xref: /dragonfly/sys/vfs/nfs/nfs_vnops.c (revision 650094e1)
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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
37  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38  */
39 
40 
41 /*
42  * vnode op calls for Sun NFS version 2 and 3
43  */
44 
45 #include "opt_inet.h"
46 
47 #include <sys/param.h>
48 #include <sys/kernel.h>
49 #include <sys/systm.h>
50 #include <sys/resourcevar.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/buf.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/namei.h>
57 #include <sys/nlookup.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/conf.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_extern.h>
69 
70 #include <sys/buf2.h>
71 
72 #include <vfs/fifofs/fifo.h>
73 #include <vfs/ufs/dir.h>
74 
75 #undef DIRBLKSIZ
76 
77 #include "rpcv2.h"
78 #include "nfsproto.h"
79 #include "nfs.h"
80 #include "nfsmount.h"
81 #include "nfsnode.h"
82 #include "xdr_subs.h"
83 #include "nfsm_subs.h"
84 
85 #include <net/if.h>
86 #include <netinet/in.h>
87 #include <netinet/in_var.h>
88 
89 #include <sys/thread2.h>
90 
91 /* Defs */
92 #define	TRUE	1
93 #define	FALSE	0
94 
95 static int	nfsfifo_read (struct vop_read_args *);
96 static int	nfsfifo_write (struct vop_write_args *);
97 static int	nfsfifo_close (struct vop_close_args *);
98 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
99 static	int	nfs_lookup (struct vop_old_lookup_args *);
100 static	int	nfs_create (struct vop_old_create_args *);
101 static	int	nfs_mknod (struct vop_old_mknod_args *);
102 static	int	nfs_open (struct vop_open_args *);
103 static	int	nfs_close (struct vop_close_args *);
104 static	int	nfs_access (struct vop_access_args *);
105 static	int	nfs_getattr (struct vop_getattr_args *);
106 static	int	nfs_setattr (struct vop_setattr_args *);
107 static	int	nfs_read (struct vop_read_args *);
108 static	int	nfs_mmap (struct vop_mmap_args *);
109 static	int	nfs_fsync (struct vop_fsync_args *);
110 static	int	nfs_remove (struct vop_old_remove_args *);
111 static	int	nfs_link (struct vop_old_link_args *);
112 static	int	nfs_rename (struct vop_old_rename_args *);
113 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
114 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
115 static	int	nfs_symlink (struct vop_old_symlink_args *);
116 static	int	nfs_readdir (struct vop_readdir_args *);
117 static	int	nfs_bmap (struct vop_bmap_args *);
118 static	int	nfs_strategy (struct vop_strategy_args *);
119 static	int	nfs_lookitup (struct vnode *, const char *, int,
120 			struct ucred *, struct thread *, struct nfsnode **);
121 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
122 static int	nfs_laccess (struct vop_access_args *);
123 static int	nfs_readlink (struct vop_readlink_args *);
124 static int	nfs_print (struct vop_print_args *);
125 static int	nfs_advlock (struct vop_advlock_args *);
126 
127 static	int	nfs_nresolve (struct vop_nresolve_args *);
128 /*
129  * Global vfs data structures for nfs
130  */
131 struct vop_ops nfsv2_vnode_vops = {
132 	.vop_default =		vop_defaultop,
133 	.vop_access =		nfs_access,
134 	.vop_advlock =		nfs_advlock,
135 	.vop_bmap =		nfs_bmap,
136 	.vop_close =		nfs_close,
137 	.vop_old_create =	nfs_create,
138 	.vop_fsync =		nfs_fsync,
139 	.vop_getattr =		nfs_getattr,
140 	.vop_getpages =		vop_stdgetpages,
141 	.vop_putpages =		vop_stdputpages,
142 	.vop_inactive =		nfs_inactive,
143 	.vop_old_link =		nfs_link,
144 	.vop_old_lookup =	nfs_lookup,
145 	.vop_old_mkdir =	nfs_mkdir,
146 	.vop_old_mknod =	nfs_mknod,
147 	.vop_mmap =		nfs_mmap,
148 	.vop_open =		nfs_open,
149 	.vop_print =		nfs_print,
150 	.vop_read =		nfs_read,
151 	.vop_readdir =		nfs_readdir,
152 	.vop_readlink =		nfs_readlink,
153 	.vop_reclaim =		nfs_reclaim,
154 	.vop_old_remove =	nfs_remove,
155 	.vop_old_rename =	nfs_rename,
156 	.vop_old_rmdir =	nfs_rmdir,
157 	.vop_setattr =		nfs_setattr,
158 	.vop_strategy =		nfs_strategy,
159 	.vop_old_symlink =	nfs_symlink,
160 	.vop_write =		nfs_write,
161 	.vop_nresolve =		nfs_nresolve
162 };
163 
164 /*
165  * Special device vnode ops
166  */
167 struct vop_ops nfsv2_spec_vops = {
168 	.vop_default =		vop_defaultop,
169 	.vop_access =		nfs_laccess,
170 	.vop_close =		nfs_close,
171 	.vop_fsync =		nfs_fsync,
172 	.vop_getattr =		nfs_getattr,
173 	.vop_inactive =		nfs_inactive,
174 	.vop_print =		nfs_print,
175 	.vop_read =		vop_stdnoread,
176 	.vop_reclaim =		nfs_reclaim,
177 	.vop_setattr =		nfs_setattr,
178 	.vop_write =		vop_stdnowrite
179 };
180 
181 struct vop_ops nfsv2_fifo_vops = {
182 	.vop_default =		fifo_vnoperate,
183 	.vop_access =		nfs_laccess,
184 	.vop_close =		nfsfifo_close,
185 	.vop_fsync =		nfs_fsync,
186 	.vop_getattr =		nfs_getattr,
187 	.vop_inactive =		nfs_inactive,
188 	.vop_print =		nfs_print,
189 	.vop_read =		nfsfifo_read,
190 	.vop_reclaim =		nfs_reclaim,
191 	.vop_setattr =		nfs_setattr,
192 	.vop_write =		nfsfifo_write
193 };
194 
195 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
196 				  struct componentname *cnp,
197 				  struct vattr *vap);
198 static int	nfs_removerpc (struct vnode *dvp, const char *name,
199 				   int namelen,
200 				   struct ucred *cred, struct thread *td);
201 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
202 				   int fnamelen, struct vnode *tdvp,
203 				   const char *tnameptr, int tnamelen,
204 				   struct ucred *cred, struct thread *td);
205 static int	nfs_renameit (struct vnode *sdvp,
206 				  struct componentname *scnp,
207 				  struct sillyrename *sp);
208 
209 SYSCTL_DECL(_vfs_nfs);
210 
211 static int nfs_flush_on_rename = 1;
212 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
213 	   &nfs_flush_on_rename, 0, "flush fvp prior to rename");
214 static int nfs_flush_on_hlink = 0;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
216 	   &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
217 
218 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
220 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
221 
222 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
224 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
225 
226 static int	nfspos_cache_timeout = NFS_MINATTRTIMO;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
228 	   &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
229 
230 static int	nfsv3_commit_on_close = 0;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
232 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
233 #if 0
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
235 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
236 
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
238 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
239 #endif
240 
241 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
242 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
243 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
244 
245 /*
246  * Returns whether a name component is a degenerate '.' or '..'.
247  */
248 static __inline
249 int
250 nlcdegenerate(struct nlcomponent *nlc)
251 {
252 	if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
253 		return(1);
254 	if (nlc->nlc_namelen == 2 &&
255 	    nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
256 		return(1);
257 	return(0);
258 }
259 
260 static int
261 nfs3_access_otw(struct vnode *vp, int wmode,
262 		struct thread *td, struct ucred *cred)
263 {
264 	struct nfsnode *np = VTONFS(vp);
265 	int attrflag;
266 	int error = 0;
267 	u_int32_t *tl;
268 	u_int32_t rmode;
269 	struct nfsm_info info;
270 
271 	info.mrep = NULL;
272 	info.v3 = 1;
273 
274 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
275 	nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
276 		     NFSX_FH(info.v3) + NFSX_UNSIGNED);
277 	ERROROUT(nfsm_fhtom(&info, vp));
278 	tl = nfsm_build(&info, NFSX_UNSIGNED);
279 	*tl = txdr_unsigned(wmode);
280 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
281 	ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
282 	if (error == 0) {
283 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
284 		rmode = fxdr_unsigned(u_int32_t, *tl);
285 		np->n_mode = rmode;
286 		np->n_modeuid = cred->cr_uid;
287 		np->n_modestamp = mycpu->gd_time_seconds;
288 	}
289 	m_freem(info.mrep);
290 	info.mrep = NULL;
291 nfsmout:
292 	return error;
293 }
294 
295 /*
296  * nfs access vnode op.
297  * For nfs version 2, just return ok. File accesses may fail later.
298  * For nfs version 3, use the access rpc to check accessibility. If file modes
299  * are changed on the server, accesses might still fail later.
300  *
301  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
302  */
303 static int
304 nfs_access(struct vop_access_args *ap)
305 {
306 	struct ucred *cred;
307 	struct vnode *vp = ap->a_vp;
308 	thread_t td = curthread;
309 	int error = 0;
310 	u_int32_t mode, wmode;
311 	struct nfsnode *np = VTONFS(vp);
312 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
313 	int v3 = NFS_ISV3(vp);
314 
315 	lwkt_gettoken(&nmp->nm_token);
316 
317 	/*
318 	 * Disallow write attempts on filesystems mounted read-only;
319 	 * unless the file is a socket, fifo, or a block or character
320 	 * device resident on the filesystem.
321 	 */
322 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
323 		switch (vp->v_type) {
324 		case VREG:
325 		case VDIR:
326 		case VLNK:
327 			lwkt_reltoken(&nmp->nm_token);
328 			return (EROFS);
329 		default:
330 			break;
331 		}
332 	}
333 
334 	/*
335 	 * The NFS protocol passes only the effective uid/gid over the wire but
336 	 * we need to check access against real ids if AT_EACCESS not set.
337 	 * Handle this case by cloning the credentials and setting the
338 	 * effective ids to the real ones.
339 	 */
340 	if (ap->a_flags & AT_EACCESS) {
341 		cred = crhold(ap->a_cred);
342 	} else {
343 		cred = crdup(ap->a_cred);
344 		cred->cr_uid = cred->cr_ruid;
345 		cred->cr_gid = cred->cr_rgid;
346 	}
347 
348 	/*
349 	 * For nfs v3, check to see if we have done this recently, and if
350 	 * so return our cached result instead of making an ACCESS call.
351 	 * If not, do an access rpc, otherwise you are stuck emulating
352 	 * ufs_access() locally using the vattr. This may not be correct,
353 	 * since the server may apply other access criteria such as
354 	 * client uid-->server uid mapping that we do not know about.
355 	 */
356 	if (v3) {
357 		if (ap->a_mode & VREAD)
358 			mode = NFSV3ACCESS_READ;
359 		else
360 			mode = 0;
361 		if (vp->v_type != VDIR) {
362 			if (ap->a_mode & VWRITE)
363 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
364 			if (ap->a_mode & VEXEC)
365 				mode |= NFSV3ACCESS_EXECUTE;
366 		} else {
367 			if (ap->a_mode & VWRITE)
368 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
369 					 NFSV3ACCESS_DELETE);
370 			if (ap->a_mode & VEXEC)
371 				mode |= NFSV3ACCESS_LOOKUP;
372 		}
373 		/* XXX safety belt, only make blanket request if caching */
374 		if (nfsaccess_cache_timeout > 0) {
375 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
376 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
377 				NFSV3ACCESS_DELETE | NFSV3ACCESS_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 		if (np->n_modestamp &&
387 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
388 		   (cred->cr_uid == np->n_modeuid) &&
389 		   ((np->n_mode & mode) == mode)) {
390 			nfsstats.accesscache_hits++;
391 		} else {
392 			/*
393 			 * Either a no, or a don't know.  Go to the wire.
394 			 */
395 			nfsstats.accesscache_misses++;
396 		        error = nfs3_access_otw(vp, wmode, td, cred);
397 			if (!error) {
398 				if ((np->n_mode & mode) != mode) {
399 					error = EACCES;
400 				}
401 			}
402 		}
403 	} else {
404 		if ((error = nfs_laccess(ap)) != 0) {
405 			crfree(cred);
406 			lwkt_reltoken(&nmp->nm_token);
407 			return (error);
408 		}
409 
410 		/*
411 		 * Attempt to prevent a mapped root from accessing a file
412 		 * which it shouldn't.  We try to read a byte from the file
413 		 * if the user is root and the file is not zero length.
414 		 * After calling nfs_laccess, we should have the correct
415 		 * file size cached.
416 		 */
417 		if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
418 		    && VTONFS(vp)->n_size > 0) {
419 			struct iovec aiov;
420 			struct uio auio;
421 			char buf[1];
422 
423 			aiov.iov_base = buf;
424 			aiov.iov_len = 1;
425 			auio.uio_iov = &aiov;
426 			auio.uio_iovcnt = 1;
427 			auio.uio_offset = 0;
428 			auio.uio_resid = 1;
429 			auio.uio_segflg = UIO_SYSSPACE;
430 			auio.uio_rw = UIO_READ;
431 			auio.uio_td = td;
432 
433 			if (vp->v_type == VREG) {
434 				error = nfs_readrpc_uio(vp, &auio);
435 			} else if (vp->v_type == VDIR) {
436 				char* bp;
437 				bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
438 				aiov.iov_base = bp;
439 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
440 				error = nfs_readdirrpc_uio(vp, &auio);
441 				kfree(bp, M_TEMP);
442 			} else if (vp->v_type == VLNK) {
443 				error = nfs_readlinkrpc_uio(vp, &auio);
444 			} else {
445 				error = EACCES;
446 			}
447 		}
448 	}
449 	/*
450 	 * [re]record creds for reading and/or writing if access
451 	 * was granted.  Assume the NFS server will grant read access
452 	 * for execute requests.
453 	 */
454 	if (error == 0) {
455 		if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
456 			crhold(cred);
457 			if (np->n_rucred)
458 				crfree(np->n_rucred);
459 			np->n_rucred = cred;
460 		}
461 		if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
462 			crhold(cred);
463 			if (np->n_wucred)
464 				crfree(np->n_wucred);
465 			np->n_wucred = cred;
466 		}
467 	}
468 	lwkt_reltoken(&nmp->nm_token);
469 	crfree(cred);
470 	return(error);
471 }
472 
473 /*
474  * nfs open vnode op
475  * Check to see if the type is ok
476  * and that deletion is not in progress.
477  * For paged in text files, you will need to flush the page cache
478  * if consistency is lost.
479  *
480  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
481  *	    struct file *a_fp)
482  */
483 /* ARGSUSED */
484 static int
485 nfs_open(struct vop_open_args *ap)
486 {
487 	struct vnode *vp = ap->a_vp;
488 	struct nfsnode *np = VTONFS(vp);
489 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
490 	struct vattr vattr;
491 	int error;
492 
493 	lwkt_gettoken(&nmp->nm_token);
494 
495 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
496 #ifdef DIAGNOSTIC
497 		kprintf("open eacces vtyp=%d\n",vp->v_type);
498 #endif
499 		lwkt_reltoken(&nmp->nm_token);
500 		return (EOPNOTSUPP);
501 	}
502 
503 	/*
504 	 * Save valid creds for reading and writing for later RPCs.
505 	 */
506 	if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
507 		crhold(ap->a_cred);
508 		if (np->n_rucred)
509 			crfree(np->n_rucred);
510 		np->n_rucred = ap->a_cred;
511 	}
512 	if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
513 		crhold(ap->a_cred);
514 		if (np->n_wucred)
515 			crfree(np->n_wucred);
516 		np->n_wucred = ap->a_cred;
517 	}
518 
519 	/*
520 	 * Clear the attribute cache only if opening with write access.  It
521 	 * is unclear if we should do this at all here, but we certainly
522 	 * should not clear the cache unconditionally simply because a file
523 	 * is being opened.
524 	 */
525 	if (ap->a_mode & FWRITE)
526 		np->n_attrstamp = 0;
527 
528 	/*
529 	 * For normal NFS, reconcile changes made locally verses
530 	 * changes made remotely.  Note that VOP_GETATTR only goes
531 	 * to the wire if the cached attribute has timed out or been
532 	 * cleared.
533 	 *
534 	 * If local modifications have been made clear the attribute
535 	 * cache to force an attribute and modified time check.  If
536 	 * GETATTR detects that the file has been changed by someone
537 	 * other then us it will set NRMODIFIED.
538 	 *
539 	 * If we are opening a directory and local changes have been
540 	 * made we have to invalidate the cache in order to ensure
541 	 * that we get the most up-to-date information from the
542 	 * server.  XXX
543 	 */
544 	if (np->n_flag & NLMODIFIED) {
545 		np->n_attrstamp = 0;
546 		if (vp->v_type == VDIR) {
547 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
548 			if (error == EINTR)
549 				return (error);
550 			nfs_invaldir(vp);
551 		}
552 	}
553 	error = VOP_GETATTR(vp, &vattr);
554 	if (error) {
555 		lwkt_reltoken(&nmp->nm_token);
556 		return (error);
557 	}
558 	if (np->n_flag & NRMODIFIED) {
559 		if (vp->v_type == VDIR)
560 			nfs_invaldir(vp);
561 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
562 		if (error == EINTR) {
563 			lwkt_reltoken(&nmp->nm_token);
564 			return (error);
565 		}
566 		np->n_flag &= ~NRMODIFIED;
567 	}
568 	error = vop_stdopen(ap);
569 	lwkt_reltoken(&nmp->nm_token);
570 
571 	return error;
572 }
573 
574 /*
575  * nfs close vnode op
576  * What an NFS client should do upon close after writing is a debatable issue.
577  * Most NFS clients push delayed writes to the server upon close, basically for
578  * two reasons:
579  * 1 - So that any write errors may be reported back to the client process
580  *     doing the close system call. By far the two most likely errors are
581  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582  * 2 - To put a worst case upper bound on cache inconsistency between
583  *     multiple clients for the file.
584  * There is also a consistency problem for Version 2 of the protocol w.r.t.
585  * not being able to tell if other clients are writing a file concurrently,
586  * since there is no way of knowing if the changed modify time in the reply
587  * is only due to the write for this client.
588  * (NFS Version 3 provides weak cache consistency data in the reply that
589  *  should be sufficient to detect and handle this case.)
590  *
591  * The current code does the following:
592  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594  *                     or commit them (this satisfies 1 and 2 except for the
595  *                     case where the server crashes after this close but
596  *                     before the commit RPC, which is felt to be "good
597  *                     enough". Changing the last argument to nfs_flush() to
598  *                     a 1 would force a commit operation, if it is felt a
599  *                     commit is necessary now.
600  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
601  *                     1 should be dealt with via an fsync() system call for
602  *                     cases where write errors are important.
603  *
604  * nfs_close(struct vnode *a_vp, int a_fflag)
605  */
606 /* ARGSUSED */
607 static int
608 nfs_close(struct vop_close_args *ap)
609 {
610 	struct vnode *vp = ap->a_vp;
611 	struct nfsnode *np = VTONFS(vp);
612 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
613 	int error = 0;
614 	thread_t td = curthread;
615 
616 	lwkt_gettoken(&nmp->nm_token);
617 
618 	if (vp->v_type == VREG) {
619 	    if (np->n_flag & NLMODIFIED) {
620 		if (NFS_ISV3(vp)) {
621 		    /*
622 		     * Under NFSv3 we have dirty buffers to dispose of.  We
623 		     * must flush them to the NFS server.  We have the option
624 		     * of waiting all the way through the commit rpc or just
625 		     * waiting for the initial write.  The default is to only
626 		     * wait through the initial write so the data is in the
627 		     * server's cache, which is roughly similar to the state
628 		     * a standard disk subsystem leaves the file in on close().
629 		     *
630 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
631 		     * potential races with other processes, and certainly
632 		     * cannot clear it if we don't commit.
633 		     */
634 		    int cm = nfsv3_commit_on_close ? 1 : 0;
635 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
636 		    /* np->n_flag &= ~NLMODIFIED; */
637 		} else {
638 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
639 		}
640 		np->n_attrstamp = 0;
641 	    }
642 	    if (np->n_flag & NWRITEERR) {
643 		np->n_flag &= ~NWRITEERR;
644 		error = np->n_error;
645 	    }
646 	}
647 	vop_stdclose(ap);
648 	lwkt_reltoken(&nmp->nm_token);
649 
650 	return (error);
651 }
652 
653 /*
654  * nfs getattr call from vfs.
655  *
656  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
657  */
658 static int
659 nfs_getattr(struct vop_getattr_args *ap)
660 {
661 	struct vnode *vp = ap->a_vp;
662 	struct nfsnode *np = VTONFS(vp);
663 	struct nfsmount *nmp;
664 	int error = 0;
665 	thread_t td = curthread;
666 	struct nfsm_info info;
667 
668 	info.mrep = NULL;
669 	info.v3 = NFS_ISV3(vp);
670 	nmp = VFSTONFS(vp->v_mount);
671 
672 	lwkt_gettoken(&nmp->nm_token);
673 
674 	/*
675 	 * Update local times for special files.
676 	 */
677 	if (np->n_flag & (NACC | NUPD))
678 		np->n_flag |= NCHG;
679 	/*
680 	 * First look in the cache.
681 	 */
682 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
683 		goto done;
684 
685 	if (info.v3 && nfsaccess_cache_timeout > 0) {
686 		nfsstats.accesscache_misses++;
687 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
688 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
689 			goto done;
690 	}
691 
692 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
693 	nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
694 	ERROROUT(nfsm_fhtom(&info, vp));
695 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
696 				nfs_vpcred(vp, ND_CHECK), &error));
697 	if (error == 0) {
698 		ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
699 	}
700 	m_freem(info.mrep);
701 	info.mrep = NULL;
702 done:
703 	/*
704 	 * NFS doesn't support chflags flags.  If the nfs mount was
705 	 * made -o cache set the UF_CACHE bit for swapcache.
706 	 */
707 	if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
708 		ap->a_vap->va_flags |= UF_CACHE;
709 nfsmout:
710 	lwkt_reltoken(&nmp->nm_token);
711 	return (error);
712 }
713 
714 /*
715  * nfs setattr call.
716  *
717  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
718  */
719 static int
720 nfs_setattr(struct vop_setattr_args *ap)
721 {
722 	struct vnode *vp = ap->a_vp;
723 	struct nfsnode *np = VTONFS(vp);
724 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
725 	struct vattr *vap = ap->a_vap;
726 	int biosize = vp->v_mount->mnt_stat.f_iosize;
727 	int error = 0;
728 	int boff;
729 	off_t tsize;
730 	thread_t td = curthread;
731 
732 #ifndef nolint
733 	tsize = (off_t)0;
734 #endif
735 	/*
736 	 * Setting of flags is not supported.
737 	 */
738 	if (vap->va_flags != VNOVAL)
739 		return (EOPNOTSUPP);
740 
741 	/*
742 	 * Disallow write attempts if the filesystem is mounted read-only.
743 	 */
744   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
745 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
746 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
747 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
748 		return (EROFS);
749 
750 	lwkt_gettoken(&nmp->nm_token);
751 
752 	if (vap->va_size != VNOVAL) {
753 		/*
754 		 * truncation requested
755 		 */
756  		switch (vp->v_type) {
757  		case VDIR:
758 			lwkt_reltoken(&nmp->nm_token);
759  			return (EISDIR);
760  		case VCHR:
761  		case VBLK:
762  		case VSOCK:
763  		case VFIFO:
764 			if (vap->va_mtime.tv_sec == VNOVAL &&
765 			    vap->va_atime.tv_sec == VNOVAL &&
766 			    vap->va_mode == (mode_t)VNOVAL &&
767 			    vap->va_uid == (uid_t)VNOVAL &&
768 			    vap->va_gid == (gid_t)VNOVAL) {
769 				lwkt_reltoken(&nmp->nm_token);
770 				return (0);
771 			}
772  			vap->va_size = VNOVAL;
773  			break;
774  		default:
775 			/*
776 			 * Disallow write attempts if the filesystem is
777 			 * mounted read-only.
778 			 */
779 			if (vp->v_mount->mnt_flag & MNT_RDONLY) {
780 				lwkt_reltoken(&nmp->nm_token);
781 				return (EROFS);
782 			}
783 
784 			tsize = np->n_size;
785 again:
786 			boff = (int)vap->va_size & (biosize - 1);
787 			error = nfs_meta_setsize(vp, td, vap->va_size, 0);
788 
789 #if 0
790  			if (np->n_flag & NLMODIFIED) {
791  			    if (vap->va_size == 0)
792  				error = nfs_vinvalbuf(vp, 0, 1);
793  			    else
794  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
795  			}
796 #endif
797 			/*
798 			 * note: this loop case almost always happens at
799 			 * least once per truncation.
800 			 */
801 			if (error == 0 && np->n_size != vap->va_size)
802 				goto again;
803 			np->n_vattr.va_size = vap->va_size;
804 			break;
805 		}
806 	} else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
807 		/*
808 		 * What to do.  If we are modifying the mtime we lose
809 		 * mtime detection of changes made by the server or other
810 		 * clients.  But programs like rsync/rdist/cpdup are going
811 		 * to call utimes a lot.  We don't want to piecemeal sync.
812 		 *
813 		 * For now sync if any prior remote changes were detected,
814 		 * but allow us to lose track of remote changes made during
815 		 * the utimes operation.
816 		 */
817 		if (np->n_flag & NRMODIFIED)
818 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
819 		if (error == EINTR)
820 			return (error);
821 		if (error == 0) {
822 			if (vap->va_mtime.tv_sec != VNOVAL) {
823 				np->n_mtime = vap->va_mtime.tv_sec;
824 			}
825 		}
826 	}
827 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
828 
829 	/*
830 	 * Sanity check if a truncation was issued.  This should only occur
831 	 * if multiple processes are racing on the same file.
832 	 */
833 	if (error == 0 && vap->va_size != VNOVAL &&
834 	    np->n_size != vap->va_size) {
835 		kprintf("NFS ftruncate: server disagrees on the file size: "
836 			"%jd/%jd/%jd\n",
837 			(intmax_t)tsize,
838 			(intmax_t)vap->va_size,
839 			(intmax_t)np->n_size);
840 		goto again;
841 	}
842 	if (error && vap->va_size != VNOVAL) {
843 		np->n_size = np->n_vattr.va_size = tsize;
844 		nfs_meta_setsize(vp, td, np->n_size, 0);
845 	}
846 	lwkt_reltoken(&nmp->nm_token);
847 
848 	return (error);
849 }
850 
851 /*
852  * Do an nfs setattr rpc.
853  */
854 static int
855 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
856 	       struct ucred *cred, struct thread *td)
857 {
858 	struct nfsv2_sattr *sp;
859 	struct nfsnode *np = VTONFS(vp);
860 	u_int32_t *tl;
861 	int error = 0, wccflag = NFSV3_WCCRATTR;
862 	struct nfsm_info info;
863 
864 	info.mrep = NULL;
865 	info.v3 = NFS_ISV3(vp);
866 
867 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
868 	nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
869 		     NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
870 	ERROROUT(nfsm_fhtom(&info, vp));
871 	if (info.v3) {
872 		nfsm_v3attrbuild(&info, vap, TRUE);
873 		tl = nfsm_build(&info, NFSX_UNSIGNED);
874 		*tl = nfs_false;
875 	} else {
876 		sp = nfsm_build(&info, NFSX_V2SATTR);
877 		if (vap->va_mode == (mode_t)VNOVAL)
878 			sp->sa_mode = nfs_xdrneg1;
879 		else
880 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
881 		if (vap->va_uid == (uid_t)VNOVAL)
882 			sp->sa_uid = nfs_xdrneg1;
883 		else
884 			sp->sa_uid = txdr_unsigned(vap->va_uid);
885 		if (vap->va_gid == (gid_t)VNOVAL)
886 			sp->sa_gid = nfs_xdrneg1;
887 		else
888 			sp->sa_gid = txdr_unsigned(vap->va_gid);
889 		sp->sa_size = txdr_unsigned(vap->va_size);
890 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
891 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
892 	}
893 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
894 	if (info.v3) {
895 		np->n_modestamp = 0;
896 		ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
897 	} else {
898 		ERROROUT(nfsm_loadattr(&info, vp, NULL));
899 	}
900 	m_freem(info.mrep);
901 	info.mrep = NULL;
902 nfsmout:
903 	return (error);
904 }
905 
906 static
907 void
908 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
909 {
910 	if (nctimeout == 0)
911 		nctimeout = 1;
912 	else
913 		nctimeout *= hz;
914 	cache_setvp(nch, vp);
915 	cache_settimeout(nch, nctimeout);
916 }
917 
918 /*
919  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
920  * nfs_lookup() until all remaining new api calls are implemented.
921  *
922  * Resolve a namecache entry.  This function is passed a locked ncp and
923  * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
924  */
925 static int
926 nfs_nresolve(struct vop_nresolve_args *ap)
927 {
928 	struct thread *td = curthread;
929 	struct namecache *ncp;
930 	struct nfsmount *nmp;
931 	struct ucred *cred;
932 	struct nfsnode *np;
933 	struct vnode *dvp;
934 	struct vnode *nvp;
935 	nfsfh_t *fhp;
936 	int attrflag;
937 	int fhsize;
938 	int error;
939 	int tmp_error;
940 	int len;
941 	struct nfsm_info info;
942 
943 	cred = ap->a_cred;
944 	dvp = ap->a_dvp;
945 	nmp = VFSTONFS(dvp->v_mount);
946 
947 	lwkt_gettoken(&nmp->nm_token);
948 
949 	if ((error = vget(dvp, LK_SHARED)) != 0) {
950 		lwkt_reltoken(&nmp->nm_token);
951 		return (error);
952 	}
953 
954 	info.mrep = NULL;
955 	info.v3 = NFS_ISV3(dvp);
956 
957 	nvp = NULL;
958 	nfsstats.lookupcache_misses++;
959 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
960 	ncp = ap->a_nch->ncp;
961 	len = ncp->nc_nlen;
962 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
963 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
964 	ERROROUT(nfsm_fhtom(&info, dvp));
965 	ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
966 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
967 				ap->a_cred, &error));
968 	if (error) {
969 		/*
970 		 * Cache negatve lookups to reduce NFS traffic, but use
971 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
972 		 * XXX we should add a namecache flag for no-caching
973 		 * to uncache the negative hit as soon as possible, but
974 		 * we cannot simply destroy the entry because it is used
975 		 * as a placeholder by the caller.
976 		 *
977 		 * The refactored nfs code will overwrite a non-zero error
978 		 * with 0 when we use ERROROUT(), so don't here.
979 		 */
980 		if (error == ENOENT)
981 			nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
982 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
983 					     NFS_LATTR_NOSHRINK);
984 		if (tmp_error) {
985 			error = tmp_error;
986 			goto nfsmout;
987 		}
988 		m_freem(info.mrep);
989 		info.mrep = NULL;
990 		goto nfsmout;
991 	}
992 
993 	/*
994 	 * Success, get the file handle, do various checks, and load
995 	 * post-operation data from the reply packet.  Theoretically
996 	 * we should never be looking up "." so, theoretically, we
997 	 * should never get the same file handle as our directory.  But
998 	 * we check anyway. XXX
999 	 *
1000 	 * Note that no timeout is set for the positive cache hit.  We
1001 	 * assume, theoretically, that ESTALE returns will be dealt with
1002 	 * properly to handle NFS races and in anycase we cannot depend
1003 	 * on a timeout to deal with NFS open/create/excl issues so instead
1004 	 * of a bad hack here the rest of the NFS client code needs to do
1005 	 * the right thing.
1006 	 */
1007 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1008 
1009 	np = VTONFS(dvp);
1010 	if (NFS_CMPFH(np, fhp, fhsize)) {
1011 		vref(dvp);
1012 		nvp = dvp;
1013 	} else {
1014 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1015 		if (error) {
1016 			m_freem(info.mrep);
1017 			info.mrep = NULL;
1018 			vput(dvp);
1019 			lwkt_reltoken(&nmp->nm_token);
1020 			return (error);
1021 		}
1022 		nvp = NFSTOV(np);
1023 	}
1024 	if (info.v3) {
1025 		ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1026 					  NFS_LATTR_NOSHRINK));
1027 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1028 					  NFS_LATTR_NOSHRINK));
1029 	} else {
1030 		ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1031 	}
1032 	nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1033 	m_freem(info.mrep);
1034 	info.mrep = NULL;
1035 nfsmout:
1036 	lwkt_reltoken(&nmp->nm_token);
1037 	vput(dvp);
1038 	if (nvp) {
1039 		if (nvp == dvp)
1040 			vrele(nvp);
1041 		else
1042 			vput(nvp);
1043 	}
1044 	return (error);
1045 }
1046 
1047 /*
1048  * 'cached' nfs directory lookup
1049  *
1050  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1051  *
1052  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1053  *	      struct componentname *a_cnp)
1054  */
1055 static int
1056 nfs_lookup(struct vop_old_lookup_args *ap)
1057 {
1058 	struct componentname *cnp = ap->a_cnp;
1059 	struct vnode *dvp = ap->a_dvp;
1060 	struct vnode **vpp = ap->a_vpp;
1061 	int flags = cnp->cn_flags;
1062 	struct vnode *newvp;
1063 	struct nfsmount *nmp;
1064 	long len;
1065 	nfsfh_t *fhp;
1066 	struct nfsnode *np;
1067 	int lockparent, wantparent, attrflag, fhsize;
1068 	int error;
1069 	int tmp_error;
1070 	struct nfsm_info info;
1071 
1072 	info.mrep = NULL;
1073 	info.v3 = NFS_ISV3(dvp);
1074 	error = 0;
1075 
1076 	/*
1077 	 * Read-only mount check and directory check.
1078 	 */
1079 	*vpp = NULLVP;
1080 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1081 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1082 		return (EROFS);
1083 
1084 	if (dvp->v_type != VDIR)
1085 		return (ENOTDIR);
1086 
1087 	/*
1088 	 * Look it up in the cache.  Note that ENOENT is only returned if we
1089 	 * previously entered a negative hit (see later on).  The additional
1090 	 * nfsneg_cache_timeout check causes previously cached results to
1091 	 * be instantly ignored if the negative caching is turned off.
1092 	 */
1093 	lockparent = flags & CNP_LOCKPARENT;
1094 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1095 	nmp = VFSTONFS(dvp->v_mount);
1096 	np = VTONFS(dvp);
1097 
1098 	lwkt_gettoken(&nmp->nm_token);
1099 
1100 	/*
1101 	 * Go to the wire.
1102 	 */
1103 	error = 0;
1104 	newvp = NULLVP;
1105 	nfsstats.lookupcache_misses++;
1106 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1107 	len = cnp->cn_namelen;
1108 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1109 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1110 	ERROROUT(nfsm_fhtom(&info, dvp));
1111 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1112 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1113 				cnp->cn_cred, &error));
1114 	if (error) {
1115 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1116 					     NFS_LATTR_NOSHRINK);
1117 		if (tmp_error) {
1118 			error = tmp_error;
1119 			goto nfsmout;
1120 		}
1121 
1122 		m_freem(info.mrep);
1123 		info.mrep = NULL;
1124 		goto nfsmout;
1125 	}
1126 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1127 
1128 	/*
1129 	 * Handle RENAME case...
1130 	 */
1131 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1132 		if (NFS_CMPFH(np, fhp, fhsize)) {
1133 			m_freem(info.mrep);
1134 			info.mrep = NULL;
1135 			lwkt_reltoken(&nmp->nm_token);
1136 			return (EISDIR);
1137 		}
1138 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1139 		if (error) {
1140 			m_freem(info.mrep);
1141 			info.mrep = NULL;
1142 			lwkt_reltoken(&nmp->nm_token);
1143 			return (error);
1144 		}
1145 		newvp = NFSTOV(np);
1146 		if (info.v3) {
1147 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1148 						  NFS_LATTR_NOSHRINK));
1149 			ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1150 						  NFS_LATTR_NOSHRINK));
1151 		} else {
1152 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1153 		}
1154 		*vpp = newvp;
1155 		m_freem(info.mrep);
1156 		info.mrep = NULL;
1157 		if (!lockparent) {
1158 			vn_unlock(dvp);
1159 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1160 		}
1161 		lwkt_reltoken(&nmp->nm_token);
1162 		return (0);
1163 	}
1164 
1165 	if (flags & CNP_ISDOTDOT) {
1166 		vn_unlock(dvp);
1167 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1168 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1169 		if (error) {
1170 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1171 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1172 			lwkt_reltoken(&nmp->nm_token);
1173 			return (error); /* NOTE: return error from nget */
1174 		}
1175 		newvp = NFSTOV(np);
1176 		if (lockparent) {
1177 			error = vn_lock(dvp, LK_EXCLUSIVE);
1178 			if (error) {
1179 				vput(newvp);
1180 				lwkt_reltoken(&nmp->nm_token);
1181 				return (error);
1182 			}
1183 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1184 		}
1185 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1186 		vref(dvp);
1187 		newvp = dvp;
1188 	} else {
1189 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1190 		if (error) {
1191 			m_freem(info.mrep);
1192 			info.mrep = NULL;
1193 			lwkt_reltoken(&nmp->nm_token);
1194 			return (error);
1195 		}
1196 		if (!lockparent) {
1197 			vn_unlock(dvp);
1198 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1199 		}
1200 		newvp = NFSTOV(np);
1201 	}
1202 	if (info.v3) {
1203 		ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1204 					  NFS_LATTR_NOSHRINK));
1205 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1206 					  NFS_LATTR_NOSHRINK));
1207 	} else {
1208 		ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1209 	}
1210 #if 0
1211 	/* XXX MOVE TO nfs_nremove() */
1212 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1213 	    cnp->cn_nameiop != NAMEI_DELETE) {
1214 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1215 	}
1216 #endif
1217 	*vpp = newvp;
1218 	m_freem(info.mrep);
1219 	info.mrep = NULL;
1220 nfsmout:
1221 	if (error) {
1222 		if (newvp != NULLVP) {
1223 			vrele(newvp);
1224 			*vpp = NULLVP;
1225 		}
1226 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1227 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1228 		    error == ENOENT) {
1229 			if (!lockparent) {
1230 				vn_unlock(dvp);
1231 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1232 			}
1233 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1234 				error = EROFS;
1235 			else
1236 				error = EJUSTRETURN;
1237 		}
1238 	}
1239 	lwkt_reltoken(&nmp->nm_token);
1240 	return (error);
1241 }
1242 
1243 /*
1244  * nfs read call.
1245  * Just call nfs_bioread() to do the work.
1246  *
1247  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1248  *	    struct ucred *a_cred)
1249  */
1250 static int
1251 nfs_read(struct vop_read_args *ap)
1252 {
1253 	struct vnode *vp = ap->a_vp;
1254 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1255 	int error;
1256 
1257 	lwkt_gettoken(&nmp->nm_token);
1258 	error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1259 	lwkt_reltoken(&nmp->nm_token);
1260 
1261 	return error;
1262 }
1263 
1264 /*
1265  * nfs readlink call
1266  *
1267  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1268  */
1269 static int
1270 nfs_readlink(struct vop_readlink_args *ap)
1271 {
1272 	struct vnode *vp = ap->a_vp;
1273 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1274 	int error;
1275 
1276 	if (vp->v_type != VLNK)
1277 		return (EINVAL);
1278 
1279 	lwkt_gettoken(&nmp->nm_token);
1280 	error = nfs_bioread(vp, ap->a_uio, 0);
1281 	lwkt_reltoken(&nmp->nm_token);
1282 
1283 	return error;
1284 }
1285 
1286 /*
1287  * Do a readlink rpc.
1288  * Called by nfs_doio() from below the buffer cache.
1289  */
1290 int
1291 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1292 {
1293 	int error = 0, len, attrflag;
1294 	struct nfsm_info info;
1295 
1296 	info.mrep = NULL;
1297 	info.v3 = NFS_ISV3(vp);
1298 
1299 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1300 	nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1301 	ERROROUT(nfsm_fhtom(&info, vp));
1302 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1303 				nfs_vpcred(vp, ND_CHECK), &error));
1304 	if (info.v3) {
1305 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1306 					  NFS_LATTR_NOSHRINK));
1307 	}
1308 	if (!error) {
1309 		NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1310 		if (len == NFS_MAXPATHLEN) {
1311 			struct nfsnode *np = VTONFS(vp);
1312 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1313 				len = np->n_size;
1314 		}
1315 		ERROROUT(nfsm_mtouio(&info, uiop, len));
1316 	}
1317 	m_freem(info.mrep);
1318 	info.mrep = NULL;
1319 nfsmout:
1320 	return (error);
1321 }
1322 
1323 /*
1324  * nfs synchronous read rpc using UIO
1325  */
1326 int
1327 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1328 {
1329 	u_int32_t *tl;
1330 	struct nfsmount *nmp;
1331 	int error = 0, len, retlen, tsiz, eof, attrflag;
1332 	struct nfsm_info info;
1333 	off_t tmp_off;
1334 
1335 	info.mrep = NULL;
1336 	info.v3 = NFS_ISV3(vp);
1337 
1338 #ifndef nolint
1339 	eof = 0;
1340 #endif
1341 	nmp = VFSTONFS(vp->v_mount);
1342 
1343 	tsiz = uiop->uio_resid;
1344 	tmp_off = uiop->uio_offset + tsiz;
1345 	if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1346 		return (EFBIG);
1347 	tmp_off = uiop->uio_offset;
1348 	while (tsiz > 0) {
1349 		nfsstats.rpccnt[NFSPROC_READ]++;
1350 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1351 		nfsm_reqhead(&info, vp, NFSPROC_READ,
1352 			     NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1353 		ERROROUT(nfsm_fhtom(&info, vp));
1354 		tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1355 		if (info.v3) {
1356 			txdr_hyper(uiop->uio_offset, tl);
1357 			*(tl + 2) = txdr_unsigned(len);
1358 		} else {
1359 			*tl++ = txdr_unsigned(uiop->uio_offset);
1360 			*tl++ = txdr_unsigned(len);
1361 			*tl = 0;
1362 		}
1363 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1364 					nfs_vpcred(vp, ND_READ), &error));
1365 		if (info.v3) {
1366 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1367 						 NFS_LATTR_NOSHRINK));
1368 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1369 			eof = fxdr_unsigned(int, *(tl + 1));
1370 		} else {
1371 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1372 		}
1373 		NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1374 		ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1375 		m_freem(info.mrep);
1376 		info.mrep = NULL;
1377 
1378 		/*
1379 		 * Handle short-read from server (NFSv3).  If EOF is not
1380 		 * flagged (and no error occurred), but retlen is less
1381 		 * then the request size, we must zero-fill the remainder.
1382 		 */
1383 		if (retlen < len && info.v3 && eof == 0) {
1384 			ERROROUT(uiomovez(len - retlen, uiop));
1385 			retlen = len;
1386 		}
1387 		tsiz -= retlen;
1388 
1389 		/*
1390 		 * Terminate loop on EOF or zero-length read.
1391 		 *
1392 		 * For NFSv2 a short-read indicates EOF, not zero-fill,
1393 		 * and also terminates the loop.
1394 		 */
1395 		if (info.v3) {
1396 			if (eof || retlen == 0)
1397 				tsiz = 0;
1398 		} else if (retlen < len) {
1399 			tsiz = 0;
1400 		}
1401 	}
1402 nfsmout:
1403 	return (error);
1404 }
1405 
1406 /*
1407  * nfs write call
1408  */
1409 int
1410 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1411 		 int *iomode, int *must_commit)
1412 {
1413 	u_int32_t *tl;
1414 	int32_t backup;
1415 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1416 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1417 	int  committed = NFSV3WRITE_FILESYNC;
1418 	struct nfsm_info info;
1419 
1420 	info.mrep = NULL;
1421 	info.v3 = NFS_ISV3(vp);
1422 
1423 #ifndef DIAGNOSTIC
1424 	if (uiop->uio_iovcnt != 1)
1425 		panic("nfs: writerpc iovcnt > 1");
1426 #endif
1427 	*must_commit = 0;
1428 	tsiz = uiop->uio_resid;
1429 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1430 		return (EFBIG);
1431 	while (tsiz > 0) {
1432 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1433 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1434 		nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1435 			     NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1436 			     nfsm_rndup(len));
1437 		ERROROUT(nfsm_fhtom(&info, vp));
1438 		if (info.v3) {
1439 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1440 			txdr_hyper(uiop->uio_offset, tl);
1441 			tl += 2;
1442 			*tl++ = txdr_unsigned(len);
1443 			*tl++ = txdr_unsigned(*iomode);
1444 			*tl = txdr_unsigned(len);
1445 		} else {
1446 			u_int32_t x;
1447 
1448 			tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1449 			/* Set both "begin" and "current" to non-garbage. */
1450 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1451 			*tl++ = x;	/* "begin offset" */
1452 			*tl++ = x;	/* "current offset" */
1453 			x = txdr_unsigned(len);
1454 			*tl++ = x;	/* total to this offset */
1455 			*tl = x;	/* size of this write */
1456 		}
1457 		ERROROUT(nfsm_uiotom(&info, uiop, len));
1458 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1459 					nfs_vpcred(vp, ND_WRITE), &error));
1460 		if (info.v3) {
1461 			/*
1462 			 * The write RPC returns a before and after mtime.  The
1463 			 * nfsm_wcc_data() macro checks the before n_mtime
1464 			 * against the before time and stores the after time
1465 			 * in the nfsnode's cached vattr and n_mtime field.
1466 			 * The NRMODIFIED bit will be set if the before
1467 			 * time did not match the original mtime.
1468 			 */
1469 			wccflag = NFSV3_WCCCHK;
1470 			ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1471 			if (error == 0) {
1472 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1473 				rlen = fxdr_unsigned(int, *tl++);
1474 				if (rlen == 0) {
1475 					error = NFSERR_IO;
1476 					m_freem(info.mrep);
1477 					info.mrep = NULL;
1478 					break;
1479 				} else if (rlen < len) {
1480 					backup = len - rlen;
1481 					uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1482 					uiop->uio_iov->iov_len += backup;
1483 					uiop->uio_offset -= backup;
1484 					uiop->uio_resid += backup;
1485 					len = rlen;
1486 				}
1487 				commit = fxdr_unsigned(int, *tl++);
1488 
1489 				/*
1490 				 * Return the lowest committment level
1491 				 * obtained by any of the RPCs.
1492 				 */
1493 				if (committed == NFSV3WRITE_FILESYNC)
1494 					committed = commit;
1495 				else if (committed == NFSV3WRITE_DATASYNC &&
1496 					commit == NFSV3WRITE_UNSTABLE)
1497 					committed = commit;
1498 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1499 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1500 					NFSX_V3WRITEVERF);
1501 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1502 				} else if (bcmp((caddr_t)tl,
1503 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1504 				    *must_commit = 1;
1505 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1506 					NFSX_V3WRITEVERF);
1507 				}
1508 			}
1509 		} else {
1510 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1511 		}
1512 		m_freem(info.mrep);
1513 		info.mrep = NULL;
1514 		if (error)
1515 			break;
1516 		tsiz -= len;
1517 	}
1518 nfsmout:
1519 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1520 		committed = NFSV3WRITE_FILESYNC;
1521 	*iomode = committed;
1522 	if (error)
1523 		uiop->uio_resid = tsiz;
1524 	return (error);
1525 }
1526 
1527 /*
1528  * nfs mknod rpc
1529  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1530  * mode set to specify the file type and the size field for rdev.
1531  */
1532 static int
1533 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1534 	     struct vattr *vap)
1535 {
1536 	struct nfsv2_sattr *sp;
1537 	u_int32_t *tl;
1538 	struct vnode *newvp = NULL;
1539 	struct nfsnode *np = NULL;
1540 	struct vattr vattr;
1541 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1542 	int rmajor, rminor;
1543 	struct nfsm_info info;
1544 
1545 	info.mrep = NULL;
1546 	info.v3 = NFS_ISV3(dvp);
1547 
1548 	if (vap->va_type == VCHR || vap->va_type == VBLK) {
1549 		rmajor = txdr_unsigned(vap->va_rmajor);
1550 		rminor = txdr_unsigned(vap->va_rminor);
1551 	} else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1552 		rmajor = nfs_xdrneg1;
1553 		rminor = nfs_xdrneg1;
1554 	} else {
1555 		return (EOPNOTSUPP);
1556 	}
1557 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1558 		return (error);
1559 	}
1560 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1561 	nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1562 		     NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1563 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1564 	ERROROUT(nfsm_fhtom(&info, dvp));
1565 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1566 			     NFS_MAXNAMLEN));
1567 	if (info.v3) {
1568 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1569 		*tl++ = vtonfsv3_type(vap->va_type);
1570 		nfsm_v3attrbuild(&info, vap, FALSE);
1571 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1572 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1573 			*tl++ = txdr_unsigned(vap->va_rmajor);
1574 			*tl = txdr_unsigned(vap->va_rminor);
1575 		}
1576 	} else {
1577 		sp = nfsm_build(&info, NFSX_V2SATTR);
1578 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1579 		sp->sa_uid = nfs_xdrneg1;
1580 		sp->sa_gid = nfs_xdrneg1;
1581 		sp->sa_size = makeudev(rmajor, rminor);
1582 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1583 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1584 	}
1585 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1586 				cnp->cn_cred, &error));
1587 	if (!error) {
1588 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1589 		if (!gotvp) {
1590 			if (newvp) {
1591 				vput(newvp);
1592 				newvp = NULL;
1593 			}
1594 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1595 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1596 			if (!error)
1597 				newvp = NFSTOV(np);
1598 		}
1599 	}
1600 	if (info.v3) {
1601 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1602 	}
1603 	m_freem(info.mrep);
1604 	info.mrep = NULL;
1605 nfsmout:
1606 	if (error) {
1607 		if (newvp)
1608 			vput(newvp);
1609 	} else {
1610 		*vpp = newvp;
1611 	}
1612 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1613 	if (!wccflag)
1614 		VTONFS(dvp)->n_attrstamp = 0;
1615 	return (error);
1616 }
1617 
1618 /*
1619  * nfs mknod vop
1620  * just call nfs_mknodrpc() to do the work.
1621  *
1622  * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1623  *	     struct componentname *a_cnp, struct vattr *a_vap)
1624  */
1625 /* ARGSUSED */
1626 static int
1627 nfs_mknod(struct vop_old_mknod_args *ap)
1628 {
1629 	struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1630 	int error;
1631 
1632 	lwkt_gettoken(&nmp->nm_token);
1633 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1634 	lwkt_reltoken(&nmp->nm_token);
1635 
1636 	return error;
1637 }
1638 
1639 static u_long create_verf;
1640 /*
1641  * nfs file create call
1642  *
1643  * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1644  *	      struct componentname *a_cnp, struct vattr *a_vap)
1645  */
1646 static int
1647 nfs_create(struct vop_old_create_args *ap)
1648 {
1649 	struct vnode *dvp = ap->a_dvp;
1650 	struct vattr *vap = ap->a_vap;
1651 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1652 	struct componentname *cnp = ap->a_cnp;
1653 	struct nfsv2_sattr *sp;
1654 	u_int32_t *tl;
1655 	struct nfsnode *np = NULL;
1656 	struct vnode *newvp = NULL;
1657 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1658 	struct vattr vattr;
1659 	struct nfsm_info info;
1660 
1661 	info.mrep = NULL;
1662 	info.v3 = NFS_ISV3(dvp);
1663 	lwkt_gettoken(&nmp->nm_token);
1664 
1665 	/*
1666 	 * Oops, not for me..
1667 	 */
1668 	if (vap->va_type == VSOCK) {
1669 		error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1670 		lwkt_reltoken(&nmp->nm_token);
1671 		return error;
1672 	}
1673 
1674 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1675 		lwkt_reltoken(&nmp->nm_token);
1676 		return (error);
1677 	}
1678 	if (vap->va_vaflags & VA_EXCLUSIVE)
1679 		fmode |= O_EXCL;
1680 again:
1681 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1682 	nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1683 		     NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1684 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1685 	ERROROUT(nfsm_fhtom(&info, dvp));
1686 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1687 			     NFS_MAXNAMLEN));
1688 	if (info.v3) {
1689 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1690 		if (fmode & O_EXCL) {
1691 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1692 			tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1693 #ifdef INET
1694 			if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1695 				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1696 			else
1697 #endif
1698 				*tl++ = create_verf;
1699 			*tl = ++create_verf;
1700 		} else {
1701 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1702 			nfsm_v3attrbuild(&info, vap, FALSE);
1703 		}
1704 	} else {
1705 		sp = nfsm_build(&info, NFSX_V2SATTR);
1706 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1707 		sp->sa_uid = nfs_xdrneg1;
1708 		sp->sa_gid = nfs_xdrneg1;
1709 		sp->sa_size = 0;
1710 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1711 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1712 	}
1713 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1714 				cnp->cn_cred, &error));
1715 	if (error == 0) {
1716 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1717 		if (!gotvp) {
1718 			if (newvp) {
1719 				vput(newvp);
1720 				newvp = NULL;
1721 			}
1722 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1723 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1724 			if (!error)
1725 				newvp = NFSTOV(np);
1726 		}
1727 	}
1728 	if (info.v3) {
1729 		if (error == 0)
1730 			error = nfsm_wcc_data(&info, dvp, &wccflag);
1731 		else
1732 			(void)nfsm_wcc_data(&info, dvp, &wccflag);
1733 	}
1734 	m_freem(info.mrep);
1735 	info.mrep = NULL;
1736 nfsmout:
1737 	if (error) {
1738 		if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1739 			KKASSERT(newvp == NULL);
1740 			fmode &= ~O_EXCL;
1741 			goto again;
1742 		}
1743 	} else if (info.v3 && (fmode & O_EXCL)) {
1744 		/*
1745 		 * We are normally called with only a partially initialized
1746 		 * VAP.  Since the NFSv3 spec says that server may use the
1747 		 * file attributes to store the verifier, the spec requires
1748 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1749 		 * in atime, but we can't really assume that all servers will
1750 		 * so we ensure that our SETATTR sets both atime and mtime.
1751 		 */
1752 		if (vap->va_mtime.tv_sec == VNOVAL)
1753 			vfs_timestamp(&vap->va_mtime);
1754 		if (vap->va_atime.tv_sec == VNOVAL)
1755 			vap->va_atime = vap->va_mtime;
1756 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1757 	}
1758 	if (error == 0) {
1759 		/*
1760 		 * The new np may have enough info for access
1761 		 * checks, make sure rucred and wucred are
1762 		 * initialized for read and write rpc's.
1763 		 */
1764 		np = VTONFS(newvp);
1765 		if (np->n_rucred == NULL)
1766 			np->n_rucred = crhold(cnp->cn_cred);
1767 		if (np->n_wucred == NULL)
1768 			np->n_wucred = crhold(cnp->cn_cred);
1769 		*ap->a_vpp = newvp;
1770 	} else if (newvp) {
1771 		vput(newvp);
1772 	}
1773 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1774 	if (!wccflag)
1775 		VTONFS(dvp)->n_attrstamp = 0;
1776 	lwkt_reltoken(&nmp->nm_token);
1777 	return (error);
1778 }
1779 
1780 /*
1781  * nfs file remove call
1782  * To try and make nfs semantics closer to ufs semantics, a file that has
1783  * other processes using the vnode is renamed instead of removed and then
1784  * removed later on the last close.
1785  * - If v_sysref.refcnt > 1
1786  *	  If a rename is not already in the works
1787  *	     call nfs_sillyrename() to set it up
1788  *     else
1789  *	  do the remove rpc
1790  *
1791  * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1792  *	      struct componentname *a_cnp)
1793  */
1794 static int
1795 nfs_remove(struct vop_old_remove_args *ap)
1796 {
1797 	struct vnode *vp = ap->a_vp;
1798 	struct vnode *dvp = ap->a_dvp;
1799 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1800 	struct componentname *cnp = ap->a_cnp;
1801 	struct nfsnode *np = VTONFS(vp);
1802 	int error = 0;
1803 	struct vattr vattr;
1804 
1805 	lwkt_gettoken(&nmp->nm_token);
1806 #ifndef DIAGNOSTIC
1807 	if (vp->v_sysref.refcnt < 1)
1808 		panic("nfs_remove: bad v_sysref.refcnt");
1809 #endif
1810 	if (vp->v_type == VDIR) {
1811 		error = EPERM;
1812 	} else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1813 		   VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1814 		/*
1815 		 * throw away biocache buffers, mainly to avoid
1816 		 * unnecessary delayed writes later.
1817 		 */
1818 		error = nfs_vinvalbuf(vp, 0, 1);
1819 		/* Do the rpc */
1820 		if (error != EINTR)
1821 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1822 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1823 		/*
1824 		 * Kludge City: If the first reply to the remove rpc is lost..
1825 		 *   the reply to the retransmitted request will be ENOENT
1826 		 *   since the file was in fact removed
1827 		 *   Therefore, we cheat and return success.
1828 		 */
1829 		if (error == ENOENT)
1830 			error = 0;
1831 	} else if (!np->n_sillyrename) {
1832 		error = nfs_sillyrename(dvp, vp, cnp);
1833 	}
1834 	np->n_attrstamp = 0;
1835 	lwkt_reltoken(&nmp->nm_token);
1836 
1837 	return (error);
1838 }
1839 
1840 /*
1841  * nfs file remove rpc called from nfs_inactive
1842  */
1843 int
1844 nfs_removeit(struct sillyrename *sp)
1845 {
1846 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1847 		sp->s_cred, NULL));
1848 }
1849 
1850 /*
1851  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1852  */
1853 static int
1854 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1855 	      struct ucred *cred, struct thread *td)
1856 {
1857 	int error = 0, wccflag = NFSV3_WCCRATTR;
1858 	struct nfsm_info info;
1859 
1860 	info.mrep = NULL;
1861 	info.v3 = NFS_ISV3(dvp);
1862 
1863 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1864 	nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1865 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1866 	ERROROUT(nfsm_fhtom(&info, dvp));
1867 	ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1868 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1869 	if (info.v3) {
1870 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1871 	}
1872 	m_freem(info.mrep);
1873 	info.mrep = NULL;
1874 nfsmout:
1875 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1876 	if (!wccflag)
1877 		VTONFS(dvp)->n_attrstamp = 0;
1878 	return (error);
1879 }
1880 
1881 /*
1882  * nfs file rename call
1883  *
1884  * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1885  *	      struct componentname *a_fcnp, struct vnode *a_tdvp,
1886  *	      struct vnode *a_tvp, struct componentname *a_tcnp)
1887  */
1888 static int
1889 nfs_rename(struct vop_old_rename_args *ap)
1890 {
1891 	struct vnode *fvp = ap->a_fvp;
1892 	struct vnode *tvp = ap->a_tvp;
1893 	struct vnode *fdvp = ap->a_fdvp;
1894 	struct vnode *tdvp = ap->a_tdvp;
1895 	struct componentname *tcnp = ap->a_tcnp;
1896 	struct componentname *fcnp = ap->a_fcnp;
1897 	struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1898 	int error;
1899 
1900 	lwkt_gettoken(&nmp->nm_token);
1901 
1902 	/* Check for cross-device rename */
1903 	if ((fvp->v_mount != tdvp->v_mount) ||
1904 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1905 		error = EXDEV;
1906 		goto out;
1907 	}
1908 
1909 	/*
1910 	 * We shouldn't have to flush fvp on rename for most server-side
1911 	 * filesystems as the file handle should not change.  Unfortunately
1912 	 * the inode for some filesystems (msdosfs) might be tied to the
1913 	 * file name or directory position so to be completely safe
1914 	 * vfs.nfs.flush_on_rename is set by default.  Clear to improve
1915 	 * performance.
1916 	 *
1917 	 * We must flush tvp on rename because it might become stale on the
1918 	 * server after the rename.
1919 	 */
1920 	if (nfs_flush_on_rename)
1921 	    VOP_FSYNC(fvp, MNT_WAIT, 0);
1922 	if (tvp)
1923 	    VOP_FSYNC(tvp, MNT_WAIT, 0);
1924 
1925 	/*
1926 	 * If the tvp exists and is in use, sillyrename it before doing the
1927 	 * rename of the new file over it.
1928 	 *
1929 	 * XXX Can't sillyrename a directory.
1930 	 *
1931 	 * We do not attempt to do any namecache purges in this old API
1932 	 * routine.  The new API compat functions have access to the actual
1933 	 * namecache structures and will do it for us.
1934 	 */
1935 	if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1936 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1937 		vput(tvp);
1938 		tvp = NULL;
1939 	} else if (tvp) {
1940 		;
1941 	}
1942 
1943 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1944 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1945 		tcnp->cn_td);
1946 
1947 out:
1948 	lwkt_reltoken(&nmp->nm_token);
1949 	if (tdvp == tvp)
1950 		vrele(tdvp);
1951 	else
1952 		vput(tdvp);
1953 	if (tvp)
1954 		vput(tvp);
1955 	vrele(fdvp);
1956 	vrele(fvp);
1957 	/*
1958 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1959 	 */
1960 	if (error == ENOENT)
1961 		error = 0;
1962 	return (error);
1963 }
1964 
1965 /*
1966  * nfs file rename rpc called from nfs_remove() above
1967  */
1968 static int
1969 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1970 	     struct sillyrename *sp)
1971 {
1972 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1973 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1974 }
1975 
1976 /*
1977  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1978  */
1979 static int
1980 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1981 	      struct vnode *tdvp, const char *tnameptr, int tnamelen,
1982 	      struct ucred *cred, struct thread *td)
1983 {
1984 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1985 	struct nfsm_info info;
1986 
1987 	info.mrep = NULL;
1988 	info.v3 = NFS_ISV3(fdvp);
1989 
1990 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1991 	nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1992 		    (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1993 		    nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1994 	ERROROUT(nfsm_fhtom(&info, fdvp));
1995 	ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1996 	ERROROUT(nfsm_fhtom(&info, tdvp));
1997 	ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1998 	NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1999 	if (info.v3) {
2000 		ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
2001 		ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
2002 	}
2003 	m_freem(info.mrep);
2004 	info.mrep = NULL;
2005 nfsmout:
2006 	VTONFS(fdvp)->n_flag |= NLMODIFIED;
2007 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2008 	if (!fwccflag)
2009 		VTONFS(fdvp)->n_attrstamp = 0;
2010 	if (!twccflag)
2011 		VTONFS(tdvp)->n_attrstamp = 0;
2012 	return (error);
2013 }
2014 
2015 /*
2016  * nfs hard link create call
2017  *
2018  * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2019  *	    struct componentname *a_cnp)
2020  */
2021 static int
2022 nfs_link(struct vop_old_link_args *ap)
2023 {
2024 	struct vnode *vp = ap->a_vp;
2025 	struct vnode *tdvp = ap->a_tdvp;
2026 	struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2027 	struct componentname *cnp = ap->a_cnp;
2028 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2029 	struct nfsm_info info;
2030 
2031 	if (vp->v_mount != tdvp->v_mount) {
2032 		return (EXDEV);
2033 	}
2034 	lwkt_gettoken(&nmp->nm_token);
2035 
2036 	/*
2037 	 * The attribute cache may get out of sync with the server on link.
2038 	 * Pushing writes to the server before handle was inherited from
2039 	 * long long ago and it is unclear if we still need to do this.
2040 	 * Defaults to off.
2041 	 */
2042 	if (nfs_flush_on_hlink)
2043 		VOP_FSYNC(vp, MNT_WAIT, 0);
2044 
2045 	info.mrep = NULL;
2046 	info.v3 = NFS_ISV3(vp);
2047 
2048 	nfsstats.rpccnt[NFSPROC_LINK]++;
2049 	nfsm_reqhead(&info, vp, NFSPROC_LINK,
2050 		     NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2051 		     nfsm_rndup(cnp->cn_namelen));
2052 	ERROROUT(nfsm_fhtom(&info, vp));
2053 	ERROROUT(nfsm_fhtom(&info, tdvp));
2054 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2055 			     NFS_MAXNAMLEN));
2056 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2057 				cnp->cn_cred, &error));
2058 	if (info.v3) {
2059 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2060 					 NFS_LATTR_NOSHRINK));
2061 		ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2062 	}
2063 	m_freem(info.mrep);
2064 	info.mrep = NULL;
2065 nfsmout:
2066 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2067 	if (!attrflag)
2068 		VTONFS(vp)->n_attrstamp = 0;
2069 	if (!wccflag)
2070 		VTONFS(tdvp)->n_attrstamp = 0;
2071 	/*
2072 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2073 	 */
2074 	if (error == EEXIST)
2075 		error = 0;
2076 	lwkt_reltoken(&nmp->nm_token);
2077 	return (error);
2078 }
2079 
2080 /*
2081  * nfs symbolic link create call
2082  *
2083  * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2084  *		struct componentname *a_cnp, struct vattr *a_vap,
2085  *		char *a_target)
2086  */
2087 static int
2088 nfs_symlink(struct vop_old_symlink_args *ap)
2089 {
2090 	struct vnode *dvp = ap->a_dvp;
2091 	struct vattr *vap = ap->a_vap;
2092 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2093 	struct componentname *cnp = ap->a_cnp;
2094 	struct nfsv2_sattr *sp;
2095 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2096 	struct vnode *newvp = NULL;
2097 	struct nfsm_info info;
2098 
2099 	info.mrep = NULL;
2100 	info.v3 = NFS_ISV3(dvp);
2101 	lwkt_gettoken(&nmp->nm_token);
2102 
2103 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2104 	slen = strlen(ap->a_target);
2105 	nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2106 		     NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2107 		     nfsm_rndup(cnp->cn_namelen) +
2108 		     nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2109 	ERROROUT(nfsm_fhtom(&info, dvp));
2110 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2111 			     NFS_MAXNAMLEN));
2112 	if (info.v3) {
2113 		nfsm_v3attrbuild(&info, vap, FALSE);
2114 	}
2115 	ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2116 	if (info.v3 == 0) {
2117 		sp = nfsm_build(&info, NFSX_V2SATTR);
2118 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2119 		sp->sa_uid = nfs_xdrneg1;
2120 		sp->sa_gid = nfs_xdrneg1;
2121 		sp->sa_size = nfs_xdrneg1;
2122 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2123 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2124 	}
2125 
2126 	/*
2127 	 * Issue the NFS request and get the rpc response.
2128 	 *
2129 	 * Only NFSv3 responses returning an error of 0 actually return
2130 	 * a file handle that can be converted into newvp without having
2131 	 * to do an extra lookup rpc.
2132 	 */
2133 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2134 				cnp->cn_cred, &error));
2135 	if (info.v3) {
2136 		if (error == 0) {
2137 		       ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2138 		}
2139 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2140 	}
2141 
2142 	/*
2143 	 * out code jumps -> here, mrep is also freed.
2144 	 */
2145 
2146 	m_freem(info.mrep);
2147 	info.mrep = NULL;
2148 nfsmout:
2149 
2150 	/*
2151 	 * If we get an EEXIST error, silently convert it to no-error
2152 	 * in case of an NFS retry.
2153 	 */
2154 	if (error == EEXIST)
2155 		error = 0;
2156 
2157 	/*
2158 	 * If we do not have (or no longer have) an error, and we could
2159 	 * not extract the newvp from the response due to the request being
2160 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
2161 	 * to obtain a newvp to return.
2162 	 */
2163 	if (error == 0 && newvp == NULL) {
2164 		struct nfsnode *np = NULL;
2165 
2166 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2167 				     cnp->cn_cred, cnp->cn_td, &np);
2168 		if (!error)
2169 			newvp = NFSTOV(np);
2170 	}
2171 	if (error) {
2172 		if (newvp)
2173 			vput(newvp);
2174 	} else {
2175 		*ap->a_vpp = newvp;
2176 	}
2177 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2178 	if (!wccflag)
2179 		VTONFS(dvp)->n_attrstamp = 0;
2180 	lwkt_reltoken(&nmp->nm_token);
2181 
2182 	return (error);
2183 }
2184 
2185 /*
2186  * nfs make dir call
2187  *
2188  * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2189  *	     struct componentname *a_cnp, struct vattr *a_vap)
2190  */
2191 static int
2192 nfs_mkdir(struct vop_old_mkdir_args *ap)
2193 {
2194 	struct vnode *dvp = ap->a_dvp;
2195 	struct vattr *vap = ap->a_vap;
2196 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2197 	struct componentname *cnp = ap->a_cnp;
2198 	struct nfsv2_sattr *sp;
2199 	struct nfsnode *np = NULL;
2200 	struct vnode *newvp = NULL;
2201 	struct vattr vattr;
2202 	int error = 0, wccflag = NFSV3_WCCRATTR;
2203 	int gotvp = 0;
2204 	int len;
2205 	struct nfsm_info info;
2206 
2207 	info.mrep = NULL;
2208 	info.v3 = NFS_ISV3(dvp);
2209 	lwkt_gettoken(&nmp->nm_token);
2210 
2211 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2212 		lwkt_reltoken(&nmp->nm_token);
2213 		return (error);
2214 	}
2215 	len = cnp->cn_namelen;
2216 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2217 	nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2218 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2219 		     nfsm_rndup(len) + NFSX_SATTR(info.v3));
2220 	ERROROUT(nfsm_fhtom(&info, dvp));
2221 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2222 	if (info.v3) {
2223 		nfsm_v3attrbuild(&info, vap, FALSE);
2224 	} else {
2225 		sp = nfsm_build(&info, NFSX_V2SATTR);
2226 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2227 		sp->sa_uid = nfs_xdrneg1;
2228 		sp->sa_gid = nfs_xdrneg1;
2229 		sp->sa_size = nfs_xdrneg1;
2230 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2231 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2232 	}
2233 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2234 		    cnp->cn_cred, &error));
2235 	if (error == 0) {
2236 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2237 	}
2238 	if (info.v3) {
2239 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2240 	}
2241 	m_freem(info.mrep);
2242 	info.mrep = NULL;
2243 nfsmout:
2244 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2245 	if (!wccflag)
2246 		VTONFS(dvp)->n_attrstamp = 0;
2247 	/*
2248 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2249 	 * if we can succeed in looking up the directory.
2250 	 */
2251 	if (error == EEXIST || (!error && !gotvp)) {
2252 		if (newvp) {
2253 			vrele(newvp);
2254 			newvp = NULL;
2255 		}
2256 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2257 			cnp->cn_td, &np);
2258 		if (!error) {
2259 			newvp = NFSTOV(np);
2260 			if (newvp->v_type != VDIR)
2261 				error = EEXIST;
2262 		}
2263 	}
2264 	if (error) {
2265 		if (newvp)
2266 			vrele(newvp);
2267 	} else {
2268 		*ap->a_vpp = newvp;
2269 	}
2270 	lwkt_reltoken(&nmp->nm_token);
2271 	return (error);
2272 }
2273 
2274 /*
2275  * nfs remove directory call
2276  *
2277  * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2278  *	     struct componentname *a_cnp)
2279  */
2280 static int
2281 nfs_rmdir(struct vop_old_rmdir_args *ap)
2282 {
2283 	struct vnode *vp = ap->a_vp;
2284 	struct vnode *dvp = ap->a_dvp;
2285 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2286 	struct componentname *cnp = ap->a_cnp;
2287 	int error = 0, wccflag = NFSV3_WCCRATTR;
2288 	struct nfsm_info info;
2289 
2290 	info.mrep = NULL;
2291 	info.v3 = NFS_ISV3(dvp);
2292 
2293 	if (dvp == vp)
2294 		return (EINVAL);
2295 
2296 	lwkt_gettoken(&nmp->nm_token);
2297 
2298 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2299 	nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2300 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2301 		     nfsm_rndup(cnp->cn_namelen));
2302 	ERROROUT(nfsm_fhtom(&info, dvp));
2303 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2304 		 NFS_MAXNAMLEN));
2305 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2306 				cnp->cn_cred, &error));
2307 	if (info.v3) {
2308 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2309 	}
2310 	m_freem(info.mrep);
2311 	info.mrep = NULL;
2312 nfsmout:
2313 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2314 	if (!wccflag)
2315 		VTONFS(dvp)->n_attrstamp = 0;
2316 	/*
2317 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2318 	 */
2319 	if (error == ENOENT)
2320 		error = 0;
2321 	lwkt_reltoken(&nmp->nm_token);
2322 
2323 	return (error);
2324 }
2325 
2326 /*
2327  * nfs readdir call
2328  *
2329  * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2330  */
2331 static int
2332 nfs_readdir(struct vop_readdir_args *ap)
2333 {
2334 	struct vnode *vp = ap->a_vp;
2335 	struct nfsnode *np = VTONFS(vp);
2336 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2337 	struct uio *uio = ap->a_uio;
2338 	int tresid, error;
2339 	struct vattr vattr;
2340 
2341 	if (vp->v_type != VDIR)
2342 		return (EPERM);
2343 
2344 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2345 		return (error);
2346 
2347 	lwkt_gettoken(&nmp->nm_token);
2348 
2349 	/*
2350 	 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2351 	 * and then check that is still valid, or if this is an NQNFS mount
2352 	 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR().  Note that
2353 	 * VOP_GETATTR() does not necessarily go to the wire.
2354 	 */
2355 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2356 	    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2357 		if (VOP_GETATTR(vp, &vattr) == 0 &&
2358 		    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2359 		) {
2360 			nfsstats.direofcache_hits++;
2361 			goto done;
2362 		}
2363 	}
2364 
2365 	/*
2366 	 * Call nfs_bioread() to do the real work.  nfs_bioread() does its
2367 	 * own cache coherency checks so we do not have to.
2368 	 */
2369 	tresid = uio->uio_resid;
2370 	error = nfs_bioread(vp, uio, 0);
2371 
2372 	if (!error && uio->uio_resid == tresid)
2373 		nfsstats.direofcache_misses++;
2374 done:
2375 	lwkt_reltoken(&nmp->nm_token);
2376 	vn_unlock(vp);
2377 
2378 	return (error);
2379 }
2380 
2381 /*
2382  * Readdir rpc call.  nfs_bioread->nfs_doio->nfs_readdirrpc.
2383  *
2384  * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2385  * offset/block and converts the nfs formatted directory entries for userland
2386  * consumption as well as deals with offsets into the middle of blocks.
2387  * nfs_doio only deals with logical blocks.  In particular, uio_offset will
2388  * be block-bounded.  It must convert to cookies for the actual RPC.
2389  */
2390 int
2391 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2392 {
2393 	int len, left;
2394 	struct nfs_dirent *dp = NULL;
2395 	u_int32_t *tl;
2396 	nfsuint64 *cookiep;
2397 	caddr_t cp;
2398 	nfsuint64 cookie;
2399 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2400 	struct nfsnode *dnp = VTONFS(vp);
2401 	u_quad_t fileno;
2402 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2403 	int attrflag;
2404 	struct nfsm_info info;
2405 
2406 	info.mrep = NULL;
2407 	info.v3 = NFS_ISV3(vp);
2408 
2409 #ifndef DIAGNOSTIC
2410 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2411 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2412 		panic("nfs readdirrpc bad uio");
2413 #endif
2414 
2415 	/*
2416 	 * If there is no cookie, assume directory was stale.
2417 	 */
2418 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2419 	if (cookiep)
2420 		cookie = *cookiep;
2421 	else
2422 		return (NFSERR_BAD_COOKIE);
2423 	/*
2424 	 * Loop around doing readdir rpc's of size nm_readdirsize
2425 	 * truncated to a multiple of DIRBLKSIZ.
2426 	 * The stopping criteria is EOF or buffer full.
2427 	 */
2428 	while (more_dirs && bigenough) {
2429 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2430 		nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2431 			     NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2432 		ERROROUT(nfsm_fhtom(&info, vp));
2433 		if (info.v3) {
2434 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2435 			*tl++ = cookie.nfsuquad[0];
2436 			*tl++ = cookie.nfsuquad[1];
2437 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2438 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2439 		} else {
2440 			/*
2441 			 * WARNING!  HAMMER DIRECTORIES WILL NOT WORK WELL
2442 			 * WITH NFSv2!!!  There's nothing I can really do
2443 			 * about it other than to hope the server supports
2444 			 * rdirplus w/NFSv2.
2445 			 */
2446 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2447 			*tl++ = cookie.nfsuquad[0];
2448 		}
2449 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2450 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2451 					uiop->uio_td,
2452 					nfs_vpcred(vp, ND_READ), &error));
2453 		if (info.v3) {
2454 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2455 						  NFS_LATTR_NOSHRINK));
2456 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2457 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2458 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2459 		}
2460 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2461 		more_dirs = fxdr_unsigned(int, *tl);
2462 
2463 		/* loop thru the dir entries, converting them to std form */
2464 		while (more_dirs && bigenough) {
2465 			if (info.v3) {
2466 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2467 				fileno = fxdr_hyper(tl);
2468 				len = fxdr_unsigned(int, *(tl + 2));
2469 			} else {
2470 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2471 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2472 				len = fxdr_unsigned(int, *tl);
2473 			}
2474 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2475 				error = EBADRPC;
2476 				m_freem(info.mrep);
2477 				info.mrep = NULL;
2478 				goto nfsmout;
2479 			}
2480 
2481 			/*
2482 			 * len is the number of bytes in the path element
2483 			 * name, not including the \0 termination.
2484 			 *
2485 			 * tlen is the number of bytes w have to reserve for
2486 			 * the path element name.
2487 			 */
2488 			tlen = nfsm_rndup(len);
2489 			if (tlen == len)
2490 				tlen += 4;	/* To ensure null termination */
2491 
2492 			/*
2493 			 * If the entry would cross a DIRBLKSIZ boundary,
2494 			 * extend the previous nfs_dirent to cover the
2495 			 * remaining space.
2496 			 */
2497 			left = DIRBLKSIZ - blksiz;
2498 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2499 				dp->nfs_reclen += left;
2500 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2501 				uiop->uio_iov->iov_len -= left;
2502 				uiop->uio_offset += left;
2503 				uiop->uio_resid -= left;
2504 				blksiz = 0;
2505 			}
2506 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2507 				bigenough = 0;
2508 			if (bigenough) {
2509 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2510 				dp->nfs_ino = fileno;
2511 				dp->nfs_namlen = len;
2512 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2513 				dp->nfs_type = DT_UNKNOWN;
2514 				blksiz += dp->nfs_reclen;
2515 				if (blksiz == DIRBLKSIZ)
2516 					blksiz = 0;
2517 				uiop->uio_offset += sizeof(struct nfs_dirent);
2518 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2519 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2520 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2521 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2522 
2523 				/*
2524 				 * The uiop has advanced by nfs_dirent + len
2525 				 * but really needs to advance by
2526 				 * nfs_dirent + tlen
2527 				 */
2528 				cp = uiop->uio_iov->iov_base;
2529 				tlen -= len;
2530 				*cp = '\0';	/* null terminate */
2531 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2532 				uiop->uio_iov->iov_len -= tlen;
2533 				uiop->uio_offset += tlen;
2534 				uiop->uio_resid -= tlen;
2535 			} else {
2536 				/*
2537 				 * NFS strings must be rounded up (nfsm_myouio
2538 				 * handled that in the bigenough case).
2539 				 */
2540 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2541 			}
2542 			if (info.v3) {
2543 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2544 			} else {
2545 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2546 			}
2547 
2548 			/*
2549 			 * If we were able to accomodate the last entry,
2550 			 * get the cookie for the next one.  Otherwise
2551 			 * hold-over the cookie for the one we were not
2552 			 * able to accomodate.
2553 			 */
2554 			if (bigenough) {
2555 				cookie.nfsuquad[0] = *tl++;
2556 				if (info.v3)
2557 					cookie.nfsuquad[1] = *tl++;
2558 			} else if (info.v3) {
2559 				tl += 2;
2560 			} else {
2561 				tl++;
2562 			}
2563 			more_dirs = fxdr_unsigned(int, *tl);
2564 		}
2565 		/*
2566 		 * If at end of rpc data, get the eof boolean
2567 		 */
2568 		if (!more_dirs) {
2569 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2570 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2571 		}
2572 		m_freem(info.mrep);
2573 		info.mrep = NULL;
2574 	}
2575 	/*
2576 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2577 	 * by increasing d_reclen for the last record.
2578 	 */
2579 	if (blksiz > 0) {
2580 		left = DIRBLKSIZ - blksiz;
2581 		dp->nfs_reclen += left;
2582 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2583 		uiop->uio_iov->iov_len -= left;
2584 		uiop->uio_offset += left;
2585 		uiop->uio_resid -= left;
2586 	}
2587 
2588 	if (bigenough) {
2589 		/*
2590 		 * We hit the end of the directory, update direofoffset.
2591 		 */
2592 		dnp->n_direofoffset = uiop->uio_offset;
2593 	} else {
2594 		/*
2595 		 * There is more to go, insert the link cookie so the
2596 		 * next block can be read.
2597 		 */
2598 		if (uiop->uio_resid > 0)
2599 			kprintf("EEK! readdirrpc resid > 0\n");
2600 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2601 		*cookiep = cookie;
2602 	}
2603 nfsmout:
2604 	return (error);
2605 }
2606 
2607 /*
2608  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2609  */
2610 int
2611 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2612 {
2613 	int len, left;
2614 	struct nfs_dirent *dp;
2615 	u_int32_t *tl;
2616 	struct vnode *newvp;
2617 	nfsuint64 *cookiep;
2618 	caddr_t dpossav1, dpossav2;
2619 	caddr_t cp;
2620 	struct mbuf *mdsav1, *mdsav2;
2621 	nfsuint64 cookie;
2622 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2623 	struct nfsnode *dnp = VTONFS(vp), *np;
2624 	nfsfh_t *fhp;
2625 	u_quad_t fileno;
2626 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2627 	int attrflag, fhsize;
2628 	struct nchandle nch;
2629 	struct nchandle dnch;
2630 	struct nlcomponent nlc;
2631 	struct nfsm_info info;
2632 
2633 	info.mrep = NULL;
2634 	info.v3 = 1;
2635 
2636 #ifndef nolint
2637 	dp = NULL;
2638 #endif
2639 #ifndef DIAGNOSTIC
2640 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2641 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2642 		panic("nfs readdirplusrpc bad uio");
2643 #endif
2644 	/*
2645 	 * Obtain the namecache record for the directory so we have something
2646 	 * to use as a basis for creating the entries.  This function will
2647 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2648 	 * from the tree and cannot be used for upward traversals, and the
2649 	 * ncp may be unnamed.  Note that other unrelated operations may
2650 	 * cause the ncp to be named at any time.
2651 	 *
2652 	 * We have to lock the ncp to prevent a lock order reversal when
2653 	 * rdirplus does nlookups of the children, because the vnode is
2654 	 * locked and has to stay that way.
2655 	 */
2656 	cache_fromdvp(vp, NULL, 0, &dnch);
2657 	bzero(&nlc, sizeof(nlc));
2658 	newvp = NULLVP;
2659 
2660 	/*
2661 	 * If there is no cookie, assume directory was stale.
2662 	 */
2663 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2664 	if (cookiep) {
2665 		cookie = *cookiep;
2666 	} else {
2667 		if (dnch.ncp)
2668 			cache_drop(&dnch);
2669 		return (NFSERR_BAD_COOKIE);
2670 	}
2671 
2672 	/*
2673 	 * Loop around doing readdir rpc's of size nm_readdirsize
2674 	 * truncated to a multiple of DIRBLKSIZ.
2675 	 * The stopping criteria is EOF or buffer full.
2676 	 */
2677 	while (more_dirs && bigenough) {
2678 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2679 		nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2680 			     NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2681 		ERROROUT(nfsm_fhtom(&info, vp));
2682 		tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2683 		*tl++ = cookie.nfsuquad[0];
2684 		*tl++ = cookie.nfsuquad[1];
2685 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2686 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2687 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2688 		*tl = txdr_unsigned(nmp->nm_rsize);
2689 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2690 					uiop->uio_td,
2691 					nfs_vpcred(vp, ND_READ), &error));
2692 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2693 					  NFS_LATTR_NOSHRINK));
2694 		NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2695 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2696 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2697 		more_dirs = fxdr_unsigned(int, *tl);
2698 
2699 		/* loop thru the dir entries, doctoring them to 4bsd form */
2700 		while (more_dirs && bigenough) {
2701 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2702 			fileno = fxdr_hyper(tl);
2703 			len = fxdr_unsigned(int, *(tl + 2));
2704 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2705 				error = EBADRPC;
2706 				m_freem(info.mrep);
2707 				info.mrep = NULL;
2708 				goto nfsmout;
2709 			}
2710 			tlen = nfsm_rndup(len);
2711 			if (tlen == len)
2712 				tlen += 4;	/* To ensure null termination*/
2713 			left = DIRBLKSIZ - blksiz;
2714 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2715 				dp->nfs_reclen += left;
2716 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2717 				uiop->uio_iov->iov_len -= left;
2718 				uiop->uio_offset += left;
2719 				uiop->uio_resid -= left;
2720 				blksiz = 0;
2721 			}
2722 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2723 				bigenough = 0;
2724 			if (bigenough) {
2725 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2726 				dp->nfs_ino = fileno;
2727 				dp->nfs_namlen = len;
2728 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2729 				dp->nfs_type = DT_UNKNOWN;
2730 				blksiz += dp->nfs_reclen;
2731 				if (blksiz == DIRBLKSIZ)
2732 					blksiz = 0;
2733 				uiop->uio_offset += sizeof(struct nfs_dirent);
2734 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2735 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2736 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2737 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2738 				nlc.nlc_namelen = len;
2739 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2740 				cp = uiop->uio_iov->iov_base;
2741 				tlen -= len;
2742 				*cp = '\0';
2743 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2744 				uiop->uio_iov->iov_len -= tlen;
2745 				uiop->uio_offset += tlen;
2746 				uiop->uio_resid -= tlen;
2747 			} else {
2748 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2749 			}
2750 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2751 			if (bigenough) {
2752 				cookie.nfsuquad[0] = *tl++;
2753 				cookie.nfsuquad[1] = *tl++;
2754 			} else {
2755 				tl += 2;
2756 			}
2757 
2758 			/*
2759 			 * Since the attributes are before the file handle
2760 			 * (sigh), we must skip over the attributes and then
2761 			 * come back and get them.
2762 			 */
2763 			attrflag = fxdr_unsigned(int, *tl);
2764 			if (attrflag) {
2765 			    dpossav1 = info.dpos;
2766 			    mdsav1 = info.md;
2767 			    ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2768 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2769 			    doit = fxdr_unsigned(int, *tl);
2770 			    if (doit) {
2771 				NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2772 			    }
2773 			    if (doit && bigenough && !nlcdegenerate(&nlc) &&
2774 				!NFS_CMPFH(dnp, fhp, fhsize)
2775 			    ) {
2776 				if (dnch.ncp) {
2777 #if 0
2778 				    kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2779 					nlc.nlc_namelen, nlc.nlc_namelen,
2780 					nlc.nlc_nameptr);
2781 #endif
2782 				    /*
2783 				     * This is a bit hokey but there isn't
2784 				     * much we can do about it.  We can't
2785 				     * hold the directory vp locked while
2786 				     * doing lookups and gets.
2787 				     */
2788 				    nch = cache_nlookup_nonblock(&dnch, &nlc);
2789 				    if (nch.ncp == NULL)
2790 					goto rdfail;
2791 				    cache_setunresolved(&nch);
2792 				    error = nfs_nget_nonblock(vp->v_mount, fhp,
2793 							      fhsize, &np);
2794 				    if (error) {
2795 					cache_put(&nch);
2796 					goto rdfail;
2797 				    }
2798 				    newvp = NFSTOV(np);
2799 				    dpossav2 = info.dpos;
2800 				    info.dpos = dpossav1;
2801 				    mdsav2 = info.md;
2802 				    info.md = mdsav1;
2803 				    ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2804 				    info.dpos = dpossav2;
2805 				    info.md = mdsav2;
2806 				    dp->nfs_type =
2807 					    IFTODT(VTTOIF(np->n_vattr.va_type));
2808 				    nfs_cache_setvp(&nch, newvp,
2809 						    nfspos_cache_timeout);
2810 				    vput(newvp);
2811 				    newvp = NULLVP;
2812 				    cache_put(&nch);
2813 				} else {
2814 rdfail:
2815 				    ;
2816 #if 0
2817 				    kprintf("Warning: NFS/rddirplus, "
2818 					    "UNABLE TO ENTER %*.*s\n",
2819 					nlc.nlc_namelen, nlc.nlc_namelen,
2820 					nlc.nlc_nameptr);
2821 #endif
2822 				}
2823 			    }
2824 			} else {
2825 			    /* Just skip over the file handle */
2826 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2827 			    i = fxdr_unsigned(int, *tl);
2828 			    ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2829 			}
2830 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2831 			more_dirs = fxdr_unsigned(int, *tl);
2832 		}
2833 		/*
2834 		 * If at end of rpc data, get the eof boolean
2835 		 */
2836 		if (!more_dirs) {
2837 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2838 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2839 		}
2840 		m_freem(info.mrep);
2841 		info.mrep = NULL;
2842 	}
2843 	/*
2844 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2845 	 * by increasing d_reclen for the last record.
2846 	 */
2847 	if (blksiz > 0) {
2848 		left = DIRBLKSIZ - blksiz;
2849 		dp->nfs_reclen += left;
2850 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2851 		uiop->uio_iov->iov_len -= left;
2852 		uiop->uio_offset += left;
2853 		uiop->uio_resid -= left;
2854 	}
2855 
2856 	/*
2857 	 * We are now either at the end of the directory or have filled the
2858 	 * block.
2859 	 */
2860 	if (bigenough) {
2861 		dnp->n_direofoffset = uiop->uio_offset;
2862 	} else {
2863 		if (uiop->uio_resid > 0)
2864 			kprintf("EEK! readdirplusrpc resid > 0\n");
2865 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2866 		*cookiep = cookie;
2867 	}
2868 nfsmout:
2869 	if (newvp != NULLVP) {
2870 	        if (newvp == vp)
2871 			vrele(newvp);
2872 		else
2873 			vput(newvp);
2874 		newvp = NULLVP;
2875 	}
2876 	if (dnch.ncp)
2877 		cache_drop(&dnch);
2878 	return (error);
2879 }
2880 
2881 /*
2882  * Silly rename. To make the NFS filesystem that is stateless look a little
2883  * more like the "ufs" a remove of an active vnode is translated to a rename
2884  * to a funny looking filename that is removed by nfs_inactive on the
2885  * nfsnode. There is the potential for another process on a different client
2886  * to create the same funny name between the nfs_lookitup() fails and the
2887  * nfs_rename() completes, but...
2888  */
2889 static int
2890 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2891 {
2892 	struct sillyrename *sp;
2893 	struct nfsnode *np;
2894 	int error;
2895 
2896 	/*
2897 	 * We previously purged dvp instead of vp.  I don't know why, it
2898 	 * completely destroys performance.  We can't do it anyway with the
2899 	 * new VFS API since we would be breaking the namecache topology.
2900 	 */
2901 	cache_purge(vp);	/* XXX */
2902 	np = VTONFS(vp);
2903 #ifndef DIAGNOSTIC
2904 	if (vp->v_type == VDIR)
2905 		panic("nfs: sillyrename dir");
2906 #endif
2907 	sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
2908 	sp->s_cred = crdup(cnp->cn_cred);
2909 	sp->s_dvp = dvp;
2910 	vref(dvp);
2911 
2912 	/* Fudge together a funny name */
2913 	sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2914 				(int)(intptr_t)cnp->cn_td);
2915 
2916 	/* Try lookitups until we get one that isn't there */
2917 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2918 		cnp->cn_td, NULL) == 0) {
2919 		sp->s_name[4]++;
2920 		if (sp->s_name[4] > 'z') {
2921 			error = EINVAL;
2922 			goto bad;
2923 		}
2924 	}
2925 	error = nfs_renameit(dvp, cnp, sp);
2926 	if (error)
2927 		goto bad;
2928 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2929 		cnp->cn_td, &np);
2930 	np->n_sillyrename = sp;
2931 	return (0);
2932 bad:
2933 	vrele(sp->s_dvp);
2934 	crfree(sp->s_cred);
2935 	kfree((caddr_t)sp, M_NFSREQ);
2936 	return (error);
2937 }
2938 
2939 /*
2940  * Look up a file name and optionally either update the file handle or
2941  * allocate an nfsnode, depending on the value of npp.
2942  * npp == NULL	--> just do the lookup
2943  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2944  *			handled too
2945  * *npp != NULL --> update the file handle in the vnode
2946  */
2947 static int
2948 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2949 	     struct thread *td, struct nfsnode **npp)
2950 {
2951 	struct vnode *newvp = NULL;
2952 	struct nfsnode *np, *dnp = VTONFS(dvp);
2953 	int error = 0, fhlen, attrflag;
2954 	nfsfh_t *nfhp;
2955 	struct nfsm_info info;
2956 
2957 	info.mrep = NULL;
2958 	info.v3 = NFS_ISV3(dvp);
2959 
2960 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2961 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2962 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2963 	ERROROUT(nfsm_fhtom(&info, dvp));
2964 	ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2965 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2966 	if (npp && !error) {
2967 		NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2968 		if (*npp) {
2969 		    np = *npp;
2970 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2971 			kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2972 			np->n_fhp = &np->n_fh;
2973 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2974 			np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2975 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2976 		    np->n_fhsize = fhlen;
2977 		    newvp = NFSTOV(np);
2978 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2979 		    vref(dvp);
2980 		    newvp = dvp;
2981 		} else {
2982 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2983 		    if (error) {
2984 			m_freem(info.mrep);
2985 			info.mrep = NULL;
2986 			return (error);
2987 		    }
2988 		    newvp = NFSTOV(np);
2989 		}
2990 		if (info.v3) {
2991 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2992 						  NFS_LATTR_NOSHRINK));
2993 			if (!attrflag && *npp == NULL) {
2994 				m_freem(info.mrep);
2995 				info.mrep = NULL;
2996 				if (newvp == dvp)
2997 					vrele(newvp);
2998 				else
2999 					vput(newvp);
3000 				return (ENOENT);
3001 			}
3002 		} else {
3003 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3004 		}
3005 	}
3006 	m_freem(info.mrep);
3007 	info.mrep = NULL;
3008 nfsmout:
3009 	if (npp && *npp == NULL) {
3010 		if (error) {
3011 			if (newvp) {
3012 				if (newvp == dvp)
3013 					vrele(newvp);
3014 				else
3015 					vput(newvp);
3016 			}
3017 		} else
3018 			*npp = np;
3019 	}
3020 	return (error);
3021 }
3022 
3023 /*
3024  * Nfs Version 3 commit rpc
3025  *
3026  * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3027  * involved.
3028  */
3029 int
3030 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3031 {
3032 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3033 	int error = 0, wccflag = NFSV3_WCCRATTR;
3034 	struct nfsm_info info;
3035 	u_int32_t *tl;
3036 
3037 	info.mrep = NULL;
3038 	info.v3 = 1;
3039 
3040 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3041 		return (0);
3042 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3043 	nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3044 	ERROROUT(nfsm_fhtom(&info, vp));
3045 	tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3046 	txdr_hyper(offset, tl);
3047 	tl += 2;
3048 	*tl = txdr_unsigned(cnt);
3049 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3050 				nfs_vpcred(vp, ND_WRITE), &error));
3051 	ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3052 	if (!error) {
3053 		NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3054 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3055 			NFSX_V3WRITEVERF)) {
3056 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3057 				NFSX_V3WRITEVERF);
3058 			error = NFSERR_STALEWRITEVERF;
3059 		}
3060 	}
3061 	m_freem(info.mrep);
3062 	info.mrep = NULL;
3063 nfsmout:
3064 	return (error);
3065 }
3066 
3067 /*
3068  * Kludge City..
3069  * - make nfs_bmap() essentially a no-op that does no translation
3070  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3071  *   (Maybe I could use the process's page mapping, but I was concerned that
3072  *    Kernel Write might not be enabled and also figured copyout() would do
3073  *    a lot more work than bcopy() and also it currently happens in the
3074  *    context of the swapper process (2).
3075  *
3076  * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3077  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
3078  */
3079 static int
3080 nfs_bmap(struct vop_bmap_args *ap)
3081 {
3082 	/* no token lock required */
3083 	if (ap->a_doffsetp != NULL)
3084 		*ap->a_doffsetp = ap->a_loffset;
3085 	if (ap->a_runp != NULL)
3086 		*ap->a_runp = 0;
3087 	if (ap->a_runb != NULL)
3088 		*ap->a_runb = 0;
3089 	return (0);
3090 }
3091 
3092 /*
3093  * Strategy routine.
3094  */
3095 static int
3096 nfs_strategy(struct vop_strategy_args *ap)
3097 {
3098 	struct bio *bio = ap->a_bio;
3099 	struct bio *nbio;
3100 	struct buf *bp __debugvar = bio->bio_buf;
3101 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3102 	struct thread *td;
3103 	int error;
3104 
3105 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
3106 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
3107 	KASSERT(BUF_REFCNT(bp) > 0,
3108 		("nfs_strategy: buffer %p not locked", bp));
3109 
3110 	if (bio->bio_flags & BIO_SYNC)
3111 		td = curthread;	/* XXX */
3112 	else
3113 		td = NULL;
3114 
3115 	lwkt_gettoken(&nmp->nm_token);
3116 
3117         /*
3118 	 * We probably don't need to push an nbio any more since no
3119 	 * block conversion is required due to the use of 64 bit byte
3120 	 * offsets, but do it anyway.
3121 	 *
3122 	 * NOTE: When NFS callers itself via this strategy routines and
3123 	 *	 sets up a synchronous I/O, it expects the I/O to run
3124 	 *	 synchronously (its bio_done routine just assumes it),
3125 	 *	 so for now we have to honor the bit.
3126          */
3127 	nbio = push_bio(bio);
3128 	nbio->bio_offset = bio->bio_offset;
3129 	nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3130 
3131 	/*
3132 	 * If the op is asynchronous and an i/o daemon is waiting
3133 	 * queue the request, wake it up and wait for completion
3134 	 * otherwise just do it ourselves.
3135 	 */
3136 	if (bio->bio_flags & BIO_SYNC) {
3137 		error = nfs_doio(ap->a_vp, nbio, td);
3138 	} else {
3139 		nfs_asyncio(ap->a_vp, nbio);
3140 		error = 0;
3141 	}
3142 	lwkt_reltoken(&nmp->nm_token);
3143 
3144 	return (error);
3145 }
3146 
3147 /*
3148  * Mmap a file
3149  *
3150  * NB Currently unsupported.
3151  *
3152  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3153  */
3154 /* ARGSUSED */
3155 static int
3156 nfs_mmap(struct vop_mmap_args *ap)
3157 {
3158 	/* no token lock required */
3159 	return (EINVAL);
3160 }
3161 
3162 /*
3163  * fsync vnode op. Just call nfs_flush() with commit == 1.
3164  *
3165  * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3166  */
3167 /* ARGSUSED */
3168 static int
3169 nfs_fsync(struct vop_fsync_args *ap)
3170 {
3171 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3172 	int error;
3173 
3174 	lwkt_gettoken(&nmp->nm_token);
3175 	error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3176 	lwkt_reltoken(&nmp->nm_token);
3177 
3178 	return error;
3179 }
3180 
3181 /*
3182  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
3183  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
3184  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
3185  * set the buffer contains data that has already been written to the server
3186  * and which now needs a commit RPC.
3187  *
3188  * If commit is 0 we only take one pass and only flush buffers containing new
3189  * dirty data.
3190  *
3191  * If commit is 1 we take two passes, issuing a commit RPC in the second
3192  * pass.
3193  *
3194  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3195  * to completely flush all pending data.
3196  *
3197  * Note that the RB_SCAN code properly handles the case where the
3198  * callback might block and directly or indirectly (another thread) cause
3199  * the RB tree to change.
3200  */
3201 
3202 #ifndef NFS_COMMITBVECSIZ
3203 #define NFS_COMMITBVECSIZ	16
3204 #endif
3205 
3206 struct nfs_flush_info {
3207 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3208 	struct thread *td;
3209 	struct vnode *vp;
3210 	int waitfor;
3211 	int slpflag;
3212 	int slptimeo;
3213 	int loops;
3214 	struct buf *bvary[NFS_COMMITBVECSIZ];
3215 	int bvsize;
3216 	off_t beg_off;
3217 	off_t end_off;
3218 };
3219 
3220 static int nfs_flush_bp(struct buf *bp, void *data);
3221 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3222 
3223 int
3224 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3225 {
3226 	struct nfsnode *np = VTONFS(vp);
3227 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3228 	struct nfs_flush_info info;
3229 	int error;
3230 
3231 	bzero(&info, sizeof(info));
3232 	info.td = td;
3233 	info.vp = vp;
3234 	info.waitfor = waitfor;
3235 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3236 	info.loops = 0;
3237 	lwkt_gettoken(&vp->v_token);
3238 
3239 	do {
3240 		/*
3241 		 * Flush mode
3242 		 */
3243 		info.mode = NFI_FLUSHNEW;
3244 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3245 				nfs_flush_bp, &info);
3246 
3247 		/*
3248 		 * Take a second pass if committing and no error occured.
3249 		 * Clean up any left over collection (whether an error
3250 		 * occurs or not).
3251 		 */
3252 		if (commit && error == 0) {
3253 			info.mode = NFI_COMMIT;
3254 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3255 					nfs_flush_bp, &info);
3256 			if (info.bvsize)
3257 				error = nfs_flush_docommit(&info, error);
3258 		}
3259 
3260 		/*
3261 		 * Wait for pending I/O to complete before checking whether
3262 		 * any further dirty buffers exist.
3263 		 */
3264 		while (waitfor == MNT_WAIT &&
3265 		       bio_track_active(&vp->v_track_write)) {
3266 			error = bio_track_wait(&vp->v_track_write,
3267 					       info.slpflag, info.slptimeo);
3268 			if (error) {
3269 				/*
3270 				 * We have to be able to break out if this
3271 				 * is an 'intr' mount.
3272 				 */
3273 				if (nfs_sigintr(nmp, NULL, td)) {
3274 					error = -EINTR;
3275 					break;
3276 				}
3277 
3278 				/*
3279 				 * Since we do not process pending signals,
3280 				 * once we get a PCATCH our tsleep() will no
3281 				 * longer sleep, switch to a fixed timeout
3282 				 * instead.
3283 				 */
3284 				if (info.slpflag == PCATCH) {
3285 					info.slpflag = 0;
3286 					info.slptimeo = 2 * hz;
3287 				}
3288 				error = 0;
3289 			}
3290 		}
3291 		++info.loops;
3292 		/*
3293 		 * Loop if we are flushing synchronous as well as committing,
3294 		 * and dirty buffers are still present.  Otherwise we might livelock.
3295 		 */
3296 	} while (waitfor == MNT_WAIT && commit &&
3297 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3298 
3299 	/*
3300 	 * The callbacks have to return a negative error to terminate the
3301 	 * RB scan.
3302 	 */
3303 	if (error < 0)
3304 		error = -error;
3305 
3306 	/*
3307 	 * Deal with any error collection
3308 	 */
3309 	if (np->n_flag & NWRITEERR) {
3310 		error = np->n_error;
3311 		np->n_flag &= ~NWRITEERR;
3312 	}
3313 	lwkt_reltoken(&vp->v_token);
3314 	return (error);
3315 }
3316 
3317 static
3318 int
3319 nfs_flush_bp(struct buf *bp, void *data)
3320 {
3321 	struct nfs_flush_info *info = data;
3322 	int lkflags;
3323 	int error;
3324 	off_t toff;
3325 
3326 	error = 0;
3327 	switch(info->mode) {
3328 	case NFI_FLUSHNEW:
3329 		error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3330 		if (error && info->loops && info->waitfor == MNT_WAIT) {
3331 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3332 			if (error) {
3333 				lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3334 				if (info->slpflag & PCATCH)
3335 					lkflags |= LK_PCATCH;
3336 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3337 						     info->slptimeo);
3338 			}
3339 		}
3340 
3341 		/*
3342 		 * Ignore locking errors
3343 		 */
3344 		if (error) {
3345 			error = 0;
3346 			break;
3347 		}
3348 
3349 		/*
3350 		 * The buffer may have changed out from under us, even if
3351 		 * we did not block (MPSAFE).  Check again now that it is
3352 		 * locked.
3353 		 */
3354 		if (bp->b_vp == info->vp &&
3355 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3356 			bremfree(bp);
3357 			bawrite(bp);
3358 		} else {
3359 			BUF_UNLOCK(bp);
3360 		}
3361 		break;
3362 	case NFI_COMMIT:
3363 		/*
3364 		 * Only process buffers in need of a commit which we can
3365 		 * immediately lock.  This may prevent a buffer from being
3366 		 * committed, but the normal flush loop will block on the
3367 		 * same buffer so we shouldn't get into an endless loop.
3368 		 */
3369 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3370 		    (B_DELWRI | B_NEEDCOMMIT)) {
3371 			break;
3372 		}
3373 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3374 			break;
3375 
3376 		/*
3377 		 * We must recheck after successfully locking the buffer.
3378 		 */
3379 		if (bp->b_vp != info->vp ||
3380 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3381 		    (B_DELWRI | B_NEEDCOMMIT)) {
3382 			BUF_UNLOCK(bp);
3383 			break;
3384 		}
3385 
3386 		/*
3387 		 * NOTE: storing the bp in the bvary[] basically sets
3388 		 * it up for a commit operation.
3389 		 *
3390 		 * We must call vfs_busy_pages() now so the commit operation
3391 		 * is interlocked with user modifications to memory mapped
3392 		 * pages.  The b_dirtyoff/b_dirtyend range is not correct
3393 		 * until after the pages have been busied.
3394 		 *
3395 		 * Note: to avoid loopback deadlocks, we do not
3396 		 * assign b_runningbufspace.
3397 		 */
3398 		bremfree(bp);
3399 		bp->b_cmd = BUF_CMD_WRITE;
3400 		vfs_busy_pages(bp->b_vp, bp);
3401 		info->bvary[info->bvsize] = bp;
3402 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3403 		if (info->bvsize == 0 || toff < info->beg_off)
3404 			info->beg_off = toff;
3405 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3406 		if (info->bvsize == 0 || toff > info->end_off)
3407 			info->end_off = toff;
3408 		++info->bvsize;
3409 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3410 			error = nfs_flush_docommit(info, 0);
3411 			KKASSERT(info->bvsize == 0);
3412 		}
3413 	}
3414 	return (error);
3415 }
3416 
3417 static
3418 int
3419 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3420 {
3421 	struct vnode *vp;
3422 	struct buf *bp;
3423 	off_t bytes;
3424 	int retv;
3425 	int i;
3426 
3427 	vp = info->vp;
3428 
3429 	if (info->bvsize > 0) {
3430 		/*
3431 		 * Commit data on the server, as required.  Note that
3432 		 * nfs_commit will use the vnode's cred for the commit.
3433 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3434 		 */
3435 		bytes = info->end_off - info->beg_off;
3436 		if (bytes > 0x40000000)
3437 			bytes = 0x40000000;
3438 		if (error) {
3439 			retv = -error;
3440 		} else {
3441 			retv = nfs_commitrpc_uio(vp, info->beg_off,
3442 						 (int)bytes, info->td);
3443 			if (retv == NFSERR_STALEWRITEVERF)
3444 				nfs_clearcommit(vp->v_mount);
3445 		}
3446 
3447 		/*
3448 		 * Now, either mark the blocks I/O done or mark the
3449 		 * blocks dirty, depending on whether the commit
3450 		 * succeeded.
3451 		 */
3452 		for (i = 0; i < info->bvsize; ++i) {
3453 			bp = info->bvary[i];
3454 			if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3455 				/*
3456 				 * Either an error or the original
3457 				 * vfs_busy_pages() cleared B_NEEDCOMMIT
3458 				 * due to finding new dirty VM pages in
3459 				 * the buffer.
3460 				 *
3461 				 * Leave B_DELWRI intact.
3462 				 */
3463 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3464 				vfs_unbusy_pages(bp);
3465 				bp->b_cmd = BUF_CMD_DONE;
3466 				bqrelse(bp);
3467 			} else {
3468 				/*
3469 				 * Success, remove B_DELWRI ( bundirty() ).
3470 				 *
3471 				 * b_dirtyoff/b_dirtyend seem to be NFS
3472 				 * specific.  We should probably move that
3473 				 * into bundirty(). XXX
3474 				 *
3475 				 * We are faking an I/O write, we have to
3476 				 * start the transaction in order to
3477 				 * immediately biodone() it.
3478 				 */
3479 				bundirty(bp);
3480 				bp->b_flags &= ~B_ERROR;
3481 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3482 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3483 				biodone(&bp->b_bio1);
3484 			}
3485 		}
3486 		info->bvsize = 0;
3487 	}
3488 	return (error);
3489 }
3490 
3491 /*
3492  * NFS advisory byte-level locks.
3493  * Currently unsupported.
3494  *
3495  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3496  *		int a_flags)
3497  */
3498 static int
3499 nfs_advlock(struct vop_advlock_args *ap)
3500 {
3501 	struct nfsnode *np = VTONFS(ap->a_vp);
3502 
3503 	/* no token lock currently required */
3504 	/*
3505 	 * The following kludge is to allow diskless support to work
3506 	 * until a real NFS lockd is implemented. Basically, just pretend
3507 	 * that this is a local lock.
3508 	 */
3509 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3510 }
3511 
3512 /*
3513  * Print out the contents of an nfsnode.
3514  *
3515  * nfs_print(struct vnode *a_vp)
3516  */
3517 static int
3518 nfs_print(struct vop_print_args *ap)
3519 {
3520 	struct vnode *vp = ap->a_vp;
3521 	struct nfsnode *np = VTONFS(vp);
3522 
3523 	kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3524 		(long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3525 	if (vp->v_type == VFIFO)
3526 		fifo_printinfo(vp);
3527 	kprintf("\n");
3528 	return (0);
3529 }
3530 
3531 /*
3532  * nfs special file access vnode op.
3533  *
3534  * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3535  */
3536 static int
3537 nfs_laccess(struct vop_access_args *ap)
3538 {
3539 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3540 	struct vattr vattr;
3541 	int error;
3542 
3543 	lwkt_gettoken(&nmp->nm_token);
3544 	error = VOP_GETATTR(ap->a_vp, &vattr);
3545 	if (error == 0) {
3546 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3547 					  vattr.va_mode, 0);
3548 	}
3549 	lwkt_reltoken(&nmp->nm_token);
3550 
3551 	return (error);
3552 }
3553 
3554 /*
3555  * Read wrapper for fifos.
3556  *
3557  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3558  *		struct ucred *a_cred)
3559  */
3560 static int
3561 nfsfifo_read(struct vop_read_args *ap)
3562 {
3563 	struct nfsnode *np = VTONFS(ap->a_vp);
3564 
3565 	/* no token access required */
3566 	/*
3567 	 * Set access flag.
3568 	 */
3569 	np->n_flag |= NACC;
3570 	getnanotime(&np->n_atim);
3571 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3572 }
3573 
3574 /*
3575  * Write wrapper for fifos.
3576  *
3577  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3578  *		 struct ucred *a_cred)
3579  */
3580 static int
3581 nfsfifo_write(struct vop_write_args *ap)
3582 {
3583 	struct nfsnode *np = VTONFS(ap->a_vp);
3584 
3585 	/* no token access required */
3586 	/*
3587 	 * Set update flag.
3588 	 */
3589 	np->n_flag |= NUPD;
3590 	getnanotime(&np->n_mtim);
3591 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3592 }
3593 
3594 /*
3595  * Close wrapper for fifos.
3596  *
3597  * Update the times on the nfsnode then do fifo close.
3598  *
3599  * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3600  */
3601 static int
3602 nfsfifo_close(struct vop_close_args *ap)
3603 {
3604 	struct vnode *vp = ap->a_vp;
3605 	struct nfsnode *np = VTONFS(vp);
3606 	struct vattr vattr;
3607 	struct timespec ts;
3608 
3609 	/* no token access required */
3610 
3611 	if (np->n_flag & (NACC | NUPD)) {
3612 		getnanotime(&ts);
3613 		if (np->n_flag & NACC)
3614 			np->n_atim = ts;
3615 		if (np->n_flag & NUPD)
3616 			np->n_mtim = ts;
3617 		np->n_flag |= NCHG;
3618 		if (vp->v_sysref.refcnt == 1 &&
3619 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3620 			VATTR_NULL(&vattr);
3621 			if (np->n_flag & NACC)
3622 				vattr.va_atime = np->n_atim;
3623 			if (np->n_flag & NUPD)
3624 				vattr.va_mtime = np->n_mtim;
3625 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3626 		}
3627 	}
3628 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3629 }
3630 
3631