xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_inet.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/jail.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68 
69 #include <fs/nfs/nfsport.h>
70 #include <fs/nfsclient/nfsnode.h>
71 #include <fs/nfsclient/nfsmount.h>
72 #include <fs/nfsclient/nfs.h>
73 
74 #include <net/if.h>
75 #include <netinet/in.h>
76 #include <netinet/in_var.h>
77 
78 #include <nfs/nfs_lock.h>
79 
80 /* Defs */
81 #define	TRUE	1
82 #define	FALSE	0
83 
84 extern struct nfsstats newnfsstats;
85 extern int nfsrv_useacl;
86 MALLOC_DECLARE(M_NEWNFSREQ);
87 
88 /*
89  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
90  * calls are not in getblk() and brelse() so that they would not be necessary
91  * here.
92  */
93 #ifndef B_VMIO
94 #define	vfs_busy_pages(bp, f)
95 #endif
96 
97 static vop_read_t	nfsfifo_read;
98 static vop_write_t	nfsfifo_write;
99 static vop_close_t	nfsfifo_close;
100 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
101 		    struct thread *);
102 static vop_lookup_t	nfs_lookup;
103 static vop_create_t	nfs_create;
104 static vop_mknod_t	nfs_mknod;
105 static vop_open_t	nfs_open;
106 static vop_pathconf_t	nfs_pathconf;
107 static vop_close_t	nfs_close;
108 static vop_access_t	nfs_access;
109 static vop_getattr_t	nfs_getattr;
110 static vop_setattr_t	nfs_setattr;
111 static vop_read_t	nfs_read;
112 static vop_fsync_t	nfs_fsync;
113 static vop_remove_t	nfs_remove;
114 static vop_link_t	nfs_link;
115 static vop_rename_t	nfs_rename;
116 static vop_mkdir_t	nfs_mkdir;
117 static vop_rmdir_t	nfs_rmdir;
118 static vop_symlink_t	nfs_symlink;
119 static vop_readdir_t	nfs_readdir;
120 static vop_strategy_t	nfs_strategy;
121 static vop_lock1_t	nfs_lock1;
122 static	int	nfs_lookitup(struct vnode *, char *, int,
123 		    struct ucred *, struct thread *, struct nfsnode **);
124 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
125 		    struct componentname *);
126 static vop_access_t	nfsspec_access;
127 static vop_readlink_t	nfs_readlink;
128 static vop_print_t	nfs_print;
129 static vop_advlock_t	nfs_advlock;
130 static vop_advlockasync_t nfs_advlockasync;
131 static vop_getacl_t nfs_getacl;
132 static vop_setacl_t nfs_setacl;
133 
134 /*
135  * Global vfs data structures for nfs
136  */
137 struct vop_vector newnfs_vnodeops = {
138 	.vop_default =		&default_vnodeops,
139 	.vop_access =		nfs_access,
140 	.vop_advlock =		nfs_advlock,
141 	.vop_advlockasync =	nfs_advlockasync,
142 	.vop_close =		nfs_close,
143 	.vop_create =		nfs_create,
144 	.vop_fsync =		nfs_fsync,
145 	.vop_getattr =		nfs_getattr,
146 	.vop_getpages =		ncl_getpages,
147 	.vop_putpages =		ncl_putpages,
148 	.vop_inactive =		ncl_inactive,
149 	.vop_link =		nfs_link,
150 	.vop_lock1 = 		nfs_lock1,
151 	.vop_lookup =		nfs_lookup,
152 	.vop_mkdir =		nfs_mkdir,
153 	.vop_mknod =		nfs_mknod,
154 	.vop_open =		nfs_open,
155 	.vop_pathconf =		nfs_pathconf,
156 	.vop_print =		nfs_print,
157 	.vop_read =		nfs_read,
158 	.vop_readdir =		nfs_readdir,
159 	.vop_readlink =		nfs_readlink,
160 	.vop_reclaim =		ncl_reclaim,
161 	.vop_remove =		nfs_remove,
162 	.vop_rename =		nfs_rename,
163 	.vop_rmdir =		nfs_rmdir,
164 	.vop_setattr =		nfs_setattr,
165 	.vop_strategy =		nfs_strategy,
166 	.vop_symlink =		nfs_symlink,
167 	.vop_write =		ncl_write,
168 	.vop_getacl =		nfs_getacl,
169 	.vop_setacl =		nfs_setacl,
170 };
171 
172 struct vop_vector newnfs_fifoops = {
173 	.vop_default =		&fifo_specops,
174 	.vop_access =		nfsspec_access,
175 	.vop_close =		nfsfifo_close,
176 	.vop_fsync =		nfs_fsync,
177 	.vop_getattr =		nfs_getattr,
178 	.vop_inactive =		ncl_inactive,
179 	.vop_print =		nfs_print,
180 	.vop_read =		nfsfifo_read,
181 	.vop_reclaim =		ncl_reclaim,
182 	.vop_setattr =		nfs_setattr,
183 	.vop_write =		nfsfifo_write,
184 };
185 
186 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
187     struct componentname *cnp, struct vattr *vap);
188 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
189     int namelen, struct ucred *cred, struct thread *td);
190 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
191     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
192     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
193 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
194     struct componentname *scnp, struct sillyrename *sp);
195 
196 /*
197  * Global variables
198  */
199 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
200 
201 SYSCTL_DECL(_vfs_nfs);
202 
203 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
204 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
205 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
206 
207 static int	nfs_prime_access_cache = 0;
208 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
209 	   &nfs_prime_access_cache, 0,
210 	   "Prime NFS ACCESS cache when fetching attributes");
211 
212 static int	newnfs_commit_on_close = 0;
213 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
214     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
215 
216 static int	nfs_clean_pages_on_close = 1;
217 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
218 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
219 
220 int newnfs_directio_enable = 0;
221 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
222 	   &newnfs_directio_enable, 0, "Enable NFS directio");
223 
224 /*
225  * This sysctl allows other processes to mmap a file that has been opened
226  * O_DIRECT by a process.  In general, having processes mmap the file while
227  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
228  * this by default to prevent DoS attacks - to prevent a malicious user from
229  * opening up files O_DIRECT preventing other users from mmap'ing these
230  * files.  "Protected" environments where stricter consistency guarantees are
231  * required can disable this knob.  The process that opened the file O_DIRECT
232  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
233  * meaningful.
234  */
235 int newnfs_directio_allow_mmap = 1;
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
237 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
238 
239 #if 0
240 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
241 	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
242 
243 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
244 	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
245 #endif
246 
247 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
248 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
249 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
250 
251 /*
252  * SMP Locking Note :
253  * The list of locks after the description of the lock is the ordering
254  * of other locks acquired with the lock held.
255  * np->n_mtx : Protects the fields in the nfsnode.
256        VM Object Lock
257        VI_MTX (acquired indirectly)
258  * nmp->nm_mtx : Protects the fields in the nfsmount.
259        rep->r_mtx
260  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
261  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
262        nmp->nm_mtx
263        rep->r_mtx
264  * rep->r_mtx : Protects the fields in an nfsreq.
265  */
266 
267 static int
268 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
269     struct ucred *cred, u_int32_t *retmode)
270 {
271 	int error = 0, attrflag, i, lrupos;
272 	u_int32_t rmode;
273 	struct nfsnode *np = VTONFS(vp);
274 	struct nfsvattr nfsva;
275 
276 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
277 	    &rmode, NULL);
278 	if (attrflag)
279 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
280 	if (!error) {
281 		lrupos = 0;
282 		mtx_lock(&np->n_mtx);
283 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
284 			if (np->n_accesscache[i].uid == cred->cr_uid) {
285 				np->n_accesscache[i].mode = rmode;
286 				np->n_accesscache[i].stamp = time_second;
287 				break;
288 			}
289 			if (i > 0 && np->n_accesscache[i].stamp <
290 			    np->n_accesscache[lrupos].stamp)
291 				lrupos = i;
292 		}
293 		if (i == NFS_ACCESSCACHESIZE) {
294 			np->n_accesscache[lrupos].uid = cred->cr_uid;
295 			np->n_accesscache[lrupos].mode = rmode;
296 			np->n_accesscache[lrupos].stamp = time_second;
297 		}
298 		mtx_unlock(&np->n_mtx);
299 		if (retmode != NULL)
300 			*retmode = rmode;
301 	} else if (NFS_ISV4(vp)) {
302 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
303 	}
304 	return (error);
305 }
306 
307 /*
308  * nfs access vnode op.
309  * For nfs version 2, just return ok. File accesses may fail later.
310  * For nfs version 3, use the access rpc to check accessibility. If file modes
311  * are changed on the server, accesses might still fail later.
312  */
313 static int
314 nfs_access(struct vop_access_args *ap)
315 {
316 	struct vnode *vp = ap->a_vp;
317 	int error = 0, i, gotahit;
318 	u_int32_t mode, wmode, rmode;
319 	int v34 = NFS_ISV34(vp);
320 	struct nfsnode *np = VTONFS(vp);
321 
322 	/*
323 	 * Disallow write attempts on filesystems mounted read-only;
324 	 * unless the file is a socket, fifo, or a block or character
325 	 * device resident on the filesystem.
326 	 */
327 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
328 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
329 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
330 		switch (vp->v_type) {
331 		case VREG:
332 		case VDIR:
333 		case VLNK:
334 			return (EROFS);
335 		default:
336 			break;
337 		}
338 	}
339 	/*
340 	 * For nfs v3 or v4, check to see if we have done this recently, and if
341 	 * so return our cached result instead of making an ACCESS call.
342 	 * If not, do an access rpc, otherwise you are stuck emulating
343 	 * ufs_access() locally using the vattr. This may not be correct,
344 	 * since the server may apply other access criteria such as
345 	 * client uid-->server uid mapping that we do not know about.
346 	 */
347 	if (v34) {
348 		if (ap->a_accmode & VREAD)
349 			mode = NFSACCESS_READ;
350 		else
351 			mode = 0;
352 		if (vp->v_type != VDIR) {
353 			if (ap->a_accmode & VWRITE)
354 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
355 			if (ap->a_accmode & VAPPEND)
356 				mode |= NFSACCESS_EXTEND;
357 			if (ap->a_accmode & VEXEC)
358 				mode |= NFSACCESS_EXECUTE;
359 			if (ap->a_accmode & VDELETE)
360 				mode |= NFSACCESS_DELETE;
361 		} else {
362 			if (ap->a_accmode & VWRITE)
363 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
364 			if (ap->a_accmode & VAPPEND)
365 				mode |= NFSACCESS_EXTEND;
366 			if (ap->a_accmode & VEXEC)
367 				mode |= NFSACCESS_LOOKUP;
368 			if (ap->a_accmode & VDELETE)
369 				mode |= NFSACCESS_DELETE;
370 			if (ap->a_accmode & VDELETE_CHILD)
371 				mode |= NFSACCESS_MODIFY;
372 		}
373 		/* XXX safety belt, only make blanket request if caching */
374 		if (nfsaccess_cache_timeout > 0) {
375 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
376 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
377 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
378 		} else {
379 			wmode = mode;
380 		}
381 
382 		/*
383 		 * Does our cached result allow us to give a definite yes to
384 		 * this request?
385 		 */
386 		gotahit = 0;
387 		mtx_lock(&np->n_mtx);
388 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
389 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
390 			    if (time_second < (np->n_accesscache[i].stamp
391 				+ nfsaccess_cache_timeout) &&
392 				(np->n_accesscache[i].mode & mode) == mode) {
393 				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
394 				gotahit = 1;
395 			    }
396 			    break;
397 			}
398 		}
399 		mtx_unlock(&np->n_mtx);
400 		if (gotahit == 0) {
401 			/*
402 			 * Either a no, or a don't know.  Go to the wire.
403 			 */
404 			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
405 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
406 			    ap->a_cred, &rmode);
407 			if (!error &&
408 			    (rmode & mode) != mode)
409 				error = EACCES;
410 		}
411 		return (error);
412 	} else {
413 		if ((error = nfsspec_access(ap)) != 0) {
414 			return (error);
415 		}
416 		/*
417 		 * Attempt to prevent a mapped root from accessing a file
418 		 * which it shouldn't.  We try to read a byte from the file
419 		 * if the user is root and the file is not zero length.
420 		 * After calling nfsspec_access, we should have the correct
421 		 * file size cached.
422 		 */
423 		mtx_lock(&np->n_mtx);
424 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
425 		    && VTONFS(vp)->n_size > 0) {
426 			struct iovec aiov;
427 			struct uio auio;
428 			char buf[1];
429 
430 			mtx_unlock(&np->n_mtx);
431 			aiov.iov_base = buf;
432 			aiov.iov_len = 1;
433 			auio.uio_iov = &aiov;
434 			auio.uio_iovcnt = 1;
435 			auio.uio_offset = 0;
436 			auio.uio_resid = 1;
437 			auio.uio_segflg = UIO_SYSSPACE;
438 			auio.uio_rw = UIO_READ;
439 			auio.uio_td = ap->a_td;
440 
441 			if (vp->v_type == VREG)
442 				error = ncl_readrpc(vp, &auio, ap->a_cred);
443 			else if (vp->v_type == VDIR) {
444 				char* bp;
445 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
446 				aiov.iov_base = bp;
447 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
448 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
449 				    ap->a_td);
450 				free(bp, M_TEMP);
451 			} else if (vp->v_type == VLNK)
452 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
453 			else
454 				error = EACCES;
455 		} else
456 			mtx_unlock(&np->n_mtx);
457 		return (error);
458 	}
459 }
460 
461 
462 /*
463  * nfs open vnode op
464  * Check to see if the type is ok
465  * and that deletion is not in progress.
466  * For paged in text files, you will need to flush the page cache
467  * if consistency is lost.
468  */
469 /* ARGSUSED */
470 static int
471 nfs_open(struct vop_open_args *ap)
472 {
473 	struct vnode *vp = ap->a_vp;
474 	struct nfsnode *np = VTONFS(vp);
475 	struct vattr vattr;
476 	int error;
477 	int fmode = ap->a_mode;
478 
479 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
480 		return (EOPNOTSUPP);
481 
482 	/*
483 	 * For NFSv4, we need to do the Open Op before cache validation,
484 	 * so that we conform to RFC3530 Sec. 9.3.1.
485 	 */
486 	if (NFS_ISV4(vp)) {
487 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
488 		if (error) {
489 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
490 			    (gid_t)0);
491 			return (error);
492 		}
493 	}
494 
495 	/*
496 	 * Now, if this Open will be doing reading, re-validate/flush the
497 	 * cache, so that Close/Open coherency is maintained.
498 	 */
499 	mtx_lock(&np->n_mtx);
500 	if (np->n_flag & NMODIFIED) {
501 		mtx_unlock(&np->n_mtx);
502 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
503 		if (error == EINTR || error == EIO) {
504 			if (NFS_ISV4(vp))
505 				(void) nfsrpc_close(vp, 0, ap->a_td);
506 			return (error);
507 		}
508 		mtx_lock(&np->n_mtx);
509 		np->n_attrstamp = 0;
510 		if (vp->v_type == VDIR)
511 			np->n_direofoffset = 0;
512 		mtx_unlock(&np->n_mtx);
513 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
514 		if (error) {
515 			if (NFS_ISV4(vp))
516 				(void) nfsrpc_close(vp, 0, ap->a_td);
517 			return (error);
518 		}
519 		mtx_lock(&np->n_mtx);
520 		np->n_mtime = vattr.va_mtime;
521 		if (NFS_ISV4(vp))
522 			np->n_change = vattr.va_filerev;
523 	} else {
524 		mtx_unlock(&np->n_mtx);
525 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
526 		if (error) {
527 			if (NFS_ISV4(vp))
528 				(void) nfsrpc_close(vp, 0, ap->a_td);
529 			return (error);
530 		}
531 		mtx_lock(&np->n_mtx);
532 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
533 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
534 			if (vp->v_type == VDIR)
535 				np->n_direofoffset = 0;
536 			mtx_unlock(&np->n_mtx);
537 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
538 			if (error == EINTR || error == EIO) {
539 				if (NFS_ISV4(vp))
540 					(void) nfsrpc_close(vp, 0, ap->a_td);
541 				return (error);
542 			}
543 			mtx_lock(&np->n_mtx);
544 			np->n_mtime = vattr.va_mtime;
545 			if (NFS_ISV4(vp))
546 				np->n_change = vattr.va_filerev;
547 		}
548 	}
549 
550 	/*
551 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
552 	 */
553 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
554 	    (vp->v_type == VREG)) {
555 		if (np->n_directio_opens == 0) {
556 			mtx_unlock(&np->n_mtx);
557 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
558 			if (error) {
559 				if (NFS_ISV4(vp))
560 					(void) nfsrpc_close(vp, 0, ap->a_td);
561 				return (error);
562 			}
563 			mtx_lock(&np->n_mtx);
564 			np->n_flag |= NNONCACHE;
565 		}
566 		np->n_directio_opens++;
567 	}
568 	mtx_unlock(&np->n_mtx);
569 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
570 	return (0);
571 }
572 
573 /*
574  * nfs close vnode op
575  * What an NFS client should do upon close after writing is a debatable issue.
576  * Most NFS clients push delayed writes to the server upon close, basically for
577  * two reasons:
578  * 1 - So that any write errors may be reported back to the client process
579  *     doing the close system call. By far the two most likely errors are
580  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
581  * 2 - To put a worst case upper bound on cache inconsistency between
582  *     multiple clients for the file.
583  * There is also a consistency problem for Version 2 of the protocol w.r.t.
584  * not being able to tell if other clients are writing a file concurrently,
585  * since there is no way of knowing if the changed modify time in the reply
586  * is only due to the write for this client.
587  * (NFS Version 3 provides weak cache consistency data in the reply that
588  *  should be sufficient to detect and handle this case.)
589  *
590  * The current code does the following:
591  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
592  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
593  *                     or commit them (this satisfies 1 and 2 except for the
594  *                     case where the server crashes after this close but
595  *                     before the commit RPC, which is felt to be "good
596  *                     enough". Changing the last argument to ncl_flush() to
597  *                     a 1 would force a commit operation, if it is felt a
598  *                     commit is necessary now.
599  * for NFS Version 4 - flush the dirty buffers and commit them, if
600  *		       nfscl_mustflush() says this is necessary.
601  *                     It is necessary if there is no write delegation held,
602  *                     in order to satisfy open/close coherency.
603  *                     If the file isn't cached on local stable storage,
604  *                     it may be necessary in order to detect "out of space"
605  *                     errors from the server, if the write delegation
606  *                     issued by the server doesn't allow the file to grow.
607  */
608 /* ARGSUSED */
609 static int
610 nfs_close(struct vop_close_args *ap)
611 {
612 	struct vnode *vp = ap->a_vp;
613 	struct nfsnode *np = VTONFS(vp);
614 	struct nfsvattr nfsva;
615 	struct ucred *cred;
616 	int error = 0, ret, localcred = 0;
617 	int fmode = ap->a_fflag;
618 
619 	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
620 		return (0);
621 	/*
622 	 * During shutdown, a_cred isn't valid, so just use root.
623 	 */
624 	if (ap->a_cred == NOCRED) {
625 		cred = newnfs_getcred();
626 		localcred = 1;
627 	} else {
628 		cred = ap->a_cred;
629 	}
630 	if (vp->v_type == VREG) {
631 	    /*
632 	     * Examine and clean dirty pages, regardless of NMODIFIED.
633 	     * This closes a major hole in close-to-open consistency.
634 	     * We want to push out all dirty pages (and buffers) on
635 	     * close, regardless of whether they were dirtied by
636 	     * mmap'ed writes or via write().
637 	     */
638 	    if (nfs_clean_pages_on_close && vp->v_object) {
639 		VM_OBJECT_LOCK(vp->v_object);
640 		vm_object_page_clean(vp->v_object, 0, 0, 0);
641 		VM_OBJECT_UNLOCK(vp->v_object);
642 	    }
643 	    mtx_lock(&np->n_mtx);
644 	    if (np->n_flag & NMODIFIED) {
645 		mtx_unlock(&np->n_mtx);
646 		if (NFS_ISV3(vp)) {
647 		    /*
648 		     * Under NFSv3 we have dirty buffers to dispose of.  We
649 		     * must flush them to the NFS server.  We have the option
650 		     * of waiting all the way through the commit rpc or just
651 		     * waiting for the initial write.  The default is to only
652 		     * wait through the initial write so the data is in the
653 		     * server's cache, which is roughly similar to the state
654 		     * a standard disk subsystem leaves the file in on close().
655 		     *
656 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
657 		     * potential races with other processes, and certainly
658 		     * cannot clear it if we don't commit.
659 		     * These races occur when there is no longer the old
660 		     * traditional vnode locking implemented for Vnode Ops.
661 		     */
662 		    int cm = newnfs_commit_on_close ? 1 : 0;
663 		    error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
664 		    /* np->n_flag &= ~NMODIFIED; */
665 		} else if (NFS_ISV4(vp)) {
666 			if (nfscl_mustflush(vp) != 0) {
667 				int cm = newnfs_commit_on_close ? 1 : 0;
668 				error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
669 				    cm, 0);
670 				/*
671 				 * as above w.r.t races when clearing
672 				 * NMODIFIED.
673 				 * np->n_flag &= ~NMODIFIED;
674 				 */
675 			}
676 		} else
677 		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
678 		mtx_lock(&np->n_mtx);
679 	    }
680  	    /*
681  	     * Invalidate the attribute cache in all cases.
682  	     * An open is going to fetch fresh attrs any way, other procs
683  	     * on this node that have file open will be forced to do an
684  	     * otw attr fetch, but this is safe.
685 	     * --> A user found that their RPC count dropped by 20% when
686 	     *     this was commented out and I can't see any requirement
687 	     *     for it, so I've disabled it when negative lookups are
688 	     *     enabled. (What does this have to do with negative lookup
689 	     *     caching? Well nothing, except it was reported by the
690 	     *     same user that needed negative lookup caching and I wanted
691 	     *     there to be a way to disable it to see if it
692 	     *     is the cause of some caching/coherency issue that might
693 	     *     crop up.)
694  	     */
695 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
696 		    np->n_attrstamp = 0;
697 	    if (np->n_flag & NWRITEERR) {
698 		np->n_flag &= ~NWRITEERR;
699 		error = np->n_error;
700 	    }
701 	    mtx_unlock(&np->n_mtx);
702 	}
703 
704 	if (NFS_ISV4(vp)) {
705 		/*
706 		 * Get attributes so "change" is up to date.
707 		 */
708 		if (error == 0 && nfscl_mustflush(vp) != 0) {
709 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
710 			    NULL);
711 			if (!ret) {
712 				np->n_change = nfsva.na_filerev;
713 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
714 				    NULL, 0, 0);
715 			}
716 		}
717 
718 		/*
719 		 * and do the close.
720 		 */
721 		ret = nfsrpc_close(vp, 0, ap->a_td);
722 		if (!error && ret)
723 			error = ret;
724 		if (error)
725 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
726 			    (gid_t)0);
727 	}
728 	if (newnfs_directio_enable)
729 		KASSERT((np->n_directio_asyncwr == 0),
730 			("nfs_close: dirty unflushed (%d) directio buffers\n",
731 			 np->n_directio_asyncwr));
732 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
733 		mtx_lock(&np->n_mtx);
734 		KASSERT((np->n_directio_opens > 0),
735 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
736 		np->n_directio_opens--;
737 		if (np->n_directio_opens == 0)
738 			np->n_flag &= ~NNONCACHE;
739 		mtx_unlock(&np->n_mtx);
740 	}
741 	if (localcred)
742 		NFSFREECRED(cred);
743 	return (error);
744 }
745 
746 /*
747  * nfs getattr call from vfs.
748  */
749 static int
750 nfs_getattr(struct vop_getattr_args *ap)
751 {
752 	struct vnode *vp = ap->a_vp;
753 	struct thread *td = curthread;	/* XXX */
754 	struct nfsnode *np = VTONFS(vp);
755 	int error = 0;
756 	struct nfsvattr nfsva;
757 	struct vattr *vap = ap->a_vap;
758 	struct vattr vattr;
759 
760 	/*
761 	 * Update local times for special files.
762 	 */
763 	mtx_lock(&np->n_mtx);
764 	if (np->n_flag & (NACC | NUPD))
765 		np->n_flag |= NCHG;
766 	mtx_unlock(&np->n_mtx);
767 	/*
768 	 * First look in the cache.
769 	 */
770 	if (ncl_getattrcache(vp, &vattr) == 0) {
771 		vap->va_type = vattr.va_type;
772 		vap->va_mode = vattr.va_mode;
773 		vap->va_nlink = vattr.va_nlink;
774 		vap->va_uid = vattr.va_uid;
775 		vap->va_gid = vattr.va_gid;
776 		vap->va_fsid = vattr.va_fsid;
777 		vap->va_fileid = vattr.va_fileid;
778 		vap->va_size = vattr.va_size;
779 		vap->va_blocksize = vattr.va_blocksize;
780 		vap->va_atime = vattr.va_atime;
781 		vap->va_mtime = vattr.va_mtime;
782 		vap->va_ctime = vattr.va_ctime;
783 		vap->va_gen = vattr.va_gen;
784 		vap->va_flags = vattr.va_flags;
785 		vap->va_rdev = vattr.va_rdev;
786 		vap->va_bytes = vattr.va_bytes;
787 		vap->va_filerev = vattr.va_filerev;
788 		/*
789 		 * Get the local modify time for the case of a write
790 		 * delegation.
791 		 */
792 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
793 		return (0);
794 	}
795 
796 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
797 	    nfsaccess_cache_timeout > 0) {
798 		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
799 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
800 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
801 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
802 			return (0);
803 		}
804 	}
805 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
806 	if (!error)
807 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
808 	if (!error) {
809 		/*
810 		 * Get the local modify time for the case of a write
811 		 * delegation.
812 		 */
813 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
814 	} else if (NFS_ISV4(vp)) {
815 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
816 	}
817 	return (error);
818 }
819 
820 /*
821  * nfs setattr call.
822  */
823 static int
824 nfs_setattr(struct vop_setattr_args *ap)
825 {
826 	struct vnode *vp = ap->a_vp;
827 	struct nfsnode *np = VTONFS(vp);
828 	struct thread *td = curthread;	/* XXX */
829 	struct vattr *vap = ap->a_vap;
830 	int error = 0;
831 	u_quad_t tsize;
832 
833 #ifndef nolint
834 	tsize = (u_quad_t)0;
835 #endif
836 
837 	/*
838 	 * Setting of flags and marking of atimes are not supported.
839 	 */
840 	if (vap->va_flags != VNOVAL)
841 		return (EOPNOTSUPP);
842 
843 	/*
844 	 * Disallow write attempts if the filesystem is mounted read-only.
845 	 */
846   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
847 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
848 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
849 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
850 		return (EROFS);
851 	if (vap->va_size != VNOVAL) {
852  		switch (vp->v_type) {
853  		case VDIR:
854  			return (EISDIR);
855  		case VCHR:
856  		case VBLK:
857  		case VSOCK:
858  		case VFIFO:
859 			if (vap->va_mtime.tv_sec == VNOVAL &&
860 			    vap->va_atime.tv_sec == VNOVAL &&
861 			    vap->va_mode == (mode_t)VNOVAL &&
862 			    vap->va_uid == (uid_t)VNOVAL &&
863 			    vap->va_gid == (gid_t)VNOVAL)
864 				return (0);
865  			vap->va_size = VNOVAL;
866  			break;
867  		default:
868 			/*
869 			 * Disallow write attempts if the filesystem is
870 			 * mounted read-only.
871 			 */
872 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
873 				return (EROFS);
874 			/*
875 			 *  We run vnode_pager_setsize() early (why?),
876 			 * we must set np->n_size now to avoid vinvalbuf
877 			 * V_SAVE races that might setsize a lower
878 			 * value.
879 			 */
880 			mtx_lock(&np->n_mtx);
881 			tsize = np->n_size;
882 			mtx_unlock(&np->n_mtx);
883 			error = ncl_meta_setsize(vp, ap->a_cred, td,
884 			    vap->va_size);
885 			mtx_lock(&np->n_mtx);
886  			if (np->n_flag & NMODIFIED) {
887 			    tsize = np->n_size;
888 			    mtx_unlock(&np->n_mtx);
889  			    if (vap->va_size == 0)
890  				error = ncl_vinvalbuf(vp, 0, td, 1);
891  			    else
892  				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
893  			    if (error) {
894 				vnode_pager_setsize(vp, tsize);
895 				return (error);
896 			    }
897 			    /*
898 			     * Call nfscl_delegmodtime() to set the modify time
899 			     * locally, as required.
900 			     */
901 			    nfscl_delegmodtime(vp);
902  			} else
903 			    mtx_unlock(&np->n_mtx);
904 			/*
905 			 * np->n_size has already been set to vap->va_size
906 			 * in ncl_meta_setsize(). We must set it again since
907 			 * nfs_loadattrcache() could be called through
908 			 * ncl_meta_setsize() and could modify np->n_size.
909 			 */
910 			mtx_lock(&np->n_mtx);
911  			np->n_vattr.na_size = np->n_size = vap->va_size;
912 			mtx_unlock(&np->n_mtx);
913   		};
914   	} else {
915 		mtx_lock(&np->n_mtx);
916 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
917 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
918 			mtx_unlock(&np->n_mtx);
919 			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
920 			    (error == EINTR || error == EIO))
921 				return (error);
922 		} else
923 			mtx_unlock(&np->n_mtx);
924 	}
925 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
926 	if (error && vap->va_size != VNOVAL) {
927 		mtx_lock(&np->n_mtx);
928 		np->n_size = np->n_vattr.na_size = tsize;
929 		vnode_pager_setsize(vp, tsize);
930 		mtx_unlock(&np->n_mtx);
931 	}
932 	return (error);
933 }
934 
935 /*
936  * Do an nfs setattr rpc.
937  */
938 static int
939 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
940     struct thread *td)
941 {
942 	struct nfsnode *np = VTONFS(vp);
943 	int error, ret, attrflag, i;
944 	struct nfsvattr nfsva;
945 
946 	if (NFS_ISV34(vp)) {
947 		mtx_lock(&np->n_mtx);
948 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
949 			np->n_accesscache[i].stamp = 0;
950 		np->n_flag |= NDELEGMOD;
951 		mtx_unlock(&np->n_mtx);
952 	}
953 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
954 	    NULL);
955 	if (attrflag) {
956 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
957 		if (ret && !error)
958 			error = ret;
959 	}
960 	if (error && NFS_ISV4(vp))
961 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
962 	return (error);
963 }
964 
965 /*
966  * nfs lookup call, one step at a time...
967  * First look in cache
968  * If not found, unlock the directory nfsnode and do the rpc
969  */
970 static int
971 nfs_lookup(struct vop_lookup_args *ap)
972 {
973 	struct componentname *cnp = ap->a_cnp;
974 	struct vnode *dvp = ap->a_dvp;
975 	struct vnode **vpp = ap->a_vpp;
976 	struct mount *mp = dvp->v_mount;
977 	int flags = cnp->cn_flags;
978 	struct vnode *newvp;
979 	struct nfsmount *nmp;
980 	struct nfsnode *np, *newnp;
981 	int error = 0, attrflag, dattrflag, ltype;
982 	struct thread *td = cnp->cn_thread;
983 	struct nfsfh *nfhp;
984 	struct nfsvattr dnfsva, nfsva;
985 	struct vattr vattr;
986 	struct timespec dmtime;
987 
988 	*vpp = NULLVP;
989 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
990 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
991 		return (EROFS);
992 	if (dvp->v_type != VDIR)
993 		return (ENOTDIR);
994 	nmp = VFSTONFS(mp);
995 	np = VTONFS(dvp);
996 
997 	/* For NFSv4, wait until any remove is done. */
998 	mtx_lock(&np->n_mtx);
999 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1000 		np->n_flag |= NREMOVEWANT;
1001 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1002 	}
1003 	mtx_unlock(&np->n_mtx);
1004 
1005 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1006 		return (error);
1007 	error = cache_lookup(dvp, vpp, cnp);
1008 	if (error > 0 && error != ENOENT)
1009 		return (error);
1010 	if (error == -1) {
1011 		/*
1012 		 * We only accept a positive hit in the cache if the
1013 		 * change time of the file matches our cached copy.
1014 		 * Otherwise, we discard the cache entry and fallback
1015 		 * to doing a lookup RPC.
1016 		 *
1017 		 * To better handle stale file handles and attributes,
1018 		 * clear the attribute cache of this node if it is a
1019 		 * leaf component, part of an open() call, and not
1020 		 * locally modified before fetching the attributes.
1021 		 * This should allow stale file handles to be detected
1022 		 * here where we can fall back to a LOOKUP RPC to
1023 		 * recover rather than having nfs_open() detect the
1024 		 * stale file handle and failing open(2) with ESTALE.
1025 		 */
1026 		newvp = *vpp;
1027 		newnp = VTONFS(newvp);
1028 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1029 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1030 		    !(newnp->n_flag & NMODIFIED)) {
1031 			mtx_lock(&newnp->n_mtx);
1032 			newnp->n_attrstamp = 0;
1033 			mtx_unlock(&newnp->n_mtx);
1034 		}
1035 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1036 		    (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1037 		    timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) {
1038 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1039 			if (cnp->cn_nameiop != LOOKUP &&
1040 			    (flags & ISLASTCN))
1041 				cnp->cn_flags |= SAVENAME;
1042 			return (0);
1043 		}
1044 		cache_purge(newvp);
1045 		if (dvp != newvp)
1046 			vput(newvp);
1047 		else
1048 			vrele(newvp);
1049 		*vpp = NULLVP;
1050 	} else if (error == ENOENT) {
1051 		if (dvp->v_iflag & VI_DOOMED)
1052 			return (ENOENT);
1053 		/*
1054 		 * We only accept a negative hit in the cache if the
1055 		 * modification time of the parent directory matches
1056 		 * our cached copy.  Otherwise, we discard all of the
1057 		 * negative cache entries for this directory. We also
1058 		 * only trust -ve cache entries for less than
1059 		 * nm_negative_namecache_timeout seconds.
1060 		 */
1061 		if ((u_int)(ticks - np->n_dmtime_ticks) <
1062 		    (nmp->nm_negnametimeo * hz) &&
1063 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1064 		    timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
1065 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1066 			return (ENOENT);
1067 		}
1068 		cache_purge_negative(dvp);
1069 		mtx_lock(&np->n_mtx);
1070 		timespecclear(&np->n_dmtime);
1071 		mtx_unlock(&np->n_mtx);
1072 	}
1073 
1074 	/*
1075 	 * Cache the modification time of the parent directory in case
1076 	 * the lookup fails and results in adding the first negative
1077 	 * name cache entry for the directory.  Since this is reading
1078 	 * a single time_t, don't bother with locking.  The
1079 	 * modification time may be a bit stale, but it must be read
1080 	 * before performing the lookup RPC to prevent a race where
1081 	 * another lookup updates the timestamp on the directory after
1082 	 * the lookup RPC has been performed on the server but before
1083 	 * n_dmtime is set at the end of this function.
1084 	 */
1085 	dmtime = np->n_vattr.na_mtime;
1086 	error = 0;
1087 	newvp = NULLVP;
1088 	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1089 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1090 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1091 	    NULL);
1092 	if (dattrflag)
1093 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1094 	if (error) {
1095 		if (newvp != NULLVP) {
1096 			vput(newvp);
1097 			*vpp = NULLVP;
1098 		}
1099 
1100 		if (error != ENOENT) {
1101 			if (NFS_ISV4(dvp))
1102 				error = nfscl_maperr(td, error, (uid_t)0,
1103 				    (gid_t)0);
1104 			return (error);
1105 		}
1106 
1107 		/* The requested file was not found. */
1108 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1109 		    (flags & ISLASTCN)) {
1110 			/*
1111 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1112 			 * VWRITE) here instead of just checking
1113 			 * MNT_RDONLY.
1114 			 */
1115 			if (mp->mnt_flag & MNT_RDONLY)
1116 				return (EROFS);
1117 			cnp->cn_flags |= SAVENAME;
1118 			return (EJUSTRETURN);
1119 		}
1120 
1121 		if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1122 			/*
1123 			 * Maintain n_dmtime as the modification time
1124 			 * of the parent directory when the oldest -ve
1125 			 * name cache entry for this directory was
1126 			 * added.  If a -ve cache entry has already
1127 			 * been added with a newer modification time
1128 			 * by a concurrent lookup, then don't bother
1129 			 * adding a cache entry.  The modification
1130 			 * time of the directory might have changed
1131 			 * due to the file this lookup failed to find
1132 			 * being created.  In that case a subsequent
1133 			 * lookup would incorrectly use the entry
1134 			 * added here instead of doing an extra
1135 			 * lookup.
1136 			 */
1137 			mtx_lock(&np->n_mtx);
1138 			if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1139 				if (!timespecisset(&np->n_dmtime)) {
1140 					np->n_dmtime = dmtime;
1141 					np->n_dmtime_ticks = ticks;
1142 				}
1143 				mtx_unlock(&np->n_mtx);
1144 				cache_enter(dvp, NULL, cnp);
1145 			} else
1146 				mtx_unlock(&np->n_mtx);
1147 		}
1148 		return (ENOENT);
1149 	}
1150 
1151 	/*
1152 	 * Handle RENAME case...
1153 	 */
1154 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1155 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1156 			FREE((caddr_t)nfhp, M_NFSFH);
1157 			return (EISDIR);
1158 		}
1159 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1160 		    LK_EXCLUSIVE);
1161 		if (error)
1162 			return (error);
1163 		newvp = NFSTOV(np);
1164 		if (attrflag)
1165 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1166 			    0, 1);
1167 		*vpp = newvp;
1168 		cnp->cn_flags |= SAVENAME;
1169 		return (0);
1170 	}
1171 
1172 	if (flags & ISDOTDOT) {
1173 		ltype = VOP_ISLOCKED(dvp);
1174 		error = vfs_busy(mp, MBF_NOWAIT);
1175 		if (error != 0) {
1176 			vfs_ref(mp);
1177 			VOP_UNLOCK(dvp, 0);
1178 			error = vfs_busy(mp, 0);
1179 			vn_lock(dvp, ltype | LK_RETRY);
1180 			vfs_rel(mp);
1181 			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1182 				vfs_unbusy(mp);
1183 				error = ENOENT;
1184 			}
1185 			if (error != 0)
1186 				return (error);
1187 		}
1188 		VOP_UNLOCK(dvp, 0);
1189 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1190 		    cnp->cn_lkflags);
1191 		if (error == 0)
1192 			newvp = NFSTOV(np);
1193 		vfs_unbusy(mp);
1194 		if (newvp != dvp)
1195 			vn_lock(dvp, ltype | LK_RETRY);
1196 		if (dvp->v_iflag & VI_DOOMED) {
1197 			if (error == 0) {
1198 				if (newvp == dvp)
1199 					vrele(newvp);
1200 				else
1201 					vput(newvp);
1202 			}
1203 			error = ENOENT;
1204 		}
1205 		if (error != 0)
1206 			return (error);
1207 		if (attrflag)
1208 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1209 			    0, 1);
1210 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1211 		FREE((caddr_t)nfhp, M_NFSFH);
1212 		VREF(dvp);
1213 		newvp = dvp;
1214 		if (attrflag)
1215 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1216 			    0, 1);
1217 	} else {
1218 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1219 		    cnp->cn_lkflags);
1220 		if (error)
1221 			return (error);
1222 		newvp = NFSTOV(np);
1223 		if (attrflag)
1224 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1225 			    0, 1);
1226 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1227 		    !(np->n_flag & NMODIFIED)) {
1228 			/*
1229 			 * Flush the attribute cache when opening a
1230 			 * leaf node to ensure that fresh attributes
1231 			 * are fetched in nfs_open() since we did not
1232 			 * fetch attributes from the LOOKUP reply.
1233 			 */
1234 			mtx_lock(&np->n_mtx);
1235 			np->n_attrstamp = 0;
1236 			mtx_unlock(&np->n_mtx);
1237 		}
1238 	}
1239 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1240 		cnp->cn_flags |= SAVENAME;
1241 	if ((cnp->cn_flags & MAKEENTRY) &&
1242 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1243 		np->n_ctime = np->n_vattr.na_vattr.va_ctime;
1244 		cache_enter(dvp, newvp, cnp);
1245 	}
1246 	*vpp = newvp;
1247 	return (0);
1248 }
1249 
1250 /*
1251  * nfs read call.
1252  * Just call ncl_bioread() to do the work.
1253  */
1254 static int
1255 nfs_read(struct vop_read_args *ap)
1256 {
1257 	struct vnode *vp = ap->a_vp;
1258 
1259 	switch (vp->v_type) {
1260 	case VREG:
1261 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1262 	case VDIR:
1263 		return (EISDIR);
1264 	default:
1265 		return (EOPNOTSUPP);
1266 	}
1267 }
1268 
1269 /*
1270  * nfs readlink call
1271  */
1272 static int
1273 nfs_readlink(struct vop_readlink_args *ap)
1274 {
1275 	struct vnode *vp = ap->a_vp;
1276 
1277 	if (vp->v_type != VLNK)
1278 		return (EINVAL);
1279 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1280 }
1281 
1282 /*
1283  * Do a readlink rpc.
1284  * Called by ncl_doio() from below the buffer cache.
1285  */
1286 int
1287 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1288 {
1289 	int error, ret, attrflag;
1290 	struct nfsvattr nfsva;
1291 
1292 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1293 	    &attrflag, NULL);
1294 	if (attrflag) {
1295 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1296 		if (ret && !error)
1297 			error = ret;
1298 	}
1299 	if (error && NFS_ISV4(vp))
1300 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1301 	return (error);
1302 }
1303 
1304 /*
1305  * nfs read rpc call
1306  * Ditto above
1307  */
1308 int
1309 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1310 {
1311 	int error, ret, attrflag;
1312 	struct nfsvattr nfsva;
1313 
1314 	error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1315 	    NULL);
1316 	if (attrflag) {
1317 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1318 		if (ret && !error)
1319 			error = ret;
1320 	}
1321 	if (error && NFS_ISV4(vp))
1322 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1323 	return (error);
1324 }
1325 
1326 /*
1327  * nfs write call
1328  */
1329 int
1330 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1331     int *iomode, int *must_commit, int called_from_strategy)
1332 {
1333 	struct nfsvattr nfsva;
1334 	int error = 0, attrflag, ret;
1335 
1336 	error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1337 	    uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1338 	if (attrflag) {
1339 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1340 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1341 			    1);
1342 		else
1343 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1344 			    1);
1345 		if (ret && !error)
1346 			error = ret;
1347 	}
1348 	if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1349 		*iomode = NFSWRITE_FILESYNC;
1350 	if (error && NFS_ISV4(vp))
1351 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1352 	return (error);
1353 }
1354 
1355 /*
1356  * nfs mknod rpc
1357  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1358  * mode set to specify the file type and the size field for rdev.
1359  */
1360 static int
1361 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1362     struct vattr *vap)
1363 {
1364 	struct nfsvattr nfsva, dnfsva;
1365 	struct vnode *newvp = NULL;
1366 	struct nfsnode *np = NULL, *dnp;
1367 	struct nfsfh *nfhp;
1368 	struct vattr vattr;
1369 	int error = 0, attrflag, dattrflag;
1370 	u_int32_t rdev;
1371 
1372 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1373 		rdev = vap->va_rdev;
1374 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1375 		rdev = 0xffffffff;
1376 	else
1377 		return (EOPNOTSUPP);
1378 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1379 		return (error);
1380 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1381 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1382 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1383 	if (!error) {
1384 		if (!nfhp)
1385 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1386 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1387 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1388 			    NULL);
1389 		if (nfhp)
1390 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1391 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1392 	}
1393 	if (dattrflag)
1394 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1395 	if (!error) {
1396 		newvp = NFSTOV(np);
1397 		if (attrflag != 0) {
1398 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1399 			    0, 1);
1400 			if (error != 0)
1401 				vput(newvp);
1402 		}
1403 	}
1404 	if (!error) {
1405 		if ((cnp->cn_flags & MAKEENTRY))
1406 			cache_enter(dvp, newvp, cnp);
1407 		*vpp = newvp;
1408 	} else if (NFS_ISV4(dvp)) {
1409 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1410 		    vap->va_gid);
1411 	}
1412 	dnp = VTONFS(dvp);
1413 	mtx_lock(&dnp->n_mtx);
1414 	dnp->n_flag |= NMODIFIED;
1415 	if (!dattrflag)
1416 		dnp->n_attrstamp = 0;
1417 	mtx_unlock(&dnp->n_mtx);
1418 	return (error);
1419 }
1420 
1421 /*
1422  * nfs mknod vop
1423  * just call nfs_mknodrpc() to do the work.
1424  */
1425 /* ARGSUSED */
1426 static int
1427 nfs_mknod(struct vop_mknod_args *ap)
1428 {
1429 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1430 }
1431 
1432 static struct mtx nfs_cverf_mtx;
1433 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1434     MTX_DEF);
1435 
1436 static nfsquad_t
1437 nfs_get_cverf(void)
1438 {
1439 	static nfsquad_t cverf;
1440 	nfsquad_t ret;
1441 	static int cverf_initialized = 0;
1442 
1443 	mtx_lock(&nfs_cverf_mtx);
1444 	if (cverf_initialized == 0) {
1445 		cverf.lval[0] = arc4random();
1446 		cverf.lval[1] = arc4random();
1447 		cverf_initialized = 1;
1448 	} else
1449 		cverf.qval++;
1450 	ret = cverf;
1451 	mtx_unlock(&nfs_cverf_mtx);
1452 
1453 	return (ret);
1454 }
1455 
1456 /*
1457  * nfs file create call
1458  */
1459 static int
1460 nfs_create(struct vop_create_args *ap)
1461 {
1462 	struct vnode *dvp = ap->a_dvp;
1463 	struct vattr *vap = ap->a_vap;
1464 	struct componentname *cnp = ap->a_cnp;
1465 	struct nfsnode *np = NULL, *dnp;
1466 	struct vnode *newvp = NULL;
1467 	struct nfsmount *nmp;
1468 	struct nfsvattr dnfsva, nfsva;
1469 	struct nfsfh *nfhp;
1470 	nfsquad_t cverf;
1471 	int error = 0, attrflag, dattrflag, fmode = 0;
1472 	struct vattr vattr;
1473 
1474 	/*
1475 	 * Oops, not for me..
1476 	 */
1477 	if (vap->va_type == VSOCK)
1478 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1479 
1480 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1481 		return (error);
1482 	if (vap->va_vaflags & VA_EXCLUSIVE)
1483 		fmode |= O_EXCL;
1484 	dnp = VTONFS(dvp);
1485 	nmp = VFSTONFS(vnode_mount(dvp));
1486 again:
1487 	/* For NFSv4, wait until any remove is done. */
1488 	mtx_lock(&dnp->n_mtx);
1489 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1490 		dnp->n_flag |= NREMOVEWANT;
1491 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1492 	}
1493 	mtx_unlock(&dnp->n_mtx);
1494 
1495 	cverf = nfs_get_cverf();
1496 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1497 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1498 	    &nfhp, &attrflag, &dattrflag, NULL);
1499 	if (!error) {
1500 		if (nfhp == NULL)
1501 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1502 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1503 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1504 			    NULL);
1505 		if (nfhp != NULL)
1506 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1507 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1508 	}
1509 	if (dattrflag)
1510 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1511 	if (!error) {
1512 		newvp = NFSTOV(np);
1513 		if (attrflag)
1514 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1515 			    0, 1);
1516 	}
1517 	if (error) {
1518 		if (newvp != NULL) {
1519 			vput(newvp);
1520 			newvp = NULL;
1521 		}
1522 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1523 		    error == NFSERR_NOTSUPP) {
1524 			fmode &= ~O_EXCL;
1525 			goto again;
1526 		}
1527 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1528 		if (nfscl_checksattr(vap, &nfsva)) {
1529 			/*
1530 			 * We are normally called with only a partially
1531 			 * initialized VAP. Since the NFSv3 spec says that
1532 			 * the server may use the file attributes to
1533 			 * store the verifier, the spec requires us to do a
1534 			 * SETATTR RPC. FreeBSD servers store the verifier in
1535 			 * atime, but we can't really assume that all servers
1536 			 * will so we ensure that our SETATTR sets both atime
1537 			 * and mtime.
1538 			 */
1539 			if (vap->va_mtime.tv_sec == VNOVAL)
1540 				vfs_timestamp(&vap->va_mtime);
1541 			if (vap->va_atime.tv_sec == VNOVAL)
1542 				vap->va_atime = vap->va_mtime;
1543 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1544 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1545 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1546 			    vap->va_gid != (gid_t)VNOVAL)) {
1547 				/* try again without setting uid/gid */
1548 				vap->va_uid = (uid_t)VNOVAL;
1549 				vap->va_gid = (uid_t)VNOVAL;
1550 				error = nfsrpc_setattr(newvp, vap, NULL,
1551 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1552 				    &attrflag, NULL);
1553 			}
1554 			if (attrflag)
1555 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1556 				    NULL, 0, 1);
1557 		}
1558 	}
1559 	if (!error) {
1560 		if (cnp->cn_flags & MAKEENTRY)
1561 			cache_enter(dvp, newvp, cnp);
1562 		*ap->a_vpp = newvp;
1563 	} else if (NFS_ISV4(dvp)) {
1564 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1565 		    vap->va_gid);
1566 	}
1567 	mtx_lock(&dnp->n_mtx);
1568 	dnp->n_flag |= NMODIFIED;
1569 	if (!dattrflag)
1570 		dnp->n_attrstamp = 0;
1571 	mtx_unlock(&dnp->n_mtx);
1572 	return (error);
1573 }
1574 
1575 /*
1576  * nfs file remove call
1577  * To try and make nfs semantics closer to ufs semantics, a file that has
1578  * other processes using the vnode is renamed instead of removed and then
1579  * removed later on the last close.
1580  * - If v_usecount > 1
1581  *	  If a rename is not already in the works
1582  *	     call nfs_sillyrename() to set it up
1583  *     else
1584  *	  do the remove rpc
1585  */
1586 static int
1587 nfs_remove(struct vop_remove_args *ap)
1588 {
1589 	struct vnode *vp = ap->a_vp;
1590 	struct vnode *dvp = ap->a_dvp;
1591 	struct componentname *cnp = ap->a_cnp;
1592 	struct nfsnode *np = VTONFS(vp);
1593 	int error = 0;
1594 	struct vattr vattr;
1595 
1596 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1597 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1598 	if (vp->v_type == VDIR)
1599 		error = EPERM;
1600 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1601 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1602 	    vattr.va_nlink > 1)) {
1603 		/*
1604 		 * Purge the name cache so that the chance of a lookup for
1605 		 * the name succeeding while the remove is in progress is
1606 		 * minimized. Without node locking it can still happen, such
1607 		 * that an I/O op returns ESTALE, but since you get this if
1608 		 * another host removes the file..
1609 		 */
1610 		cache_purge(vp);
1611 		/*
1612 		 * throw away biocache buffers, mainly to avoid
1613 		 * unnecessary delayed writes later.
1614 		 */
1615 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1616 		/* Do the rpc */
1617 		if (error != EINTR && error != EIO)
1618 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1619 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1620 		/*
1621 		 * Kludge City: If the first reply to the remove rpc is lost..
1622 		 *   the reply to the retransmitted request will be ENOENT
1623 		 *   since the file was in fact removed
1624 		 *   Therefore, we cheat and return success.
1625 		 */
1626 		if (error == ENOENT)
1627 			error = 0;
1628 	} else if (!np->n_sillyrename)
1629 		error = nfs_sillyrename(dvp, vp, cnp);
1630 	mtx_lock(&np->n_mtx);
1631 	np->n_attrstamp = 0;
1632 	mtx_unlock(&np->n_mtx);
1633 	return (error);
1634 }
1635 
1636 /*
1637  * nfs file remove rpc called from nfs_inactive
1638  */
1639 int
1640 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1641 {
1642 	/*
1643 	 * Make sure that the directory vnode is still valid.
1644 	 * XXX we should lock sp->s_dvp here.
1645 	 */
1646 	if (sp->s_dvp->v_type == VBAD)
1647 		return (0);
1648 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1649 	    sp->s_cred, NULL));
1650 }
1651 
1652 /*
1653  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1654  */
1655 static int
1656 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1657     int namelen, struct ucred *cred, struct thread *td)
1658 {
1659 	struct nfsvattr dnfsva;
1660 	struct nfsnode *dnp = VTONFS(dvp);
1661 	int error = 0, dattrflag;
1662 
1663 	mtx_lock(&dnp->n_mtx);
1664 	dnp->n_flag |= NREMOVEINPROG;
1665 	mtx_unlock(&dnp->n_mtx);
1666 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1667 	    &dattrflag, NULL);
1668 	mtx_lock(&dnp->n_mtx);
1669 	if ((dnp->n_flag & NREMOVEWANT)) {
1670 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1671 		mtx_unlock(&dnp->n_mtx);
1672 		wakeup((caddr_t)dnp);
1673 	} else {
1674 		dnp->n_flag &= ~NREMOVEINPROG;
1675 		mtx_unlock(&dnp->n_mtx);
1676 	}
1677 	if (dattrflag)
1678 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1679 	mtx_lock(&dnp->n_mtx);
1680 	dnp->n_flag |= NMODIFIED;
1681 	if (!dattrflag)
1682 		dnp->n_attrstamp = 0;
1683 	mtx_unlock(&dnp->n_mtx);
1684 	if (error && NFS_ISV4(dvp))
1685 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1686 	return (error);
1687 }
1688 
1689 /*
1690  * nfs file rename call
1691  */
1692 static int
1693 nfs_rename(struct vop_rename_args *ap)
1694 {
1695 	struct vnode *fvp = ap->a_fvp;
1696 	struct vnode *tvp = ap->a_tvp;
1697 	struct vnode *fdvp = ap->a_fdvp;
1698 	struct vnode *tdvp = ap->a_tdvp;
1699 	struct componentname *tcnp = ap->a_tcnp;
1700 	struct componentname *fcnp = ap->a_fcnp;
1701 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1702 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1703 	struct nfsv4node *newv4 = NULL;
1704 	int error;
1705 
1706 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1707 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1708 	/* Check for cross-device rename */
1709 	if ((fvp->v_mount != tdvp->v_mount) ||
1710 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1711 		error = EXDEV;
1712 		goto out;
1713 	}
1714 
1715 	if (fvp == tvp) {
1716 		ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1717 		error = 0;
1718 		goto out;
1719 	}
1720 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1721 		goto out;
1722 
1723 	/*
1724 	 * We have to flush B_DELWRI data prior to renaming
1725 	 * the file.  If we don't, the delayed-write buffers
1726 	 * can be flushed out later after the file has gone stale
1727 	 * under NFSV3.  NFSV2 does not have this problem because
1728 	 * ( as far as I can tell ) it flushes dirty buffers more
1729 	 * often.
1730 	 *
1731 	 * Skip the rename operation if the fsync fails, this can happen
1732 	 * due to the server's volume being full, when we pushed out data
1733 	 * that was written back to our cache earlier. Not checking for
1734 	 * this condition can result in potential (silent) data loss.
1735 	 */
1736 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1737 	VOP_UNLOCK(fvp, 0);
1738 	if (!error && tvp)
1739 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1740 	if (error)
1741 		goto out;
1742 
1743 	/*
1744 	 * If the tvp exists and is in use, sillyrename it before doing the
1745 	 * rename of the new file over it.
1746 	 * XXX Can't sillyrename a directory.
1747 	 */
1748 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1749 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1750 		vput(tvp);
1751 		tvp = NULL;
1752 	}
1753 
1754 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1755 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1756 	    tcnp->cn_thread);
1757 
1758 	if (error == 0 && NFS_ISV4(tdvp)) {
1759 		/*
1760 		 * For NFSv4, check to see if it is the same name and
1761 		 * replace the name, if it is different.
1762 		 */
1763 		MALLOC(newv4, struct nfsv4node *,
1764 		    sizeof (struct nfsv4node) +
1765 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1766 		    M_NFSV4NODE, M_WAITOK);
1767 		mtx_lock(&tdnp->n_mtx);
1768 		mtx_lock(&fnp->n_mtx);
1769 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1770 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1771 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1772 		      tcnp->cn_namelen) ||
1773 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1774 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1775 			tdnp->n_fhp->nfh_len))) {
1776 #ifdef notdef
1777 { char nnn[100]; int nnnl;
1778 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1779 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1780 nnn[nnnl] = '\0';
1781 printf("ren replace=%s\n",nnn);
1782 }
1783 #endif
1784 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1785 			fnp->n_v4 = newv4;
1786 			newv4 = NULL;
1787 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1788 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1789 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1790 			    tdnp->n_fhp->nfh_len);
1791 			NFSBCOPY(tcnp->cn_nameptr,
1792 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1793 		}
1794 		mtx_unlock(&tdnp->n_mtx);
1795 		mtx_unlock(&fnp->n_mtx);
1796 		if (newv4 != NULL)
1797 			FREE((caddr_t)newv4, M_NFSV4NODE);
1798 	}
1799 
1800 	if (fvp->v_type == VDIR) {
1801 		if (tvp != NULL && tvp->v_type == VDIR)
1802 			cache_purge(tdvp);
1803 		cache_purge(fdvp);
1804 	}
1805 
1806 out:
1807 	if (tdvp == tvp)
1808 		vrele(tdvp);
1809 	else
1810 		vput(tdvp);
1811 	if (tvp)
1812 		vput(tvp);
1813 	vrele(fdvp);
1814 	vrele(fvp);
1815 	/*
1816 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1817 	 */
1818 	if (error == ENOENT)
1819 		error = 0;
1820 	return (error);
1821 }
1822 
1823 /*
1824  * nfs file rename rpc called from nfs_remove() above
1825  */
1826 static int
1827 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1828     struct sillyrename *sp)
1829 {
1830 
1831 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1832 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1833 	    scnp->cn_thread));
1834 }
1835 
1836 /*
1837  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1838  */
1839 static int
1840 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1841     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1842     int tnamelen, struct ucred *cred, struct thread *td)
1843 {
1844 	struct nfsvattr fnfsva, tnfsva;
1845 	struct nfsnode *fdnp = VTONFS(fdvp);
1846 	struct nfsnode *tdnp = VTONFS(tdvp);
1847 	int error = 0, fattrflag, tattrflag;
1848 
1849 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1850 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1851 	    &tattrflag, NULL, NULL);
1852 	mtx_lock(&fdnp->n_mtx);
1853 	fdnp->n_flag |= NMODIFIED;
1854 	if (fattrflag != 0) {
1855 		mtx_unlock(&fdnp->n_mtx);
1856 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1857 	} else {
1858 		fdnp->n_attrstamp = 0;
1859 		mtx_unlock(&fdnp->n_mtx);
1860 	}
1861 	mtx_lock(&tdnp->n_mtx);
1862 	tdnp->n_flag |= NMODIFIED;
1863 	if (tattrflag != 0) {
1864 		mtx_unlock(&tdnp->n_mtx);
1865 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1866 	} else {
1867 		tdnp->n_attrstamp = 0;
1868 		mtx_unlock(&tdnp->n_mtx);
1869 	}
1870 	if (error && NFS_ISV4(fdvp))
1871 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1872 	return (error);
1873 }
1874 
1875 /*
1876  * nfs hard link create call
1877  */
1878 static int
1879 nfs_link(struct vop_link_args *ap)
1880 {
1881 	struct vnode *vp = ap->a_vp;
1882 	struct vnode *tdvp = ap->a_tdvp;
1883 	struct componentname *cnp = ap->a_cnp;
1884 	struct nfsnode *np, *tdnp;
1885 	struct nfsvattr nfsva, dnfsva;
1886 	int error = 0, attrflag, dattrflag;
1887 
1888 	if (vp->v_mount != tdvp->v_mount) {
1889 		return (EXDEV);
1890 	}
1891 
1892 	/*
1893 	 * Push all writes to the server, so that the attribute cache
1894 	 * doesn't get "out of sync" with the server.
1895 	 * XXX There should be a better way!
1896 	 */
1897 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1898 
1899 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1900 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1901 	    &dattrflag, NULL);
1902 	tdnp = VTONFS(tdvp);
1903 	mtx_lock(&tdnp->n_mtx);
1904 	tdnp->n_flag |= NMODIFIED;
1905 	if (dattrflag != 0) {
1906 		mtx_unlock(&tdnp->n_mtx);
1907 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1908 	} else {
1909 		tdnp->n_attrstamp = 0;
1910 		mtx_unlock(&tdnp->n_mtx);
1911 	}
1912 	if (attrflag)
1913 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1914 	else {
1915 		np = VTONFS(vp);
1916 		mtx_lock(&np->n_mtx);
1917 		np->n_attrstamp = 0;
1918 		mtx_unlock(&np->n_mtx);
1919 	}
1920 	/*
1921 	 * If negative lookup caching is enabled, I might as well
1922 	 * add an entry for this node. Not necessary for correctness,
1923 	 * but if negative caching is enabled, then the system
1924 	 * must care about lookup caching hit rate, so...
1925 	 */
1926 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1927 	    (cnp->cn_flags & MAKEENTRY))
1928 		cache_enter(tdvp, vp, cnp);
1929 	if (error && NFS_ISV4(vp))
1930 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1931 		    (gid_t)0);
1932 	return (error);
1933 }
1934 
1935 /*
1936  * nfs symbolic link create call
1937  */
1938 static int
1939 nfs_symlink(struct vop_symlink_args *ap)
1940 {
1941 	struct vnode *dvp = ap->a_dvp;
1942 	struct vattr *vap = ap->a_vap;
1943 	struct componentname *cnp = ap->a_cnp;
1944 	struct nfsvattr nfsva, dnfsva;
1945 	struct nfsfh *nfhp;
1946 	struct nfsnode *np = NULL, *dnp;
1947 	struct vnode *newvp = NULL;
1948 	int error = 0, attrflag, dattrflag, ret;
1949 
1950 	vap->va_type = VLNK;
1951 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1952 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1953 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1954 	if (nfhp) {
1955 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1956 		    &np, NULL, LK_EXCLUSIVE);
1957 		if (!ret)
1958 			newvp = NFSTOV(np);
1959 		else if (!error)
1960 			error = ret;
1961 	}
1962 	if (newvp != NULL) {
1963 		if (attrflag)
1964 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1965 			    0, 1);
1966 	} else if (!error) {
1967 		/*
1968 		 * If we do not have an error and we could not extract the
1969 		 * newvp from the response due to the request being NFSv2, we
1970 		 * have to do a lookup in order to obtain a newvp to return.
1971 		 */
1972 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1973 		    cnp->cn_cred, cnp->cn_thread, &np);
1974 		if (!error)
1975 			newvp = NFSTOV(np);
1976 	}
1977 	if (error) {
1978 		if (newvp)
1979 			vput(newvp);
1980 		if (NFS_ISV4(dvp))
1981 			error = nfscl_maperr(cnp->cn_thread, error,
1982 			    vap->va_uid, vap->va_gid);
1983 	} else {
1984 		/*
1985 		 * If negative lookup caching is enabled, I might as well
1986 		 * add an entry for this node. Not necessary for correctness,
1987 		 * but if negative caching is enabled, then the system
1988 		 * must care about lookup caching hit rate, so...
1989 		 */
1990 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1991 		    (cnp->cn_flags & MAKEENTRY))
1992 			cache_enter(dvp, newvp, cnp);
1993 		*ap->a_vpp = newvp;
1994 	}
1995 
1996 	dnp = VTONFS(dvp);
1997 	mtx_lock(&dnp->n_mtx);
1998 	dnp->n_flag |= NMODIFIED;
1999 	if (dattrflag != 0) {
2000 		mtx_unlock(&dnp->n_mtx);
2001 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2002 	} else {
2003 		dnp->n_attrstamp = 0;
2004 		mtx_unlock(&dnp->n_mtx);
2005 	}
2006 	return (error);
2007 }
2008 
2009 /*
2010  * nfs make dir call
2011  */
2012 static int
2013 nfs_mkdir(struct vop_mkdir_args *ap)
2014 {
2015 	struct vnode *dvp = ap->a_dvp;
2016 	struct vattr *vap = ap->a_vap;
2017 	struct componentname *cnp = ap->a_cnp;
2018 	struct nfsnode *np = NULL, *dnp;
2019 	struct vnode *newvp = NULL;
2020 	struct vattr vattr;
2021 	struct nfsfh *nfhp;
2022 	struct nfsvattr nfsva, dnfsva;
2023 	int error = 0, attrflag, dattrflag, ret;
2024 
2025 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2026 		return (error);
2027 	vap->va_type = VDIR;
2028 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2029 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2030 	    &attrflag, &dattrflag, NULL);
2031 	dnp = VTONFS(dvp);
2032 	mtx_lock(&dnp->n_mtx);
2033 	dnp->n_flag |= NMODIFIED;
2034 	if (dattrflag != 0) {
2035 		mtx_unlock(&dnp->n_mtx);
2036 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2037 	} else {
2038 		dnp->n_attrstamp = 0;
2039 		mtx_unlock(&dnp->n_mtx);
2040 	}
2041 	if (nfhp) {
2042 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2043 		    &np, NULL, LK_EXCLUSIVE);
2044 		if (!ret) {
2045 			newvp = NFSTOV(np);
2046 			if (attrflag)
2047 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2048 				NULL, 0, 1);
2049 		} else if (!error)
2050 			error = ret;
2051 	}
2052 	if (!error && newvp == NULL) {
2053 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2054 		    cnp->cn_cred, cnp->cn_thread, &np);
2055 		if (!error) {
2056 			newvp = NFSTOV(np);
2057 			if (newvp->v_type != VDIR)
2058 				error = EEXIST;
2059 		}
2060 	}
2061 	if (error) {
2062 		if (newvp)
2063 			vput(newvp);
2064 		if (NFS_ISV4(dvp))
2065 			error = nfscl_maperr(cnp->cn_thread, error,
2066 			    vap->va_uid, vap->va_gid);
2067 	} else {
2068 		/*
2069 		 * If negative lookup caching is enabled, I might as well
2070 		 * add an entry for this node. Not necessary for correctness,
2071 		 * but if negative caching is enabled, then the system
2072 		 * must care about lookup caching hit rate, so...
2073 		 */
2074 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2075 		    (cnp->cn_flags & MAKEENTRY))
2076 			cache_enter(dvp, newvp, cnp);
2077 		*ap->a_vpp = newvp;
2078 	}
2079 	return (error);
2080 }
2081 
2082 /*
2083  * nfs remove directory call
2084  */
2085 static int
2086 nfs_rmdir(struct vop_rmdir_args *ap)
2087 {
2088 	struct vnode *vp = ap->a_vp;
2089 	struct vnode *dvp = ap->a_dvp;
2090 	struct componentname *cnp = ap->a_cnp;
2091 	struct nfsnode *dnp;
2092 	struct nfsvattr dnfsva;
2093 	int error, dattrflag;
2094 
2095 	if (dvp == vp)
2096 		return (EINVAL);
2097 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2098 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2099 	dnp = VTONFS(dvp);
2100 	mtx_lock(&dnp->n_mtx);
2101 	dnp->n_flag |= NMODIFIED;
2102 	if (dattrflag != 0) {
2103 		mtx_unlock(&dnp->n_mtx);
2104 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2105 	} else {
2106 		dnp->n_attrstamp = 0;
2107 		mtx_unlock(&dnp->n_mtx);
2108 	}
2109 
2110 	cache_purge(dvp);
2111 	cache_purge(vp);
2112 	if (error && NFS_ISV4(dvp))
2113 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2114 		    (gid_t)0);
2115 	/*
2116 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2117 	 */
2118 	if (error == ENOENT)
2119 		error = 0;
2120 	return (error);
2121 }
2122 
2123 /*
2124  * nfs readdir call
2125  */
2126 static int
2127 nfs_readdir(struct vop_readdir_args *ap)
2128 {
2129 	struct vnode *vp = ap->a_vp;
2130 	struct nfsnode *np = VTONFS(vp);
2131 	struct uio *uio = ap->a_uio;
2132 	int tresid, error = 0;
2133 	struct vattr vattr;
2134 
2135 	if (vp->v_type != VDIR)
2136 		return(EPERM);
2137 
2138 	/*
2139 	 * First, check for hit on the EOF offset cache
2140 	 */
2141 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2142 	    (np->n_flag & NMODIFIED) == 0) {
2143 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2144 			mtx_lock(&np->n_mtx);
2145 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2146 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2147 				mtx_unlock(&np->n_mtx);
2148 				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2149 				return (0);
2150 			} else
2151 				mtx_unlock(&np->n_mtx);
2152 		}
2153 	}
2154 
2155 	/*
2156 	 * Call ncl_bioread() to do the real work.
2157 	 */
2158 	tresid = uio->uio_resid;
2159 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2160 
2161 	if (!error && uio->uio_resid == tresid)
2162 		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2163 	return (error);
2164 }
2165 
2166 /*
2167  * Readdir rpc call.
2168  * Called from below the buffer cache by ncl_doio().
2169  */
2170 int
2171 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2172     struct thread *td)
2173 {
2174 	struct nfsvattr nfsva;
2175 	nfsuint64 *cookiep, cookie;
2176 	struct nfsnode *dnp = VTONFS(vp);
2177 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2178 	int error = 0, eof, attrflag;
2179 
2180 	KASSERT(uiop->uio_iovcnt == 1 &&
2181 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2182 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2183 	    ("nfs readdirrpc bad uio"));
2184 
2185 	/*
2186 	 * If there is no cookie, assume directory was stale.
2187 	 */
2188 	ncl_dircookie_lock(dnp);
2189 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2190 	if (cookiep) {
2191 		cookie = *cookiep;
2192 		ncl_dircookie_unlock(dnp);
2193 	} else {
2194 		ncl_dircookie_unlock(dnp);
2195 		return (NFSERR_BAD_COOKIE);
2196 	}
2197 
2198 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2199 		(void)ncl_fsinfo(nmp, vp, cred, td);
2200 
2201 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2202 	    &attrflag, &eof, NULL);
2203 	if (attrflag)
2204 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2205 
2206 	if (!error) {
2207 		/*
2208 		 * We are now either at the end of the directory or have filled
2209 		 * the block.
2210 		 */
2211 		if (eof)
2212 			dnp->n_direofoffset = uiop->uio_offset;
2213 		else {
2214 			if (uiop->uio_resid > 0)
2215 				ncl_printf("EEK! readdirrpc resid > 0\n");
2216 			ncl_dircookie_lock(dnp);
2217 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2218 			*cookiep = cookie;
2219 			ncl_dircookie_unlock(dnp);
2220 		}
2221 	} else if (NFS_ISV4(vp)) {
2222 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2223 	}
2224 	return (error);
2225 }
2226 
2227 /*
2228  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2229  */
2230 int
2231 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2232     struct thread *td)
2233 {
2234 	struct nfsvattr nfsva;
2235 	nfsuint64 *cookiep, cookie;
2236 	struct nfsnode *dnp = VTONFS(vp);
2237 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2238 	int error = 0, attrflag, eof;
2239 
2240 	KASSERT(uiop->uio_iovcnt == 1 &&
2241 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2242 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2243 	    ("nfs readdirplusrpc bad uio"));
2244 
2245 	/*
2246 	 * If there is no cookie, assume directory was stale.
2247 	 */
2248 	ncl_dircookie_lock(dnp);
2249 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2250 	if (cookiep) {
2251 		cookie = *cookiep;
2252 		ncl_dircookie_unlock(dnp);
2253 	} else {
2254 		ncl_dircookie_unlock(dnp);
2255 		return (NFSERR_BAD_COOKIE);
2256 	}
2257 
2258 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2259 		(void)ncl_fsinfo(nmp, vp, cred, td);
2260 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2261 	    &attrflag, &eof, NULL);
2262 	if (attrflag)
2263 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2264 
2265 	if (!error) {
2266 		/*
2267 		 * We are now either at end of the directory or have filled the
2268 		 * the block.
2269 		 */
2270 		if (eof)
2271 			dnp->n_direofoffset = uiop->uio_offset;
2272 		else {
2273 			if (uiop->uio_resid > 0)
2274 				ncl_printf("EEK! readdirplusrpc resid > 0\n");
2275 			ncl_dircookie_lock(dnp);
2276 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2277 			*cookiep = cookie;
2278 			ncl_dircookie_unlock(dnp);
2279 		}
2280 	} else if (NFS_ISV4(vp)) {
2281 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2282 	}
2283 	return (error);
2284 }
2285 
2286 /*
2287  * Silly rename. To make the NFS filesystem that is stateless look a little
2288  * more like the "ufs" a remove of an active vnode is translated to a rename
2289  * to a funny looking filename that is removed by nfs_inactive on the
2290  * nfsnode. There is the potential for another process on a different client
2291  * to create the same funny name between the nfs_lookitup() fails and the
2292  * nfs_rename() completes, but...
2293  */
2294 static int
2295 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2296 {
2297 	struct sillyrename *sp;
2298 	struct nfsnode *np;
2299 	int error;
2300 	short pid;
2301 	unsigned int lticks;
2302 
2303 	cache_purge(dvp);
2304 	np = VTONFS(vp);
2305 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2306 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2307 	    M_NEWNFSREQ, M_WAITOK);
2308 	sp->s_cred = crhold(cnp->cn_cred);
2309 	sp->s_dvp = dvp;
2310 	VREF(dvp);
2311 
2312 	/*
2313 	 * Fudge together a funny name.
2314 	 * Changing the format of the funny name to accomodate more
2315 	 * sillynames per directory.
2316 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2317 	 * CPU ticks since boot.
2318 	 */
2319 	pid = cnp->cn_thread->td_proc->p_pid;
2320 	lticks = (unsigned int)ticks;
2321 	for ( ; ; ) {
2322 		sp->s_namlen = sprintf(sp->s_name,
2323 				       ".nfs.%08x.%04x4.4", lticks,
2324 				       pid);
2325 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2326 				 cnp->cn_thread, NULL))
2327 			break;
2328 		lticks++;
2329 	}
2330 	error = nfs_renameit(dvp, vp, cnp, sp);
2331 	if (error)
2332 		goto bad;
2333 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2334 		cnp->cn_thread, &np);
2335 	np->n_sillyrename = sp;
2336 	return (0);
2337 bad:
2338 	vrele(sp->s_dvp);
2339 	crfree(sp->s_cred);
2340 	free((caddr_t)sp, M_NEWNFSREQ);
2341 	return (error);
2342 }
2343 
2344 /*
2345  * Look up a file name and optionally either update the file handle or
2346  * allocate an nfsnode, depending on the value of npp.
2347  * npp == NULL	--> just do the lookup
2348  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2349  *			handled too
2350  * *npp != NULL --> update the file handle in the vnode
2351  */
2352 static int
2353 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2354     struct thread *td, struct nfsnode **npp)
2355 {
2356 	struct vnode *newvp = NULL, *vp;
2357 	struct nfsnode *np, *dnp = VTONFS(dvp);
2358 	struct nfsfh *nfhp, *onfhp;
2359 	struct nfsvattr nfsva, dnfsva;
2360 	struct componentname cn;
2361 	int error = 0, attrflag, dattrflag;
2362 	u_int hash;
2363 
2364 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2365 	    &nfhp, &attrflag, &dattrflag, NULL);
2366 	if (dattrflag)
2367 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2368 	if (npp && !error) {
2369 		if (*npp != NULL) {
2370 		    np = *npp;
2371 		    vp = NFSTOV(np);
2372 		    /*
2373 		     * For NFSv4, check to see if it is the same name and
2374 		     * replace the name, if it is different.
2375 		     */
2376 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2377 			(np->n_v4->n4_namelen != len ||
2378 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2379 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2380 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2381 			 dnp->n_fhp->nfh_len))) {
2382 #ifdef notdef
2383 { char nnn[100]; int nnnl;
2384 nnnl = (len < 100) ? len : 99;
2385 bcopy(name, nnn, nnnl);
2386 nnn[nnnl] = '\0';
2387 printf("replace=%s\n",nnn);
2388 }
2389 #endif
2390 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2391 			    MALLOC(np->n_v4, struct nfsv4node *,
2392 				sizeof (struct nfsv4node) +
2393 				dnp->n_fhp->nfh_len + len - 1,
2394 				M_NFSV4NODE, M_WAITOK);
2395 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2396 			    np->n_v4->n4_namelen = len;
2397 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2398 				dnp->n_fhp->nfh_len);
2399 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2400 		    }
2401 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2402 			FNV1_32_INIT);
2403 		    onfhp = np->n_fhp;
2404 		    /*
2405 		     * Rehash node for new file handle.
2406 		     */
2407 		    vfs_hash_rehash(vp, hash);
2408 		    np->n_fhp = nfhp;
2409 		    if (onfhp != NULL)
2410 			FREE((caddr_t)onfhp, M_NFSFH);
2411 		    newvp = NFSTOV(np);
2412 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2413 		    FREE((caddr_t)nfhp, M_NFSFH);
2414 		    VREF(dvp);
2415 		    newvp = dvp;
2416 		} else {
2417 		    cn.cn_nameptr = name;
2418 		    cn.cn_namelen = len;
2419 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2420 			&np, NULL, LK_EXCLUSIVE);
2421 		    if (error)
2422 			return (error);
2423 		    newvp = NFSTOV(np);
2424 		}
2425 		if (!attrflag && *npp == NULL) {
2426 			if (newvp == dvp)
2427 				vrele(newvp);
2428 			else
2429 				vput(newvp);
2430 			return (ENOENT);
2431 		}
2432 		if (attrflag)
2433 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2434 			    0, 1);
2435 	}
2436 	if (npp && *npp == NULL) {
2437 		if (error) {
2438 			if (newvp) {
2439 				if (newvp == dvp)
2440 					vrele(newvp);
2441 				else
2442 					vput(newvp);
2443 			}
2444 		} else
2445 			*npp = np;
2446 	}
2447 	if (error && NFS_ISV4(dvp))
2448 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2449 	return (error);
2450 }
2451 
2452 /*
2453  * Nfs Version 3 and 4 commit rpc
2454  */
2455 int
2456 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2457    struct thread *td)
2458 {
2459 	struct nfsvattr nfsva;
2460 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2461 	int error, attrflag;
2462 	u_char verf[NFSX_VERF];
2463 
2464 	mtx_lock(&nmp->nm_mtx);
2465 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2466 		mtx_unlock(&nmp->nm_mtx);
2467 		return (0);
2468 	}
2469 	mtx_unlock(&nmp->nm_mtx);
2470 	error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2471 	    &attrflag, NULL);
2472 	if (!error) {
2473 		mtx_lock(&nmp->nm_mtx);
2474 		if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2475 			NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2476 			error = NFSERR_STALEWRITEVERF;
2477 		}
2478 		mtx_unlock(&nmp->nm_mtx);
2479 		if (!error && attrflag)
2480 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2481 			    0, 1);
2482 	} else if (NFS_ISV4(vp)) {
2483 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2484 	}
2485 	return (error);
2486 }
2487 
2488 /*
2489  * Strategy routine.
2490  * For async requests when nfsiod(s) are running, queue the request by
2491  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2492  * request.
2493  */
2494 static int
2495 nfs_strategy(struct vop_strategy_args *ap)
2496 {
2497 	struct buf *bp = ap->a_bp;
2498 	struct ucred *cr;
2499 
2500 	KASSERT(!(bp->b_flags & B_DONE),
2501 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2502 	BUF_ASSERT_HELD(bp);
2503 
2504 	if (bp->b_iocmd == BIO_READ)
2505 		cr = bp->b_rcred;
2506 	else
2507 		cr = bp->b_wcred;
2508 
2509 	/*
2510 	 * If the op is asynchronous and an i/o daemon is waiting
2511 	 * queue the request, wake it up and wait for completion
2512 	 * otherwise just do it ourselves.
2513 	 */
2514 	if ((bp->b_flags & B_ASYNC) == 0 ||
2515 	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2516 		(void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2517 	return (0);
2518 }
2519 
2520 /*
2521  * fsync vnode op. Just call ncl_flush() with commit == 1.
2522  */
2523 /* ARGSUSED */
2524 static int
2525 nfs_fsync(struct vop_fsync_args *ap)
2526 {
2527 	return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2528 }
2529 
2530 /*
2531  * Flush all the blocks associated with a vnode.
2532  * 	Walk through the buffer pool and push any dirty pages
2533  *	associated with the vnode.
2534  * If the called_from_renewthread argument is TRUE, it has been called
2535  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2536  * waiting for a buffer write to complete.
2537  */
2538 int
2539 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2540     int commit, int called_from_renewthread)
2541 {
2542 	struct nfsnode *np = VTONFS(vp);
2543 	struct buf *bp;
2544 	int i;
2545 	struct buf *nbp;
2546 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2547 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2548 	int passone = 1, trycnt = 0;
2549 	u_quad_t off, endoff, toff;
2550 	struct ucred* wcred = NULL;
2551 	struct buf **bvec = NULL;
2552 	struct bufobj *bo;
2553 #ifndef NFS_COMMITBVECSIZ
2554 #define	NFS_COMMITBVECSIZ	20
2555 #endif
2556 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2557 	int bvecsize = 0, bveccount;
2558 
2559 	if (called_from_renewthread != 0)
2560 		slptimeo = hz;
2561 	if (nmp->nm_flag & NFSMNT_INT)
2562 		slpflag = NFS_PCATCH;
2563 	if (!commit)
2564 		passone = 0;
2565 	bo = &vp->v_bufobj;
2566 	/*
2567 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2568 	 * server, but has not been committed to stable storage on the server
2569 	 * yet. On the first pass, the byte range is worked out and the commit
2570 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2571 	 * job.
2572 	 */
2573 again:
2574 	off = (u_quad_t)-1;
2575 	endoff = 0;
2576 	bvecpos = 0;
2577 	if (NFS_ISV34(vp) && commit) {
2578 		if (bvec != NULL && bvec != bvec_on_stack)
2579 			free(bvec, M_TEMP);
2580 		/*
2581 		 * Count up how many buffers waiting for a commit.
2582 		 */
2583 		bveccount = 0;
2584 		BO_LOCK(bo);
2585 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2586 			if (!BUF_ISLOCKED(bp) &&
2587 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2588 				== (B_DELWRI | B_NEEDCOMMIT))
2589 				bveccount++;
2590 		}
2591 		/*
2592 		 * Allocate space to remember the list of bufs to commit.  It is
2593 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2594 		 * If we can't get memory (for whatever reason), we will end up
2595 		 * committing the buffers one-by-one in the loop below.
2596 		 */
2597 		if (bveccount > NFS_COMMITBVECSIZ) {
2598 			/*
2599 			 * Release the vnode interlock to avoid a lock
2600 			 * order reversal.
2601 			 */
2602 			BO_UNLOCK(bo);
2603 			bvec = (struct buf **)
2604 				malloc(bveccount * sizeof(struct buf *),
2605 				       M_TEMP, M_NOWAIT);
2606 			BO_LOCK(bo);
2607 			if (bvec == NULL) {
2608 				bvec = bvec_on_stack;
2609 				bvecsize = NFS_COMMITBVECSIZ;
2610 			} else
2611 				bvecsize = bveccount;
2612 		} else {
2613 			bvec = bvec_on_stack;
2614 			bvecsize = NFS_COMMITBVECSIZ;
2615 		}
2616 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2617 			if (bvecpos >= bvecsize)
2618 				break;
2619 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2620 				nbp = TAILQ_NEXT(bp, b_bobufs);
2621 				continue;
2622 			}
2623 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2624 			    (B_DELWRI | B_NEEDCOMMIT)) {
2625 				BUF_UNLOCK(bp);
2626 				nbp = TAILQ_NEXT(bp, b_bobufs);
2627 				continue;
2628 			}
2629 			BO_UNLOCK(bo);
2630 			bremfree(bp);
2631 			/*
2632 			 * Work out if all buffers are using the same cred
2633 			 * so we can deal with them all with one commit.
2634 			 *
2635 			 * NOTE: we are not clearing B_DONE here, so we have
2636 			 * to do it later on in this routine if we intend to
2637 			 * initiate I/O on the bp.
2638 			 *
2639 			 * Note: to avoid loopback deadlocks, we do not
2640 			 * assign b_runningbufspace.
2641 			 */
2642 			if (wcred == NULL)
2643 				wcred = bp->b_wcred;
2644 			else if (wcred != bp->b_wcred)
2645 				wcred = NOCRED;
2646 			vfs_busy_pages(bp, 1);
2647 
2648 			BO_LOCK(bo);
2649 			/*
2650 			 * bp is protected by being locked, but nbp is not
2651 			 * and vfs_busy_pages() may sleep.  We have to
2652 			 * recalculate nbp.
2653 			 */
2654 			nbp = TAILQ_NEXT(bp, b_bobufs);
2655 
2656 			/*
2657 			 * A list of these buffers is kept so that the
2658 			 * second loop knows which buffers have actually
2659 			 * been committed. This is necessary, since there
2660 			 * may be a race between the commit rpc and new
2661 			 * uncommitted writes on the file.
2662 			 */
2663 			bvec[bvecpos++] = bp;
2664 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2665 				bp->b_dirtyoff;
2666 			if (toff < off)
2667 				off = toff;
2668 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2669 			if (toff > endoff)
2670 				endoff = toff;
2671 		}
2672 		BO_UNLOCK(bo);
2673 	}
2674 	if (bvecpos > 0) {
2675 		/*
2676 		 * Commit data on the server, as required.
2677 		 * If all bufs are using the same wcred, then use that with
2678 		 * one call for all of them, otherwise commit each one
2679 		 * separately.
2680 		 */
2681 		if (wcred != NOCRED)
2682 			retv = ncl_commit(vp, off, (int)(endoff - off),
2683 					  wcred, td);
2684 		else {
2685 			retv = 0;
2686 			for (i = 0; i < bvecpos; i++) {
2687 				off_t off, size;
2688 				bp = bvec[i];
2689 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2690 					bp->b_dirtyoff;
2691 				size = (u_quad_t)(bp->b_dirtyend
2692 						  - bp->b_dirtyoff);
2693 				retv = ncl_commit(vp, off, (int)size,
2694 						  bp->b_wcred, td);
2695 				if (retv) break;
2696 			}
2697 		}
2698 
2699 		if (retv == NFSERR_STALEWRITEVERF)
2700 			ncl_clearcommit(vp->v_mount);
2701 
2702 		/*
2703 		 * Now, either mark the blocks I/O done or mark the
2704 		 * blocks dirty, depending on whether the commit
2705 		 * succeeded.
2706 		 */
2707 		for (i = 0; i < bvecpos; i++) {
2708 			bp = bvec[i];
2709 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2710 			if (retv) {
2711 				/*
2712 				 * Error, leave B_DELWRI intact
2713 				 */
2714 				vfs_unbusy_pages(bp);
2715 				brelse(bp);
2716 			} else {
2717 				/*
2718 				 * Success, remove B_DELWRI ( bundirty() ).
2719 				 *
2720 				 * b_dirtyoff/b_dirtyend seem to be NFS
2721 				 * specific.  We should probably move that
2722 				 * into bundirty(). XXX
2723 				 */
2724 				bufobj_wref(bo);
2725 				bp->b_flags |= B_ASYNC;
2726 				bundirty(bp);
2727 				bp->b_flags &= ~B_DONE;
2728 				bp->b_ioflags &= ~BIO_ERROR;
2729 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2730 				bufdone(bp);
2731 			}
2732 		}
2733 	}
2734 
2735 	/*
2736 	 * Start/do any write(s) that are required.
2737 	 */
2738 loop:
2739 	BO_LOCK(bo);
2740 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2741 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2742 			if (waitfor != MNT_WAIT || passone)
2743 				continue;
2744 
2745 			error = BUF_TIMELOCK(bp,
2746 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2747 			    BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2748 			if (error == 0) {
2749 				BUF_UNLOCK(bp);
2750 				goto loop;
2751 			}
2752 			if (error == ENOLCK) {
2753 				error = 0;
2754 				goto loop;
2755 			}
2756 			if (called_from_renewthread != 0) {
2757 				/*
2758 				 * Return EIO so the flush will be retried
2759 				 * later.
2760 				 */
2761 				error = EIO;
2762 				goto done;
2763 			}
2764 			if (newnfs_sigintr(nmp, td)) {
2765 				error = EINTR;
2766 				goto done;
2767 			}
2768 			if (slpflag & PCATCH) {
2769 				slpflag = 0;
2770 				slptimeo = 2 * hz;
2771 			}
2772 			goto loop;
2773 		}
2774 		if ((bp->b_flags & B_DELWRI) == 0)
2775 			panic("nfs_fsync: not dirty");
2776 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2777 			BUF_UNLOCK(bp);
2778 			continue;
2779 		}
2780 		BO_UNLOCK(bo);
2781 		bremfree(bp);
2782 		if (passone || !commit)
2783 		    bp->b_flags |= B_ASYNC;
2784 		else
2785 		    bp->b_flags |= B_ASYNC;
2786 		bwrite(bp);
2787 		if (newnfs_sigintr(nmp, td)) {
2788 			error = EINTR;
2789 			goto done;
2790 		}
2791 		goto loop;
2792 	}
2793 	if (passone) {
2794 		passone = 0;
2795 		BO_UNLOCK(bo);
2796 		goto again;
2797 	}
2798 	if (waitfor == MNT_WAIT) {
2799 		while (bo->bo_numoutput) {
2800 			error = bufobj_wwait(bo, slpflag, slptimeo);
2801 			if (error) {
2802 			    BO_UNLOCK(bo);
2803 			    if (called_from_renewthread != 0) {
2804 				/*
2805 				 * Return EIO so that the flush will be
2806 				 * retried later.
2807 				 */
2808 				error = EIO;
2809 				goto done;
2810 			    }
2811 			    error = newnfs_sigintr(nmp, td);
2812 			    if (error)
2813 				goto done;
2814 			    if (slpflag & PCATCH) {
2815 				slpflag = 0;
2816 				slptimeo = 2 * hz;
2817 			    }
2818 			    BO_LOCK(bo);
2819 			}
2820 		}
2821 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2822 			BO_UNLOCK(bo);
2823 			goto loop;
2824 		}
2825 		/*
2826 		 * Wait for all the async IO requests to drain
2827 		 */
2828 		BO_UNLOCK(bo);
2829 		mtx_lock(&np->n_mtx);
2830 		while (np->n_directio_asyncwr > 0) {
2831 			np->n_flag |= NFSYNCWAIT;
2832 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
2833 			    &np->n_mtx, slpflag | (PRIBIO + 1),
2834 			    "nfsfsync", 0);
2835 			if (error) {
2836 				if (newnfs_sigintr(nmp, td)) {
2837 					mtx_unlock(&np->n_mtx);
2838 					error = EINTR;
2839 					goto done;
2840 				}
2841 			}
2842 		}
2843 		mtx_unlock(&np->n_mtx);
2844 	} else
2845 		BO_UNLOCK(bo);
2846 	mtx_lock(&np->n_mtx);
2847 	if (np->n_flag & NWRITEERR) {
2848 		error = np->n_error;
2849 		np->n_flag &= ~NWRITEERR;
2850 	}
2851   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2852 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2853   		np->n_flag &= ~NMODIFIED;
2854 	mtx_unlock(&np->n_mtx);
2855 done:
2856 	if (bvec != NULL && bvec != bvec_on_stack)
2857 		free(bvec, M_TEMP);
2858 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2859 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2860 	     np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2861 		/* try, try again... */
2862 		passone = 1;
2863 		wcred = NULL;
2864 		bvec = NULL;
2865 		bvecsize = 0;
2866 printf("try%d\n", trycnt);
2867 		goto again;
2868 	}
2869 	return (error);
2870 }
2871 
2872 /*
2873  * NFS advisory byte-level locks.
2874  */
2875 static int
2876 nfs_advlock(struct vop_advlock_args *ap)
2877 {
2878 	struct vnode *vp = ap->a_vp;
2879 	struct ucred *cred;
2880 	struct nfsnode *np = VTONFS(ap->a_vp);
2881 	struct proc *p = (struct proc *)ap->a_id;
2882 	struct thread *td = curthread;	/* XXX */
2883 	struct vattr va;
2884 	int ret, error = EOPNOTSUPP;
2885 	u_quad_t size;
2886 
2887 	if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2888 		cred = p->p_ucred;
2889 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2890 		if (vp->v_iflag & VI_DOOMED) {
2891 			VOP_UNLOCK(vp, 0);
2892 			return (EBADF);
2893 		}
2894 
2895 		/*
2896 		 * If this is unlocking a write locked region, flush and
2897 		 * commit them before unlocking. This is required by
2898 		 * RFC3530 Sec. 9.3.2.
2899 		 */
2900 		if (ap->a_op == F_UNLCK &&
2901 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
2902 			(void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2903 
2904 		/*
2905 		 * Loop around doing the lock op, while a blocking lock
2906 		 * must wait for the lock op to succeed.
2907 		 */
2908 		do {
2909 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2910 			    ap->a_fl, 0, cred, td);
2911 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2912 			    ap->a_op == F_SETLK) {
2913 				VOP_UNLOCK(vp, 0);
2914 				error = nfs_catnap(PZERO | PCATCH, ret,
2915 				    "ncladvl");
2916 				if (error)
2917 					return (EINTR);
2918 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2919 				if (vp->v_iflag & VI_DOOMED) {
2920 					VOP_UNLOCK(vp, 0);
2921 					return (EBADF);
2922 				}
2923 			}
2924 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2925 		     ap->a_op == F_SETLK);
2926 		if (ret == NFSERR_DENIED) {
2927 			VOP_UNLOCK(vp, 0);
2928 			return (EAGAIN);
2929 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2930 			VOP_UNLOCK(vp, 0);
2931 			return (ret);
2932 		} else if (ret != 0) {
2933 			VOP_UNLOCK(vp, 0);
2934 			return (EACCES);
2935 		}
2936 
2937 		/*
2938 		 * Now, if we just got a lock, invalidate data in the buffer
2939 		 * cache, as required, so that the coherency conforms with
2940 		 * RFC3530 Sec. 9.3.2.
2941 		 */
2942 		if (ap->a_op == F_SETLK) {
2943 			if ((np->n_flag & NMODIFIED) == 0) {
2944 				np->n_attrstamp = 0;
2945 				ret = VOP_GETATTR(vp, &va, cred);
2946 			}
2947 			if ((np->n_flag & NMODIFIED) || ret ||
2948 			    np->n_change != va.va_filerev) {
2949 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2950 				np->n_attrstamp = 0;
2951 				ret = VOP_GETATTR(vp, &va, cred);
2952 				if (!ret) {
2953 					np->n_mtime = va.va_mtime;
2954 					np->n_change = va.va_filerev;
2955 				}
2956 			}
2957 		}
2958 		VOP_UNLOCK(vp, 0);
2959 		return (0);
2960 	} else if (!NFS_ISV4(vp)) {
2961 		error = vn_lock(vp, LK_SHARED);
2962 		if (error)
2963 			return (error);
2964 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2965 			size = VTONFS(vp)->n_size;
2966 			VOP_UNLOCK(vp, 0);
2967 			error = lf_advlock(ap, &(vp->v_lockf), size);
2968 		} else {
2969 			if (nfs_advlock_p != NULL)
2970 				error = nfs_advlock_p(ap);
2971 			else {
2972 				VOP_UNLOCK(vp, 0);
2973 				error = ENOLCK;
2974 			}
2975 		}
2976 	}
2977 	return (error);
2978 }
2979 
2980 /*
2981  * NFS advisory byte-level locks.
2982  */
2983 static int
2984 nfs_advlockasync(struct vop_advlockasync_args *ap)
2985 {
2986 	struct vnode *vp = ap->a_vp;
2987 	u_quad_t size;
2988 	int error;
2989 
2990 	if (NFS_ISV4(vp))
2991 		return (EOPNOTSUPP);
2992 	error = vn_lock(vp, LK_SHARED);
2993 	if (error)
2994 		return (error);
2995 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2996 		size = VTONFS(vp)->n_size;
2997 		VOP_UNLOCK(vp, 0);
2998 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
2999 	} else {
3000 		VOP_UNLOCK(vp, 0);
3001 		error = EOPNOTSUPP;
3002 	}
3003 	return (error);
3004 }
3005 
3006 /*
3007  * Print out the contents of an nfsnode.
3008  */
3009 static int
3010 nfs_print(struct vop_print_args *ap)
3011 {
3012 	struct vnode *vp = ap->a_vp;
3013 	struct nfsnode *np = VTONFS(vp);
3014 
3015 	ncl_printf("\tfileid %ld fsid 0x%x",
3016 	   np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3017 	if (vp->v_type == VFIFO)
3018 		fifo_printinfo(vp);
3019 	printf("\n");
3020 	return (0);
3021 }
3022 
3023 /*
3024  * This is the "real" nfs::bwrite(struct buf*).
3025  * We set B_CACHE if this is a VMIO buffer.
3026  */
3027 int
3028 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3029 {
3030 	int s;
3031 	int oldflags = bp->b_flags;
3032 #if 0
3033 	int retv = 1;
3034 	off_t off;
3035 #endif
3036 
3037 	BUF_ASSERT_HELD(bp);
3038 
3039 	if (bp->b_flags & B_INVAL) {
3040 		brelse(bp);
3041 		return(0);
3042 	}
3043 
3044 	bp->b_flags |= B_CACHE;
3045 
3046 	/*
3047 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3048 	 */
3049 
3050 	s = splbio();
3051 	bundirty(bp);
3052 	bp->b_flags &= ~B_DONE;
3053 	bp->b_ioflags &= ~BIO_ERROR;
3054 	bp->b_iocmd = BIO_WRITE;
3055 
3056 	bufobj_wref(bp->b_bufobj);
3057 	curthread->td_ru.ru_oublock++;
3058 	splx(s);
3059 
3060 	/*
3061 	 * Note: to avoid loopback deadlocks, we do not
3062 	 * assign b_runningbufspace.
3063 	 */
3064 	vfs_busy_pages(bp, 1);
3065 
3066 	BUF_KERNPROC(bp);
3067 	bp->b_iooffset = dbtob(bp->b_blkno);
3068 	bstrategy(bp);
3069 
3070 	if( (oldflags & B_ASYNC) == 0) {
3071 		int rtval = bufwait(bp);
3072 
3073 		if (oldflags & B_DELWRI) {
3074 			s = splbio();
3075 			reassignbuf(bp);
3076 			splx(s);
3077 		}
3078 		brelse(bp);
3079 		return (rtval);
3080 	}
3081 
3082 	return (0);
3083 }
3084 
3085 /*
3086  * nfs special file access vnode op.
3087  * Essentially just get vattr and then imitate iaccess() since the device is
3088  * local to the client.
3089  */
3090 static int
3091 nfsspec_access(struct vop_access_args *ap)
3092 {
3093 	struct vattr *vap;
3094 	struct ucred *cred = ap->a_cred;
3095 	struct vnode *vp = ap->a_vp;
3096 	accmode_t accmode = ap->a_accmode;
3097 	struct vattr vattr;
3098 	int error;
3099 
3100 	/*
3101 	 * Disallow write attempts on filesystems mounted read-only;
3102 	 * unless the file is a socket, fifo, or a block or character
3103 	 * device resident on the filesystem.
3104 	 */
3105 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3106 		switch (vp->v_type) {
3107 		case VREG:
3108 		case VDIR:
3109 		case VLNK:
3110 			return (EROFS);
3111 		default:
3112 			break;
3113 		}
3114 	}
3115 	vap = &vattr;
3116 	error = VOP_GETATTR(vp, vap, cred);
3117 	if (error)
3118 		goto out;
3119 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3120 	    accmode, cred, NULL);
3121 out:
3122 	return error;
3123 }
3124 
3125 /*
3126  * Read wrapper for fifos.
3127  */
3128 static int
3129 nfsfifo_read(struct vop_read_args *ap)
3130 {
3131 	struct nfsnode *np = VTONFS(ap->a_vp);
3132 	int error;
3133 
3134 	/*
3135 	 * Set access flag.
3136 	 */
3137 	mtx_lock(&np->n_mtx);
3138 	np->n_flag |= NACC;
3139 	getnanotime(&np->n_atim);
3140 	mtx_unlock(&np->n_mtx);
3141 	error = fifo_specops.vop_read(ap);
3142 	return error;
3143 }
3144 
3145 /*
3146  * Write wrapper for fifos.
3147  */
3148 static int
3149 nfsfifo_write(struct vop_write_args *ap)
3150 {
3151 	struct nfsnode *np = VTONFS(ap->a_vp);
3152 
3153 	/*
3154 	 * Set update flag.
3155 	 */
3156 	mtx_lock(&np->n_mtx);
3157 	np->n_flag |= NUPD;
3158 	getnanotime(&np->n_mtim);
3159 	mtx_unlock(&np->n_mtx);
3160 	return(fifo_specops.vop_write(ap));
3161 }
3162 
3163 /*
3164  * Close wrapper for fifos.
3165  *
3166  * Update the times on the nfsnode then do fifo close.
3167  */
3168 static int
3169 nfsfifo_close(struct vop_close_args *ap)
3170 {
3171 	struct vnode *vp = ap->a_vp;
3172 	struct nfsnode *np = VTONFS(vp);
3173 	struct vattr vattr;
3174 	struct timespec ts;
3175 
3176 	mtx_lock(&np->n_mtx);
3177 	if (np->n_flag & (NACC | NUPD)) {
3178 		getnanotime(&ts);
3179 		if (np->n_flag & NACC)
3180 			np->n_atim = ts;
3181 		if (np->n_flag & NUPD)
3182 			np->n_mtim = ts;
3183 		np->n_flag |= NCHG;
3184 		if (vrefcnt(vp) == 1 &&
3185 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3186 			VATTR_NULL(&vattr);
3187 			if (np->n_flag & NACC)
3188 				vattr.va_atime = np->n_atim;
3189 			if (np->n_flag & NUPD)
3190 				vattr.va_mtime = np->n_mtim;
3191 			mtx_unlock(&np->n_mtx);
3192 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3193 			goto out;
3194 		}
3195 	}
3196 	mtx_unlock(&np->n_mtx);
3197 out:
3198 	return (fifo_specops.vop_close(ap));
3199 }
3200 
3201 /*
3202  * Just call ncl_writebp() with the force argument set to 1.
3203  *
3204  * NOTE: B_DONE may or may not be set in a_bp on call.
3205  */
3206 static int
3207 nfs_bwrite(struct buf *bp)
3208 {
3209 
3210 	return (ncl_writebp(bp, 1, curthread));
3211 }
3212 
3213 struct buf_ops buf_ops_newnfs = {
3214 	.bop_name	=	"buf_ops_nfs",
3215 	.bop_write	=	nfs_bwrite,
3216 	.bop_strategy	=	bufstrategy,
3217 	.bop_sync	=	bufsync,
3218 	.bop_bdflush	=	bufbdflush,
3219 };
3220 
3221 /*
3222  * Cloned from vop_stdlock(), and then the ugly hack added.
3223  */
3224 static int
3225 nfs_lock1(struct vop_lock1_args *ap)
3226 {
3227 	struct vnode *vp = ap->a_vp;
3228 	int error = 0;
3229 
3230 	/*
3231 	 * Since vfs_hash_get() calls vget() and it will no longer work
3232 	 * for FreeBSD8 with flags == 0, I can only think of this horrible
3233 	 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3234 	 * and then handle it here. All I want for this case is a v_usecount
3235 	 * on the vnode to use for recovery, while another thread might
3236 	 * hold a lock on the vnode. I have the other threads blocked, so
3237 	 * there isn't any race problem.
3238 	 */
3239 	if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3240 		if ((ap->a_flags & LK_INTERLOCK) == 0)
3241 			panic("ncllock1");
3242 		if ((vp->v_iflag & VI_DOOMED))
3243 			error = ENOENT;
3244 		VI_UNLOCK(vp);
3245 		return (error);
3246 	}
3247 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3248 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3249 	    ap->a_line));
3250 }
3251 
3252 static int
3253 nfs_getacl(struct vop_getacl_args *ap)
3254 {
3255 	int error;
3256 
3257 	if (ap->a_type != ACL_TYPE_NFS4)
3258 		return (EOPNOTSUPP);
3259 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3260 	    NULL);
3261 	if (error > NFSERR_STALE) {
3262 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3263 		error = EPERM;
3264 	}
3265 	return (error);
3266 }
3267 
3268 static int
3269 nfs_setacl(struct vop_setacl_args *ap)
3270 {
3271 	int error;
3272 
3273 	if (ap->a_type != ACL_TYPE_NFS4)
3274 		return (EOPNOTSUPP);
3275 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3276 	    NULL);
3277 	if (error > NFSERR_STALE) {
3278 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3279 		error = EPERM;
3280 	}
3281 	return (error);
3282 }
3283 
3284 /*
3285  * Return POSIX pathconf information applicable to nfs filesystems.
3286  */
3287 static int
3288 nfs_pathconf(struct vop_pathconf_args *ap)
3289 {
3290 	struct nfsv3_pathconf pc;
3291 	struct nfsvattr nfsva;
3292 	struct vnode *vp = ap->a_vp;
3293 	struct thread *td = curthread;
3294 	int attrflag, error;
3295 
3296 	if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3297 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3298 	    ap->a_name == _PC_NO_TRUNC))) {
3299 		/*
3300 		 * Since only the above 4 a_names are returned by the NFSv3
3301 		 * Pathconf RPC, there is no point in doing it for others.
3302 		 */
3303 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3304 		    &attrflag, NULL);
3305 		if (attrflag != 0)
3306 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3307 			    1);
3308 		if (error != 0)
3309 			return (error);
3310 	} else {
3311 		/*
3312 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3313 		 * just fake them.
3314 		 */
3315 		pc.pc_linkmax = LINK_MAX;
3316 		pc.pc_namemax = NFS_MAXNAMLEN;
3317 		pc.pc_notrunc = 1;
3318 		pc.pc_chownrestricted = 1;
3319 		pc.pc_caseinsensitive = 0;
3320 		pc.pc_casepreserving = 1;
3321 		error = 0;
3322 	}
3323 	switch (ap->a_name) {
3324 	case _PC_LINK_MAX:
3325 		*ap->a_retval = pc.pc_linkmax;
3326 		break;
3327 	case _PC_NAME_MAX:
3328 		*ap->a_retval = pc.pc_namemax;
3329 		break;
3330 	case _PC_PATH_MAX:
3331 		*ap->a_retval = PATH_MAX;
3332 		break;
3333 	case _PC_PIPE_BUF:
3334 		*ap->a_retval = PIPE_BUF;
3335 		break;
3336 	case _PC_CHOWN_RESTRICTED:
3337 		*ap->a_retval = pc.pc_chownrestricted;
3338 		break;
3339 	case _PC_NO_TRUNC:
3340 		*ap->a_retval = pc.pc_notrunc;
3341 		break;
3342 	case _PC_ACL_EXTENDED:
3343 		*ap->a_retval = 0;
3344 		break;
3345 	case _PC_ACL_NFS4:
3346 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3347 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3348 			*ap->a_retval = 1;
3349 		else
3350 			*ap->a_retval = 0;
3351 		break;
3352 	case _PC_ACL_PATH_MAX:
3353 		if (NFS_ISV4(vp))
3354 			*ap->a_retval = ACL_MAX_ENTRIES;
3355 		else
3356 			*ap->a_retval = 3;
3357 		break;
3358 	case _PC_MAC_PRESENT:
3359 		*ap->a_retval = 0;
3360 		break;
3361 	case _PC_ASYNC_IO:
3362 		/* _PC_ASYNC_IO should have been handled by upper layers. */
3363 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3364 		error = EINVAL;
3365 		break;
3366 	case _PC_PRIO_IO:
3367 		*ap->a_retval = 0;
3368 		break;
3369 	case _PC_SYNC_IO:
3370 		*ap->a_retval = 0;
3371 		break;
3372 	case _PC_ALLOC_SIZE_MIN:
3373 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3374 		break;
3375 	case _PC_FILESIZEBITS:
3376 		if (NFS_ISV34(vp))
3377 			*ap->a_retval = 64;
3378 		else
3379 			*ap->a_retval = 32;
3380 		break;
3381 	case _PC_REC_INCR_XFER_SIZE:
3382 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3383 		break;
3384 	case _PC_REC_MAX_XFER_SIZE:
3385 		*ap->a_retval = -1; /* means ``unlimited'' */
3386 		break;
3387 	case _PC_REC_MIN_XFER_SIZE:
3388 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3389 		break;
3390 	case _PC_REC_XFER_ALIGN:
3391 		*ap->a_retval = PAGE_SIZE;
3392 		break;
3393 	case _PC_SYMLINK_MAX:
3394 		*ap->a_retval = NFS_MAXPATHLEN;
3395 		break;
3396 
3397 	default:
3398 		error = EINVAL;
3399 		break;
3400 	}
3401 	return (error);
3402 }
3403 
3404