xref: /dflybsd-src/sys/vfs/nfs/nfs_vnops.c (revision 8d1ea4cc63e8cedbe8f670822d4606adf2cf7d45)
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. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
34  */
35 
36 
37 /*
38  * vnode op calls for Sun NFS version 2 and 3
39  */
40 
41 #include "opt_inet.h"
42 
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/resourcevar.h>
47 #include <sys/proc.h>
48 #include <sys/mount.h>
49 #include <sys/buf.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/namei.h>
53 #include <sys/nlookup.h>
54 #include <sys/socket.h>
55 #include <sys/vnode.h>
56 #include <sys/dirent.h>
57 #include <sys/fcntl.h>
58 #include <sys/lockf.h>
59 #include <sys/stat.h>
60 #include <sys/sysctl.h>
61 #include <sys/conf.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_extern.h>
65 
66 #include <sys/buf2.h>
67 
68 #include <vfs/fifofs/fifo.h>
69 #include <vfs/ufs/dir.h>
70 
71 #undef DIRBLKSIZ
72 
73 #include "rpcv2.h"
74 #include "nfsproto.h"
75 #include "nfs.h"
76 #include "nfsmount.h"
77 #include "nfsnode.h"
78 #include "xdr_subs.h"
79 #include "nfsm_subs.h"
80 
81 #include <net/if.h>
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 
85 /* Defs */
86 #define	TRUE	1
87 #define	FALSE	0
88 
89 static int	nfsfifo_read (struct vop_read_args *);
90 static int	nfsfifo_write (struct vop_write_args *);
91 static int	nfsfifo_close (struct vop_close_args *);
92 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
93 static	int	nfs_lookup (struct vop_old_lookup_args *);
94 static	int	nfs_create (struct vop_old_create_args *);
95 static	int	nfs_mknod (struct vop_old_mknod_args *);
96 static	int	nfs_open (struct vop_open_args *);
97 static	int	nfs_close (struct vop_close_args *);
98 static	int	nfs_access (struct vop_access_args *);
99 static	int	nfs_getattr (struct vop_getattr_args *);
100 static	int	nfs_setattr (struct vop_setattr_args *);
101 static	int	nfs_read (struct vop_read_args *);
102 static	int	nfs_mmap (struct vop_mmap_args *);
103 static	int	nfs_fsync (struct vop_fsync_args *);
104 static	int	nfs_remove (struct vop_old_remove_args *);
105 static	int	nfs_link (struct vop_old_link_args *);
106 static	int	nfs_rename (struct vop_old_rename_args *);
107 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
108 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
109 static	int	nfs_symlink (struct vop_old_symlink_args *);
110 static	int	nfs_readdir (struct vop_readdir_args *);
111 static	int	nfs_bmap (struct vop_bmap_args *);
112 static	int	nfs_strategy (struct vop_strategy_args *);
113 static	int	nfs_lookitup (struct vnode *, const char *, int,
114 			struct ucred *, struct thread *, struct nfsnode **);
115 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
116 static int	nfs_laccess (struct vop_access_args *);
117 static int	nfs_readlink (struct vop_readlink_args *);
118 static int	nfs_print (struct vop_print_args *);
119 static int	nfs_advlock (struct vop_advlock_args *);
120 static int	nfs_kqfilter (struct vop_kqfilter_args *ap);
121 
122 static	int	nfs_nresolve (struct vop_nresolve_args *);
123 /*
124  * Global vfs data structures for nfs
125  */
126 struct vop_ops nfsv2_vnode_vops = {
127 	.vop_default =		vop_defaultop,
128 	.vop_access =		nfs_access,
129 	.vop_advlock =		nfs_advlock,
130 	.vop_bmap =		nfs_bmap,
131 	.vop_close =		nfs_close,
132 	.vop_old_create =	nfs_create,
133 	.vop_fsync =		nfs_fsync,
134 	.vop_getattr =		nfs_getattr,
135 	.vop_getpages =		vop_stdgetpages,
136 	.vop_putpages =		vop_stdputpages,
137 	.vop_inactive =		nfs_inactive,
138 	.vop_old_link =		nfs_link,
139 	.vop_old_lookup =	nfs_lookup,
140 	.vop_old_mkdir =	nfs_mkdir,
141 	.vop_old_mknod =	nfs_mknod,
142 	.vop_mmap =		nfs_mmap,
143 	.vop_open =		nfs_open,
144 	.vop_print =		nfs_print,
145 	.vop_read =		nfs_read,
146 	.vop_readdir =		nfs_readdir,
147 	.vop_readlink =		nfs_readlink,
148 	.vop_reclaim =		nfs_reclaim,
149 	.vop_old_remove =	nfs_remove,
150 	.vop_old_rename =	nfs_rename,
151 	.vop_old_rmdir =	nfs_rmdir,
152 	.vop_setattr =		nfs_setattr,
153 	.vop_strategy =		nfs_strategy,
154 	.vop_old_symlink =	nfs_symlink,
155 	.vop_write =		nfs_write,
156 	.vop_nresolve =		nfs_nresolve,
157 	.vop_kqfilter =		nfs_kqfilter
158 };
159 
160 /*
161  * Special device vnode ops
162  */
163 struct vop_ops nfsv2_spec_vops = {
164 	.vop_default =		vop_defaultop,
165 	.vop_access =		nfs_laccess,
166 	.vop_close =		nfs_close,
167 	.vop_fsync =		nfs_fsync,
168 	.vop_getattr =		nfs_getattr,
169 	.vop_inactive =		nfs_inactive,
170 	.vop_print =		nfs_print,
171 	.vop_read =		vop_stdnoread,
172 	.vop_reclaim =		nfs_reclaim,
173 	.vop_setattr =		nfs_setattr,
174 	.vop_write =		vop_stdnowrite
175 };
176 
177 struct vop_ops nfsv2_fifo_vops = {
178 	.vop_default =		fifo_vnoperate,
179 	.vop_access =		nfs_laccess,
180 	.vop_close =		nfsfifo_close,
181 	.vop_fsync =		nfs_fsync,
182 	.vop_getattr =		nfs_getattr,
183 	.vop_inactive =		nfs_inactive,
184 	.vop_print =		nfs_print,
185 	.vop_read =		nfsfifo_read,
186 	.vop_reclaim =		nfs_reclaim,
187 	.vop_setattr =		nfs_setattr,
188 	.vop_write =		nfsfifo_write
189 };
190 
191 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
192 				  struct componentname *cnp,
193 				  struct vattr *vap);
194 static int	nfs_removerpc (struct vnode *dvp, const char *name,
195 				   int namelen,
196 				   struct ucred *cred, struct thread *td);
197 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
198 				   int fnamelen, struct vnode *tdvp,
199 				   const char *tnameptr, int tnamelen,
200 				   struct ucred *cred, struct thread *td);
201 static int	nfs_renameit (struct vnode *sdvp,
202 				  struct componentname *scnp,
203 				  struct sillyrename *sp);
204 
205 SYSCTL_DECL(_vfs_nfs);
206 
207 static int nfs_flush_on_rename = 1;
208 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
209 	   &nfs_flush_on_rename, 0, "flush fvp prior to rename");
210 static int nfs_flush_on_hlink = 0;
211 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
212 	   &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
213 
214 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
216 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
217 
218 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
220 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
221 
222 static int	nfspos_cache_timeout = NFS_MINATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
224 	   &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
225 
226 static int	nfsv3_commit_on_close = 0;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
228 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
229 #if 0
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
231 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
232 
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
234 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
235 #endif
236 
237 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
238 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
239 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
240 
241 static __inline
242 void
243 nfs_knote(struct vnode *vp, int flags)
244 {
245 	if (flags)
246 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
247 }
248 
249 /*
250  * Returns whether a name component is a degenerate '.' or '..'.
251  */
252 static __inline
253 int
254 nlcdegenerate(struct nlcomponent *nlc)
255 {
256 	if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
257 		return(1);
258 	if (nlc->nlc_namelen == 2 &&
259 	    nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
260 		return(1);
261 	return(0);
262 }
263 
264 static int
265 nfs3_access_otw(struct vnode *vp, int wmode,
266 		struct thread *td, struct ucred *cred)
267 {
268 	struct nfsnode *np = VTONFS(vp);
269 	int attrflag;
270 	int error = 0;
271 	u_int32_t *tl;
272 	u_int32_t rmode;
273 	struct nfsm_info info;
274 
275 	info.mrep = NULL;
276 	info.v3 = 1;
277 
278 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
279 	nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
280 		     NFSX_FH(info.v3) + NFSX_UNSIGNED);
281 	ERROROUT(nfsm_fhtom(&info, vp));
282 	tl = nfsm_build(&info, NFSX_UNSIGNED);
283 	*tl = txdr_unsigned(wmode);
284 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
285 	ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
286 	if (error == 0) {
287 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
288 		rmode = fxdr_unsigned(u_int32_t, *tl);
289 		np->n_mode = rmode;
290 		np->n_modeuid = cred->cr_uid;
291 		np->n_modestamp = mycpu->gd_time_seconds;
292 	}
293 	m_freem(info.mrep);
294 	info.mrep = NULL;
295 nfsmout:
296 	return error;
297 }
298 
299 /*
300  * nfs access vnode op.
301  * For nfs version 2, just return ok. File accesses may fail later.
302  * For nfs version 3, use the access rpc to check accessibility. If file modes
303  * are changed on the server, accesses might still fail later.
304  *
305  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
306  */
307 static int
308 nfs_access(struct vop_access_args *ap)
309 {
310 	struct ucred *cred;
311 	struct vnode *vp = ap->a_vp;
312 	thread_t td = curthread;
313 	int error = 0;
314 	u_int32_t mode, wmode;
315 	struct nfsnode *np = VTONFS(vp);
316 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
317 	int v3 = NFS_ISV3(vp);
318 
319 	lwkt_gettoken(&nmp->nm_token);
320 
321 	/*
322 	 * Disallow write attempts on filesystems mounted read-only;
323 	 * unless the file is a socket, fifo, or a block or character
324 	 * device resident on the filesystem.
325 	 */
326 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
327 		switch (vp->v_type) {
328 		case VREG:
329 		case VDIR:
330 		case VLNK:
331 			lwkt_reltoken(&nmp->nm_token);
332 			return (EROFS);
333 		default:
334 			break;
335 		}
336 	}
337 
338 	/*
339 	 * The NFS protocol passes only the effective uid/gid over the wire but
340 	 * we need to check access against real ids if AT_EACCESS not set.
341 	 * Handle this case by cloning the credentials and setting the
342 	 * effective ids to the real ones.
343 	 *
344 	 * The crdup() here can cause a lot of ucred structures to build-up
345 	 * (up to maxvnodes), so do our best to avoid it.
346 	 */
347 	if (ap->a_flags & AT_EACCESS) {
348 		cred = crhold(ap->a_cred);
349 	} else {
350 		cred = ap->a_cred;
351 		if (cred->cr_uid == cred->cr_ruid &&
352 		    cred->cr_gid == cred->cr_rgid) {
353 			cred = crhold(ap->a_cred);
354 		} else {
355 			cred = crdup(ap->a_cred);
356 			cred->cr_uid = cred->cr_ruid;
357 			cred->cr_gid = cred->cr_rgid;
358 		}
359 	}
360 
361 	/*
362 	 * For nfs v3, check to see if we have done this recently, and if
363 	 * so return our cached result instead of making an ACCESS call.
364 	 * If not, do an access rpc, otherwise you are stuck emulating
365 	 * ufs_access() locally using the vattr. This may not be correct,
366 	 * since the server may apply other access criteria such as
367 	 * client uid-->server uid mapping that we do not know about.
368 	 */
369 	if (v3) {
370 		if (ap->a_mode & VREAD)
371 			mode = NFSV3ACCESS_READ;
372 		else
373 			mode = 0;
374 		if (vp->v_type != VDIR) {
375 			if (ap->a_mode & VWRITE)
376 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
377 			if (ap->a_mode & VEXEC)
378 				mode |= NFSV3ACCESS_EXECUTE;
379 		} else {
380 			if (ap->a_mode & VWRITE)
381 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
382 					 NFSV3ACCESS_DELETE);
383 			if (ap->a_mode & VEXEC)
384 				mode |= NFSV3ACCESS_LOOKUP;
385 		}
386 		/* XXX safety belt, only make blanket request if caching */
387 		if (nfsaccess_cache_timeout > 0) {
388 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
389 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
390 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
391 		} else {
392 			wmode = mode;
393 		}
394 
395 		/*
396 		 * Does our cached result allow us to give a definite yes to
397 		 * this request?
398 		 */
399 		if (np->n_modestamp &&
400 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
401 		   (cred->cr_uid == np->n_modeuid) &&
402 		   ((np->n_mode & mode) == mode)) {
403 			nfsstats.accesscache_hits++;
404 		} else {
405 			/*
406 			 * Either a no, or a don't know.  Go to the wire.
407 			 */
408 			nfsstats.accesscache_misses++;
409 		        error = nfs3_access_otw(vp, wmode, td, cred);
410 			if (!error) {
411 				if ((np->n_mode & mode) != mode) {
412 					error = EACCES;
413 				}
414 			}
415 		}
416 	} else {
417 		if ((error = nfs_laccess(ap)) != 0) {
418 			crfree(cred);
419 			lwkt_reltoken(&nmp->nm_token);
420 			return (error);
421 		}
422 
423 		/*
424 		 * Attempt to prevent a mapped root from accessing a file
425 		 * which it shouldn't.  We try to read a byte from the file
426 		 * if the user is root and the file is not zero length.
427 		 * After calling nfs_laccess, we should have the correct
428 		 * file size cached.
429 		 */
430 		if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
431 		    && VTONFS(vp)->n_size > 0) {
432 			struct iovec aiov;
433 			struct uio auio;
434 			char buf[1];
435 
436 			aiov.iov_base = buf;
437 			aiov.iov_len = 1;
438 			auio.uio_iov = &aiov;
439 			auio.uio_iovcnt = 1;
440 			auio.uio_offset = 0;
441 			auio.uio_resid = 1;
442 			auio.uio_segflg = UIO_SYSSPACE;
443 			auio.uio_rw = UIO_READ;
444 			auio.uio_td = td;
445 
446 			if (vp->v_type == VREG) {
447 				error = nfs_readrpc_uio(vp, &auio);
448 			} else if (vp->v_type == VDIR) {
449 				char* bp;
450 				bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
451 				aiov.iov_base = bp;
452 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
453 				error = nfs_readdirrpc_uio(vp, &auio);
454 				kfree(bp, M_TEMP);
455 			} else if (vp->v_type == VLNK) {
456 				error = nfs_readlinkrpc_uio(vp, &auio);
457 			} else {
458 				error = EACCES;
459 			}
460 		}
461 	}
462 	/*
463 	 * [re]record creds for reading and/or writing if access
464 	 * was granted.  Assume the NFS server will grant read access
465 	 * for execute requests.
466 	 */
467 	if (error == 0) {
468 		if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
469 			crhold(cred);
470 			if (np->n_rucred)
471 				crfree(np->n_rucred);
472 			np->n_rucred = cred;
473 		}
474 		if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
475 			crhold(cred);
476 			if (np->n_wucred)
477 				crfree(np->n_wucred);
478 			np->n_wucred = cred;
479 		}
480 	}
481 	lwkt_reltoken(&nmp->nm_token);
482 	crfree(cred);
483 
484 	return(error);
485 }
486 
487 /*
488  * nfs open vnode op
489  * Check to see if the type is ok
490  * and that deletion is not in progress.
491  * For paged in text files, you will need to flush the page cache
492  * if consistency is lost.
493  *
494  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
495  *	    struct file *a_fp)
496  */
497 /* ARGSUSED */
498 static int
499 nfs_open(struct vop_open_args *ap)
500 {
501 	struct vnode *vp = ap->a_vp;
502 	struct nfsnode *np = VTONFS(vp);
503 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
504 	struct vattr vattr;
505 	int error;
506 
507 	lwkt_gettoken(&nmp->nm_token);
508 
509 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
510 #ifdef DIAGNOSTIC
511 		kprintf("open eacces vtyp=%d\n",vp->v_type);
512 #endif
513 		lwkt_reltoken(&nmp->nm_token);
514 		return (EOPNOTSUPP);
515 	}
516 
517 	/*
518 	 * Save valid creds for reading and writing for later RPCs.
519 	 */
520 	if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
521 		crhold(ap->a_cred);
522 		if (np->n_rucred)
523 			crfree(np->n_rucred);
524 		np->n_rucred = ap->a_cred;
525 	}
526 	if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
527 		crhold(ap->a_cred);
528 		if (np->n_wucred)
529 			crfree(np->n_wucred);
530 		np->n_wucred = ap->a_cred;
531 	}
532 
533 	/*
534 	 * Clear the attribute cache only if opening with write access.  It
535 	 * is unclear if we should do this at all here, but we certainly
536 	 * should not clear the cache unconditionally simply because a file
537 	 * is being opened.
538 	 */
539 	if (ap->a_mode & FWRITE)
540 		np->n_attrstamp = 0;
541 
542 	/*
543 	 * For normal NFS, reconcile changes made locally verses
544 	 * changes made remotely.  Note that VOP_GETATTR only goes
545 	 * to the wire if the cached attribute has timed out or been
546 	 * cleared.
547 	 *
548 	 * If local modifications have been made clear the attribute
549 	 * cache to force an attribute and modified time check.  If
550 	 * GETATTR detects that the file has been changed by someone
551 	 * other then us it will set NRMODIFIED.
552 	 *
553 	 * If we are opening a directory and local changes have been
554 	 * made we have to invalidate the cache in order to ensure
555 	 * that we get the most up-to-date information from the
556 	 * server.  XXX
557 	 */
558 	if (np->n_flag & NLMODIFIED) {
559 		np->n_attrstamp = 0;
560 		if (vp->v_type == VDIR) {
561 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
562 			if (error == EINTR) {
563 				lwkt_reltoken(&nmp->nm_token);
564 				return (error);
565 			}
566 			nfs_invaldir(vp);
567 		}
568 	}
569 	error = VOP_GETATTR(vp, &vattr);
570 	if (error) {
571 		lwkt_reltoken(&nmp->nm_token);
572 		return (error);
573 	}
574 	if (np->n_flag & NRMODIFIED) {
575 		if (vp->v_type == VDIR)
576 			nfs_invaldir(vp);
577 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
578 		if (error == EINTR) {
579 			lwkt_reltoken(&nmp->nm_token);
580 			return (error);
581 		}
582 		np->n_flag &= ~NRMODIFIED;
583 	}
584 	error = vop_stdopen(ap);
585 	lwkt_reltoken(&nmp->nm_token);
586 
587 	return error;
588 }
589 
590 /*
591  * nfs close vnode op
592  * What an NFS client should do upon close after writing is a debatable issue.
593  * Most NFS clients push delayed writes to the server upon close, basically for
594  * two reasons:
595  * 1 - So that any write errors may be reported back to the client process
596  *     doing the close system call. By far the two most likely errors are
597  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
598  * 2 - To put a worst case upper bound on cache inconsistency between
599  *     multiple clients for the file.
600  * There is also a consistency problem for Version 2 of the protocol w.r.t.
601  * not being able to tell if other clients are writing a file concurrently,
602  * since there is no way of knowing if the changed modify time in the reply
603  * is only due to the write for this client.
604  * (NFS Version 3 provides weak cache consistency data in the reply that
605  *  should be sufficient to detect and handle this case.)
606  *
607  * The current code does the following:
608  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
609  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
610  *                     or commit them (this satisfies 1 and 2 except for the
611  *                     case where the server crashes after this close but
612  *                     before the commit RPC, which is felt to be "good
613  *                     enough". Changing the last argument to nfs_flush() to
614  *                     a 1 would force a commit operation, if it is felt a
615  *                     commit is necessary now.
616  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
617  *                     1 should be dealt with via an fsync() system call for
618  *                     cases where write errors are important.
619  *
620  * nfs_close(struct vnode *a_vp, int a_fflag)
621  */
622 /* ARGSUSED */
623 static int
624 nfs_close(struct vop_close_args *ap)
625 {
626 	struct vnode *vp = ap->a_vp;
627 	struct nfsnode *np = VTONFS(vp);
628 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
629 	int error = 0;
630 	thread_t td = curthread;
631 
632 	vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */
633 	lwkt_gettoken(&nmp->nm_token);
634 
635 	if (vp->v_type == VREG) {
636 	    if (np->n_flag & NLMODIFIED) {
637 		if (NFS_ISV3(vp)) {
638 		    /*
639 		     * Under NFSv3 we have dirty buffers to dispose of.  We
640 		     * must flush them to the NFS server.  We have the option
641 		     * of waiting all the way through the commit rpc or just
642 		     * waiting for the initial write.  The default is to only
643 		     * wait through the initial write so the data is in the
644 		     * server's cache, which is roughly similar to the state
645 		     * a standard disk subsystem leaves the file in on close().
646 		     *
647 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
648 		     * potential races with other processes, and certainly
649 		     * cannot clear it if we don't commit.
650 		     */
651 		    int cm = nfsv3_commit_on_close ? 1 : 0;
652 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
653 		    /* np->n_flag &= ~NLMODIFIED; */
654 		} else {
655 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
656 		}
657 		np->n_attrstamp = 0;
658 	    }
659 	    if (np->n_flag & NWRITEERR) {
660 		np->n_flag &= ~NWRITEERR;
661 		error = np->n_error;
662 	    }
663 	}
664 	vop_stdclose(ap);
665 	lwkt_reltoken(&nmp->nm_token);
666 
667 	return (error);
668 }
669 
670 /*
671  * nfs getattr call from vfs.
672  *
673  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
674  */
675 static int
676 nfs_getattr(struct vop_getattr_args *ap)
677 {
678 	struct vnode *vp = ap->a_vp;
679 	struct nfsnode *np = VTONFS(vp);
680 	struct nfsmount *nmp;
681 	int error = 0;
682 	thread_t td = curthread;
683 	struct nfsm_info info;
684 
685 	info.mrep = NULL;
686 	info.v3 = NFS_ISV3(vp);
687 	nmp = VFSTONFS(vp->v_mount);
688 
689 	lwkt_gettoken(&nmp->nm_token);
690 
691 	/*
692 	 * Update local times for special files.
693 	 */
694 	if (np->n_flag & (NACC | NUPD))
695 		np->n_flag |= NCHG;
696 	/*
697 	 * First look in the cache.
698 	 */
699 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
700 		goto done;
701 
702 	if (info.v3 && nfsaccess_cache_timeout > 0) {
703 		nfsstats.accesscache_misses++;
704 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
705 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
706 			goto done;
707 	}
708 
709 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
710 	nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
711 	ERROROUT(nfsm_fhtom(&info, vp));
712 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
713 				nfs_vpcred(vp, ND_CHECK), &error));
714 	if (error == 0) {
715 		ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
716 	}
717 	m_freem(info.mrep);
718 	info.mrep = NULL;
719 done:
720 	/*
721 	 * NFS doesn't support chflags flags.  If the nfs mount was
722 	 * made -o cache set the UF_CACHE bit for swapcache.
723 	 */
724 	if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
725 		ap->a_vap->va_flags |= UF_CACHE;
726 nfsmout:
727 	lwkt_reltoken(&nmp->nm_token);
728 	return (error);
729 }
730 
731 /*
732  * nfs setattr call.
733  *
734  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
735  */
736 static int
737 nfs_setattr(struct vop_setattr_args *ap)
738 {
739 	struct vnode *vp = ap->a_vp;
740 	struct nfsnode *np = VTONFS(vp);
741 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
742 	struct vattr *vap = ap->a_vap;
743 	int error = 0;
744 	int kflags = 0;
745 	off_t tsize;
746 	thread_t td = curthread;
747 
748 #ifndef nolint
749 	tsize = (off_t)0;
750 #endif
751 	/*
752 	 * Setting of flags is not supported.
753 	 */
754 	if (vap->va_flags != VNOVAL)
755 		return (EOPNOTSUPP);
756 
757 	/*
758 	 * Disallow write attempts if the filesystem is mounted read-only.
759 	 */
760   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
761 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
762 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
763 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
764 		return (EROFS);
765 
766 	lwkt_gettoken(&nmp->nm_token);
767 
768 	if (vap->va_size != VNOVAL) {
769 		/*
770 		 * truncation requested
771 		 */
772  		switch (vp->v_type) {
773  		case VDIR:
774 			lwkt_reltoken(&nmp->nm_token);
775  			return (EISDIR);
776  		case VCHR:
777  		case VBLK:
778  		case VSOCK:
779  		case VFIFO:
780 			if (vap->va_mtime.tv_sec == VNOVAL &&
781 			    vap->va_atime.tv_sec == VNOVAL &&
782 			    vap->va_mode == (mode_t)VNOVAL &&
783 			    vap->va_uid == (uid_t)VNOVAL &&
784 			    vap->va_gid == (gid_t)VNOVAL) {
785 				lwkt_reltoken(&nmp->nm_token);
786 				return (0);
787 			}
788  			vap->va_size = VNOVAL;
789  			break;
790  		default:
791 			/*
792 			 * Disallow write attempts if the filesystem is
793 			 * mounted read-only.
794 			 */
795 			if (vp->v_mount->mnt_flag & MNT_RDONLY) {
796 				lwkt_reltoken(&nmp->nm_token);
797 				return (EROFS);
798 			}
799 
800 			tsize = np->n_size;
801 again:
802 			error = nfs_meta_setsize(vp, td, vap->va_size, 0);
803 
804 #if 0
805  			if (np->n_flag & NLMODIFIED) {
806  			    if (vap->va_size == 0)
807  				error = nfs_vinvalbuf(vp, 0, 1);
808  			    else
809  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
810  			}
811 #endif
812 			/*
813 			 * note: this loop case almost always happens at
814 			 * least once per truncation.
815 			 */
816 			if (error == 0 && np->n_size != vap->va_size)
817 				goto again;
818 			np->n_vattr.va_size = vap->va_size;
819 			kflags |= NOTE_WRITE;
820 			if (tsize < vap->va_size)
821 				kflags |= NOTE_EXTEND;
822 			break;
823 		}
824 	} else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
825 		/*
826 		 * What to do.  If we are modifying the mtime we lose
827 		 * mtime detection of changes made by the server or other
828 		 * clients.  But programs like rsync/rdist/cpdup are going
829 		 * to call utimes a lot.  We don't want to piecemeal sync.
830 		 *
831 		 * For now sync if any prior remote changes were detected,
832 		 * but allow us to lose track of remote changes made during
833 		 * the utimes operation.
834 		 */
835 		if (np->n_flag & NRMODIFIED)
836 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
837 		if (error == EINTR) {
838 			lwkt_reltoken(&nmp->nm_token);
839 			return (error);
840 		}
841 		if (error == 0) {
842 			if (vap->va_mtime.tv_sec != VNOVAL) {
843 				np->n_mtime = vap->va_mtime.tv_sec;
844 			}
845 		}
846 	}
847 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
848 	if (error == 0)
849 		kflags |= NOTE_EXTEND;
850 
851 	/*
852 	 * Sanity check if a truncation was issued.  This should only occur
853 	 * if multiple processes are racing on the same file.
854 	 */
855 	if (error == 0 && vap->va_size != VNOVAL &&
856 	    np->n_size != vap->va_size) {
857 		kprintf("NFS ftruncate: server disagrees on the file size: "
858 			"%jd/%jd/%jd\n",
859 			(intmax_t)tsize,
860 			(intmax_t)vap->va_size,
861 			(intmax_t)np->n_size);
862 		goto again;
863 	}
864 	if (error && vap->va_size != VNOVAL) {
865 		np->n_size = np->n_vattr.va_size = tsize;
866 		nfs_meta_setsize(vp, td, np->n_size, 0);
867 	}
868 	lwkt_reltoken(&nmp->nm_token);
869 	nfs_knote(vp, kflags);
870 
871 	return (error);
872 }
873 
874 /*
875  * Do an nfs setattr rpc.
876  */
877 static int
878 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
879 	       struct ucred *cred, struct thread *td)
880 {
881 	struct nfsv2_sattr *sp;
882 	struct nfsnode *np = VTONFS(vp);
883 	u_int32_t *tl;
884 	int error = 0, wccflag = NFSV3_WCCRATTR;
885 	struct nfsm_info info;
886 
887 	info.mrep = NULL;
888 	info.v3 = NFS_ISV3(vp);
889 
890 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
891 	nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
892 		     NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
893 	ERROROUT(nfsm_fhtom(&info, vp));
894 	if (info.v3) {
895 		nfsm_v3attrbuild(&info, vap, TRUE);
896 		tl = nfsm_build(&info, NFSX_UNSIGNED);
897 		*tl = nfs_false;
898 	} else {
899 		sp = nfsm_build(&info, NFSX_V2SATTR);
900 		if (vap->va_mode == (mode_t)VNOVAL)
901 			sp->sa_mode = nfs_xdrneg1;
902 		else
903 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
904 		if (vap->va_uid == (uid_t)VNOVAL)
905 			sp->sa_uid = nfs_xdrneg1;
906 		else
907 			sp->sa_uid = txdr_unsigned(vap->va_uid);
908 		if (vap->va_gid == (gid_t)VNOVAL)
909 			sp->sa_gid = nfs_xdrneg1;
910 		else
911 			sp->sa_gid = txdr_unsigned(vap->va_gid);
912 		sp->sa_size = txdr_unsigned(vap->va_size);
913 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
914 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
915 	}
916 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
917 	if (info.v3) {
918 		np->n_modestamp = 0;
919 		ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
920 	} else {
921 		ERROROUT(nfsm_loadattr(&info, vp, NULL));
922 	}
923 	m_freem(info.mrep);
924 	info.mrep = NULL;
925 nfsmout:
926 	return (error);
927 }
928 
929 static
930 void
931 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
932 {
933 	if (nctimeout == 0)
934 		nctimeout = 1;
935 	else
936 		nctimeout *= hz;
937 	cache_setvp(nch, vp);
938 	cache_settimeout(nch, nctimeout);
939 }
940 
941 /*
942  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
943  * nfs_lookup() until all remaining new api calls are implemented.
944  *
945  * Resolve a namecache entry.  This function is passed a locked ncp and
946  * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
947  */
948 static int
949 nfs_nresolve(struct vop_nresolve_args *ap)
950 {
951 	struct thread *td = curthread;
952 	struct namecache *ncp;
953 	struct nfsmount *nmp;
954 	struct nfsnode *np;
955 	struct vnode *dvp;
956 	struct vnode *nvp;
957 	nfsfh_t *fhp;
958 	int attrflag;
959 	int fhsize;
960 	int error;
961 	int tmp_error;
962 	int len;
963 	struct nfsm_info info;
964 
965 	dvp = ap->a_dvp;
966 	nmp = VFSTONFS(dvp->v_mount);
967 
968 	lwkt_gettoken(&nmp->nm_token);
969 
970 	if ((error = vget(dvp, LK_SHARED)) != 0) {
971 		lwkt_reltoken(&nmp->nm_token);
972 		return (error);
973 	}
974 
975 	info.mrep = NULL;
976 	info.v3 = NFS_ISV3(dvp);
977 
978 	nvp = NULL;
979 	nfsstats.lookupcache_misses++;
980 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
981 	ncp = ap->a_nch->ncp;
982 	len = ncp->nc_nlen;
983 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
984 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
985 	ERROROUT(nfsm_fhtom(&info, dvp));
986 	ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
987 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
988 				ap->a_cred, &error));
989 	if (error) {
990 		/*
991 		 * Cache negatve lookups to reduce NFS traffic, but use
992 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
993 		 * XXX we should add a namecache flag for no-caching
994 		 * to uncache the negative hit as soon as possible, but
995 		 * we cannot simply destroy the entry because it is used
996 		 * as a placeholder by the caller.
997 		 *
998 		 * The refactored nfs code will overwrite a non-zero error
999 		 * with 0 when we use ERROROUT(), so don't here.
1000 		 */
1001 		if (error == ENOENT)
1002 			nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
1003 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1004 					     NFS_LATTR_NOSHRINK);
1005 		if (tmp_error) {
1006 			error = tmp_error;
1007 			goto nfsmout;
1008 		}
1009 		m_freem(info.mrep);
1010 		info.mrep = NULL;
1011 		goto nfsmout;
1012 	}
1013 
1014 	/*
1015 	 * Success, get the file handle, do various checks, and load
1016 	 * post-operation data from the reply packet.  Theoretically
1017 	 * we should never be looking up "." so, theoretically, we
1018 	 * should never get the same file handle as our directory.  But
1019 	 * we check anyway. XXX
1020 	 *
1021 	 * Note that no timeout is set for the positive cache hit.  We
1022 	 * assume, theoretically, that ESTALE returns will be dealt with
1023 	 * properly to handle NFS races and in anycase we cannot depend
1024 	 * on a timeout to deal with NFS open/create/excl issues so instead
1025 	 * of a bad hack here the rest of the NFS client code needs to do
1026 	 * the right thing.
1027 	 */
1028 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1029 
1030 	np = VTONFS(dvp);
1031 	if (NFS_CMPFH(np, fhp, fhsize)) {
1032 		vref(dvp);
1033 		nvp = dvp;
1034 	} else {
1035 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, NULL);
1036 		if (error) {
1037 			m_freem(info.mrep);
1038 			info.mrep = NULL;
1039 			vput(dvp);
1040 			lwkt_reltoken(&nmp->nm_token);
1041 			return (error);
1042 		}
1043 		nvp = NFSTOV(np);
1044 	}
1045 	if (info.v3) {
1046 		ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1047 					  NFS_LATTR_NOSHRINK));
1048 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1049 					  NFS_LATTR_NOSHRINK));
1050 	} else {
1051 		ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1052 	}
1053 	nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1054 	m_freem(info.mrep);
1055 	info.mrep = NULL;
1056 nfsmout:
1057 	lwkt_reltoken(&nmp->nm_token);
1058 	vput(dvp);
1059 	if (nvp) {
1060 		if (nvp == dvp)
1061 			vrele(nvp);
1062 		else
1063 			vput(nvp);
1064 	}
1065 	return (error);
1066 }
1067 
1068 /*
1069  * 'cached' nfs directory lookup
1070  *
1071  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1072  *
1073  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1074  *	      struct componentname *a_cnp)
1075  */
1076 static int
1077 nfs_lookup(struct vop_old_lookup_args *ap)
1078 {
1079 	struct componentname *cnp = ap->a_cnp;
1080 	struct vnode *dvp = ap->a_dvp;
1081 	struct vnode **vpp = ap->a_vpp;
1082 	int flags = cnp->cn_flags;
1083 	struct vnode *newvp;
1084 	struct vnode *notvp;
1085 	struct nfsmount *nmp;
1086 	long len;
1087 	nfsfh_t *fhp;
1088 	struct nfsnode *np;
1089 	int lockparent, wantparent, attrflag, fhsize;
1090 	int error;
1091 	int tmp_error;
1092 	struct nfsm_info info;
1093 
1094 	info.mrep = NULL;
1095 	info.v3 = NFS_ISV3(dvp);
1096 	error = 0;
1097 
1098 	notvp = (cnp->cn_flags & CNP_NOTVP) ? cnp->cn_notvp : NULL;
1099 
1100 	/*
1101 	 * Read-only mount check and directory check.
1102 	 */
1103 	*vpp = NULLVP;
1104 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1105 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1106 		return (EROFS);
1107 
1108 	if (dvp->v_type != VDIR)
1109 		return (ENOTDIR);
1110 
1111 	/*
1112 	 * Look it up in the cache.  Note that ENOENT is only returned if we
1113 	 * previously entered a negative hit (see later on).  The additional
1114 	 * nfsneg_cache_timeout check causes previously cached results to
1115 	 * be instantly ignored if the negative caching is turned off.
1116 	 */
1117 	lockparent = flags & CNP_LOCKPARENT;
1118 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1119 	nmp = VFSTONFS(dvp->v_mount);
1120 	np = VTONFS(dvp);
1121 
1122 	lwkt_gettoken(&nmp->nm_token);
1123 
1124 	/*
1125 	 * Go to the wire.
1126 	 */
1127 	error = 0;
1128 	newvp = NULLVP;
1129 	nfsstats.lookupcache_misses++;
1130 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1131 	len = cnp->cn_namelen;
1132 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1133 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1134 	ERROROUT(nfsm_fhtom(&info, dvp));
1135 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1136 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1137 				cnp->cn_cred, &error));
1138 	if (error) {
1139 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1140 					     NFS_LATTR_NOSHRINK);
1141 		if (tmp_error) {
1142 			error = tmp_error;
1143 			goto nfsmout;
1144 		}
1145 
1146 		m_freem(info.mrep);
1147 		info.mrep = NULL;
1148 		goto nfsmout;
1149 	}
1150 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1151 
1152 	/*
1153 	 * Handle RENAME case...
1154 	 */
1155 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1156 		if (NFS_CMPFH(np, fhp, fhsize)) {
1157 			m_freem(info.mrep);
1158 			info.mrep = NULL;
1159 			lwkt_reltoken(&nmp->nm_token);
1160 			return (EISDIR);
1161 		}
1162 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1163 		if (error) {
1164 			m_freem(info.mrep);
1165 			info.mrep = NULL;
1166 			lwkt_reltoken(&nmp->nm_token);
1167 			return (error);
1168 		}
1169 		newvp = NFSTOV(np);
1170 		if (info.v3) {
1171 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1172 						  NFS_LATTR_NOSHRINK));
1173 			ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1174 						  NFS_LATTR_NOSHRINK));
1175 		} else {
1176 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1177 		}
1178 		*vpp = newvp;
1179 		m_freem(info.mrep);
1180 		info.mrep = NULL;
1181 		if (!lockparent) {
1182 			vn_unlock(dvp);
1183 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1184 		}
1185 		lwkt_reltoken(&nmp->nm_token);
1186 		return (0);
1187 	}
1188 
1189 	if (flags & CNP_ISDOTDOT) {
1190 		vn_unlock(dvp);
1191 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1192 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1193 		if (error) {
1194 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1195 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1196 			lwkt_reltoken(&nmp->nm_token);
1197 			return (error); /* NOTE: return error from nget */
1198 		}
1199 		newvp = NFSTOV(np);
1200 		if (lockparent) {
1201 			error = vn_lock(dvp, LK_EXCLUSIVE | LK_FAILRECLAIM);
1202 			if (error) {
1203 				vput(newvp);
1204 				lwkt_reltoken(&nmp->nm_token);
1205 				return (error);
1206 			}
1207 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1208 		}
1209 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1210 		vref(dvp);
1211 		newvp = dvp;
1212 	} else {
1213 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1214 		if (error) {
1215 			m_freem(info.mrep);
1216 			info.mrep = NULL;
1217 			lwkt_reltoken(&nmp->nm_token);
1218 			return (error);
1219 		}
1220 		if (!lockparent) {
1221 			vn_unlock(dvp);
1222 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1223 		}
1224 		newvp = NFSTOV(np);
1225 	}
1226 	if (info.v3) {
1227 		ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1228 					  NFS_LATTR_NOSHRINK));
1229 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1230 					  NFS_LATTR_NOSHRINK));
1231 	} else {
1232 		ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1233 	}
1234 #if 0
1235 	/* XXX MOVE TO nfs_nremove() */
1236 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1237 	    cnp->cn_nameiop != NAMEI_DELETE) {
1238 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1239 	}
1240 #endif
1241 	*vpp = newvp;
1242 	m_freem(info.mrep);
1243 	info.mrep = NULL;
1244 nfsmout:
1245 	if (error) {
1246 		if (newvp != NULLVP) {
1247 			vrele(newvp);
1248 			*vpp = NULLVP;
1249 		}
1250 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1251 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1252 		    error == ENOENT) {
1253 			if (!lockparent) {
1254 				vn_unlock(dvp);
1255 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1256 			}
1257 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1258 				error = EROFS;
1259 			else
1260 				error = EJUSTRETURN;
1261 		}
1262 	}
1263 	lwkt_reltoken(&nmp->nm_token);
1264 	return (error);
1265 }
1266 
1267 /*
1268  * nfs read call.
1269  * Just call nfs_bioread() to do the work.
1270  *
1271  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1272  *	    struct ucred *a_cred)
1273  */
1274 static int
1275 nfs_read(struct vop_read_args *ap)
1276 {
1277 	struct vnode *vp = ap->a_vp;
1278 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1279 	int error;
1280 
1281 	lwkt_gettoken(&nmp->nm_token);
1282 	error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1283 	lwkt_reltoken(&nmp->nm_token);
1284 
1285 	return error;
1286 }
1287 
1288 /*
1289  * nfs readlink call
1290  *
1291  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1292  */
1293 static int
1294 nfs_readlink(struct vop_readlink_args *ap)
1295 {
1296 	struct vnode *vp = ap->a_vp;
1297 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1298 	int error;
1299 
1300 	if (vp->v_type != VLNK)
1301 		return (EINVAL);
1302 
1303 	lwkt_gettoken(&nmp->nm_token);
1304 	error = nfs_bioread(vp, ap->a_uio, 0);
1305 	lwkt_reltoken(&nmp->nm_token);
1306 
1307 	return error;
1308 }
1309 
1310 /*
1311  * Do a readlink rpc.
1312  * Called by nfs_doio() from below the buffer cache.
1313  */
1314 int
1315 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1316 {
1317 	int error = 0, len, attrflag;
1318 	struct nfsm_info info;
1319 
1320 	info.mrep = NULL;
1321 	info.v3 = NFS_ISV3(vp);
1322 
1323 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1324 	nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1325 	ERROROUT(nfsm_fhtom(&info, vp));
1326 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1327 				nfs_vpcred(vp, ND_CHECK), &error));
1328 	if (info.v3) {
1329 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1330 					  NFS_LATTR_NOSHRINK));
1331 	}
1332 	if (!error) {
1333 		NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1334 		if (len == NFS_MAXPATHLEN) {
1335 			struct nfsnode *np = VTONFS(vp);
1336 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1337 				len = np->n_size;
1338 		}
1339 		ERROROUT(nfsm_mtouio(&info, uiop, len));
1340 	}
1341 	m_freem(info.mrep);
1342 	info.mrep = NULL;
1343 nfsmout:
1344 	return (error);
1345 }
1346 
1347 /*
1348  * nfs synchronous read rpc using UIO
1349  */
1350 int
1351 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1352 {
1353 	u_int32_t *tl;
1354 	struct nfsmount *nmp;
1355 	int error = 0, len, retlen, tsiz, eof, attrflag;
1356 	struct nfsm_info info;
1357 	off_t tmp_off;
1358 
1359 	info.mrep = NULL;
1360 	info.v3 = NFS_ISV3(vp);
1361 
1362 #ifndef nolint
1363 	eof = 0;
1364 #endif
1365 	nmp = VFSTONFS(vp->v_mount);
1366 
1367 	tsiz = uiop->uio_resid;
1368 	tmp_off = uiop->uio_offset + tsiz;
1369 	if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1370 		return (EFBIG);
1371 	tmp_off = uiop->uio_offset;
1372 	while (tsiz > 0) {
1373 		nfsstats.rpccnt[NFSPROC_READ]++;
1374 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1375 		nfsm_reqhead(&info, vp, NFSPROC_READ,
1376 			     NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1377 		ERROROUT(nfsm_fhtom(&info, vp));
1378 		tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1379 		if (info.v3) {
1380 			txdr_hyper(uiop->uio_offset, tl);
1381 			*(tl + 2) = txdr_unsigned(len);
1382 		} else {
1383 			*tl++ = txdr_unsigned(uiop->uio_offset);
1384 			*tl++ = txdr_unsigned(len);
1385 			*tl = 0;
1386 		}
1387 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1388 					nfs_vpcred(vp, ND_READ), &error));
1389 		if (info.v3) {
1390 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1391 						 NFS_LATTR_NOSHRINK));
1392 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1393 			eof = fxdr_unsigned(int, *(tl + 1));
1394 		} else {
1395 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1396 		}
1397 		NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1398 		ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1399 		m_freem(info.mrep);
1400 		info.mrep = NULL;
1401 
1402 		/*
1403 		 * Handle short-read from server (NFSv3).  If EOF is not
1404 		 * flagged (and no error occurred), but retlen is less
1405 		 * then the request size, we must zero-fill the remainder.
1406 		 */
1407 		if (retlen < len && info.v3 && eof == 0) {
1408 			ERROROUT(uiomovez(len - retlen, uiop));
1409 			retlen = len;
1410 		}
1411 		tsiz -= retlen;
1412 
1413 		/*
1414 		 * Terminate loop on EOF or zero-length read.
1415 		 *
1416 		 * For NFSv2 a short-read indicates EOF, not zero-fill,
1417 		 * and also terminates the loop.
1418 		 */
1419 		if (info.v3) {
1420 			if (eof || retlen == 0)
1421 				tsiz = 0;
1422 		} else if (retlen < len) {
1423 			tsiz = 0;
1424 		}
1425 	}
1426 nfsmout:
1427 	return (error);
1428 }
1429 
1430 /*
1431  * nfs write call
1432  */
1433 int
1434 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1435 		 int *iomode, int *must_commit)
1436 {
1437 	u_int32_t *tl;
1438 	int32_t backup;
1439 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1440 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1441 	int  committed = NFSV3WRITE_FILESYNC;
1442 	struct nfsm_info info;
1443 
1444 	info.mrep = NULL;
1445 	info.v3 = NFS_ISV3(vp);
1446 
1447 #ifndef DIAGNOSTIC
1448 	if (uiop->uio_iovcnt != 1)
1449 		panic("nfs: writerpc iovcnt > 1");
1450 #endif
1451 	*must_commit = 0;
1452 	tsiz = uiop->uio_resid;
1453 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1454 		return (EFBIG);
1455 	while (tsiz > 0) {
1456 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1457 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1458 		nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1459 			     NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1460 			     nfsm_rndup(len));
1461 		ERROROUT(nfsm_fhtom(&info, vp));
1462 		if (info.v3) {
1463 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1464 			txdr_hyper(uiop->uio_offset, tl);
1465 			tl += 2;
1466 			*tl++ = txdr_unsigned(len);
1467 			*tl++ = txdr_unsigned(*iomode);
1468 			*tl = txdr_unsigned(len);
1469 		} else {
1470 			u_int32_t x;
1471 
1472 			tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1473 			/* Set both "begin" and "current" to non-garbage. */
1474 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1475 			*tl++ = x;	/* "begin offset" */
1476 			*tl++ = x;	/* "current offset" */
1477 			x = txdr_unsigned(len);
1478 			*tl++ = x;	/* total to this offset */
1479 			*tl = x;	/* size of this write */
1480 		}
1481 		ERROROUT(nfsm_uiotom(&info, uiop, len));
1482 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1483 					nfs_vpcred(vp, ND_WRITE), &error));
1484 		if (info.v3) {
1485 			/*
1486 			 * The write RPC returns a before and after mtime.  The
1487 			 * nfsm_wcc_data() macro checks the before n_mtime
1488 			 * against the before time and stores the after time
1489 			 * in the nfsnode's cached vattr and n_mtime field.
1490 			 * The NRMODIFIED bit will be set if the before
1491 			 * time did not match the original mtime.
1492 			 */
1493 			wccflag = NFSV3_WCCCHK;
1494 			ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1495 			if (error == 0) {
1496 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1497 				rlen = fxdr_unsigned(int, *tl++);
1498 				if (rlen == 0) {
1499 					error = NFSERR_IO;
1500 					m_freem(info.mrep);
1501 					info.mrep = NULL;
1502 					break;
1503 				} else if (rlen < len) {
1504 					backup = len - rlen;
1505 					uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1506 					uiop->uio_iov->iov_len += backup;
1507 					uiop->uio_offset -= backup;
1508 					uiop->uio_resid += backup;
1509 					len = rlen;
1510 				}
1511 				commit = fxdr_unsigned(int, *tl++);
1512 
1513 				/*
1514 				 * Return the lowest committment level
1515 				 * obtained by any of the RPCs.
1516 				 */
1517 				if (committed == NFSV3WRITE_FILESYNC)
1518 					committed = commit;
1519 				else if (committed == NFSV3WRITE_DATASYNC &&
1520 					commit == NFSV3WRITE_UNSTABLE)
1521 					committed = commit;
1522 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1523 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1524 					NFSX_V3WRITEVERF);
1525 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1526 				} else if (bcmp((caddr_t)tl,
1527 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1528 				    *must_commit = 1;
1529 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1530 					NFSX_V3WRITEVERF);
1531 				}
1532 			}
1533 		} else {
1534 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1535 		}
1536 		m_freem(info.mrep);
1537 		info.mrep = NULL;
1538 		if (error)
1539 			break;
1540 		tsiz -= len;
1541 	}
1542 nfsmout:
1543 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1544 		committed = NFSV3WRITE_FILESYNC;
1545 	*iomode = committed;
1546 	if (error)
1547 		uiop->uio_resid = tsiz;
1548 	return (error);
1549 }
1550 
1551 /*
1552  * nfs mknod rpc
1553  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1554  * mode set to specify the file type and the size field for rdev.
1555  */
1556 static int
1557 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1558 	     struct vattr *vap)
1559 {
1560 	struct nfsv2_sattr *sp;
1561 	u_int32_t *tl;
1562 	struct vnode *newvp = NULL;
1563 	struct nfsnode *np = NULL;
1564 	struct vattr vattr;
1565 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1566 	int rmajor, rminor;
1567 	struct nfsm_info info;
1568 
1569 	info.mrep = NULL;
1570 	info.v3 = NFS_ISV3(dvp);
1571 
1572 	if (vap->va_type == VCHR || vap->va_type == VBLK) {
1573 		rmajor = txdr_unsigned(vap->va_rmajor);
1574 		rminor = txdr_unsigned(vap->va_rminor);
1575 	} else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1576 		rmajor = nfs_xdrneg1;
1577 		rminor = nfs_xdrneg1;
1578 	} else {
1579 		return (EOPNOTSUPP);
1580 	}
1581 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1582 		return (error);
1583 	}
1584 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1585 	nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1586 		     NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1587 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1588 	ERROROUT(nfsm_fhtom(&info, dvp));
1589 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1590 			     NFS_MAXNAMLEN));
1591 	if (info.v3) {
1592 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1593 		*tl++ = vtonfsv3_type(vap->va_type);
1594 		nfsm_v3attrbuild(&info, vap, FALSE);
1595 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1596 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1597 			*tl++ = txdr_unsigned(vap->va_rmajor);
1598 			*tl = txdr_unsigned(vap->va_rminor);
1599 		}
1600 	} else {
1601 		sp = nfsm_build(&info, NFSX_V2SATTR);
1602 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1603 		sp->sa_uid = nfs_xdrneg1;
1604 		sp->sa_gid = nfs_xdrneg1;
1605 		sp->sa_size = makeudev(rmajor, rminor);
1606 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1607 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1608 	}
1609 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1610 				cnp->cn_cred, &error));
1611 	if (!error) {
1612 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1613 		if (!gotvp) {
1614 			if (newvp) {
1615 				vput(newvp);
1616 				newvp = NULL;
1617 			}
1618 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1619 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1620 			if (!error)
1621 				newvp = NFSTOV(np);
1622 		}
1623 	}
1624 	if (info.v3) {
1625 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1626 	}
1627 	m_freem(info.mrep);
1628 	info.mrep = NULL;
1629 nfsmout:
1630 	if (error) {
1631 		if (newvp)
1632 			vput(newvp);
1633 	} else {
1634 		*vpp = newvp;
1635 	}
1636 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1637 	if (!wccflag)
1638 		VTONFS(dvp)->n_attrstamp = 0;
1639 	return (error);
1640 }
1641 
1642 /*
1643  * nfs mknod vop
1644  * just call nfs_mknodrpc() to do the work.
1645  *
1646  * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1647  *	     struct componentname *a_cnp, struct vattr *a_vap)
1648  */
1649 /* ARGSUSED */
1650 static int
1651 nfs_mknod(struct vop_old_mknod_args *ap)
1652 {
1653 	struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1654 	int error;
1655 
1656 	lwkt_gettoken(&nmp->nm_token);
1657 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1658 	lwkt_reltoken(&nmp->nm_token);
1659 	if (error == 0)
1660 		nfs_knote(ap->a_dvp, NOTE_WRITE);
1661 
1662 	return error;
1663 }
1664 
1665 static u_long create_verf;
1666 /*
1667  * nfs file create call
1668  *
1669  * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1670  *	      struct componentname *a_cnp, struct vattr *a_vap)
1671  */
1672 static int
1673 nfs_create(struct vop_old_create_args *ap)
1674 {
1675 	struct vnode *dvp = ap->a_dvp;
1676 	struct vattr *vap = ap->a_vap;
1677 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1678 	struct componentname *cnp = ap->a_cnp;
1679 	struct nfsv2_sattr *sp;
1680 	u_int32_t *tl;
1681 	struct nfsnode *np = NULL;
1682 	struct vnode *newvp = NULL;
1683 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1684 	struct vattr vattr;
1685 	struct nfsm_info info;
1686 
1687 	info.mrep = NULL;
1688 	info.v3 = NFS_ISV3(dvp);
1689 	lwkt_gettoken(&nmp->nm_token);
1690 
1691 	/*
1692 	 * Oops, not for me..
1693 	 */
1694 	if (vap->va_type == VSOCK) {
1695 		error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1696 		lwkt_reltoken(&nmp->nm_token);
1697 		return error;
1698 	}
1699 
1700 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1701 		lwkt_reltoken(&nmp->nm_token);
1702 		return (error);
1703 	}
1704 	if (vap->va_vaflags & VA_EXCLUSIVE)
1705 		fmode |= O_EXCL;
1706 again:
1707 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1708 	nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1709 		     NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1710 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1711 	ERROROUT(nfsm_fhtom(&info, dvp));
1712 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1713 			     NFS_MAXNAMLEN));
1714 	if (info.v3) {
1715 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1716 		if (fmode & O_EXCL) {
1717 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1718 			tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1719 #ifdef INET
1720 			if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1721 				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1722 			else
1723 #endif
1724 				*tl++ = create_verf;
1725 			*tl = ++create_verf;
1726 		} else {
1727 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1728 			nfsm_v3attrbuild(&info, vap, FALSE);
1729 		}
1730 	} else {
1731 		sp = nfsm_build(&info, NFSX_V2SATTR);
1732 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1733 		sp->sa_uid = nfs_xdrneg1;
1734 		sp->sa_gid = nfs_xdrneg1;
1735 		sp->sa_size = 0;
1736 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1737 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1738 	}
1739 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1740 				cnp->cn_cred, &error));
1741 	if (error == 0) {
1742 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1743 		if (!gotvp) {
1744 			if (newvp) {
1745 				vput(newvp);
1746 				newvp = NULL;
1747 			}
1748 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1749 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1750 			if (!error)
1751 				newvp = NFSTOV(np);
1752 		}
1753 	}
1754 	if (info.v3) {
1755 		if (error == 0)
1756 			error = nfsm_wcc_data(&info, dvp, &wccflag);
1757 		else
1758 			(void)nfsm_wcc_data(&info, dvp, &wccflag);
1759 	}
1760 	m_freem(info.mrep);
1761 	info.mrep = NULL;
1762 nfsmout:
1763 	if (error) {
1764 		if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1765 			KKASSERT(newvp == NULL);
1766 			fmode &= ~O_EXCL;
1767 			goto again;
1768 		}
1769 	} else if (info.v3 && (fmode & O_EXCL)) {
1770 		/*
1771 		 * We are normally called with only a partially initialized
1772 		 * VAP.  Since the NFSv3 spec says that server may use the
1773 		 * file attributes to store the verifier, the spec requires
1774 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1775 		 * in atime, but we can't really assume that all servers will
1776 		 * so we ensure that our SETATTR sets both atime and mtime.
1777 		 */
1778 		if (vap->va_mtime.tv_sec == VNOVAL)
1779 			vfs_timestamp(&vap->va_mtime);
1780 		if (vap->va_atime.tv_sec == VNOVAL)
1781 			vap->va_atime = vap->va_mtime;
1782 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1783 	}
1784 	if (error == 0) {
1785 		/*
1786 		 * The new np may have enough info for access
1787 		 * checks, make sure rucred and wucred are
1788 		 * initialized for read and write rpc's.
1789 		 */
1790 		np = VTONFS(newvp);
1791 		if (np->n_rucred == NULL)
1792 			np->n_rucred = crhold(cnp->cn_cred);
1793 		if (np->n_wucred == NULL)
1794 			np->n_wucred = crhold(cnp->cn_cred);
1795 		*ap->a_vpp = newvp;
1796 		nfs_knote(dvp, NOTE_WRITE);
1797 	} else if (newvp) {
1798 		vput(newvp);
1799 	}
1800 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1801 	if (!wccflag)
1802 		VTONFS(dvp)->n_attrstamp = 0;
1803 	lwkt_reltoken(&nmp->nm_token);
1804 	return (error);
1805 }
1806 
1807 /*
1808  * nfs file remove call
1809  * To try and make nfs semantics closer to ufs semantics, a file that has
1810  * other processes using the vnode is renamed instead of removed and then
1811  * removed later on the last close.
1812  * - If v_refcnt > 1
1813  *	  If a rename is not already in the works
1814  *	     call nfs_sillyrename() to set it up
1815  *     else
1816  *	  do the remove rpc
1817  *
1818  * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1819  *	      struct componentname *a_cnp)
1820  */
1821 static int
1822 nfs_remove(struct vop_old_remove_args *ap)
1823 {
1824 	struct vnode *vp = ap->a_vp;
1825 	struct vnode *dvp = ap->a_dvp;
1826 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1827 	struct componentname *cnp = ap->a_cnp;
1828 	struct nfsnode *np = VTONFS(vp);
1829 	int error = 0;
1830 	struct vattr vattr;
1831 
1832 	lwkt_gettoken(&nmp->nm_token);
1833 #ifndef DIAGNOSTIC
1834 	if (VREFCNT(vp) < 1)
1835 		panic("nfs_remove: bad v_refcnt");
1836 #endif
1837 	if (vp->v_type == VDIR) {
1838 		error = EPERM;
1839 	} else if (VREFCNT(vp) == 1 || (np->n_sillyrename &&
1840 		   VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1841 		/*
1842 		 * Force finalization so the VOP_INACTIVE() call is not delayed.
1843 		 * This prevents cred structures from building up in nfsnodes
1844 		 * for deleted files.
1845 		 */
1846 		atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1847 		np->n_flag |= NREMOVED;
1848 
1849 		/*
1850 		 * Throw away biocache buffers, mainly to avoid
1851 		 * unnecessary delayed writes later.
1852 		 */
1853 		error = nfs_vinvalbuf(vp, 0, 1);
1854 		/* Do the rpc */
1855 		if (error != EINTR) {
1856 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1857 					      cnp->cn_namelen,
1858 					      cnp->cn_cred, cnp->cn_td);
1859 		}
1860 
1861 		/*
1862 		 * Kludge City: If the first reply to the remove rpc is lost..
1863 		 *   the reply to the retransmitted request will be ENOENT
1864 		 *   since the file was in fact removed
1865 		 *   Therefore, we cheat and return success.
1866 		 */
1867 		if (error == ENOENT)
1868 			error = 0;
1869 	} else if (!np->n_sillyrename) {
1870 		error = nfs_sillyrename(dvp, vp, cnp);
1871 	}
1872 	np->n_attrstamp = 0;
1873 	lwkt_reltoken(&nmp->nm_token);
1874 	if (error == 0) {
1875 		nfs_knote(vp, NOTE_DELETE);
1876 		nfs_knote(dvp, NOTE_WRITE);
1877 	}
1878 
1879 	return (error);
1880 }
1881 
1882 /*
1883  * nfs file remove rpc called from nfs_inactive
1884  *
1885  * NOTE: s_dvp can be VBAD during a forced unmount.
1886  */
1887 int
1888 nfs_removeit(struct sillyrename *sp)
1889 {
1890 	if (sp->s_dvp->v_type == VBAD)
1891 		return(0);
1892 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1893 		sp->s_cred, NULL));
1894 }
1895 
1896 /*
1897  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1898  */
1899 static int
1900 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1901 	      struct ucred *cred, struct thread *td)
1902 {
1903 	int error = 0, wccflag = NFSV3_WCCRATTR;
1904 	struct nfsm_info info;
1905 
1906 	info.mrep = NULL;
1907 	info.v3 = NFS_ISV3(dvp);
1908 
1909 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1910 	nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1911 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1912 	ERROROUT(nfsm_fhtom(&info, dvp));
1913 	ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1914 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1915 	if (info.v3) {
1916 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1917 	}
1918 	m_freem(info.mrep);
1919 	info.mrep = NULL;
1920 nfsmout:
1921 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1922 	if (!wccflag)
1923 		VTONFS(dvp)->n_attrstamp = 0;
1924 	return (error);
1925 }
1926 
1927 /*
1928  * nfs file rename call
1929  *
1930  * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1931  *	      struct componentname *a_fcnp, struct vnode *a_tdvp,
1932  *	      struct vnode *a_tvp, struct componentname *a_tcnp)
1933  */
1934 static int
1935 nfs_rename(struct vop_old_rename_args *ap)
1936 {
1937 	struct vnode *fvp = ap->a_fvp;
1938 	struct vnode *tvp = ap->a_tvp;
1939 	struct vnode *fdvp = ap->a_fdvp;
1940 	struct vnode *tdvp = ap->a_tdvp;
1941 	struct componentname *tcnp = ap->a_tcnp;
1942 	struct componentname *fcnp = ap->a_fcnp;
1943 	struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1944 	int error;
1945 
1946 	lwkt_gettoken(&nmp->nm_token);
1947 
1948 	/*
1949 	 * Force finalization so the VOP_INACTIVE() call is not delayed.
1950 	 * This prevents cred structures from building up in nfsnodes
1951 	 * for deleted files.
1952 	 */
1953 	if (tvp) {
1954 		atomic_set_int(&tvp->v_refcnt, VREF_FINALIZE);
1955 		if (VTONFS(tvp))
1956 			VTONFS(tvp)->n_flag |= NREMOVED;
1957 	}
1958 
1959 	/* Check for cross-device rename */
1960 	if ((fvp->v_mount != tdvp->v_mount) ||
1961 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1962 		error = EXDEV;
1963 		goto out;
1964 	}
1965 
1966 	/*
1967 	 * We shouldn't have to flush fvp on rename for most server-side
1968 	 * filesystems as the file handle should not change.  Unfortunately
1969 	 * the inode for some filesystems (msdosfs) might be tied to the
1970 	 * file name or directory position so to be completely safe
1971 	 * vfs.nfs.flush_on_rename is set by default.  Clear to improve
1972 	 * performance.
1973 	 *
1974 	 * We must flush tvp on rename because it might become stale on the
1975 	 * server after the rename.
1976 	 */
1977 	if (nfs_flush_on_rename)
1978 	    VOP_FSYNC(fvp, MNT_WAIT, 0);
1979 	if (tvp)
1980 	    VOP_FSYNC(tvp, MNT_WAIT, 0);
1981 
1982 	/*
1983 	 * If the tvp exists and is in use, sillyrename it before doing the
1984 	 * rename of the new file over it.
1985 	 *
1986 	 * XXX Can't sillyrename a directory.
1987 	 *
1988 	 * We do not attempt to do any namecache purges in this old API
1989 	 * routine.  The new API compat functions have access to the actual
1990 	 * namecache structures and will do it for us.
1991 	 */
1992 	if (tvp && VREFCNT(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1993 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1994 		nfs_knote(tvp, NOTE_DELETE);
1995 		vput(tvp);
1996 		tvp = NULL;
1997 	} else if (tvp) {
1998 		nfs_knote(tvp, NOTE_DELETE);
1999 	}
2000 
2001 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2002 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2003 		tcnp->cn_td);
2004 
2005 out:
2006 	if (error == 0) {
2007 		nfs_knote(fdvp, NOTE_WRITE);
2008 		nfs_knote(tdvp, NOTE_WRITE);
2009 		nfs_knote(fvp, NOTE_RENAME);
2010 	}
2011 	lwkt_reltoken(&nmp->nm_token);
2012 	if (tdvp == tvp)
2013 		vrele(tdvp);
2014 	else
2015 		vput(tdvp);
2016 	if (tvp)
2017 		vput(tvp);
2018 	vrele(fdvp);
2019 	vrele(fvp);
2020 	/*
2021 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2022 	 */
2023 	if (error == ENOENT)
2024 		error = 0;
2025 	return (error);
2026 }
2027 
2028 /*
2029  * nfs file rename rpc called from nfs_remove() above
2030  */
2031 static int
2032 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
2033 	     struct sillyrename *sp)
2034 {
2035 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2036 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
2037 }
2038 
2039 /*
2040  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2041  */
2042 static int
2043 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
2044 	      struct vnode *tdvp, const char *tnameptr, int tnamelen,
2045 	      struct ucred *cred, struct thread *td)
2046 {
2047 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2048 	struct nfsm_info info;
2049 
2050 	info.mrep = NULL;
2051 	info.v3 = NFS_ISV3(fdvp);
2052 
2053 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2054 	nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
2055 		    (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
2056 		    nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
2057 	ERROROUT(nfsm_fhtom(&info, fdvp));
2058 	ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
2059 	ERROROUT(nfsm_fhtom(&info, tdvp));
2060 	ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
2061 	NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
2062 	if (info.v3) {
2063 		ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
2064 		ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
2065 	}
2066 	m_freem(info.mrep);
2067 	info.mrep = NULL;
2068 nfsmout:
2069 	VTONFS(fdvp)->n_flag |= NLMODIFIED;
2070 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2071 	if (!fwccflag)
2072 		VTONFS(fdvp)->n_attrstamp = 0;
2073 	if (!twccflag)
2074 		VTONFS(tdvp)->n_attrstamp = 0;
2075 	return (error);
2076 }
2077 
2078 /*
2079  * nfs hard link create call
2080  *
2081  * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2082  *	    struct componentname *a_cnp)
2083  */
2084 static int
2085 nfs_link(struct vop_old_link_args *ap)
2086 {
2087 	struct vnode *vp = ap->a_vp;
2088 	struct vnode *tdvp = ap->a_tdvp;
2089 	struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2090 	struct componentname *cnp = ap->a_cnp;
2091 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2092 	struct nfsm_info info;
2093 
2094 	if (vp->v_mount != tdvp->v_mount) {
2095 		return (EXDEV);
2096 	}
2097 	lwkt_gettoken(&nmp->nm_token);
2098 
2099 	/*
2100 	 * The attribute cache may get out of sync with the server on link.
2101 	 * Pushing writes to the server before handle was inherited from
2102 	 * long long ago and it is unclear if we still need to do this.
2103 	 * Defaults to off.
2104 	 */
2105 	if (nfs_flush_on_hlink)
2106 		VOP_FSYNC(vp, MNT_WAIT, 0);
2107 
2108 	info.mrep = NULL;
2109 	info.v3 = NFS_ISV3(vp);
2110 
2111 	nfsstats.rpccnt[NFSPROC_LINK]++;
2112 	nfsm_reqhead(&info, vp, NFSPROC_LINK,
2113 		     NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2114 		     nfsm_rndup(cnp->cn_namelen));
2115 	ERROROUT(nfsm_fhtom(&info, vp));
2116 	ERROROUT(nfsm_fhtom(&info, tdvp));
2117 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2118 			     NFS_MAXNAMLEN));
2119 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2120 				cnp->cn_cred, &error));
2121 	if (info.v3) {
2122 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2123 					 NFS_LATTR_NOSHRINK));
2124 		ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2125 	}
2126 	m_freem(info.mrep);
2127 	info.mrep = NULL;
2128 nfsmout:
2129 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2130 	if (!attrflag)
2131 		VTONFS(vp)->n_attrstamp = 0;
2132 	if (!wccflag)
2133 		VTONFS(tdvp)->n_attrstamp = 0;
2134 	/*
2135 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2136 	 */
2137 	if (error == EEXIST)
2138 		error = 0;
2139 	lwkt_reltoken(&nmp->nm_token);
2140 	if (error == 0) {
2141 		nfs_knote(vp, NOTE_LINK);
2142 		nfs_knote(tdvp, NOTE_WRITE);
2143 	}
2144 
2145 	return (error);
2146 }
2147 
2148 /*
2149  * nfs symbolic link create call
2150  *
2151  * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2152  *		struct componentname *a_cnp, struct vattr *a_vap,
2153  *		char *a_target)
2154  */
2155 static int
2156 nfs_symlink(struct vop_old_symlink_args *ap)
2157 {
2158 	struct vnode *dvp = ap->a_dvp;
2159 	struct vattr *vap = ap->a_vap;
2160 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2161 	struct componentname *cnp = ap->a_cnp;
2162 	struct nfsv2_sattr *sp;
2163 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2164 	struct vnode *newvp = NULL;
2165 	struct nfsm_info info;
2166 
2167 	info.mrep = NULL;
2168 	info.v3 = NFS_ISV3(dvp);
2169 	lwkt_gettoken(&nmp->nm_token);
2170 
2171 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2172 	slen = strlen(ap->a_target);
2173 	nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2174 		     NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2175 		     nfsm_rndup(cnp->cn_namelen) +
2176 		     nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2177 	ERROROUT(nfsm_fhtom(&info, dvp));
2178 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2179 			     NFS_MAXNAMLEN));
2180 	if (info.v3) {
2181 		nfsm_v3attrbuild(&info, vap, FALSE);
2182 	}
2183 	ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2184 	if (info.v3 == 0) {
2185 		sp = nfsm_build(&info, NFSX_V2SATTR);
2186 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2187 		sp->sa_uid = nfs_xdrneg1;
2188 		sp->sa_gid = nfs_xdrneg1;
2189 		sp->sa_size = nfs_xdrneg1;
2190 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2191 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2192 	}
2193 
2194 	/*
2195 	 * Issue the NFS request and get the rpc response.
2196 	 *
2197 	 * Only NFSv3 responses returning an error of 0 actually return
2198 	 * a file handle that can be converted into newvp without having
2199 	 * to do an extra lookup rpc.
2200 	 */
2201 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2202 				cnp->cn_cred, &error));
2203 	if (info.v3) {
2204 		if (error == 0) {
2205 		       ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2206 		}
2207 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2208 	}
2209 
2210 	/*
2211 	 * out code jumps -> here, mrep is also freed.
2212 	 */
2213 
2214 	m_freem(info.mrep);
2215 	info.mrep = NULL;
2216 nfsmout:
2217 
2218 	/*
2219 	 * If we get an EEXIST error, silently convert it to no-error
2220 	 * in case of an NFS retry.
2221 	 */
2222 	if (error == EEXIST)
2223 		error = 0;
2224 
2225 	/*
2226 	 * If we do not have (or no longer have) an error, and we could
2227 	 * not extract the newvp from the response due to the request being
2228 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
2229 	 * to obtain a newvp to return.
2230 	 */
2231 	if (error == 0 && newvp == NULL) {
2232 		struct nfsnode *np = NULL;
2233 
2234 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2235 				     cnp->cn_cred, cnp->cn_td, &np);
2236 		if (!error)
2237 			newvp = NFSTOV(np);
2238 	}
2239 	if (error) {
2240 		if (newvp)
2241 			vput(newvp);
2242 	} else {
2243 		*ap->a_vpp = newvp;
2244 	}
2245 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2246 	if (!wccflag)
2247 		VTONFS(dvp)->n_attrstamp = 0;
2248 	if (error == 0 && *ap->a_vpp)
2249 		nfs_knote(*ap->a_vpp, NOTE_WRITE);
2250 	lwkt_reltoken(&nmp->nm_token);
2251 
2252 	return (error);
2253 }
2254 
2255 /*
2256  * nfs make dir call
2257  *
2258  * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2259  *	     struct componentname *a_cnp, struct vattr *a_vap)
2260  */
2261 static int
2262 nfs_mkdir(struct vop_old_mkdir_args *ap)
2263 {
2264 	struct vnode *dvp = ap->a_dvp;
2265 	struct vattr *vap = ap->a_vap;
2266 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2267 	struct componentname *cnp = ap->a_cnp;
2268 	struct nfsv2_sattr *sp;
2269 	struct nfsnode *np = NULL;
2270 	struct vnode *newvp = NULL;
2271 	struct vattr vattr;
2272 	int error = 0, wccflag = NFSV3_WCCRATTR;
2273 	int gotvp = 0;
2274 	int len;
2275 	struct nfsm_info info;
2276 
2277 	info.mrep = NULL;
2278 	info.v3 = NFS_ISV3(dvp);
2279 	lwkt_gettoken(&nmp->nm_token);
2280 
2281 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2282 		lwkt_reltoken(&nmp->nm_token);
2283 		return (error);
2284 	}
2285 	len = cnp->cn_namelen;
2286 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2287 	nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2288 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2289 		     nfsm_rndup(len) + NFSX_SATTR(info.v3));
2290 	ERROROUT(nfsm_fhtom(&info, dvp));
2291 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2292 	if (info.v3) {
2293 		nfsm_v3attrbuild(&info, vap, FALSE);
2294 	} else {
2295 		sp = nfsm_build(&info, NFSX_V2SATTR);
2296 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2297 		sp->sa_uid = nfs_xdrneg1;
2298 		sp->sa_gid = nfs_xdrneg1;
2299 		sp->sa_size = nfs_xdrneg1;
2300 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2301 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2302 	}
2303 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2304 		    cnp->cn_cred, &error));
2305 	if (error == 0) {
2306 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2307 	}
2308 	if (info.v3) {
2309 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2310 	}
2311 	m_freem(info.mrep);
2312 	info.mrep = NULL;
2313 nfsmout:
2314 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2315 	if (!wccflag)
2316 		VTONFS(dvp)->n_attrstamp = 0;
2317 	/*
2318 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2319 	 * if we can succeed in looking up the directory.
2320 	 */
2321 	if (error == EEXIST || (!error && !gotvp)) {
2322 		if (newvp) {
2323 			vrele(newvp);
2324 			newvp = NULL;
2325 		}
2326 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2327 			cnp->cn_td, &np);
2328 		if (!error) {
2329 			newvp = NFSTOV(np);
2330 			if (newvp->v_type != VDIR)
2331 				error = EEXIST;
2332 		}
2333 	}
2334 	if (error) {
2335 		if (newvp)
2336 			vrele(newvp);
2337 	} else {
2338 		nfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
2339 		*ap->a_vpp = newvp;
2340 	}
2341 	lwkt_reltoken(&nmp->nm_token);
2342 	return (error);
2343 }
2344 
2345 /*
2346  * nfs remove directory call
2347  *
2348  * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2349  *	     struct componentname *a_cnp)
2350  */
2351 static int
2352 nfs_rmdir(struct vop_old_rmdir_args *ap)
2353 {
2354 	struct vnode *vp = ap->a_vp;
2355 	struct vnode *dvp = ap->a_dvp;
2356 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2357 	struct componentname *cnp = ap->a_cnp;
2358 	int error = 0, wccflag = NFSV3_WCCRATTR;
2359 	struct nfsm_info info;
2360 
2361 	info.mrep = NULL;
2362 	info.v3 = NFS_ISV3(dvp);
2363 
2364 	if (dvp == vp)
2365 		return (EINVAL);
2366 
2367 	lwkt_gettoken(&nmp->nm_token);
2368 
2369 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2370 	nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2371 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2372 		     nfsm_rndup(cnp->cn_namelen));
2373 	ERROROUT(nfsm_fhtom(&info, dvp));
2374 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2375 		 NFS_MAXNAMLEN));
2376 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2377 				cnp->cn_cred, &error));
2378 	if (info.v3) {
2379 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2380 	}
2381 	m_freem(info.mrep);
2382 	info.mrep = NULL;
2383 nfsmout:
2384 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2385 	if (!wccflag)
2386 		VTONFS(dvp)->n_attrstamp = 0;
2387 	/*
2388 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2389 	 */
2390 	if (error == ENOENT)
2391 		error = 0;
2392 	else
2393 		nfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
2394 	lwkt_reltoken(&nmp->nm_token);
2395 
2396 	return (error);
2397 }
2398 
2399 /*
2400  * nfs readdir call
2401  *
2402  * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2403  */
2404 static int
2405 nfs_readdir(struct vop_readdir_args *ap)
2406 {
2407 	struct vnode *vp = ap->a_vp;
2408 	struct nfsnode *np = VTONFS(vp);
2409 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2410 	struct uio *uio = ap->a_uio;
2411 	int tresid, error;
2412 	struct vattr vattr;
2413 
2414 	if (vp->v_type != VDIR)
2415 		return (EPERM);
2416 
2417 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2418 	if (error)
2419 		return (error);
2420 
2421 	lwkt_gettoken(&nmp->nm_token);
2422 
2423 	/*
2424 	 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2425 	 * and then check that is still valid, or if this is an NQNFS mount
2426 	 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR().  Note that
2427 	 * VOP_GETATTR() does not necessarily go to the wire.
2428 	 */
2429 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2430 	    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2431 		if (VOP_GETATTR(vp, &vattr) == 0 &&
2432 		    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2433 		) {
2434 			nfsstats.direofcache_hits++;
2435 			goto done;
2436 		}
2437 	}
2438 
2439 	/*
2440 	 * Call nfs_bioread() to do the real work.  nfs_bioread() does its
2441 	 * own cache coherency checks so we do not have to.
2442 	 */
2443 	tresid = uio->uio_resid;
2444 	error = nfs_bioread(vp, uio, 0);
2445 
2446 	if (!error && uio->uio_resid == tresid)
2447 		nfsstats.direofcache_misses++;
2448 done:
2449 	lwkt_reltoken(&nmp->nm_token);
2450 	vn_unlock(vp);
2451 
2452 	return (error);
2453 }
2454 
2455 /*
2456  * Readdir rpc call.  nfs_bioread->nfs_doio->nfs_readdirrpc.
2457  *
2458  * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2459  * offset/block and converts the nfs formatted directory entries for userland
2460  * consumption as well as deals with offsets into the middle of blocks.
2461  * nfs_doio only deals with logical blocks.  In particular, uio_offset will
2462  * be block-bounded.  It must convert to cookies for the actual RPC.
2463  */
2464 int
2465 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2466 {
2467 	int len, left;
2468 	struct nfs_dirent *dp = NULL;
2469 	u_int32_t *tl;
2470 	nfsuint64 *cookiep;
2471 	caddr_t cp;
2472 	nfsuint64 cookie;
2473 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2474 	struct nfsnode *dnp = VTONFS(vp);
2475 	u_quad_t fileno;
2476 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2477 	int attrflag;
2478 	struct nfsm_info info;
2479 
2480 	info.mrep = NULL;
2481 	info.v3 = NFS_ISV3(vp);
2482 
2483 #ifndef DIAGNOSTIC
2484 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2485 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2486 		panic("nfs readdirrpc bad uio");
2487 #endif
2488 
2489 	/*
2490 	 * If there is no cookie, assume directory was stale.
2491 	 */
2492 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2493 	if (cookiep)
2494 		cookie = *cookiep;
2495 	else
2496 		return (NFSERR_BAD_COOKIE);
2497 	/*
2498 	 * Loop around doing readdir rpc's of size nm_readdirsize
2499 	 * truncated to a multiple of DIRBLKSIZ.
2500 	 * The stopping criteria is EOF or buffer full.
2501 	 */
2502 	while (more_dirs && bigenough) {
2503 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2504 		nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2505 			     NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2506 		ERROROUT(nfsm_fhtom(&info, vp));
2507 		if (info.v3) {
2508 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2509 			*tl++ = cookie.nfsuquad[0];
2510 			*tl++ = cookie.nfsuquad[1];
2511 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2512 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2513 		} else {
2514 			/*
2515 			 * WARNING!  HAMMER DIRECTORIES WILL NOT WORK WELL
2516 			 * WITH NFSv2!!!  There's nothing I can really do
2517 			 * about it other than to hope the server supports
2518 			 * rdirplus w/NFSv2.
2519 			 */
2520 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2521 			*tl++ = cookie.nfsuquad[0];
2522 		}
2523 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2524 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2525 					uiop->uio_td,
2526 					nfs_vpcred(vp, ND_READ), &error));
2527 		if (info.v3) {
2528 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2529 						  NFS_LATTR_NOSHRINK));
2530 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2531 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2532 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2533 		}
2534 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2535 		more_dirs = fxdr_unsigned(int, *tl);
2536 
2537 		/* loop thru the dir entries, converting them to std form */
2538 		while (more_dirs && bigenough) {
2539 			if (info.v3) {
2540 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2541 				fileno = fxdr_hyper(tl);
2542 				len = fxdr_unsigned(int, *(tl + 2));
2543 			} else {
2544 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2545 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2546 				len = fxdr_unsigned(int, *tl);
2547 			}
2548 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2549 				error = EBADRPC;
2550 				m_freem(info.mrep);
2551 				info.mrep = NULL;
2552 				goto nfsmout;
2553 			}
2554 
2555 			/*
2556 			 * len is the number of bytes in the path element
2557 			 * name, not including the \0 termination.
2558 			 *
2559 			 * tlen is the number of bytes w have to reserve for
2560 			 * the path element name.
2561 			 */
2562 			tlen = nfsm_rndup(len);
2563 			if (tlen == len)
2564 				tlen += 4;	/* To ensure null termination */
2565 
2566 			/*
2567 			 * If the entry would cross a DIRBLKSIZ boundary,
2568 			 * extend the previous nfs_dirent to cover the
2569 			 * remaining space.
2570 			 */
2571 			left = DIRBLKSIZ - blksiz;
2572 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2573 				dp->nfs_reclen += left;
2574 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2575 				uiop->uio_iov->iov_len -= left;
2576 				uiop->uio_offset += left;
2577 				uiop->uio_resid -= left;
2578 				blksiz = 0;
2579 			}
2580 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2581 				bigenough = 0;
2582 			if (bigenough) {
2583 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2584 				dp->nfs_ino = fileno;
2585 				dp->nfs_namlen = len;
2586 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2587 				dp->nfs_type = DT_UNKNOWN;
2588 				blksiz += dp->nfs_reclen;
2589 				if (blksiz == DIRBLKSIZ)
2590 					blksiz = 0;
2591 				uiop->uio_offset += sizeof(struct nfs_dirent);
2592 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2593 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2594 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2595 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2596 
2597 				/*
2598 				 * The uiop has advanced by nfs_dirent + len
2599 				 * but really needs to advance by
2600 				 * nfs_dirent + tlen
2601 				 */
2602 				cp = uiop->uio_iov->iov_base;
2603 				tlen -= len;
2604 				*cp = '\0';	/* null terminate */
2605 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2606 				uiop->uio_iov->iov_len -= tlen;
2607 				uiop->uio_offset += tlen;
2608 				uiop->uio_resid -= tlen;
2609 			} else {
2610 				/*
2611 				 * NFS strings must be rounded up (nfsm_myouio
2612 				 * handled that in the bigenough case).
2613 				 */
2614 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2615 			}
2616 			if (info.v3) {
2617 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2618 			} else {
2619 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2620 			}
2621 
2622 			/*
2623 			 * If we were able to accomodate the last entry,
2624 			 * get the cookie for the next one.  Otherwise
2625 			 * hold-over the cookie for the one we were not
2626 			 * able to accomodate.
2627 			 */
2628 			if (bigenough) {
2629 				cookie.nfsuquad[0] = *tl++;
2630 				if (info.v3)
2631 					cookie.nfsuquad[1] = *tl++;
2632 			} else if (info.v3) {
2633 				tl += 2;
2634 			} else {
2635 				tl++;
2636 			}
2637 			more_dirs = fxdr_unsigned(int, *tl);
2638 		}
2639 		/*
2640 		 * If at end of rpc data, get the eof boolean
2641 		 */
2642 		if (!more_dirs) {
2643 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2644 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2645 		}
2646 		m_freem(info.mrep);
2647 		info.mrep = NULL;
2648 	}
2649 	/*
2650 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2651 	 * by increasing d_reclen for the last record.
2652 	 */
2653 	if (blksiz > 0) {
2654 		left = DIRBLKSIZ - blksiz;
2655 		dp->nfs_reclen += left;
2656 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2657 		uiop->uio_iov->iov_len -= left;
2658 		uiop->uio_offset += left;
2659 		uiop->uio_resid -= left;
2660 	}
2661 
2662 	if (bigenough) {
2663 		/*
2664 		 * We hit the end of the directory, update direofoffset.
2665 		 */
2666 		dnp->n_direofoffset = uiop->uio_offset;
2667 	} else {
2668 		/*
2669 		 * There is more to go, insert the link cookie so the
2670 		 * next block can be read.
2671 		 */
2672 		if (uiop->uio_resid > 0)
2673 			kprintf("EEK! readdirrpc resid > 0\n");
2674 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2675 		*cookiep = cookie;
2676 	}
2677 nfsmout:
2678 	return (error);
2679 }
2680 
2681 /*
2682  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2683  */
2684 int
2685 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2686 {
2687 	int len, left;
2688 	struct nfs_dirent *dp;
2689 	u_int32_t *tl;
2690 	struct vnode *newvp;
2691 	nfsuint64 *cookiep;
2692 	caddr_t dpossav1, dpossav2;
2693 	caddr_t cp;
2694 	struct mbuf *mdsav1, *mdsav2;
2695 	nfsuint64 cookie;
2696 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2697 	struct nfsnode *dnp = VTONFS(vp), *np;
2698 	nfsfh_t *fhp;
2699 	u_quad_t fileno;
2700 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2701 	int attrflag, fhsize;
2702 	struct nchandle nch;
2703 	struct nchandle dnch;
2704 	struct nlcomponent nlc;
2705 	struct nfsm_info info;
2706 
2707 	info.mrep = NULL;
2708 	info.v3 = 1;
2709 
2710 #ifndef nolint
2711 	dp = NULL;
2712 #endif
2713 #ifndef DIAGNOSTIC
2714 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2715 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2716 		panic("nfs readdirplusrpc bad uio");
2717 #endif
2718 	/*
2719 	 * Obtain the namecache record for the directory so we have something
2720 	 * to use as a basis for creating the entries.  This function will
2721 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2722 	 * from the tree and cannot be used for upward traversals, and the
2723 	 * ncp may be unnamed.  Note that other unrelated operations may
2724 	 * cause the ncp to be named at any time.
2725 	 *
2726 	 * We have to lock the ncp to prevent a lock order reversal when
2727 	 * rdirplus does nlookups of the children, because the vnode is
2728 	 * locked and has to stay that way.
2729 	 */
2730 	cache_fromdvp(vp, NULL, 0, &dnch);
2731 	bzero(&nlc, sizeof(nlc));
2732 	newvp = NULLVP;
2733 
2734 	/*
2735 	 * If there is no cookie, assume directory was stale.
2736 	 */
2737 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2738 	if (cookiep) {
2739 		cookie = *cookiep;
2740 	} else {
2741 		if (dnch.ncp)
2742 			cache_drop(&dnch);
2743 		return (NFSERR_BAD_COOKIE);
2744 	}
2745 
2746 	/*
2747 	 * Loop around doing readdir rpc's of size nm_readdirsize
2748 	 * truncated to a multiple of DIRBLKSIZ.
2749 	 * The stopping criteria is EOF or buffer full.
2750 	 */
2751 	while (more_dirs && bigenough) {
2752 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2753 		nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2754 			     NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2755 		ERROROUT(nfsm_fhtom(&info, vp));
2756 		tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2757 		*tl++ = cookie.nfsuquad[0];
2758 		*tl++ = cookie.nfsuquad[1];
2759 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2760 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2761 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2762 		*tl = txdr_unsigned(nmp->nm_rsize);
2763 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2764 					uiop->uio_td,
2765 					nfs_vpcred(vp, ND_READ), &error));
2766 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2767 					  NFS_LATTR_NOSHRINK));
2768 		NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2769 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2770 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2771 		more_dirs = fxdr_unsigned(int, *tl);
2772 
2773 		/* loop thru the dir entries, doctoring them to 4bsd form */
2774 		while (more_dirs && bigenough) {
2775 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2776 			fileno = fxdr_hyper(tl);
2777 			len = fxdr_unsigned(int, *(tl + 2));
2778 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2779 				error = EBADRPC;
2780 				m_freem(info.mrep);
2781 				info.mrep = NULL;
2782 				goto nfsmout;
2783 			}
2784 			tlen = nfsm_rndup(len);
2785 			if (tlen == len)
2786 				tlen += 4;	/* To ensure null termination*/
2787 			left = DIRBLKSIZ - blksiz;
2788 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2789 				dp->nfs_reclen += left;
2790 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2791 				uiop->uio_iov->iov_len -= left;
2792 				uiop->uio_offset += left;
2793 				uiop->uio_resid -= left;
2794 				blksiz = 0;
2795 			}
2796 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2797 				bigenough = 0;
2798 			if (bigenough) {
2799 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2800 				dp->nfs_ino = fileno;
2801 				dp->nfs_namlen = len;
2802 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2803 				dp->nfs_type = DT_UNKNOWN;
2804 				blksiz += dp->nfs_reclen;
2805 				if (blksiz == DIRBLKSIZ)
2806 					blksiz = 0;
2807 				uiop->uio_offset += sizeof(struct nfs_dirent);
2808 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2809 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2810 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2811 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2812 				nlc.nlc_namelen = len;
2813 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2814 				cp = uiop->uio_iov->iov_base;
2815 				tlen -= len;
2816 				*cp = '\0';
2817 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2818 				uiop->uio_iov->iov_len -= tlen;
2819 				uiop->uio_offset += tlen;
2820 				uiop->uio_resid -= tlen;
2821 			} else {
2822 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2823 			}
2824 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2825 			if (bigenough) {
2826 				cookie.nfsuquad[0] = *tl++;
2827 				cookie.nfsuquad[1] = *tl++;
2828 			} else {
2829 				tl += 2;
2830 			}
2831 
2832 			/*
2833 			 * Since the attributes are before the file handle
2834 			 * (sigh), we must skip over the attributes and then
2835 			 * come back and get them.
2836 			 */
2837 			attrflag = fxdr_unsigned(int, *tl);
2838 			if (attrflag) {
2839 			    dpossav1 = info.dpos;
2840 			    mdsav1 = info.md;
2841 			    ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2842 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2843 			    doit = fxdr_unsigned(int, *tl);
2844 			    if (doit) {
2845 				NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2846 			    }
2847 			    if (doit && bigenough && !nlcdegenerate(&nlc) &&
2848 				!NFS_CMPFH(dnp, fhp, fhsize)
2849 			    ) {
2850 				if (dnch.ncp) {
2851 #if 0
2852 				    kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2853 					nlc.nlc_namelen, nlc.nlc_namelen,
2854 					nlc.nlc_nameptr);
2855 #endif
2856 				    /*
2857 				     * This is a bit hokey but there isn't
2858 				     * much we can do about it.  We can't
2859 				     * hold the directory vp locked while
2860 				     * doing lookups and gets.
2861 				     */
2862 				    nch = cache_nlookup_nonblock(&dnch, &nlc);
2863 				    if (nch.ncp == NULL)
2864 					goto rdfail;
2865 				    cache_setunresolved(&nch);
2866 				    error = nfs_nget_nonblock(vp->v_mount, fhp,
2867 							      fhsize, &np,
2868 							      NULL);
2869 				    if (error) {
2870 					cache_put(&nch);
2871 					goto rdfail;
2872 				    }
2873 				    newvp = NFSTOV(np);
2874 				    dpossav2 = info.dpos;
2875 				    info.dpos = dpossav1;
2876 				    mdsav2 = info.md;
2877 				    info.md = mdsav1;
2878 				    ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2879 				    info.dpos = dpossav2;
2880 				    info.md = mdsav2;
2881 				    dp->nfs_type =
2882 					    IFTODT(VTTOIF(np->n_vattr.va_type));
2883 				    nfs_cache_setvp(&nch, newvp,
2884 						    nfspos_cache_timeout);
2885 				    vput(newvp);
2886 				    newvp = NULLVP;
2887 				    cache_put(&nch);
2888 				} else {
2889 rdfail:
2890 				    ;
2891 #if 0
2892 				    kprintf("Warning: NFS/rddirplus, "
2893 					    "UNABLE TO ENTER %*.*s\n",
2894 					nlc.nlc_namelen, nlc.nlc_namelen,
2895 					nlc.nlc_nameptr);
2896 #endif
2897 				}
2898 			    }
2899 			} else {
2900 			    /* Just skip over the file handle */
2901 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2902 			    i = fxdr_unsigned(int, *tl);
2903 			    ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2904 			}
2905 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2906 			more_dirs = fxdr_unsigned(int, *tl);
2907 		}
2908 		/*
2909 		 * If at end of rpc data, get the eof boolean
2910 		 */
2911 		if (!more_dirs) {
2912 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2913 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2914 		}
2915 		m_freem(info.mrep);
2916 		info.mrep = NULL;
2917 	}
2918 	/*
2919 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2920 	 * by increasing d_reclen for the last record.
2921 	 */
2922 	if (blksiz > 0) {
2923 		left = DIRBLKSIZ - blksiz;
2924 		dp->nfs_reclen += left;
2925 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2926 		uiop->uio_iov->iov_len -= left;
2927 		uiop->uio_offset += left;
2928 		uiop->uio_resid -= left;
2929 	}
2930 
2931 	/*
2932 	 * We are now either at the end of the directory or have filled the
2933 	 * block.
2934 	 */
2935 	if (bigenough) {
2936 		dnp->n_direofoffset = uiop->uio_offset;
2937 	} else {
2938 		if (uiop->uio_resid > 0)
2939 			kprintf("EEK! readdirplusrpc resid > 0\n");
2940 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2941 		*cookiep = cookie;
2942 	}
2943 nfsmout:
2944 	if (newvp != NULLVP) {
2945 	        if (newvp == vp)
2946 			vrele(newvp);
2947 		else
2948 			vput(newvp);
2949 		newvp = NULLVP;
2950 	}
2951 	if (dnch.ncp)
2952 		cache_drop(&dnch);
2953 	return (error);
2954 }
2955 
2956 /*
2957  * Silly rename. To make the NFS filesystem that is stateless look a little
2958  * more like the "ufs" a remove of an active vnode is translated to a rename
2959  * to a funny looking filename that is removed by nfs_inactive on the
2960  * nfsnode. There is the potential for another process on a different client
2961  * to create the same funny name between the nfs_lookitup() fails and the
2962  * nfs_rename() completes, but...
2963  */
2964 static int
2965 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2966 {
2967 	struct sillyrename *sp;
2968 	struct nfsnode *np;
2969 	int error;
2970 
2971 	/*
2972 	 * Force finalization so the VOP_INACTIVE() call is not delayed.
2973 	 * This prevents cred structures from building up in nfsnodes
2974 	 * for deleted files.
2975 	 */
2976 	atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
2977 	np = VTONFS(vp);
2978 	np->n_flag |= NREMOVED;
2979 
2980 	/*
2981 	 * We previously purged dvp instead of vp.  I don't know why, it
2982 	 * completely destroys performance.  We can't do it anyway with the
2983 	 * new VFS API since we would be breaking the namecache topology.
2984 	 */
2985 	cache_purge(vp);	/* XXX */
2986 #ifndef DIAGNOSTIC
2987 	if (vp->v_type == VDIR)
2988 		panic("nfs: sillyrename dir");
2989 #endif
2990 	sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
2991 	sp->s_cred = crdup(cnp->cn_cred);
2992 	sp->s_dvp = dvp;
2993 	vref(dvp);
2994 
2995 	/* Fudge together a funny name */
2996 	sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2997 				(int)(intptr_t)cnp->cn_td);
2998 
2999 	/* Try lookitups until we get one that isn't there */
3000 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3001 		cnp->cn_td, NULL) == 0) {
3002 		sp->s_name[4]++;
3003 		if (sp->s_name[4] > 'z') {
3004 			error = EINVAL;
3005 			goto bad;
3006 		}
3007 	}
3008 	error = nfs_renameit(dvp, cnp, sp);
3009 	if (error)
3010 		goto bad;
3011 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3012 		cnp->cn_td, &np);
3013 	np->n_sillyrename = sp;
3014 	return (0);
3015 bad:
3016 	vrele(sp->s_dvp);
3017 	crfree(sp->s_cred);
3018 	kfree((caddr_t)sp, M_NFSREQ);
3019 
3020 	return (error);
3021 }
3022 
3023 /*
3024  * Look up a file name and optionally either update the file handle or
3025  * allocate an nfsnode, depending on the value of npp.
3026  * npp == NULL	--> just do the lookup
3027  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3028  *			handled too
3029  * *npp != NULL --> update the file handle in the vnode
3030  */
3031 static int
3032 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
3033 	     struct thread *td, struct nfsnode **npp)
3034 {
3035 	struct vnode *newvp = NULL;
3036 	struct nfsnode *np, *dnp = VTONFS(dvp);
3037 	int error = 0, fhlen, attrflag;
3038 	nfsfh_t *nfhp;
3039 	struct nfsm_info info;
3040 
3041 	info.mrep = NULL;
3042 	info.v3 = NFS_ISV3(dvp);
3043 
3044 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3045 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
3046 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3047 	ERROROUT(nfsm_fhtom(&info, dvp));
3048 	ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
3049 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
3050 	if (npp && !error) {
3051 		NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
3052 		if (*npp) {
3053 		    np = *npp;
3054 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3055 			kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
3056 			np->n_fhp = &np->n_fh;
3057 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
3058 			np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
3059 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
3060 		    np->n_fhsize = fhlen;
3061 		    newvp = NFSTOV(np);
3062 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3063 		    vref(dvp);
3064 		    newvp = dvp;
3065 		} else {
3066 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, NULL);
3067 		    if (error) {
3068 			m_freem(info.mrep);
3069 			info.mrep = NULL;
3070 			return (error);
3071 		    }
3072 		    newvp = NFSTOV(np);
3073 		}
3074 		if (info.v3) {
3075 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
3076 						  NFS_LATTR_NOSHRINK));
3077 			if (!attrflag && *npp == NULL) {
3078 				m_freem(info.mrep);
3079 				info.mrep = NULL;
3080 				if (newvp == dvp)
3081 					vrele(newvp);
3082 				else
3083 					vput(newvp);
3084 				return (ENOENT);
3085 			}
3086 		} else {
3087 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3088 		}
3089 	}
3090 	m_freem(info.mrep);
3091 	info.mrep = NULL;
3092 nfsmout:
3093 	if (npp && *npp == NULL) {
3094 		if (error) {
3095 			if (newvp) {
3096 				if (newvp == dvp)
3097 					vrele(newvp);
3098 				else
3099 					vput(newvp);
3100 			}
3101 		} else
3102 			*npp = np;
3103 	}
3104 	return (error);
3105 }
3106 
3107 /*
3108  * Nfs Version 3 commit rpc
3109  *
3110  * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3111  * involved.
3112  */
3113 int
3114 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3115 {
3116 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3117 	int error = 0, wccflag = NFSV3_WCCRATTR;
3118 	struct nfsm_info info;
3119 	u_int32_t *tl;
3120 
3121 	info.mrep = NULL;
3122 	info.v3 = 1;
3123 
3124 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3125 		return (0);
3126 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3127 	nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3128 	ERROROUT(nfsm_fhtom(&info, vp));
3129 	tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3130 	txdr_hyper(offset, tl);
3131 	tl += 2;
3132 	*tl = txdr_unsigned(cnt);
3133 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3134 				nfs_vpcred(vp, ND_WRITE), &error));
3135 	ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3136 	if (!error) {
3137 		NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3138 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3139 			NFSX_V3WRITEVERF)) {
3140 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3141 				NFSX_V3WRITEVERF);
3142 			error = NFSERR_STALEWRITEVERF;
3143 		}
3144 	}
3145 	m_freem(info.mrep);
3146 	info.mrep = NULL;
3147 nfsmout:
3148 	return (error);
3149 }
3150 
3151 /*
3152  * Kludge City..
3153  * - make nfs_bmap() essentially a no-op that does no translation
3154  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3155  *   (Maybe I could use the process's page mapping, but I was concerned that
3156  *    Kernel Write might not be enabled and also figured copyout() would do
3157  *    a lot more work than bcopy() and also it currently happens in the
3158  *    context of the swapper process (2).
3159  *
3160  * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3161  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
3162  */
3163 static int
3164 nfs_bmap(struct vop_bmap_args *ap)
3165 {
3166 	/* no token lock required */
3167 	if (ap->a_doffsetp != NULL)
3168 		*ap->a_doffsetp = ap->a_loffset;
3169 	if (ap->a_runp != NULL)
3170 		*ap->a_runp = 0;
3171 	if (ap->a_runb != NULL)
3172 		*ap->a_runb = 0;
3173 	return (0);
3174 }
3175 
3176 /*
3177  * Strategy routine.
3178  */
3179 static int
3180 nfs_strategy(struct vop_strategy_args *ap)
3181 {
3182 	struct bio *bio = ap->a_bio;
3183 	struct bio *nbio;
3184 	struct buf *bp __debugvar = bio->bio_buf;
3185 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3186 	struct thread *td;
3187 	int error;
3188 
3189 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
3190 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
3191 	KASSERT(BUF_LOCKINUSE(bp),
3192 		("nfs_strategy: buffer %p not locked", bp));
3193 
3194 	if (bio->bio_flags & BIO_SYNC)
3195 		td = curthread;	/* XXX */
3196 	else
3197 		td = NULL;
3198 
3199 	lwkt_gettoken(&nmp->nm_token);
3200 
3201         /*
3202 	 * We probably don't need to push an nbio any more since no
3203 	 * block conversion is required due to the use of 64 bit byte
3204 	 * offsets, but do it anyway.
3205 	 *
3206 	 * NOTE: When NFS callers itself via this strategy routines and
3207 	 *	 sets up a synchronous I/O, it expects the I/O to run
3208 	 *	 synchronously (its bio_done routine just assumes it),
3209 	 *	 so for now we have to honor the bit.
3210          */
3211 	nbio = push_bio(bio);
3212 	nbio->bio_offset = bio->bio_offset;
3213 	nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3214 
3215 	/*
3216 	 * If the op is asynchronous and an i/o daemon is waiting
3217 	 * queue the request, wake it up and wait for completion
3218 	 * otherwise just do it ourselves.
3219 	 */
3220 	if (bio->bio_flags & BIO_SYNC) {
3221 		error = nfs_doio(ap->a_vp, nbio, td);
3222 	} else {
3223 		nfs_asyncio(ap->a_vp, nbio);
3224 		error = 0;
3225 	}
3226 	lwkt_reltoken(&nmp->nm_token);
3227 
3228 	return (error);
3229 }
3230 
3231 /*
3232  * Mmap a file
3233  *
3234  * NB Currently unsupported.
3235  *
3236  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3237  */
3238 /* ARGSUSED */
3239 static int
3240 nfs_mmap(struct vop_mmap_args *ap)
3241 {
3242 	/* no token lock required */
3243 	return (EINVAL);
3244 }
3245 
3246 /*
3247  * fsync vnode op. Just call nfs_flush() with commit == 1.
3248  *
3249  * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3250  */
3251 /* ARGSUSED */
3252 static int
3253 nfs_fsync(struct vop_fsync_args *ap)
3254 {
3255 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3256 	int error;
3257 
3258 	lwkt_gettoken(&nmp->nm_token);
3259 
3260 	/*
3261 	 * NOTE: Because attributes are set synchronously we currently
3262 	 *	 do not have to implement vsetisdirty()/vclrisdirty().
3263 	 */
3264 	error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3265 
3266 	lwkt_reltoken(&nmp->nm_token);
3267 
3268 	return error;
3269 }
3270 
3271 /*
3272  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
3273  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
3274  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
3275  * set the buffer contains data that has already been written to the server
3276  * and which now needs a commit RPC.
3277  *
3278  * If commit is 0 we only take one pass and only flush buffers containing new
3279  * dirty data.
3280  *
3281  * If commit is 1 we take two passes, issuing a commit RPC in the second
3282  * pass.
3283  *
3284  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3285  * to completely flush all pending data.
3286  *
3287  * Note that the RB_SCAN code properly handles the case where the
3288  * callback might block and directly or indirectly (another thread) cause
3289  * the RB tree to change.
3290  */
3291 
3292 #ifndef NFS_COMMITBVECSIZ
3293 #define NFS_COMMITBVECSIZ	16
3294 #endif
3295 
3296 struct nfs_flush_info {
3297 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3298 	struct thread *td;
3299 	struct vnode *vp;
3300 	int waitfor;
3301 	int slpflag;
3302 	int slptimeo;
3303 	int loops;
3304 	struct buf *bvary[NFS_COMMITBVECSIZ];
3305 	int bvsize;
3306 	off_t beg_off;
3307 	off_t end_off;
3308 };
3309 
3310 static int nfs_flush_bp(struct buf *bp, void *data);
3311 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3312 
3313 int
3314 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3315 {
3316 	struct nfsnode *np = VTONFS(vp);
3317 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3318 	struct nfs_flush_info info;
3319 	int error;
3320 
3321 	bzero(&info, sizeof(info));
3322 	info.td = td;
3323 	info.vp = vp;
3324 	info.waitfor = waitfor;
3325 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3326 	info.loops = 0;
3327 	lwkt_gettoken(&vp->v_token);
3328 
3329 	do {
3330 		/*
3331 		 * Flush mode
3332 		 */
3333 		info.mode = NFI_FLUSHNEW;
3334 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3335 				nfs_flush_bp, &info);
3336 
3337 		/*
3338 		 * Take a second pass if committing and no error occured.
3339 		 * Clean up any left over collection (whether an error
3340 		 * occurs or not).
3341 		 */
3342 		if (commit && error == 0) {
3343 			info.mode = NFI_COMMIT;
3344 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3345 					nfs_flush_bp, &info);
3346 			if (info.bvsize)
3347 				error = nfs_flush_docommit(&info, error);
3348 		}
3349 
3350 		/*
3351 		 * Wait for pending I/O to complete before checking whether
3352 		 * any further dirty buffers exist.
3353 		 */
3354 		while (waitfor == MNT_WAIT &&
3355 		       bio_track_active(&vp->v_track_write)) {
3356 			error = bio_track_wait(&vp->v_track_write,
3357 					       info.slpflag, info.slptimeo);
3358 			if (error) {
3359 				/*
3360 				 * We have to be able to break out if this
3361 				 * is an 'intr' mount.
3362 				 */
3363 				if (nfs_sigintr(nmp, NULL, td)) {
3364 					error = -EINTR;
3365 					break;
3366 				}
3367 
3368 				/*
3369 				 * Since we do not process pending signals,
3370 				 * once we get a PCATCH our tsleep() will no
3371 				 * longer sleep, switch to a fixed timeout
3372 				 * instead.
3373 				 */
3374 				if (info.slpflag == PCATCH) {
3375 					info.slpflag = 0;
3376 					info.slptimeo = 2 * hz;
3377 				}
3378 				error = 0;
3379 			}
3380 		}
3381 		++info.loops;
3382 		/*
3383 		 * Loop if we are flushing synchronous as well as committing,
3384 		 * and dirty buffers are still present.  Otherwise we might livelock.
3385 		 */
3386 	} while (waitfor == MNT_WAIT && commit &&
3387 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3388 
3389 	/*
3390 	 * The callbacks have to return a negative error to terminate the
3391 	 * RB scan.
3392 	 */
3393 	if (error < 0)
3394 		error = -error;
3395 
3396 	/*
3397 	 * Deal with any error collection
3398 	 */
3399 	if (np->n_flag & NWRITEERR) {
3400 		error = np->n_error;
3401 		np->n_flag &= ~NWRITEERR;
3402 	}
3403 	lwkt_reltoken(&vp->v_token);
3404 	return (error);
3405 }
3406 
3407 static
3408 int
3409 nfs_flush_bp(struct buf *bp, void *data)
3410 {
3411 	struct nfs_flush_info *info = data;
3412 	int lkflags;
3413 	int error;
3414 	off_t toff;
3415 
3416 	error = 0;
3417 	switch(info->mode) {
3418 	case NFI_FLUSHNEW:
3419 		error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3420 		if (error && info->loops && info->waitfor == MNT_WAIT) {
3421 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3422 			if (error) {
3423 				lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3424 				if (info->slpflag & PCATCH)
3425 					lkflags |= LK_PCATCH;
3426 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3427 						     info->slptimeo);
3428 			}
3429 		}
3430 
3431 		/*
3432 		 * Ignore locking errors
3433 		 */
3434 		if (error) {
3435 			error = 0;
3436 			break;
3437 		}
3438 
3439 		/*
3440 		 * The buffer may have changed out from under us, even if
3441 		 * we did not block (MPSAFE).  Check again now that it is
3442 		 * locked.
3443 		 */
3444 		if (bp->b_vp == info->vp &&
3445 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3446 			bremfree(bp);
3447 			bawrite(bp);
3448 		} else {
3449 			BUF_UNLOCK(bp);
3450 		}
3451 		break;
3452 	case NFI_COMMIT:
3453 		/*
3454 		 * Only process buffers in need of a commit which we can
3455 		 * immediately lock.  This may prevent a buffer from being
3456 		 * committed, but the normal flush loop will block on the
3457 		 * same buffer so we shouldn't get into an endless loop.
3458 		 */
3459 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3460 		    (B_DELWRI | B_NEEDCOMMIT)) {
3461 			break;
3462 		}
3463 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3464 			break;
3465 
3466 		/*
3467 		 * We must recheck after successfully locking the buffer.
3468 		 */
3469 		if (bp->b_vp != info->vp ||
3470 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3471 		    (B_DELWRI | B_NEEDCOMMIT)) {
3472 			BUF_UNLOCK(bp);
3473 			break;
3474 		}
3475 
3476 		/*
3477 		 * NOTE: storing the bp in the bvary[] basically sets
3478 		 * it up for a commit operation.
3479 		 *
3480 		 * We must call vfs_busy_pages() now so the commit operation
3481 		 * is interlocked with user modifications to memory mapped
3482 		 * pages.  The b_dirtyoff/b_dirtyend range is not correct
3483 		 * until after the pages have been busied.
3484 		 *
3485 		 * Note: to avoid loopback deadlocks, we do not
3486 		 * assign b_runningbufspace.
3487 		 */
3488 		bremfree(bp);
3489 		bp->b_cmd = BUF_CMD_WRITE;
3490 		vfs_busy_pages(bp->b_vp, bp);
3491 		info->bvary[info->bvsize] = bp;
3492 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3493 		if (info->bvsize == 0 || toff < info->beg_off)
3494 			info->beg_off = toff;
3495 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3496 		if (info->bvsize == 0 || toff > info->end_off)
3497 			info->end_off = toff;
3498 		++info->bvsize;
3499 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3500 			error = nfs_flush_docommit(info, 0);
3501 			KKASSERT(info->bvsize == 0);
3502 		}
3503 	}
3504 	return (error);
3505 }
3506 
3507 static
3508 int
3509 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3510 {
3511 	struct vnode *vp;
3512 	struct buf *bp;
3513 	off_t bytes;
3514 	int retv;
3515 	int i;
3516 
3517 	vp = info->vp;
3518 
3519 	if (info->bvsize > 0) {
3520 		/*
3521 		 * Commit data on the server, as required.  Note that
3522 		 * nfs_commit will use the vnode's cred for the commit.
3523 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3524 		 */
3525 		bytes = info->end_off - info->beg_off;
3526 		if (bytes > 0x40000000)
3527 			bytes = 0x40000000;
3528 		if (error) {
3529 			retv = -error;
3530 		} else {
3531 			retv = nfs_commitrpc_uio(vp, info->beg_off,
3532 						 (int)bytes, info->td);
3533 			if (retv == NFSERR_STALEWRITEVERF)
3534 				nfs_clearcommit(vp->v_mount);
3535 		}
3536 
3537 		/*
3538 		 * Now, either mark the blocks I/O done or mark the
3539 		 * blocks dirty, depending on whether the commit
3540 		 * succeeded.
3541 		 */
3542 		for (i = 0; i < info->bvsize; ++i) {
3543 			bp = info->bvary[i];
3544 			if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3545 				/*
3546 				 * Either an error or the original
3547 				 * vfs_busy_pages() cleared B_NEEDCOMMIT
3548 				 * due to finding new dirty VM pages in
3549 				 * the buffer.
3550 				 *
3551 				 * Leave B_DELWRI intact.
3552 				 */
3553 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3554 				vfs_unbusy_pages(bp);
3555 				bp->b_cmd = BUF_CMD_DONE;
3556 				bqrelse(bp);
3557 			} else {
3558 				/*
3559 				 * Success, remove B_DELWRI ( bundirty() ).
3560 				 *
3561 				 * b_dirtyoff/b_dirtyend seem to be NFS
3562 				 * specific.  We should probably move that
3563 				 * into bundirty(). XXX
3564 				 *
3565 				 * We are faking an I/O write, we have to
3566 				 * start the transaction in order to
3567 				 * immediately biodone() it.
3568 				 */
3569 				bundirty(bp);
3570 				bp->b_flags &= ~B_ERROR;
3571 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3572 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3573 				biodone(&bp->b_bio1);
3574 			}
3575 		}
3576 		info->bvsize = 0;
3577 	}
3578 	return (error);
3579 }
3580 
3581 /*
3582  * NFS advisory byte-level locks.
3583  * Currently unsupported.
3584  *
3585  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3586  *		int a_flags)
3587  */
3588 static int
3589 nfs_advlock(struct vop_advlock_args *ap)
3590 {
3591 	struct nfsnode *np = VTONFS(ap->a_vp);
3592 
3593 	/* no token lock currently required */
3594 	/*
3595 	 * The following kludge is to allow diskless support to work
3596 	 * until a real NFS lockd is implemented. Basically, just pretend
3597 	 * that this is a local lock.
3598 	 */
3599 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3600 }
3601 
3602 /*
3603  * Print out the contents of an nfsnode.
3604  *
3605  * nfs_print(struct vnode *a_vp)
3606  */
3607 static int
3608 nfs_print(struct vop_print_args *ap)
3609 {
3610 	struct vnode *vp = ap->a_vp;
3611 	struct nfsnode *np = VTONFS(vp);
3612 
3613 	kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3614 		(long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3615 	if (vp->v_type == VFIFO)
3616 		fifo_printinfo(vp);
3617 	kprintf("\n");
3618 	return (0);
3619 }
3620 
3621 /*
3622  * nfs special file access vnode op.
3623  *
3624  * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3625  */
3626 static int
3627 nfs_laccess(struct vop_access_args *ap)
3628 {
3629 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3630 	struct vattr vattr;
3631 	int error;
3632 
3633 	lwkt_gettoken(&nmp->nm_token);
3634 	error = VOP_GETATTR(ap->a_vp, &vattr);
3635 	if (error == 0) {
3636 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3637 					  vattr.va_mode, 0);
3638 	}
3639 	lwkt_reltoken(&nmp->nm_token);
3640 
3641 	return (error);
3642 }
3643 
3644 /*
3645  * Read wrapper for fifos.
3646  *
3647  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3648  *		struct ucred *a_cred)
3649  */
3650 static int
3651 nfsfifo_read(struct vop_read_args *ap)
3652 {
3653 	struct nfsnode *np = VTONFS(ap->a_vp);
3654 
3655 	/* no token access required */
3656 	/*
3657 	 * Set access flag.
3658 	 */
3659 	np->n_flag |= NACC;
3660 	getnanotime(&np->n_atim);
3661 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3662 }
3663 
3664 /*
3665  * Write wrapper for fifos.
3666  *
3667  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3668  *		 struct ucred *a_cred)
3669  */
3670 static int
3671 nfsfifo_write(struct vop_write_args *ap)
3672 {
3673 	struct nfsnode *np = VTONFS(ap->a_vp);
3674 
3675 	/* no token access required */
3676 	/*
3677 	 * Set update flag.
3678 	 */
3679 	np->n_flag |= NUPD;
3680 	getnanotime(&np->n_mtim);
3681 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3682 }
3683 
3684 /*
3685  * Close wrapper for fifos.
3686  *
3687  * Update the times on the nfsnode then do fifo close.
3688  *
3689  * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3690  */
3691 static int
3692 nfsfifo_close(struct vop_close_args *ap)
3693 {
3694 	struct vnode *vp = ap->a_vp;
3695 	struct nfsnode *np = VTONFS(vp);
3696 	struct vattr vattr;
3697 	struct timespec ts;
3698 
3699 	/* no token access required */
3700 
3701 	vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */
3702 	if (np->n_flag & (NACC | NUPD)) {
3703 		getnanotime(&ts);
3704 		if (np->n_flag & NACC)
3705 			np->n_atim = ts;
3706 		if (np->n_flag & NUPD)
3707 			np->n_mtim = ts;
3708 		np->n_flag |= NCHG;
3709 		if (VREFCNT(vp) == 1 &&
3710 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3711 			VATTR_NULL(&vattr);
3712 			if (np->n_flag & NACC)
3713 				vattr.va_atime = np->n_atim;
3714 			if (np->n_flag & NUPD)
3715 				vattr.va_mtime = np->n_mtim;
3716 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3717 		}
3718 	}
3719 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3720 }
3721 
3722 /************************************************************************
3723  *                          KQFILTER OPS                                *
3724  ************************************************************************/
3725 
3726 static void filt_nfsdetach(struct knote *kn);
3727 static int filt_nfsread(struct knote *kn, long hint);
3728 static int filt_nfswrite(struct knote *kn, long hint);
3729 static int filt_nfsvnode(struct knote *kn, long hint);
3730 
3731 static struct filterops nfsread_filtops =
3732 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
3733 	  NULL, filt_nfsdetach, filt_nfsread };
3734 static struct filterops nfswrite_filtops =
3735 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
3736 	  NULL, filt_nfsdetach, filt_nfswrite };
3737 static struct filterops nfsvnode_filtops =
3738 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
3739 	  NULL, filt_nfsdetach, filt_nfsvnode };
3740 
3741 static int
3742 nfs_kqfilter (struct vop_kqfilter_args *ap)
3743 {
3744 	struct vnode *vp = ap->a_vp;
3745 	struct knote *kn = ap->a_kn;
3746 
3747 	switch (kn->kn_filter) {
3748 	case EVFILT_READ:
3749 		kn->kn_fop = &nfsread_filtops;
3750 		break;
3751 	case EVFILT_WRITE:
3752 		kn->kn_fop = &nfswrite_filtops;
3753 		break;
3754 	case EVFILT_VNODE:
3755 		kn->kn_fop = &nfsvnode_filtops;
3756 		break;
3757 	default:
3758 		return (EOPNOTSUPP);
3759 	}
3760 
3761 	kn->kn_hook = (caddr_t)vp;
3762 
3763 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
3764 
3765 	return(0);
3766 }
3767 
3768 static void
3769 filt_nfsdetach(struct knote *kn)
3770 {
3771 	struct vnode *vp = (void *)kn->kn_hook;
3772 
3773 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
3774 }
3775 
3776 static int
3777 filt_nfsread(struct knote *kn, long hint)
3778 {
3779 	struct vnode *vp = (void *)kn->kn_hook;
3780 	struct nfsnode *node = VTONFS(vp);
3781 	off_t off;
3782 
3783 	if (hint == NOTE_REVOKE) {
3784 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
3785 		return(1);
3786 	}
3787 
3788 	/*
3789 	 * Interlock against MP races when performing this function. XXX
3790 	 */
3791 	/* TMPFS_NODE_LOCK_SH(node); */
3792 	off = node->n_size - kn->kn_fp->f_offset;
3793 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
3794 	if (kn->kn_sfflags & NOTE_OLDAPI) {
3795 		/* TMPFS_NODE_UNLOCK(node); */
3796 		return(1);
3797 	}
3798 	if (kn->kn_data == 0) {
3799 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
3800 	}
3801 	/* TMPFS_NODE_UNLOCK(node); */
3802 	return (kn->kn_data != 0);
3803 }
3804 
3805 static int
3806 filt_nfswrite(struct knote *kn, long hint)
3807 {
3808 	if (hint == NOTE_REVOKE)
3809 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
3810 	kn->kn_data = 0;
3811 	return (1);
3812 }
3813 
3814 static int
3815 filt_nfsvnode(struct knote *kn, long hint)
3816 {
3817 	if (kn->kn_sfflags & hint)
3818 		kn->kn_fflags |= hint;
3819 	if (hint == NOTE_REVOKE) {
3820 		kn->kn_flags |= (EV_EOF | EV_NODATA);
3821 		return (1);
3822 	}
3823 	return (kn->kn_fflags != 0);
3824 }
3825