readdir_readahead++;
3174 #endif
3175 				nfs_async_readdir(vp, rrdc, cr, nfsreaddir);
3176 				return (error);
3177 			}
3178 		}
3179 
3180 		mutex_exit(&rp->r_statelock);
3181 		rddir_cache_rele(rdc);
3182 		return (error);
3183 	}
3184 
3185 	/*
3186 	 * Didn't find an entry in the cache.  Construct a new empty
3187 	 * entry and link it into the cache.  Other processes attempting
3188 	 * to access this entry will need to wait until it is filled in.
3189 	 *
3190 	 * Since kmem_alloc may block, another pass through the cache
3191 	 * will need to be taken to make sure that another process
3192 	 * hasn't already added an entry to the cache for this request.
3193 	 */
3194 	if (nrdc == NULL) {
3195 		mutex_exit(&rp->r_statelock);
3196 		nrdc = rddir_cache_alloc(KM_SLEEP);
3197 		nrdc->nfs_cookie = uiop->uio_offset;
3198 		nrdc->buflen = count;
3199 		goto top;
3200 	}
3201 
3202 	/*
3203 	 * Add this entry to the cache.
3204 	 */
3205 	avl_insert(&rp->r_dir, nrdc, where);
3206 	rddir_cache_hold(nrdc);
3207 	mutex_exit(&rp->r_statelock);
3208 
3209 bottom:
3210 #ifdef DEBUG
3211 	missed = 1;
3212 	nfs_readdir_cache_misses++;
3213 #endif
3214 	/*
3215 	 * Do the readdir.
3216 	 */
3217 	error = nfsreaddir(vp, nrdc, cr);
3218 
3219 	/*
3220 	 * If this operation failed, just return the error which occurred.
3221 	 */
3222 	if (error != 0)
3223 		return (error);
3224 
3225 	/*
3226 	 * Since the RPC operation will have taken sometime and blocked
3227 	 * this process, another pass through the cache will need to be
3228 	 * taken to find the correct cache entry.  It is possible that
3229 	 * the correct cache entry will not be there (although one was
3230 	 * added) because the directory changed during the RPC operation
3231 	 * and the readdir cache was flushed.  In this case, just start
3232 	 * over.  It is hoped that this will not happen too often... :-)
3233 	 */
3234 	nrdc = NULL;
3235 	goto top;
3236 	/* NOTREACHED */
3237 }
3238 
3239 static int
3240 nfsreaddir(vnode_t *vp, rddir_cache *rdc, cred_t *cr)
3241 {
3242 	int error;
3243 	struct nfsrddirargs rda;
3244 	struct nfsrddirres rd;
3245 	rnode_t *rp;
3246 	mntinfo_t *mi;
3247 	uint_t count;
3248 	int douprintf;
3249 	failinfo_t fi, *fip;
3250 
3251 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3252 	count = rdc->buflen;
3253 
3254 	rp = VTOR(vp);
3255 	mi = VTOMI(vp);
3256 
3257 	rda.rda_fh = *VTOFH(vp);
3258 	rda.rda_offset = rdc->nfs_cookie;
3259 
3260 	/*
3261 	 * NFS client failover support
3262 	 * suppress failover unless we have a zero cookie
3263 	 */
3264 	if (rdc->nfs_cookie == (off_t)0) {
3265 		fi.vp = vp;
3266 		fi.fhp = (caddr_t)&rda.rda_fh;
3267 		fi.copyproc = nfscopyfh;
3268 		fi.lookupproc = nfslookup;
3269 		fi.xattrdirproc = acl_getxattrdir2;
3270 		fip = &fi;
3271 	} else {
3272 		fip = NULL;
3273 	}
3274 
3275 	rd.rd_entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3276 	rd.rd_size = count;
3277 	rd.rd_offset = rda.rda_offset;
3278 
3279 	douprintf = 1;
3280 
3281 	if (mi->mi_io_kstats) {
3282 		mutex_enter(&mi->mi_lock);
3283 		kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
3284 		mutex_exit(&mi->mi_lock);
3285 	}
3286 
3287 	do {
3288 		rda.rda_count = MIN(count, mi->mi_curread);
3289 		error = rfs2call(mi, RFS_READDIR,
3290 		    xdr_rddirargs, (caddr_t)&rda,
3291 		    xdr_getrddirres, (caddr_t)&rd, cr,
3292 		    &douprintf, &rd.rd_status, 0, fip);
3293 	} while (error == ENFS_TRYAGAIN);
3294 
3295 	if (mi->mi_io_kstats) {
3296 		mutex_enter(&mi->mi_lock);
3297 		kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
3298 		mutex_exit(&mi->mi_lock);
3299 	}
3300 
3301 	/*
3302 	 * Since we are actually doing a READDIR RPC, we must have
3303 	 * exclusive access to the cache entry being filled.  Thus,
3304 	 * it is safe to update all fields except for the flags
3305 	 * field.  The r_statelock in the rnode must be held to
3306 	 * prevent two different threads from simultaneously
3307 	 * attempting to update the flags field.  This can happen
3308 	 * if we are turning off RDDIR and the other thread is
3309 	 * trying to set RDDIRWAIT.
3310 	 */
3311 	ASSERT(rdc->flags & RDDIR);
3312 	if (!error) {
3313 		error = geterrno(rd.rd_status);
3314 		if (!error) {
3315 			rdc->nfs_ncookie = rd.rd_offset;
3316 			rdc->eof = rd.rd_eof ? 1 : 0;
3317 			rdc->entlen = rd.rd_size;
3318 			ASSERT(rdc->entlen <= rdc->buflen);
3319 #ifdef DEBUG
3320 			rdc->entries = rddir_cache_buf_alloc(rdc->buflen,
3321 			    KM_SLEEP);
3322 #else
3323 			rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3324 #endif
3325 			bcopy(rd.rd_entries, rdc->entries, rdc->entlen);
3326 			rdc->error = 0;
3327 			if (mi->mi_io_kstats) {
3328 				mutex_enter(&mi->mi_lock);
3329 				KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
3330 				KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
3331 				    rd.rd_size;
3332 				mutex_exit(&mi->mi_lock);
3333 			}
3334 		} else {
3335 			PURGE_STALE_FH(error, vp, cr);
3336 		}
3337 	}
3338 	if (error) {
3339 		rdc->entries = NULL;
3340 		rdc->error = error;
3341 	}
3342 	kmem_free(rd.rd_entries, rdc->buflen);
3343 
3344 	mutex_enter(&rp->r_statelock);
3345 	rdc->flags &= ~RDDIR;
3346 	if (rdc->flags & RDDIRWAIT) {
3347 		rdc->flags &= ~RDDIRWAIT;
3348 		cv_broadcast(&rdc->cv);
3349 	}
3350 	if (error)
3351 		rdc->flags |= RDDIRREQ;
3352 	mutex_exit(&rp->r_statelock);
3353 
3354 	rddir_cache_rele(rdc);
3355 
3356 	return (error);
3357 }
3358 
3359 #ifdef DEBUG
3360 static int nfs_bio_do_stop = 0;
3361 #endif
3362 
3363 static int
3364 nfs_bio(struct buf *bp, cred_t *cr)
3365 {
3366 	rnode_t *rp = VTOR(bp->b_vp);
3367 	int count;
3368 	int error;
3369 	cred_t *cred;
3370 	uint_t offset;
3371 
3372 	DTRACE_IO1(start, struct buf *, bp);
3373 
3374 	ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone);
3375 	offset = dbtob(bp->b_blkno);
3376 
3377 	if (bp->b_flags & B_READ) {
3378 		mutex_enter(&rp->r_statelock);
3379 		if (rp->r_cred != NULL) {
3380 			cred = rp->r_cred;
3381 			crhold(cred);
3382 		} else {
3383 			rp->r_cred = cr;
3384 			crhold(cr);
3385 			cred = cr;
3386 			crhold(cred);
3387 		}
3388 		mutex_exit(&rp->r_statelock);
3389 	read_again:
3390 		error = bp->b_error = nfsread(bp->b_vp, bp->b_un.b_addr,
3391 		    offset, bp->b_bcount, &bp->b_resid, cred);
3392 
3393 		crfree(cred);
3394 		if (!error) {
3395 			if (bp->b_resid) {
3396 				/*
3397 				 * Didn't get it all because we hit EOF,
3398 				 * zero all the memory beyond the EOF.
3399 				 */
3400 				/* bzero(rdaddr + */
3401 				bzero(bp->b_un.b_addr +
3402 				    bp->b_bcount - bp->b_resid, bp->b_resid);
3403 			}
3404 			mutex_enter(&rp->r_statelock);
3405 			if (bp->b_resid == bp->b_bcount &&
3406 			    offset >= rp->r_size) {
3407 				/*
3408 				 * We didn't read anything at all as we are
3409 				 * past EOF.  Return an error indicator back
3410 				 * but don't destroy the pages (yet).
3411 				 */
3412 				error = NFS_EOF;
3413 			}
3414 			mutex_exit(&rp->r_statelock);
3415 		} else if (error == EACCES) {
3416 			mutex_enter(&rp->r_statelock);
3417 			if (cred != cr) {
3418 				if (rp->r_cred != NULL)
3419 					crfree(rp->r_cred);
3420 				rp->r_cred = cr;
3421 				crhold(cr);
3422 				cred = cr;
3423 				crhold(cred);
3424 				mutex_exit(&rp->r_statelock);
3425 				goto read_again;
3426 			}
3427 			mutex_exit(&rp->r_statelock);
3428 		}
3429 	} else {
3430 		if (!(rp->r_flags & RSTALE)) {
3431 			mutex_enter(&rp->r_statelock);
3432 			if (rp->r_cred != NULL) {
3433 				cred = rp->r_cred;
3434 				crhold(cred);
3435 			} else {
3436 				rp->r_cred = cr;
3437 				crhold(cr);
3438 				cred = cr;
3439 				crhold(cred);
3440 			}
3441 			mutex_exit(&rp->r_statelock);
3442 		write_again:
3443 			mutex_enter(&rp->r_statelock);
3444 			count = MIN(bp->b_bcount, rp->r_size - offset);
3445 			mutex_exit(&rp->r_statelock);
3446 			if (count < 0)
3447 				cmn_err(CE_PANIC, "nfs_bio: write count < 0");
3448 #ifdef DEBUG
3449 			if (count == 0) {
3450 				zcmn_err(getzoneid(), CE_WARN,
3451 				    "nfs_bio: zero length write at %d",
3452 				    offset);
3453 				nfs_printfhandle(&rp->r_fh);
3454 				if (nfs_bio_do_stop)
3455 					debug_enter("nfs_bio");
3456 			}
3457 #endif
3458 			error = nfswrite(bp->b_vp, bp->b_un.b_addr, offset,
3459 			    count, cred);
3460 			if (error == EACCES) {
3461 				mutex_enter(&rp->r_statelock);
3462 				if (cred != cr) {
3463 					if (rp->r_cred != NULL)
3464 						crfree(rp->r_cred);
3465 					rp->r_cred = cr;
3466 					crhold(cr);
3467 					crfree(cred);
3468 					cred = cr;
3469 					crhold(cred);
3470 					mutex_exit(&rp->r_statelock);
3471 					goto write_again;
3472 				}
3473 				mutex_exit(&rp->r_statelock);
3474 			}
3475 			bp->b_error = error;
3476 			if (error && error != EINTR) {
3477 				/*
3478 				 * Don't print EDQUOT errors on the console.
3479 				 * Don't print asynchronous EACCES errors.
3480 				 * Don't print EFBIG errors.
3481 				 * Print all other write errors.
3482 				 */
3483 				if (error != EDQUOT && error != EFBIG &&
3484 				    (error != EACCES ||
3485 				    !(bp->b_flags & B_ASYNC)))
3486 					nfs_write_error(bp->b_vp, error, cred);
3487 				/*
3488 				 * Update r_error and r_flags as appropriate.
3489 				 * If the error was ESTALE, then mark the
3490 				 * rnode as not being writeable and save
3491 				 * the error status.  Otherwise, save any
3492 				 * errors which occur from asynchronous
3493 				 * page invalidations.  Any errors occurring
3494 				 * from other operations should be saved
3495 				 * by the caller.
3496 				 */
3497 				mutex_enter(&rp->r_statelock);
3498 				if (error == ESTALE) {
3499 					rp->r_flags |= RSTALE;
3500 					if (!rp->r_error)
3501 						rp->r_error = error;
3502 				} else if (!rp->r_error &&
3503 				    (bp->b_flags &
3504 				    (B_INVAL|B_FORCE|B_ASYNC)) ==
3505 				    (B_INVAL|B_FORCE|B_ASYNC)) {
3506 					rp->r_error = error;
3507 				}
3508 				mutex_exit(&rp->r_statelock);
3509 			}
3510 			crfree(cred);
3511 		} else {
3512 			error = rp->r_error;
3513 			/*
3514 			 * A close may have cleared r_error, if so,
3515 			 * propagate ESTALE error return properly
3516 			 */
3517 			if (error == 0)
3518 				error = ESTALE;
3519 		}
3520 	}
3521 
3522 	if (error != 0 && error != NFS_EOF)
3523 		bp->b_flags |= B_ERROR;
3524 
3525 	DTRACE_IO1(done, struct buf *, bp);
3526 
3527 	return (error);
3528 }
3529 
3530 /* ARGSUSED */
3531 static int
3532 nfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3533 {
3534 	struct nfs_fid *fp;
3535 	rnode_t *rp;
3536 
3537 	rp = VTOR(vp);
3538 
3539 	if (fidp->fid_len < (sizeof (struct nfs_fid) - sizeof (short))) {
3540 		fidp->fid_len = sizeof (struct nfs_fid) - sizeof (short);
3541 		return (ENOSPC);
3542 	}
3543 	fp = (struct nfs_fid *)fidp;
3544 	fp->nf_pad = 0;
3545 	fp->nf_len = sizeof (struct nfs_fid) - sizeof (short);
3546 	bcopy(rp->r_fh.fh_buf, fp->nf_data, NFS_FHSIZE);
3547 	return (0);
3548 }
3549 
3550 /* ARGSUSED2 */
3551 static int
3552 nfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3553 {
3554 	rnode_t *rp = VTOR(vp);
3555 
3556 	if (!write_lock) {
3557 		(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3558 		return (V_WRITELOCK_FALSE);
3559 	}
3560 
3561 	if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) {
3562 		(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3563 		if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp))
3564 			return (V_WRITELOCK_FALSE);
3565 		nfs_rw_exit(&rp->r_rwlock);
3566 	}
3567 
3568 	(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE);
3569 	return (V_WRITELOCK_TRUE);
3570 }
3571 
3572 /* ARGSUSED */
3573 static void
3574 nfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3575 {
3576 	rnode_t *rp = VTOR(vp);
3577 
3578 	nfs_rw_exit(&rp->r_rwlock);
3579 }
3580 
3581 /* ARGSUSED */
3582 static int
3583 nfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
3584 {
3585 
3586 	/*
3587 	 * Because we stuff the readdir cookie into the offset field
3588 	 * someone may attempt to do an lseek with the cookie which
3589 	 * we want to succeed.
3590 	 */
3591 	if (vp->v_type == VDIR)
3592 		return (0);
3593 	if (*noffp < 0 || *noffp > MAXOFF32_T)
3594 		return (EINVAL);
3595 	return (0);
3596 }
3597 
3598 /*
3599  * number of NFS_MAXDATA blocks to read ahead
3600  * optimized for 100 base-T.
3601  */
3602 static int nfs_nra = 4;
3603 
3604 #ifdef DEBUG
3605 static int nfs_lostpage = 0;	/* number of times we lost original page */
3606 #endif
3607 
3608 /*
3609  * Return all the pages from [off..off+len) in file
3610  */
3611 /* ARGSUSED */
3612 static int
3613 nfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3614 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3615 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3616 {
3617 	rnode_t *rp;
3618 	int error;
3619 	mntinfo_t *mi;
3620 
3621 	if (vp->v_flag & VNOMAP)
3622 		return (ENOSYS);
3623 
3624 	ASSERT(off <= MAXOFF32_T);
3625 	if (nfs_zone() != VTOMI(vp)->mi_zone)
3626 		return (EIO);
3627 	if (protp != NULL)
3628 		*protp = PROT_ALL;
3629 
3630 	/*
3631 	 * Now valididate that the caches are up to date.
3632 	 */
3633 	error = nfs_validate_caches(vp, cr);
3634 	if (error)
3635 		return (error);
3636 
3637 	rp = VTOR(vp);
3638 	mi = VTOMI(vp);
3639 retry:
3640 	mutex_enter(&rp->r_statelock);
3641 
3642 	/*
3643 	 * Don't create dirty pages faster than they
3644 	 * can be cleaned so that the system doesn't
3645 	 * get imbalanced.  If the async queue is
3646 	 * maxed out, then wait for it to drain before
3647 	 * creating more dirty pages.  Also, wait for
3648 	 * any threads doing pagewalks in the vop_getattr
3649 	 * entry points so that they don't block for
3650 	 * long periods.
3651 	 */
3652 	if (rw == S_CREATE) {
3653 		while ((mi->mi_max_threads != 0 &&
3654 		    rp->r_awcount > 2 * mi->mi_max_threads) ||
3655 		    rp->r_gcount > 0)
3656 			cv_wait(&rp->r_cv, &rp->r_statelock);
3657 	}
3658 
3659 	/*
3660 	 * If we are getting called as a side effect of an nfs_write()
3661 	 * operation the local file size might not be extended yet.
3662 	 * In this case we want to be able to return pages of zeroes.
3663 	 */
3664 	if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) {
3665 		mutex_exit(&rp->r_statelock);
3666 		return (EFAULT);		/* beyond EOF */
3667 	}
3668 
3669 	mutex_exit(&rp->r_statelock);
3670 
3671 	if (len <= PAGESIZE) {
3672 		error = nfs_getapage(vp, off, len, protp, pl, plsz,
3673 		    seg, addr, rw, cr);
3674 	} else {
3675 		error = pvn_getpages(nfs_getapage, vp, off, len, protp,
3676 		    pl, plsz, seg, addr, rw, cr);
3677 	}
3678 
3679 	switch (error) {
3680 	case NFS_EOF:
3681 		nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr);
3682 		goto retry;
3683 	case ESTALE:
3684 		PURGE_STALE_FH(error, vp, cr);
3685 	}
3686 
3687 	return (error);
3688 }
3689 
3690 /*
3691  * Called from pvn_getpages or nfs_getpage to get a particular page.
3692  */
3693 /* ARGSUSED */
3694 static int
3695 nfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
3696 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3697 	enum seg_rw rw, cred_t *cr)
3698 {
3699 	rnode_t *rp;
3700 	uint_t bsize;
3701 	struct buf *bp;
3702 	page_t *pp;
3703 	u_offset_t lbn;
3704 	u_offset_t io_off;
3705 	u_offset_t blkoff;
3706 	u_offset_t rablkoff;
3707 	size_t io_len;
3708 	uint_t blksize;
3709 	int error;
3710 	int readahead;
3711 	int readahead_issued = 0;
3712 	int ra_window; /* readahead window */
3713 	page_t *pagefound;
3714 
3715 	if (nfs_zone() != VTOMI(vp)->mi_zone)
3716 		return (EIO);
3717 	rp = VTOR(vp);
3718 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3719 
3720 reread:
3721 	bp = NULL;
3722 	pp = NULL;
3723 	pagefound = NULL;
3724 
3725 	if (pl != NULL)
3726 		pl[0] = NULL;
3727 
3728 	error = 0;
3729 	lbn = off / bsize;
3730 	blkoff = lbn * bsize;
3731 
3732 	/*
3733 	 * Queueing up the readahead before doing the synchronous read
3734 	 * results in a significant increase in read throughput because
3735 	 * of the increased parallelism between the async threads and
3736 	 * the process context.
3737 	 */
3738 	if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 &&
3739 	    rw != S_CREATE &&
3740 	    !(vp->v_flag & VNOCACHE)) {
3741 		mutex_enter(&rp->r_statelock);
3742 
3743 		/*
3744 		 * Calculate the number of readaheads to do.
3745 		 * a) No readaheads at offset = 0.
3746 		 * b) Do maximum(nfs_nra) readaheads when the readahead
3747 		 *    window is closed.
3748 		 * c) Do readaheads between 1 to (nfs_nra - 1) depending
3749 		 *    upon how far the readahead window is open or close.
3750 		 * d) No readaheads if rp->r_nextr is not within the scope
3751 		 *    of the readahead window (random i/o).
3752 		 */
3753 
3754 		if (off == 0)
3755 			readahead = 0;
3756 		else if (blkoff == rp->r_nextr)
3757 			readahead = nfs_nra;
3758 		else if (rp->r_nextr > blkoff &&
3759 		    ((ra_window = (rp->r_nextr - blkoff) / bsize)
3760 		    <= (nfs_nra - 1)))
3761 			readahead = nfs_nra - ra_window;
3762 		else
3763 			readahead = 0;
3764 
3765 		rablkoff = rp->r_nextr;
3766 		while (readahead > 0 && rablkoff + bsize < rp->r_size) {
3767 			mutex_exit(&rp->r_statelock);
3768 			if (nfs_async_readahead(vp, rablkoff + bsize,
3769 			    addr + (rablkoff + bsize - off), seg, cr,
3770 			    nfs_readahead) < 0) {
3771 				mutex_enter(&rp->r_statelock);
3772 				break;
3773 			}
3774 			readahead--;
3775 			rablkoff += bsize;
3776 			/*
3777 			 * Indicate that we did a readahead so
3778 			 * readahead offset is not updated
3779 			 * by the synchronous read below.
3780 			 */
3781 			readahead_issued = 1;
3782 			mutex_enter(&rp->r_statelock);
3783 			/*
3784 			 * set readahead offset to
3785 			 * offset of last async readahead
3786 			 * request.
3787 			 */
3788 			rp->r_nextr = rablkoff;
3789 		}
3790 		mutex_exit(&rp->r_statelock);
3791 	}
3792 
3793 again:
3794 	if ((pagefound = page_exists(vp, off)) == NULL) {
3795 		if (pl == NULL) {
3796 			(void) nfs_async_readahead(vp, blkoff, addr, seg, cr,
3797 			    nfs_readahead);
3798 		} else if (rw == S_CREATE) {
3799 			/*
3800 			 * Block for this page is not allocated, or the offset
3801 			 * is beyond the current allocation size, or we're
3802 			 * allocating a swap slot and the page was not found,
3803 			 * so allocate it and return a zero page.
3804 			 */
3805 			if ((pp = page_create_va(vp, off,
3806 			    PAGESIZE, PG_WAIT, seg, addr)) == NULL)
3807 				cmn_err(CE_PANIC, "nfs_getapage: page_create");
3808 			io_len = PAGESIZE;
3809 			mutex_enter(&rp->r_statelock);
3810 			rp->r_nextr = off + PAGESIZE;
3811 			mutex_exit(&rp->r_statelock);
3812 		} else {
3813 			/*
3814 			 * Need to go to server to get a BLOCK, exception to
3815 			 * that being while reading at offset = 0 or doing
3816 			 * random i/o, in that case read only a PAGE.
3817 			 */
3818 			mutex_enter(&rp->r_statelock);
3819 			if (blkoff < rp->r_size &&
3820 			    blkoff + bsize >= rp->r_size) {
3821 				/*
3822 				 * If only a block or less is left in
3823 				 * the file, read all that is remaining.
3824 				 */
3825 				if (rp->r_size <= off) {
3826 					/*
3827 					 * Trying to access beyond EOF,
3828 					 * set up to get at least one page.
3829 					 */
3830 					blksize = off + PAGESIZE - blkoff;
3831 				} else
3832 					blksize = rp->r_size - blkoff;
3833 			} else if ((off == 0) ||
3834 			    (off != rp->r_nextr && !readahead_issued)) {
3835 				blksize = PAGESIZE;
3836 				blkoff = off; /* block = page here */
3837 			} else
3838 				blksize = bsize;
3839 			mutex_exit(&rp->r_statelock);
3840 
3841 			pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3842 			    &io_len, blkoff, blksize, 0);
3843 
3844 			/*
3845 			 * Some other thread has entered the page,
3846 			 * so just use it.
3847 			 */
3848 			if (pp == NULL)
3849 				goto again;
3850 
3851 			/*
3852 			 * Now round the request size up to page boundaries.
3853 			 * This ensures that the entire page will be
3854 			 * initialized to zeroes if EOF is encountered.
3855 			 */
3856 			io_len = ptob(btopr(io_len));
3857 
3858 			bp = pageio_setup(pp, io_len, vp, B_READ);
3859 			ASSERT(bp != NULL);
3860 
3861 			/*
3862 			 * pageio_setup should have set b_addr to 0.  This
3863 			 * is correct since we want to do I/O on a page
3864 			 * boundary.  bp_mapin will use this addr to calculate
3865 			 * an offset, and then set b_addr to the kernel virtual
3866 			 * address it allocated for us.
3867 			 */
3868 			ASSERT(bp->b_un.b_addr == 0);
3869 
3870 			bp->b_edev = 0;
3871 			bp->b_dev = 0;
3872 			bp->b_lblkno = lbtodb(io_off);
3873 			bp->b_file = vp;
3874 			bp->b_offset = (offset_t)off;
3875 			bp_mapin(bp);
3876 
3877 			/*
3878 			 * If doing a write beyond what we believe is EOF,
3879 			 * don't bother trying to read the pages from the
3880 			 * server, we'll just zero the pages here.  We
3881 			 * don't check that the rw flag is S_WRITE here
3882 			 * because some implementations may attempt a
3883 			 * read access to the buffer before copying data.
3884 			 */
3885 			mutex_enter(&rp->r_statelock);
3886 			if (io_off >= rp->r_size && seg == segkmap) {
3887 				mutex_exit(&rp->r_statelock);
3888 				bzero(bp->b_un.b_addr, io_len);
3889 			} else {
3890 				mutex_exit(&rp->r_statelock);
3891 				error = nfs_bio(bp, cr);
3892 			}
3893 
3894 			/*
3895 			 * Unmap the buffer before freeing it.
3896 			 */
3897 			bp_mapout(bp);
3898 			pageio_done(bp);
3899 
3900 			if (error == NFS_EOF) {
3901 				/*
3902 				 * If doing a write system call just return
3903 				 * zeroed pages, else user tried to get pages
3904 				 * beyond EOF, return error.  We don't check
3905 				 * that the rw flag is S_WRITE here because
3906 				 * some implementations may attempt a read
3907 				 * access to the buffer before copying data.
3908 				 */
3909 				if (seg == segkmap)
3910 					error = 0;
3911 				else
3912 					error = EFAULT;
3913 			}
3914 
3915 			if (!readahead_issued && !error) {
3916 				mutex_enter(&rp->r_statelock);
3917 				rp->r_nextr = io_off + io_len;
3918 				mutex_exit(&rp->r_statelock);
3919 			}
3920 		}
3921 	}
3922 
3923 out:
3924 	if (pl == NULL)
3925 		return (error);
3926 
3927 	if (error) {
3928 		if (pp != NULL)
3929 			pvn_read_done(pp, B_ERROR);
3930 		return (error);
3931 	}
3932 
3933 	if (pagefound) {
3934 		se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED);
3935 
3936 		/*
3937 		 * Page exists in the cache, acquire the appropriate lock.
3938 		 * If this fails, start all over again.
3939 		 */
3940 		if ((pp = page_lookup(vp, off, se)) == NULL) {
3941 #ifdef DEBUG
3942 			nfs_lostpage++;
3943 #endif
3944 			goto reread;
3945 		}
3946 		pl[0] = pp;
3947 		pl[1] = NULL;
3948 		return (0);
3949 	}
3950 
3951 	if (pp != NULL)
3952 		pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3953 
3954 	return (error);
3955 }
3956 
3957 static void
3958 nfs_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg,
3959 	cred_t *cr)
3960 {
3961 	int error;
3962 	page_t *pp;
3963 	u_offset_t io_off;
3964 	size_t io_len;
3965 	struct buf *bp;
3966 	uint_t bsize, blksize;
3967 	rnode_t *rp = VTOR(vp);
3968 
3969 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3970 
3971 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3972 
3973 	mutex_enter(&rp->r_statelock);
3974 	if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) {
3975 		/*
3976 		 * If less than a block left in file read less
3977 		 * than a block.
3978 		 */
3979 		blksize = rp->r_size - blkoff;
3980 	} else
3981 		blksize = bsize;
3982 	mutex_exit(&rp->r_statelock);
3983 
3984 	pp = pvn_read_kluster(vp, blkoff, segkmap, addr,
3985 	    &io_off, &io_len, blkoff, blksize, 1);
3986 	/*
3987 	 * The isra flag passed to the kluster function is 1, we may have
3988 	 * gotten a return value of NULL for a variety of reasons (# of free
3989 	 * pages < minfree, someone entered the page on the vnode etc). In all
3990 	 * cases, we want to punt on the readahead.
3991 	 */
3992 	if (pp == NULL)
3993 		return;
3994 
3995 	/*
3996 	 * Now round the request size up to page boundaries.
3997 	 * This ensures that the entire page will be
3998 	 * initialized to zeroes if EOF is encountered.
3999 	 */
4000 	io_len = ptob(btopr(io_len));
4001 
4002 	bp = pageio_setup(pp, io_len, vp, B_READ);
4003 	ASSERT(bp != NULL);
4004 
4005 	/*
4006 	 * pageio_setup should have set b_addr to 0.  This is correct since
4007 	 * we want to do I/O on a page boundary. bp_mapin() will use this addr
4008 	 * to calculate an offset, and then set b_addr to the kernel virtual
4009 	 * address it allocated for us.
4010 	 */
4011 	ASSERT(bp->b_un.b_addr == 0);
4012 
4013 	bp->b_edev = 0;
4014 	bp->b_dev = 0;
4015 	bp->b_lblkno = lbtodb(io_off);
4016 	bp->b_file = vp;
4017 	bp->b_offset = (offset_t)blkoff;
4018 	bp_mapin(bp);
4019 
4020 	/*
4021 	 * If doing a write beyond what we believe is EOF, don't bother trying
4022 	 * to read the pages from the server, we'll just zero the pages here.
4023 	 * We don't check that the rw flag is S_WRITE here because some
4024 	 * implementations may attempt a read access to the buffer before
4025 	 * copying data.
4026 	 */
4027 	mutex_enter(&rp->r_statelock);
4028 	if (io_off >= rp->r_size && seg == segkmap) {
4029 		mutex_exit(&rp->r_statelock);
4030 		bzero(bp->b_un.b_addr, io_len);
4031 		error = 0;
4032 	} else {
4033 		mutex_exit(&rp->r_statelock);
4034 		error = nfs_bio(bp, cr);
4035 		if (error == NFS_EOF)
4036 			error = 0;
4037 	}
4038 
4039 	/*
4040 	 * Unmap the buffer before freeing it.
4041 	 */
4042 	bp_mapout(bp);
4043 	pageio_done(bp);
4044 
4045 	pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ);
4046 
4047 	/*
4048 	 * In case of error set readahead offset
4049 	 * to the lowest offset.
4050 	 * pvn_read_done() calls VN_DISPOSE to destroy the pages
4051 	 */
4052 	if (error && rp->r_nextr > io_off) {
4053 		mutex_enter(&rp->r_statelock);
4054 		if (rp->r_nextr > io_off)
4055 			rp->r_nextr = io_off;
4056 		mutex_exit(&rp->r_statelock);
4057 	}
4058 }
4059 
4060 /*
4061  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
4062  * If len == 0, do from off to EOF.
4063  *
4064  * The normal cases should be len == 0 && off == 0 (entire vp list),
4065  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
4066  * (from pageout).
4067  */
4068 /* ARGSUSED */
4069 static int
4070 nfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4071 	caller_context_t *ct)
4072 {
4073 	int error;
4074 	rnode_t *rp;
4075 
4076 	ASSERT(cr != NULL);
4077 
4078 	/*
4079 	 * XXX - Why should this check be made here?
4080 	 */
4081 	if (vp->v_flag & VNOMAP)
4082 		return (ENOSYS);
4083 
4084 	if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp))
4085 		return (0);
4086 
4087 	if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone)
4088 		return (EIO);
4089 	ASSERT(off <= MAXOFF32_T);
4090 
4091 	rp = VTOR(vp);
4092 	mutex_enter(&rp->r_statelock);
4093 	rp->r_count++;
4094 	mutex_exit(&rp->r_statelock);
4095 	error = nfs_putpages(vp, off, len, flags, cr);
4096 	mutex_enter(&rp->r_statelock);
4097 	rp->r_count--;
4098 	cv_broadcast(&rp->r_cv);
4099 	mutex_exit(&rp->r_statelock);
4100 
4101 	return (error);
4102 }
4103 
4104 /*
4105  * Write out a single page, possibly klustering adjacent dirty pages.
4106  */
4107 int
4108 nfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
4109 	int flags, cred_t *cr)
4110 {
4111 	u_offset_t io_off;
4112 	u_offset_t lbn_off;
4113 	u_offset_t lbn;
4114 	size_t io_len;
4115 	uint_t bsize;
4116 	int error;
4117 	rnode_t *rp;
4118 
4119 	ASSERT(!vn_is_readonly(vp));
4120 	ASSERT(pp != NULL);
4121 	ASSERT(cr != NULL);
4122 	ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone);
4123 
4124 	rp = VTOR(vp);
4125 	ASSERT(rp->r_count > 0);
4126 
4127 	ASSERT(pp->p_offset <= MAXOFF32_T);
4128 
4129 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
4130 	lbn = pp->p_offset / bsize;
4131 	lbn_off = lbn * bsize;
4132 
4133 	/*
4134 	 * Find a kluster that fits in one block, or in
4135 	 * one page if pages are bigger than blocks.  If
4136 	 * there is less file space allocated than a whole
4137 	 * page, we'll shorten the i/o request below.
4138 	 */
4139 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off,
4140 	    roundup(bsize, PAGESIZE), flags);
4141 
4142 	/*
4143 	 * pvn_write_kluster shouldn't have returned a page with offset
4144 	 * behind the original page we were given.  Verify that.
4145 	 */
4146 	ASSERT((pp->p_offset / bsize) >= lbn);
4147 
4148 	/*
4149 	 * Now pp will have the list of kept dirty pages marked for
4150 	 * write back.  It will also handle invalidation and freeing
4151 	 * of pages that are not dirty.  Check for page length rounding
4152 	 * problems.
4153 	 */
4154 	if (io_off + io_len > lbn_off + bsize) {
4155 		ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE);
4156 		io_len = lbn_off + bsize - io_off;
4157 	}
4158 	/*
4159 	 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4160 	 * consistent value of r_size. RMODINPROGRESS is set in writerp().
4161 	 * When RMODINPROGRESS is set it indicates that a uiomove() is in
4162 	 * progress and the r_size has not been made consistent with the
4163 	 * new size of the file. When the uiomove() completes the r_size is
4164 	 * updated and the RMODINPROGRESS flag is cleared.
4165 	 *
4166 	 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4167 	 * consistent value of r_size. Without this handshaking, it is
4168 	 * possible that nfs(3)_bio() picks  up the old value of r_size
4169 	 * before the uiomove() in writerp() completes. This will result
4170 	 * in the write through nfs(3)_bio() being dropped.
4171 	 *
4172 	 * More precisely, there is a window between the time the uiomove()
4173 	 * completes and the time the r_size is updated. If a VOP_PUTPAGE()
4174 	 * operation intervenes in this window, the page will be picked up,
4175 	 * because it is dirty (it will be unlocked, unless it was
4176 	 * pagecreate'd). When the page is picked up as dirty, the dirty
4177 	 * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is
4178 	 * checked. This will still be the old size. Therefore the page will
4179 	 * not be written out. When segmap_release() calls VOP_PUTPAGE(),
4180 	 * the page will be found to be clean and the write will be dropped.
4181 	 */
4182 	if (rp->r_flags & RMODINPROGRESS) {
4183 		mutex_enter(&rp->r_statelock);
4184 		if ((rp->r_flags & RMODINPROGRESS) &&
4185 		    rp->r_modaddr + MAXBSIZE > io_off &&
4186 		    rp->r_modaddr < io_off + io_len) {
4187 			page_t *plist;
4188 			/*
4189 			 * A write is in progress for this region of the file.
4190 			 * If we did not detect RMODINPROGRESS here then this
4191 			 * path through nfs_putapage() would eventually go to
4192 			 * nfs(3)_bio() and may not write out all of the data
4193 			 * in the pages. We end up losing data. So we decide
4194 			 * to set the modified bit on each page in the page
4195 			 * list and mark the rnode with RDIRTY. This write
4196 			 * will be restarted at some later time.
4197 			 */
4198 			plist = pp;
4199 			while (plist != NULL) {
4200 				pp = plist;
4201 				page_sub(&plist, pp);
4202 				hat_setmod(pp);
4203 				page_io_unlock(pp);
4204 				page_unlock(pp);
4205 			}
4206 			rp->r_flags |= RDIRTY;
4207 			mutex_exit(&rp->r_statelock);
4208 			if (offp)
4209 				*offp = io_off;
4210 			if (lenp)
4211 				*lenp = io_len;
4212 			return (0);
4213 		}
4214 		mutex_exit(&rp->r_statelock);
4215 	}
4216 
4217 	if (flags & B_ASYNC) {
4218 		error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr,
4219 		    nfs_sync_putapage);
4220 	} else
4221 		error = nfs_sync_putapage(vp, pp, io_off, io_len, flags, cr);
4222 
4223 	if (offp)
4224 		*offp = io_off;
4225 	if (lenp)
4226 		*lenp = io_len;
4227 	return (error);
4228 }
4229 
4230 static int
4231 nfs_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4232 	int flags, cred_t *cr)
4233 {
4234 	int error;
4235 	rnode_t *rp;
4236 
4237 	flags |= B_WRITE;
4238 
4239 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4240 	error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4241 
4242 	rp = VTOR(vp);
4243 
4244 	if ((error == ENOSPC || error == EDQUOT || error == EACCES) &&
4245 	    (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) {
4246 		if (!(rp->r_flags & ROUTOFSPACE)) {
4247 			mutex_enter(&rp->r_statelock);
4248 			rp->r_flags |= ROUTOFSPACE;
4249 			mutex_exit(&rp->r_statelock);
4250 		}
4251 		flags |= B_ERROR;
4252 		pvn_write_done(pp, flags);
4253 		/*
4254 		 * If this was not an async thread, then try again to
4255 		 * write out the pages, but this time, also destroy
4256 		 * them whether or not the write is successful.  This
4257 		 * will prevent memory from filling up with these
4258 		 * pages and destroying them is the only alternative
4259 		 * if they can't be written out.
4260 		 *
4261 		 * Don't do this if this is an async thread because
4262 		 * when the pages are unlocked in pvn_write_done,
4263 		 * some other thread could have come along, locked
4264 		 * them, and queued for an async thread.  It would be
4265 		 * possible for all of the async threads to be tied
4266 		 * up waiting to lock the pages again and they would
4267 		 * all already be locked and waiting for an async
4268 		 * thread to handle them.  Deadlock.
4269 		 */
4270 		if (!(flags & B_ASYNC)) {
4271 			error = nfs_putpage(vp, io_off, io_len,
4272 			    B_INVAL | B_FORCE, cr, NULL);
4273 		}
4274 	} else {
4275 		if (error)
4276 			flags |= B_ERROR;
4277 		else if (rp->r_flags & ROUTOFSPACE) {
4278 			mutex_enter(&rp->r_statelock);
4279 			rp->r_flags &= ~ROUTOFSPACE;
4280 			mutex_exit(&rp->r_statelock);
4281 		}
4282 		pvn_write_done(pp, flags);
4283 	}
4284 
4285 	return (error);
4286 }
4287 
4288 /* ARGSUSED */
4289 static int
4290 nfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4291 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4292 	caller_context_t *ct)
4293 {
4294 	struct segvn_crargs vn_a;
4295 	int error;
4296 	rnode_t *rp;
4297 	struct vattr va;
4298 
4299 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4300 		return (EIO);
4301 
4302 	if (vp->v_flag & VNOMAP)
4303 		return (ENOSYS);
4304 
4305 	if (off > MAXOFF32_T)
4306 		return (EFBIG);
4307 
4308 	if (off < 0 || off + len < 0)
4309 		return (ENXIO);
4310 
4311 	if (vp->v_type != VREG)
4312 		return (ENODEV);
4313 
4314 	/*
4315 	 * If there is cached data and if close-to-open consistency
4316 	 * checking is not turned off and if the file system is not
4317 	 * mounted readonly, then force an over the wire getattr.
4318 	 * Otherwise, just invoke nfsgetattr to get a copy of the
4319 	 * attributes.  The attribute cache will be used unless it
4320 	 * is timed out and if it is, then an over the wire getattr
4321 	 * will be issued.
4322 	 */
4323 	va.va_mask = AT_ALL;
4324 	if (vn_has_cached_data(vp) &&
4325 	    !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp))
4326 		error = nfs_getattr_otw(vp, &va, cr);
4327 	else
4328 		error = nfsgetattr(vp, &va, cr);
4329 	if (error)
4330 		return (error);
4331 
4332 	/*
4333 	 * Check to see if the vnode is currently marked as not cachable.
4334 	 * This means portions of the file are locked (through VOP_FRLOCK).
4335 	 * In this case the map request must be refused.  We use
4336 	 * rp->r_lkserlock to avoid a race with concurrent lock requests.
4337 	 */
4338 	rp = VTOR(vp);
4339 
4340 	/*
4341 	 * Atomically increment r_inmap after acquiring r_rwlock. The
4342 	 * idea here is to acquire r_rwlock to block read/write and
4343 	 * not to protect r_inmap. r_inmap will inform nfs_read/write()
4344 	 * that we are in nfs_map(). Now, r_rwlock is acquired in order
4345 	 * and we can prevent the deadlock that would have occurred
4346 	 * when nfs_addmap() would have acquired it out of order.
4347 	 *
4348 	 * Since we are not protecting r_inmap by any lock, we do not
4349 	 * hold any lock when we decrement it. We atomically decrement
4350 	 * r_inmap after we release r_lkserlock.
4351 	 */
4352 
4353 	if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp)))
4354 		return (EINTR);
4355 	atomic_add_int(&rp->r_inmap, 1);
4356 	nfs_rw_exit(&rp->r_rwlock);
4357 
4358 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) {
4359 		atomic_add_int(&rp->r_inmap, -1);
4360 		return (EINTR);
4361 	}
4362 	if (vp->v_flag & VNOCACHE) {
4363 		error = EAGAIN;
4364 		goto done;
4365 	}
4366 
4367 	/*
4368 	 * Don't allow concurrent locks and mapping if mandatory locking is
4369 	 * enabled.
4370 	 */
4371 	if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) &&
4372 	    MANDLOCK(vp, va.va_mode)) {
4373 		error = EAGAIN;
4374 		goto done;
4375 	}
4376 
4377 	as_rangelock(as);
4378 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4379 	if (error != 0) {
4380 		as_rangeunlock(as);
4381 		goto done;
4382 	}
4383 
4384 	vn_a.vp = vp;
4385 	vn_a.offset = off;
4386 	vn_a.type = (flags & MAP_TYPE);
4387 	vn_a.prot = (uchar_t)prot;
4388 	vn_a.maxprot = (uchar_t)maxprot;
4389 	vn_a.flags = (flags & ~MAP_TYPE);
4390 	vn_a.cred = cr;
4391 	vn_a.amp = NULL;
4392 	vn_a.szc = 0;
4393 	vn_a.lgrp_mem_policy_flags = 0;
4394 
4395 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4396 	as_rangeunlock(as);
4397 
4398 done:
4399 	nfs_rw_exit(&rp->r_lkserlock);
4400 	atomic_add_int(&rp->r_inmap, -1);
4401 	return (error);
4402 }
4403 
4404 /* ARGSUSED */
4405 static int
4406 nfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4407 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4408 	caller_context_t *ct)
4409 {
4410 	rnode_t *rp;
4411 
4412 	if (vp->v_flag & VNOMAP)
4413 		return (ENOSYS);
4414 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4415 		return (EIO);
4416 
4417 	rp = VTOR(vp);
4418 	atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len));
4419 
4420 	return (0);
4421 }
4422 
4423 /* ARGSUSED */
4424 static int
4425 nfs_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, offset_t offset,
4426 	struct flk_callback *flk_cbp, cred_t *cr, caller_context_t *ct)
4427 {
4428 	netobj lm_fh;
4429 	int rc;
4430 	u_offset_t start, end;
4431 	rnode_t *rp;
4432 	int error = 0, intr = INTR(vp);
4433 
4434 	/* check for valid cmd parameter */
4435 	if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW)
4436 		return (EINVAL);
4437 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4438 		return (EIO);
4439 
4440 	/* Verify l_type. */
4441 	switch (bfp->l_type) {
4442 	case F_RDLCK:
4443 		if (cmd != F_GETLK && !(flag & FREAD))
4444 			return (EBADF);
4445 		break;
4446 	case F_WRLCK:
4447 		if (cmd != F_GETLK && !(flag & FWRITE))
4448 			return (EBADF);
4449 		break;
4450 	case F_UNLCK:
4451 		intr = 0;
4452 		break;
4453 
4454 	default:
4455 		return (EINVAL);
4456 	}
4457 
4458 	/* check the validity of the lock range */
4459 	if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset))
4460 		return (rc);
4461 	if (rc = flk_check_lock_data(start, end, MAXOFF32_T))
4462 		return (rc);
4463 
4464 	/*
4465 	 * If the filesystem is mounted using local locking, pass the
4466 	 * request off to the local locking code.
4467 	 */
4468 	if (VTOMI(vp)->mi_flags & MI_LLOCK) {
4469 		if (offset > MAXOFF32_T)
4470 			return (EFBIG);
4471 		if (cmd == F_SETLK || cmd == F_SETLKW) {
4472 			/*
4473 			 * For complete safety, we should be holding
4474 			 * r_lkserlock.  However, we can't call
4475 			 * lm_safelock and then fs_frlock while
4476 			 * holding r_lkserlock, so just invoke
4477 			 * lm_safelock and expect that this will
4478 			 * catch enough of the cases.
4479 			 */
4480 			if (!lm_safelock(vp, bfp, cr))
4481 				return (EAGAIN);
4482 		}
4483 		return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4484 	}
4485 
4486 	rp = VTOR(vp);
4487 
4488 	/*
4489 	 * Check whether the given lock request can proceed, given the
4490 	 * current file mappings.
4491 	 */
4492 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr))
4493 		return (EINTR);
4494 	if (cmd == F_SETLK || cmd == F_SETLKW) {
4495 		if (!lm_safelock(vp, bfp, cr)) {
4496 			rc = EAGAIN;
4497 			goto done;
4498 		}
4499 	}
4500 
4501 	/*
4502 	 * Flush the cache after waiting for async I/O to finish.  For new
4503 	 * locks, this is so that the process gets the latest bits from the
4504 	 * server.  For unlocks, this is so that other clients see the
4505 	 * latest bits once the file has been unlocked.  If currently dirty
4506 	 * pages can't be flushed, then don't allow a lock to be set.  But
4507 	 * allow unlocks to succeed, to avoid having orphan locks on the
4508 	 * server.
4509 	 */
4510 	if (cmd != F_GETLK) {
4511 		mutex_enter(&rp->r_statelock);
4512 		while (rp->r_count > 0) {
4513 			if (intr) {
4514 				klwp_t *lwp = ttolwp(curthread);
4515 
4516 				if (lwp != NULL)
4517 					lwp->lwp_nostop++;
4518 				if (cv_wait_sig(&rp->r_cv, &rp->r_statelock)
4519 				    == 0) {
4520 					if (lwp != NULL)
4521 						lwp->lwp_nostop--;
4522 					rc = EINTR;
4523 					break;
4524 				}
4525 				if (lwp != NULL)
4526 					lwp->lwp_nostop--;
4527 			} else
4528 			cv_wait(&rp->r_cv, &rp->r_statelock);
4529 		}
4530 		mutex_exit(&rp->r_statelock);
4531 		if (rc != 0)
4532 			goto done;
4533 		error = nfs_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct);
4534 		if (error) {
4535 			if (error == ENOSPC || error == EDQUOT) {
4536 				mutex_enter(&rp->r_statelock);
4537 				if (!rp->r_error)
4538 					rp->r_error = error;
4539 				mutex_exit(&rp->r_statelock);
4540 			}
4541 			if (bfp->l_type != F_UNLCK) {
4542 				rc = ENOLCK;
4543 				goto done;
4544 			}
4545 		}
4546 	}
4547 
4548 	lm_fh.n_len = sizeof (fhandle_t);
4549 	lm_fh.n_bytes = (char *)VTOFH(vp);
4550 
4551 	/*
4552 	 * Call the lock manager to do the real work of contacting
4553 	 * the server and obtaining the lock.
4554 	 */
4555 	rc = lm_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh, flk_cbp);
4556 
4557 	if (rc == 0)
4558 		nfs_lockcompletion(vp, cmd);
4559 
4560 done:
4561 	nfs_rw_exit(&rp->r_lkserlock);
4562 	return (rc);
4563 }
4564 
4565 /*
4566  * Free storage space associated with the specified vnode.  The portion
4567  * to be freed is specified by bfp->l_start and bfp->l_len (already
4568  * normalized to a "whence" of 0).
4569  *
4570  * This is an experimental facility whose continued existence is not
4571  * guaranteed.  Currently, we only support the special case
4572  * of l_len == 0, meaning free to end of file.
4573  */
4574 /* ARGSUSED */
4575 static int
4576 nfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
4577 	offset_t offset, cred_t *cr, caller_context_t *ct)
4578 {
4579 	int error;
4580 
4581 	ASSERT(vp->v_type == VREG);
4582 	if (cmd != F_FREESP)
4583 		return (EINVAL);
4584 
4585 	if (offset > MAXOFF32_T)
4586 		return (EFBIG);
4587 
4588 	if ((bfp->l_start > MAXOFF32_T) || (bfp->l_end > MAXOFF32_T) ||
4589 	    (bfp->l_len > MAXOFF32_T))
4590 		return (EFBIG);
4591 
4592 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4593 		return (EIO);
4594 
4595 	error = convoff(vp, bfp, 0, offset);
4596 	if (!error) {
4597 		ASSERT(bfp->l_start >= 0);
4598 		if (bfp->l_len == 0) {
4599 			struct vattr va;
4600 
4601 			/*
4602 			 * ftruncate should not change the ctime and
4603 			 * mtime if we truncate the file to its
4604 			 * previous size.
4605 			 */
4606 			va.va_mask = AT_SIZE;
4607 			error = nfsgetattr(vp, &va, cr);
4608 			if (error || va.va_size == bfp->l_start)
4609 				return (error);
4610 			va.va_mask = AT_SIZE;
4611 			va.va_size = bfp->l_start;
4612 			error = nfssetattr(vp, &va, 0, cr);
4613 		} else
4614 			error = EINVAL;
4615 	}
4616 
4617 	return (error);
4618 }
4619 
4620 /* ARGSUSED */
4621 static int
4622 nfs_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
4623 {
4624 
4625 	return (EINVAL);
4626 }
4627 
4628 /*
4629  * Setup and add an address space callback to do the work of the delmap call.
4630  * The callback will (and must be) deleted in the actual callback function.
4631  *
4632  * This is done in order to take care of the problem that we have with holding
4633  * the address space's a_lock for a long period of time (e.g. if the NFS server
4634  * is down).  Callbacks will be executed in the address space code while the
4635  * a_lock is not held.	Holding the address space's a_lock causes things such
4636  * as ps and fork to hang because they are trying to acquire this lock as well.
4637  */
4638 /* ARGSUSED */
4639 static int
4640 nfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4641 	size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4642 	caller_context_t *ct)
4643 {
4644 	int			caller_found;
4645 	int			error;
4646 	rnode_t			*rp;
4647 	nfs_delmap_args_t	*dmapp;
4648 	nfs_delmapcall_t	*delmap_call;
4649 
4650 	if (vp->v_flag & VNOMAP)
4651 		return (ENOSYS);
4652 	/*
4653 	 * A process may not change zones if it has NFS pages mmap'ed
4654 	 * in, so we can't legitimately get here from the wrong zone.
4655 	 */
4656 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4657 
4658 	rp = VTOR(vp);
4659 
4660 	/*
4661 	 * The way that the address space of this process deletes its mapping
4662 	 * of this file is via the following call chains:
4663 	 * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4664 	 * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4665 	 *
4666 	 * With the use of address space callbacks we are allowed to drop the
4667 	 * address space lock, a_lock, while executing the NFS operations that
4668 	 * need to go over the wire.  Returning EAGAIN to the caller of this
4669 	 * function is what drives the execution of the callback that we add
4670 	 * below.  The callback will be executed by the address space code
4671 	 * after dropping the a_lock.  When the callback is finished, since
4672 	 * we dropped the a_lock, it must be re-acquired and segvn_unmap()
4673 	 * is called again on the same segment to finish the rest of the work
4674 	 * that needs to happen during unmapping.
4675 	 *
4676 	 * This action of calling back into the segment driver causes
4677 	 * nfs_delmap() to get called again, but since the callback was
4678 	 * already executed at this point, it already did the work and there
4679 	 * is nothing left for us to do.
4680 	 *
4681 	 * To Summarize:
4682 	 * - The first time nfs_delmap is called by the current thread is when
4683 	 * we add the caller associated with this delmap to the delmap caller
4684 	 * list, add the callback, and return EAGAIN.
4685 	 * - The second time in this call chain when nfs_delmap is called we
4686 	 * will find this caller in the delmap caller list and realize there
4687 	 * is no more work to do thus removing this caller from the list and
4688 	 * returning the error that was set in the callback execution.
4689 	 */
4690 	caller_found = nfs_find_and_delete_delmapcall(rp, &error);
4691 	if (caller_found) {
4692 		/*
4693 		 * 'error' is from the actual delmap operations.  To avoid
4694 		 * hangs, we need to handle the return of EAGAIN differently
4695 		 * since this is what drives the callback execution.
4696 		 * In this case, we don't want to return EAGAIN and do the
4697 		 * callback execution because there are none to execute.
4698 		 */
4699 		if (error == EAGAIN)
4700 			return (0);
4701 		else
4702 			return (error);
4703 	}
4704 
4705 	/* current caller was not in the list */
4706 	delmap_call = nfs_init_delmapcall();
4707 
4708 	mutex_enter(&rp->r_statelock);
4709 	list_insert_tail(&rp->r_indelmap, delmap_call);
4710 	mutex_exit(&rp->r_statelock);
4711 
4712 	dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP);
4713 
4714 	dmapp->vp = vp;
4715 	dmapp->off = off;
4716 	dmapp->addr = addr;
4717 	dmapp->len = len;
4718 	dmapp->prot = prot;
4719 	dmapp->maxprot = maxprot;
4720 	dmapp->flags = flags;
4721 	dmapp->cr = cr;
4722 	dmapp->caller = delmap_call;
4723 
4724 	error = as_add_callback(as, nfs_delmap_callback, dmapp,
4725 	    AS_UNMAP_EVENT, addr, len, KM_SLEEP);
4726 
4727 	return (error ? error : EAGAIN);
4728 }
4729 
4730 /*
4731  * Remove some pages from an mmap'd vnode.  Just update the
4732  * count of pages.  If doing close-to-open, then flush all
4733  * of the pages associated with this file.  Otherwise, start
4734  * an asynchronous page flush to write out any dirty pages.
4735  * This will also associate a credential with the rnode which
4736  * can be used to write the pages.
4737  */
4738 /* ARGSUSED */
4739 static void
4740 nfs_delmap_callback(struct as *as, void *arg, uint_t event)
4741 {
4742 	int			error;
4743 	rnode_t			*rp;
4744 	mntinfo_t		*mi;
4745 	nfs_delmap_args_t	*dmapp = (nfs_delmap_args_t *)arg;
4746 
4747 	rp = VTOR(dmapp->vp);
4748 	mi = VTOMI(dmapp->vp);
4749 
4750 	atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len));
4751 	ASSERT(rp->r_mapcnt >= 0);
4752 
4753 	/*
4754 	 * Initiate a page flush if there are pages, the file system
4755 	 * was not mounted readonly, the segment was mapped shared, and
4756 	 * the pages themselves were writeable.
4757 	 */
4758 	if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) &&
4759 	    dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) {
4760 		mutex_enter(&rp->r_statelock);
4761 		rp->r_flags |= RDIRTY;
4762 		mutex_exit(&rp->r_statelock);
4763 		/*
4764 		 * If this is a cross-zone access a sync putpage won't work, so
4765 		 * the best we can do is try an async putpage.  That seems
4766 		 * better than something more draconian such as discarding the
4767 		 * dirty pages.
4768 		 */
4769 		if ((mi->mi_flags & MI_NOCTO) ||
4770 		    nfs_zone() != mi->mi_zone)
4771 			error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4772 			    B_ASYNC, dmapp->cr, NULL);
4773 		else
4774 			error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4775 			    0, dmapp->cr, NULL);
4776 		if (!error) {
4777 			mutex_enter(&rp->r_statelock);
4778 			error = rp->r_error;
4779 			rp->r_error = 0;
4780 			mutex_exit(&rp->r_statelock);
4781 		}
4782 	} else
4783 		error = 0;
4784 
4785 	if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO))
4786 		(void) nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4787 		    B_INVAL, dmapp->cr, NULL);
4788 
4789 	dmapp->caller->error = error;
4790 	(void) as_delete_callback(as, arg);
4791 	kmem_free(dmapp, sizeof (nfs_delmap_args_t));
4792 }
4793 
4794 /* ARGSUSED */
4795 static int
4796 nfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4797 	caller_context_t *ct)
4798 {
4799 	int error = 0;
4800 
4801 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4802 		return (EIO);
4803 	/*
4804 	 * This looks a little weird because it's written in a general
4805 	 * manner but we make little use of cases.  If cntl() ever gets
4806 	 * widely used, the outer switch will make more sense.
4807 	 */
4808 
4809 	switch (cmd) {
4810 
4811 	/*
4812 	 * Large file spec - need to base answer new query with
4813 	 * hardcoded constant based on the protocol.
4814 	 */
4815 	case _PC_FILESIZEBITS:
4816 		*valp = 32;
4817 		return (0);
4818 
4819 	case _PC_LINK_MAX:
4820 	case _PC_NAME_MAX:
4821 	case _PC_PATH_MAX:
4822 	case _PC_SYMLINK_MAX:
4823 	case _PC_CHOWN_RESTRICTED:
4824 	case _PC_NO_TRUNC: {
4825 		mntinfo_t *mi;
4826 		struct pathcnf *pc;
4827 
4828 		if ((mi = VTOMI(vp)) == NULL || (pc = mi->mi_pathconf) == NULL)
4829 			return (EINVAL);
4830 		error = _PC_ISSET(cmd, pc->pc_mask);    /* error or bool */
4831 		switch (cmd) {
4832 		case _PC_LINK_MAX:
4833 			*valp = pc->pc_link_max;
4834 			break;
4835 		case _PC_NAME_MAX:
4836 			*valp = pc->pc_name_max;
4837 			break;
4838 		case _PC_PATH_MAX:
4839 		case _PC_SYMLINK_MAX:
4840 			*valp = pc->pc_path_max;
4841 			break;
4842 		case _PC_CHOWN_RESTRICTED:
4843 			/*
4844 			 * if we got here, error is really a boolean which
4845 			 * indicates whether cmd is set or not.
4846 			 */
4847 			*valp = error ? 1 : 0;	/* see above */
4848 			error = 0;
4849 			break;
4850 		case _PC_NO_TRUNC:
4851 			/*
4852 			 * if we got here, error is really a boolean which
4853 			 * indicates whether cmd is set or not.
4854 			 */
4855 			*valp = error ? 1 : 0;	/* see above */
4856 			error = 0;
4857 			break;
4858 		}
4859 		return (error ? EINVAL : 0);
4860 		}
4861 
4862 	case _PC_XATTR_EXISTS:
4863 		*valp = 0;
4864 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
4865 			vnode_t *avp;
4866 			rnode_t *rp;
4867 			mntinfo_t *mi = VTOMI(vp);
4868 
4869 			if (!(mi->mi_flags & MI_EXTATTR))
4870 				return (0);
4871 
4872 			rp = VTOR(vp);
4873 			if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER,
4874 			    INTR(vp)))
4875 				return (EINTR);
4876 
4877 			error = nfslookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr);
4878 			if (error || avp == NULL)
4879 				error = acl_getxattrdir2(vp, &avp, 0, cr, 0);
4880 
4881 			nfs_rw_exit(&rp->r_rwlock);
4882 
4883 			if (error == 0 && avp != NULL) {
4884 				error = do_xattr_exists_check(avp, valp, cr);
4885 				VN_RELE(avp);
4886 			}
4887 		}
4888 		return (error ? EINVAL : 0);
4889 
4890 	case _PC_ACL_ENABLED:
4891 		*valp = _ACL_ACLENT_ENABLED;
4892 		return (0);
4893 
4894 	default:
4895 		return (EINVAL);
4896 	}
4897 }
4898 
4899 /*
4900  * Called by async thread to do synchronous pageio. Do the i/o, wait
4901  * for it to complete, and cleanup the page list when done.
4902  */
4903 static int
4904 nfs_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4905 	int flags, cred_t *cr)
4906 {
4907 	int error;
4908 
4909 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4910 	error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4911 	if (flags & B_READ)
4912 		pvn_read_done(pp, (error ? B_ERROR : 0) | flags);
4913 	else
4914 		pvn_write_done(pp, (error ? B_ERROR : 0) | flags);
4915 	return (error);
4916 }
4917 
4918 /* ARGSUSED */
4919 static int
4920 nfs_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4921 	int flags, cred_t *cr, caller_context_t *ct)
4922 {
4923 	int error;
4924 	rnode_t *rp;
4925 
4926 	if (pp == NULL)
4927 		return (EINVAL);
4928 
4929 	if (io_off > MAXOFF32_T)
4930 		return (EFBIG);
4931 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4932 		return (EIO);
4933 	rp = VTOR(vp);
4934 	mutex_enter(&rp->r_statelock);
4935 	rp->r_count++;
4936 	mutex_exit(&rp->r_statelock);
4937 
4938 	if (flags & B_ASYNC) {
4939 		error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr,
4940 		    nfs_sync_pageio);
4941 	} else
4942 		error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4943 	mutex_enter(&rp->r_statelock);
4944 	rp->r_count--;
4945 	cv_broadcast(&rp->r_cv);
4946 	mutex_exit(&rp->r_statelock);
4947 	return (error);
4948 }
4949 
4950 /* ARGSUSED */
4951 static int
4952 nfs_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4953 	caller_context_t *ct)
4954 {
4955 	int error;
4956 	mntinfo_t *mi;
4957 
4958 	mi = VTOMI(vp);
4959 
4960 	if (nfs_zone() != mi->mi_zone)
4961 		return (EIO);
4962 	if (mi->mi_flags & MI_ACL) {
4963 		error = acl_setacl2(vp, vsecattr, flag, cr);
4964 		if (mi->mi_flags & MI_ACL)
4965 			return (error);
4966 	}
4967 
4968 	return (ENOSYS);
4969 }
4970 
4971 /* ARGSUSED */
4972 static int
4973 nfs_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4974 	caller_context_t *ct)
4975 {
4976 	int error;
4977 	mntinfo_t *mi;
4978 
4979 	mi = VTOMI(vp);
4980 
4981 	if (nfs_zone() != mi->mi_zone)
4982 		return (EIO);
4983 	if (mi->mi_flags & MI_ACL) {
4984 		error = acl_getacl2(vp, vsecattr, flag, cr);
4985 		if (mi->mi_flags & MI_ACL)
4986 			return (error);
4987 	}
4988 
4989 	return (fs_fab_acl(vp, vsecattr, flag, cr, ct));
4990 }
4991 
4992 /* ARGSUSED */
4993 static int
4994 nfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
4995 	caller_context_t *ct)
4996 {
4997 	int error;
4998 	struct shrlock nshr;
4999 	struct nfs_owner nfs_owner;
5000 	netobj lm_fh;
5001 
5002 	if (nfs_zone() != VTOMI(vp)->mi_zone)
5003 		return (EIO);
5004 
5005 	/*
5006 	 * check for valid cmd parameter
5007 	 */
5008 	if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS)
5009 		return (EINVAL);
5010 
5011 	/*
5012 	 * Check access permissions
5013 	 */
5014 	if (cmd == F_SHARE &&
5015 	    (((shr->s_access & F_RDACC) && !(flag & FREAD)) ||
5016 	    ((shr->s_access & F_WRACC) && !(flag & FWRITE))))
5017 		return (EBADF);
5018 
5019 	/*
5020 	 * If the filesystem is mounted using local locking, pass the
5021 	 * request off to the local share code.
5022 	 */
5023 	if (VTOMI(vp)->mi_flags & MI_LLOCK)
5024 		return (fs_shrlock(vp, cmd, shr, flag, cr, ct));
5025 
5026 	switch (cmd) {
5027 	case F_SHARE:
5028 	case F_UNSHARE:
5029 		lm_fh.n_len = sizeof (fhandle_t);
5030 		lm_fh.n_bytes = (char *)VTOFH(vp);
5031 
5032 		/*
5033 		 * If passed an owner that is too large to fit in an
5034 		 * nfs_owner it is likely a recursive call from the
5035 		 * lock manager client and pass it straight through.  If
5036 		 * it is not a nfs_owner then simply return an error.
5037 		 */
5038 		if (shr->s_own_len > sizeof (nfs_owner.lowner)) {
5039 			if (((struct nfs_owner *)shr->s_owner)->magic !=
5040 			    NFS_OWNER_MAGIC)
5041 				return (EINVAL);
5042 
5043 			if (error = lm_shrlock(vp, cmd, shr, flag, &lm_fh)) {
5044 				error = set_errno(error);
5045 			}
5046 			return (error);
5047 		}
5048 		/*
5049 		 * Remote share reservations owner is a combination of
5050 		 * a magic number, hostname, and the local owner
5051 		 */
5052 		bzero(&nfs_owner, sizeof (nfs_owner));
5053 		nfs_owner.magic = NFS_OWNER_MAGIC;
5054 		(void) strncpy(nfs_owner.hname, uts_nodename(),
5055 		    sizeof (nfs_owner.hname));
5056 		bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len);
5057 		nshr.s_access = shr->s_access;
5058 		nshr.s_deny = shr->s_deny;
5059 		nshr.s_sysid = 0;
5060 		nshr.s_pid = ttoproc(curthread)->p_pid;
5061 		nshr.s_own_len = sizeof (nfs_owner);
5062 		nshr.s_owner = (caddr_t)&nfs_owner;
5063 
5064 		if (error = lm_shrlock(vp, cmd, &nshr, flag, &lm_fh)) {
5065 			error = set_errno(error);
5066 		}
5067 
5068 		break;
5069 
5070 	case F_HASREMOTELOCKS:
5071 		/*
5072 		 * NFS client can't store remote locks itself
5073 		 */
5074 		shr->s_access = 0;
5075 		error = 0;
5076 		break;
5077 
5078 	default:
5079 		error = EINVAL;
5080 		break;
5081 	}
5082 
5083 	return (error);
5084 }
5085