xref: /dragonfly/sys/vfs/nfs/nfs_vnops.c (revision 8a0bcd56)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
37  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38  * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 dillon Exp $
39  */
40 
41 
42 /*
43  * vnode op calls for Sun NFS version 2 and 3
44  */
45 
46 #include "opt_inet.h"
47 
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/nlookup.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
64 #include <sys/stat.h>
65 #include <sys/sysctl.h>
66 #include <sys/conf.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
71 
72 #include <sys/buf2.h>
73 
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
76 
77 #undef DIRBLKSIZ
78 
79 #include "rpcv2.h"
80 #include "nfsproto.h"
81 #include "nfs.h"
82 #include "nfsmount.h"
83 #include "nfsnode.h"
84 #include "xdr_subs.h"
85 #include "nfsm_subs.h"
86 
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 
91 #include <sys/thread2.h>
92 
93 /* Defs */
94 #define	TRUE	1
95 #define	FALSE	0
96 
97 static int	nfsfifo_read (struct vop_read_args *);
98 static int	nfsfifo_write (struct vop_write_args *);
99 static int	nfsfifo_close (struct vop_close_args *);
100 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
101 static	int	nfs_lookup (struct vop_old_lookup_args *);
102 static	int	nfs_create (struct vop_old_create_args *);
103 static	int	nfs_mknod (struct vop_old_mknod_args *);
104 static	int	nfs_open (struct vop_open_args *);
105 static	int	nfs_close (struct vop_close_args *);
106 static	int	nfs_access (struct vop_access_args *);
107 static	int	nfs_getattr (struct vop_getattr_args *);
108 static	int	nfs_setattr (struct vop_setattr_args *);
109 static	int	nfs_read (struct vop_read_args *);
110 static	int	nfs_mmap (struct vop_mmap_args *);
111 static	int	nfs_fsync (struct vop_fsync_args *);
112 static	int	nfs_remove (struct vop_old_remove_args *);
113 static	int	nfs_link (struct vop_old_link_args *);
114 static	int	nfs_rename (struct vop_old_rename_args *);
115 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
116 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
117 static	int	nfs_symlink (struct vop_old_symlink_args *);
118 static	int	nfs_readdir (struct vop_readdir_args *);
119 static	int	nfs_bmap (struct vop_bmap_args *);
120 static	int	nfs_strategy (struct vop_strategy_args *);
121 static	int	nfs_lookitup (struct vnode *, const char *, int,
122 			struct ucred *, struct thread *, struct nfsnode **);
123 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
124 static int	nfs_laccess (struct vop_access_args *);
125 static int	nfs_readlink (struct vop_readlink_args *);
126 static int	nfs_print (struct vop_print_args *);
127 static int	nfs_advlock (struct vop_advlock_args *);
128 
129 static	int	nfs_nresolve (struct vop_nresolve_args *);
130 /*
131  * Global vfs data structures for nfs
132  */
133 struct vop_ops nfsv2_vnode_vops = {
134 	.vop_default =		vop_defaultop,
135 	.vop_access =		nfs_access,
136 	.vop_advlock =		nfs_advlock,
137 	.vop_bmap =		nfs_bmap,
138 	.vop_close =		nfs_close,
139 	.vop_old_create =	nfs_create,
140 	.vop_fsync =		nfs_fsync,
141 	.vop_getattr =		nfs_getattr,
142 	.vop_getpages =		vop_stdgetpages,
143 	.vop_putpages =		vop_stdputpages,
144 	.vop_inactive =		nfs_inactive,
145 	.vop_old_link =		nfs_link,
146 	.vop_old_lookup =	nfs_lookup,
147 	.vop_old_mkdir =	nfs_mkdir,
148 	.vop_old_mknod =	nfs_mknod,
149 	.vop_mmap =		nfs_mmap,
150 	.vop_open =		nfs_open,
151 	.vop_print =		nfs_print,
152 	.vop_read =		nfs_read,
153 	.vop_readdir =		nfs_readdir,
154 	.vop_readlink =		nfs_readlink,
155 	.vop_reclaim =		nfs_reclaim,
156 	.vop_old_remove =	nfs_remove,
157 	.vop_old_rename =	nfs_rename,
158 	.vop_old_rmdir =	nfs_rmdir,
159 	.vop_setattr =		nfs_setattr,
160 	.vop_strategy =		nfs_strategy,
161 	.vop_old_symlink =	nfs_symlink,
162 	.vop_write =		nfs_write,
163 	.vop_nresolve =		nfs_nresolve
164 };
165 
166 /*
167  * Special device vnode ops
168  */
169 struct vop_ops nfsv2_spec_vops = {
170 	.vop_default =		vop_defaultop,
171 	.vop_access =		nfs_laccess,
172 	.vop_close =		nfs_close,
173 	.vop_fsync =		nfs_fsync,
174 	.vop_getattr =		nfs_getattr,
175 	.vop_inactive =		nfs_inactive,
176 	.vop_print =		nfs_print,
177 	.vop_read =		vop_stdnoread,
178 	.vop_reclaim =		nfs_reclaim,
179 	.vop_setattr =		nfs_setattr,
180 	.vop_write =		vop_stdnowrite
181 };
182 
183 struct vop_ops nfsv2_fifo_vops = {
184 	.vop_default =		fifo_vnoperate,
185 	.vop_access =		nfs_laccess,
186 	.vop_close =		nfsfifo_close,
187 	.vop_fsync =		nfs_fsync,
188 	.vop_getattr =		nfs_getattr,
189 	.vop_inactive =		nfs_inactive,
190 	.vop_print =		nfs_print,
191 	.vop_read =		nfsfifo_read,
192 	.vop_reclaim =		nfs_reclaim,
193 	.vop_setattr =		nfs_setattr,
194 	.vop_write =		nfsfifo_write
195 };
196 
197 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
198 				  struct componentname *cnp,
199 				  struct vattr *vap);
200 static int	nfs_removerpc (struct vnode *dvp, const char *name,
201 				   int namelen,
202 				   struct ucred *cred, struct thread *td);
203 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
204 				   int fnamelen, struct vnode *tdvp,
205 				   const char *tnameptr, int tnamelen,
206 				   struct ucred *cred, struct thread *td);
207 static int	nfs_renameit (struct vnode *sdvp,
208 				  struct componentname *scnp,
209 				  struct sillyrename *sp);
210 
211 SYSCTL_DECL(_vfs_nfs);
212 
213 static int nfs_flush_on_rename = 1;
214 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
215 	   &nfs_flush_on_rename, 0, "flush fvp prior to rename");
216 static int nfs_flush_on_hlink = 0;
217 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
218 	   &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
219 
220 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
221 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
222 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
223 
224 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
226 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
227 
228 static int	nfspos_cache_timeout = NFS_MINATTRTIMO;
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
230 	   &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
231 
232 static int	nfsv3_commit_on_close = 0;
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
234 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
235 #if 0
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
237 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
238 
239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
240 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
241 #endif
242 
243 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
244 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
245 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
246 
247 /*
248  * Returns whether a name component is a degenerate '.' or '..'.
249  */
250 static __inline
251 int
252 nlcdegenerate(struct nlcomponent *nlc)
253 {
254 	if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
255 		return(1);
256 	if (nlc->nlc_namelen == 2 &&
257 	    nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
258 		return(1);
259 	return(0);
260 }
261 
262 static int
263 nfs3_access_otw(struct vnode *vp, int wmode,
264 		struct thread *td, struct ucred *cred)
265 {
266 	struct nfsnode *np = VTONFS(vp);
267 	int attrflag;
268 	int error = 0;
269 	u_int32_t *tl;
270 	u_int32_t rmode;
271 	struct nfsm_info info;
272 
273 	info.mrep = NULL;
274 	info.v3 = 1;
275 
276 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
277 	nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
278 		     NFSX_FH(info.v3) + NFSX_UNSIGNED);
279 	ERROROUT(nfsm_fhtom(&info, vp));
280 	tl = nfsm_build(&info, NFSX_UNSIGNED);
281 	*tl = txdr_unsigned(wmode);
282 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
283 	ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
284 	if (error == 0) {
285 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
286 		rmode = fxdr_unsigned(u_int32_t, *tl);
287 		np->n_mode = rmode;
288 		np->n_modeuid = cred->cr_uid;
289 		np->n_modestamp = mycpu->gd_time_seconds;
290 	}
291 	m_freem(info.mrep);
292 	info.mrep = NULL;
293 nfsmout:
294 	return error;
295 }
296 
297 /*
298  * nfs access vnode op.
299  * For nfs version 2, just return ok. File accesses may fail later.
300  * For nfs version 3, use the access rpc to check accessibility. If file modes
301  * are changed on the server, accesses might still fail later.
302  *
303  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
304  */
305 static int
306 nfs_access(struct vop_access_args *ap)
307 {
308 	struct ucred *cred;
309 	struct vnode *vp = ap->a_vp;
310 	thread_t td = curthread;
311 	int error = 0;
312 	u_int32_t mode, wmode;
313 	struct nfsnode *np = VTONFS(vp);
314 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
315 	int v3 = NFS_ISV3(vp);
316 
317 	lwkt_gettoken(&nmp->nm_token);
318 
319 	/*
320 	 * Disallow write attempts on filesystems mounted read-only;
321 	 * unless the file is a socket, fifo, or a block or character
322 	 * device resident on the filesystem.
323 	 */
324 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
325 		switch (vp->v_type) {
326 		case VREG:
327 		case VDIR:
328 		case VLNK:
329 			lwkt_reltoken(&nmp->nm_token);
330 			return (EROFS);
331 		default:
332 			break;
333 		}
334 	}
335 
336 	/*
337 	 * The NFS protocol passes only the effective uid/gid over the wire but
338 	 * we need to check access against real ids if AT_EACCESS not set.
339 	 * Handle this case by cloning the credentials and setting the
340 	 * effective ids to the real ones.
341 	 */
342 	if (ap->a_flags & AT_EACCESS) {
343 		cred = crhold(ap->a_cred);
344 	} else {
345 		cred = crdup(ap->a_cred);
346 		cred->cr_uid = cred->cr_ruid;
347 		cred->cr_gid = cred->cr_rgid;
348 	}
349 
350 	/*
351 	 * For nfs v3, check to see if we have done this recently, and if
352 	 * so return our cached result instead of making an ACCESS call.
353 	 * If not, do an access rpc, otherwise you are stuck emulating
354 	 * ufs_access() locally using the vattr. This may not be correct,
355 	 * since the server may apply other access criteria such as
356 	 * client uid-->server uid mapping that we do not know about.
357 	 */
358 	if (v3) {
359 		if (ap->a_mode & VREAD)
360 			mode = NFSV3ACCESS_READ;
361 		else
362 			mode = 0;
363 		if (vp->v_type != VDIR) {
364 			if (ap->a_mode & VWRITE)
365 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
366 			if (ap->a_mode & VEXEC)
367 				mode |= NFSV3ACCESS_EXECUTE;
368 		} else {
369 			if (ap->a_mode & VWRITE)
370 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
371 					 NFSV3ACCESS_DELETE);
372 			if (ap->a_mode & VEXEC)
373 				mode |= NFSV3ACCESS_LOOKUP;
374 		}
375 		/* XXX safety belt, only make blanket request if caching */
376 		if (nfsaccess_cache_timeout > 0) {
377 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
378 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
379 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
380 		} else {
381 			wmode = mode;
382 		}
383 
384 		/*
385 		 * Does our cached result allow us to give a definite yes to
386 		 * this request?
387 		 */
388 		if (np->n_modestamp &&
389 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
390 		   (cred->cr_uid == np->n_modeuid) &&
391 		   ((np->n_mode & mode) == mode)) {
392 			nfsstats.accesscache_hits++;
393 		} else {
394 			/*
395 			 * Either a no, or a don't know.  Go to the wire.
396 			 */
397 			nfsstats.accesscache_misses++;
398 		        error = nfs3_access_otw(vp, wmode, td, cred);
399 			if (!error) {
400 				if ((np->n_mode & mode) != mode) {
401 					error = EACCES;
402 				}
403 			}
404 		}
405 	} else {
406 		if ((error = nfs_laccess(ap)) != 0) {
407 			crfree(cred);
408 			lwkt_reltoken(&nmp->nm_token);
409 			return (error);
410 		}
411 
412 		/*
413 		 * Attempt to prevent a mapped root from accessing a file
414 		 * which it shouldn't.  We try to read a byte from the file
415 		 * if the user is root and the file is not zero length.
416 		 * After calling nfs_laccess, we should have the correct
417 		 * file size cached.
418 		 */
419 		if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
420 		    && VTONFS(vp)->n_size > 0) {
421 			struct iovec aiov;
422 			struct uio auio;
423 			char buf[1];
424 
425 			aiov.iov_base = buf;
426 			aiov.iov_len = 1;
427 			auio.uio_iov = &aiov;
428 			auio.uio_iovcnt = 1;
429 			auio.uio_offset = 0;
430 			auio.uio_resid = 1;
431 			auio.uio_segflg = UIO_SYSSPACE;
432 			auio.uio_rw = UIO_READ;
433 			auio.uio_td = td;
434 
435 			if (vp->v_type == VREG) {
436 				error = nfs_readrpc_uio(vp, &auio);
437 			} else if (vp->v_type == VDIR) {
438 				char* bp;
439 				bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
440 				aiov.iov_base = bp;
441 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
442 				error = nfs_readdirrpc_uio(vp, &auio);
443 				kfree(bp, M_TEMP);
444 			} else if (vp->v_type == VLNK) {
445 				error = nfs_readlinkrpc_uio(vp, &auio);
446 			} else {
447 				error = EACCES;
448 			}
449 		}
450 	}
451 	/*
452 	 * [re]record creds for reading and/or writing if access
453 	 * was granted.  Assume the NFS server will grant read access
454 	 * for execute requests.
455 	 */
456 	if (error == 0) {
457 		if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
458 			crhold(cred);
459 			if (np->n_rucred)
460 				crfree(np->n_rucred);
461 			np->n_rucred = cred;
462 		}
463 		if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
464 			crhold(cred);
465 			if (np->n_wucred)
466 				crfree(np->n_wucred);
467 			np->n_wucred = cred;
468 		}
469 	}
470 	lwkt_reltoken(&nmp->nm_token);
471 	crfree(cred);
472 	return(error);
473 }
474 
475 /*
476  * nfs open vnode op
477  * Check to see if the type is ok
478  * and that deletion is not in progress.
479  * For paged in text files, you will need to flush the page cache
480  * if consistency is lost.
481  *
482  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
483  *	    struct file *a_fp)
484  */
485 /* ARGSUSED */
486 static int
487 nfs_open(struct vop_open_args *ap)
488 {
489 	struct vnode *vp = ap->a_vp;
490 	struct nfsnode *np = VTONFS(vp);
491 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
492 	struct vattr vattr;
493 	int error;
494 
495 	lwkt_gettoken(&nmp->nm_token);
496 
497 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
498 #ifdef DIAGNOSTIC
499 		kprintf("open eacces vtyp=%d\n",vp->v_type);
500 #endif
501 		lwkt_reltoken(&nmp->nm_token);
502 		return (EOPNOTSUPP);
503 	}
504 
505 	/*
506 	 * Save valid creds for reading and writing for later RPCs.
507 	 */
508 	if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
509 		crhold(ap->a_cred);
510 		if (np->n_rucred)
511 			crfree(np->n_rucred);
512 		np->n_rucred = ap->a_cred;
513 	}
514 	if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
515 		crhold(ap->a_cred);
516 		if (np->n_wucred)
517 			crfree(np->n_wucred);
518 		np->n_wucred = ap->a_cred;
519 	}
520 
521 	/*
522 	 * Clear the attribute cache only if opening with write access.  It
523 	 * is unclear if we should do this at all here, but we certainly
524 	 * should not clear the cache unconditionally simply because a file
525 	 * is being opened.
526 	 */
527 	if (ap->a_mode & FWRITE)
528 		np->n_attrstamp = 0;
529 
530 	/*
531 	 * For normal NFS, reconcile changes made locally verses
532 	 * changes made remotely.  Note that VOP_GETATTR only goes
533 	 * to the wire if the cached attribute has timed out or been
534 	 * cleared.
535 	 *
536 	 * If local modifications have been made clear the attribute
537 	 * cache to force an attribute and modified time check.  If
538 	 * GETATTR detects that the file has been changed by someone
539 	 * other then us it will set NRMODIFIED.
540 	 *
541 	 * If we are opening a directory and local changes have been
542 	 * made we have to invalidate the cache in order to ensure
543 	 * that we get the most up-to-date information from the
544 	 * server.  XXX
545 	 */
546 	if (np->n_flag & NLMODIFIED) {
547 		np->n_attrstamp = 0;
548 		if (vp->v_type == VDIR) {
549 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
550 			if (error == EINTR)
551 				return (error);
552 			nfs_invaldir(vp);
553 		}
554 	}
555 	error = VOP_GETATTR(vp, &vattr);
556 	if (error) {
557 		lwkt_reltoken(&nmp->nm_token);
558 		return (error);
559 	}
560 	if (np->n_flag & NRMODIFIED) {
561 		if (vp->v_type == VDIR)
562 			nfs_invaldir(vp);
563 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
564 		if (error == EINTR) {
565 			lwkt_reltoken(&nmp->nm_token);
566 			return (error);
567 		}
568 		np->n_flag &= ~NRMODIFIED;
569 	}
570 	error = vop_stdopen(ap);
571 	lwkt_reltoken(&nmp->nm_token);
572 
573 	return error;
574 }
575 
576 /*
577  * nfs close vnode op
578  * What an NFS client should do upon close after writing is a debatable issue.
579  * Most NFS clients push delayed writes to the server upon close, basically for
580  * two reasons:
581  * 1 - So that any write errors may be reported back to the client process
582  *     doing the close system call. By far the two most likely errors are
583  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
584  * 2 - To put a worst case upper bound on cache inconsistency between
585  *     multiple clients for the file.
586  * There is also a consistency problem for Version 2 of the protocol w.r.t.
587  * not being able to tell if other clients are writing a file concurrently,
588  * since there is no way of knowing if the changed modify time in the reply
589  * is only due to the write for this client.
590  * (NFS Version 3 provides weak cache consistency data in the reply that
591  *  should be sufficient to detect and handle this case.)
592  *
593  * The current code does the following:
594  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
595  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
596  *                     or commit them (this satisfies 1 and 2 except for the
597  *                     case where the server crashes after this close but
598  *                     before the commit RPC, which is felt to be "good
599  *                     enough". Changing the last argument to nfs_flush() to
600  *                     a 1 would force a commit operation, if it is felt a
601  *                     commit is necessary now.
602  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
603  *                     1 should be dealt with via an fsync() system call for
604  *                     cases where write errors are important.
605  *
606  * nfs_close(struct vnode *a_vp, int a_fflag)
607  */
608 /* ARGSUSED */
609 static int
610 nfs_close(struct vop_close_args *ap)
611 {
612 	struct vnode *vp = ap->a_vp;
613 	struct nfsnode *np = VTONFS(vp);
614 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
615 	int error = 0;
616 	thread_t td = curthread;
617 
618 	lwkt_gettoken(&nmp->nm_token);
619 
620 	if (vp->v_type == VREG) {
621 	    if (np->n_flag & NLMODIFIED) {
622 		if (NFS_ISV3(vp)) {
623 		    /*
624 		     * Under NFSv3 we have dirty buffers to dispose of.  We
625 		     * must flush them to the NFS server.  We have the option
626 		     * of waiting all the way through the commit rpc or just
627 		     * waiting for the initial write.  The default is to only
628 		     * wait through the initial write so the data is in the
629 		     * server's cache, which is roughly similar to the state
630 		     * a standard disk subsystem leaves the file in on close().
631 		     *
632 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
633 		     * potential races with other processes, and certainly
634 		     * cannot clear it if we don't commit.
635 		     */
636 		    int cm = nfsv3_commit_on_close ? 1 : 0;
637 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
638 		    /* np->n_flag &= ~NLMODIFIED; */
639 		} else {
640 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
641 		}
642 		np->n_attrstamp = 0;
643 	    }
644 	    if (np->n_flag & NWRITEERR) {
645 		np->n_flag &= ~NWRITEERR;
646 		error = np->n_error;
647 	    }
648 	}
649 	vop_stdclose(ap);
650 	lwkt_reltoken(&nmp->nm_token);
651 
652 	return (error);
653 }
654 
655 /*
656  * nfs getattr call from vfs.
657  *
658  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
659  */
660 static int
661 nfs_getattr(struct vop_getattr_args *ap)
662 {
663 	struct vnode *vp = ap->a_vp;
664 	struct nfsnode *np = VTONFS(vp);
665 	struct nfsmount *nmp;
666 	int error = 0;
667 	thread_t td = curthread;
668 	struct nfsm_info info;
669 
670 	info.mrep = NULL;
671 	info.v3 = NFS_ISV3(vp);
672 	nmp = VFSTONFS(vp->v_mount);
673 
674 	lwkt_gettoken(&nmp->nm_token);
675 
676 	/*
677 	 * Update local times for special files.
678 	 */
679 	if (np->n_flag & (NACC | NUPD))
680 		np->n_flag |= NCHG;
681 	/*
682 	 * First look in the cache.
683 	 */
684 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
685 		goto done;
686 
687 	if (info.v3 && nfsaccess_cache_timeout > 0) {
688 		nfsstats.accesscache_misses++;
689 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
690 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
691 			goto done;
692 	}
693 
694 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
695 	nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
696 	ERROROUT(nfsm_fhtom(&info, vp));
697 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
698 				nfs_vpcred(vp, ND_CHECK), &error));
699 	if (error == 0) {
700 		ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
701 	}
702 	m_freem(info.mrep);
703 	info.mrep = NULL;
704 done:
705 	/*
706 	 * NFS doesn't support chflags flags.  If the nfs mount was
707 	 * made -o cache set the UF_CACHE bit for swapcache.
708 	 */
709 	if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
710 		ap->a_vap->va_flags |= UF_CACHE;
711 nfsmout:
712 	lwkt_reltoken(&nmp->nm_token);
713 	return (error);
714 }
715 
716 /*
717  * nfs setattr call.
718  *
719  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
720  */
721 static int
722 nfs_setattr(struct vop_setattr_args *ap)
723 {
724 	struct vnode *vp = ap->a_vp;
725 	struct nfsnode *np = VTONFS(vp);
726 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
727 	struct vattr *vap = ap->a_vap;
728 	int biosize = vp->v_mount->mnt_stat.f_iosize;
729 	int error = 0;
730 	int boff;
731 	off_t tsize;
732 	thread_t td = curthread;
733 
734 #ifndef nolint
735 	tsize = (off_t)0;
736 #endif
737 	/*
738 	 * Setting of flags is not supported.
739 	 */
740 	if (vap->va_flags != VNOVAL)
741 		return (EOPNOTSUPP);
742 
743 	/*
744 	 * Disallow write attempts if the filesystem is mounted read-only.
745 	 */
746   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
747 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
748 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
749 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
750 		return (EROFS);
751 
752 	lwkt_gettoken(&nmp->nm_token);
753 
754 	if (vap->va_size != VNOVAL) {
755 		/*
756 		 * truncation requested
757 		 */
758  		switch (vp->v_type) {
759  		case VDIR:
760 			lwkt_reltoken(&nmp->nm_token);
761  			return (EISDIR);
762  		case VCHR:
763  		case VBLK:
764  		case VSOCK:
765  		case VFIFO:
766 			if (vap->va_mtime.tv_sec == VNOVAL &&
767 			    vap->va_atime.tv_sec == VNOVAL &&
768 			    vap->va_mode == (mode_t)VNOVAL &&
769 			    vap->va_uid == (uid_t)VNOVAL &&
770 			    vap->va_gid == (gid_t)VNOVAL) {
771 				lwkt_reltoken(&nmp->nm_token);
772 				return (0);
773 			}
774  			vap->va_size = VNOVAL;
775  			break;
776  		default:
777 			/*
778 			 * Disallow write attempts if the filesystem is
779 			 * mounted read-only.
780 			 */
781 			if (vp->v_mount->mnt_flag & MNT_RDONLY) {
782 				lwkt_reltoken(&nmp->nm_token);
783 				return (EROFS);
784 			}
785 
786 			tsize = np->n_size;
787 again:
788 			boff = (int)vap->va_size & (biosize - 1);
789 			error = nfs_meta_setsize(vp, td, vap->va_size, 0);
790 
791 #if 0
792  			if (np->n_flag & NLMODIFIED) {
793  			    if (vap->va_size == 0)
794  				error = nfs_vinvalbuf(vp, 0, 1);
795  			    else
796  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
797  			}
798 #endif
799 			/*
800 			 * note: this loop case almost always happens at
801 			 * least once per truncation.
802 			 */
803 			if (error == 0 && np->n_size != vap->va_size)
804 				goto again;
805 			np->n_vattr.va_size = vap->va_size;
806 			break;
807 		}
808 	} else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
809 		/*
810 		 * What to do.  If we are modifying the mtime we lose
811 		 * mtime detection of changes made by the server or other
812 		 * clients.  But programs like rsync/rdist/cpdup are going
813 		 * to call utimes a lot.  We don't want to piecemeal sync.
814 		 *
815 		 * For now sync if any prior remote changes were detected,
816 		 * but allow us to lose track of remote changes made during
817 		 * the utimes operation.
818 		 */
819 		if (np->n_flag & NRMODIFIED)
820 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
821 		if (error == EINTR)
822 			return (error);
823 		if (error == 0) {
824 			if (vap->va_mtime.tv_sec != VNOVAL) {
825 				np->n_mtime = vap->va_mtime.tv_sec;
826 			}
827 		}
828 	}
829 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
830 
831 	/*
832 	 * Sanity check if a truncation was issued.  This should only occur
833 	 * if multiple processes are racing on the same file.
834 	 */
835 	if (error == 0 && vap->va_size != VNOVAL &&
836 	    np->n_size != vap->va_size) {
837 		kprintf("NFS ftruncate: server disagrees on the file size: "
838 			"%jd/%jd/%jd\n",
839 			(intmax_t)tsize,
840 			(intmax_t)vap->va_size,
841 			(intmax_t)np->n_size);
842 		goto again;
843 	}
844 	if (error && vap->va_size != VNOVAL) {
845 		np->n_size = np->n_vattr.va_size = tsize;
846 		nfs_meta_setsize(vp, td, np->n_size, 0);
847 	}
848 	lwkt_reltoken(&nmp->nm_token);
849 
850 	return (error);
851 }
852 
853 /*
854  * Do an nfs setattr rpc.
855  */
856 static int
857 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
858 	       struct ucred *cred, struct thread *td)
859 {
860 	struct nfsv2_sattr *sp;
861 	struct nfsnode *np = VTONFS(vp);
862 	u_int32_t *tl;
863 	int error = 0, wccflag = NFSV3_WCCRATTR;
864 	struct nfsm_info info;
865 
866 	info.mrep = NULL;
867 	info.v3 = NFS_ISV3(vp);
868 
869 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
870 	nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
871 		     NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
872 	ERROROUT(nfsm_fhtom(&info, vp));
873 	if (info.v3) {
874 		nfsm_v3attrbuild(&info, vap, TRUE);
875 		tl = nfsm_build(&info, NFSX_UNSIGNED);
876 		*tl = nfs_false;
877 	} else {
878 		sp = nfsm_build(&info, NFSX_V2SATTR);
879 		if (vap->va_mode == (mode_t)VNOVAL)
880 			sp->sa_mode = nfs_xdrneg1;
881 		else
882 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
883 		if (vap->va_uid == (uid_t)VNOVAL)
884 			sp->sa_uid = nfs_xdrneg1;
885 		else
886 			sp->sa_uid = txdr_unsigned(vap->va_uid);
887 		if (vap->va_gid == (gid_t)VNOVAL)
888 			sp->sa_gid = nfs_xdrneg1;
889 		else
890 			sp->sa_gid = txdr_unsigned(vap->va_gid);
891 		sp->sa_size = txdr_unsigned(vap->va_size);
892 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
893 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
894 	}
895 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
896 	if (info.v3) {
897 		np->n_modestamp = 0;
898 		ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
899 	} else {
900 		ERROROUT(nfsm_loadattr(&info, vp, NULL));
901 	}
902 	m_freem(info.mrep);
903 	info.mrep = NULL;
904 nfsmout:
905 	return (error);
906 }
907 
908 static
909 void
910 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
911 {
912 	if (nctimeout == 0)
913 		nctimeout = 1;
914 	else
915 		nctimeout *= hz;
916 	cache_setvp(nch, vp);
917 	cache_settimeout(nch, nctimeout);
918 }
919 
920 /*
921  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
922  * nfs_lookup() until all remaining new api calls are implemented.
923  *
924  * Resolve a namecache entry.  This function is passed a locked ncp and
925  * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
926  */
927 static int
928 nfs_nresolve(struct vop_nresolve_args *ap)
929 {
930 	struct thread *td = curthread;
931 	struct namecache *ncp;
932 	struct nfsmount *nmp;
933 	struct ucred *cred;
934 	struct nfsnode *np;
935 	struct vnode *dvp;
936 	struct vnode *nvp;
937 	nfsfh_t *fhp;
938 	int attrflag;
939 	int fhsize;
940 	int error;
941 	int tmp_error;
942 	int len;
943 	struct nfsm_info info;
944 
945 	cred = ap->a_cred;
946 	dvp = ap->a_dvp;
947 	nmp = VFSTONFS(dvp->v_mount);
948 
949 	lwkt_gettoken(&nmp->nm_token);
950 
951 	if ((error = vget(dvp, LK_SHARED)) != 0) {
952 		lwkt_reltoken(&nmp->nm_token);
953 		return (error);
954 	}
955 
956 	info.mrep = NULL;
957 	info.v3 = NFS_ISV3(dvp);
958 
959 	nvp = NULL;
960 	nfsstats.lookupcache_misses++;
961 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
962 	ncp = ap->a_nch->ncp;
963 	len = ncp->nc_nlen;
964 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
965 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
966 	ERROROUT(nfsm_fhtom(&info, dvp));
967 	ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
968 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
969 				ap->a_cred, &error));
970 	if (error) {
971 		/*
972 		 * Cache negatve lookups to reduce NFS traffic, but use
973 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
974 		 * XXX we should add a namecache flag for no-caching
975 		 * to uncache the negative hit as soon as possible, but
976 		 * we cannot simply destroy the entry because it is used
977 		 * as a placeholder by the caller.
978 		 *
979 		 * The refactored nfs code will overwrite a non-zero error
980 		 * with 0 when we use ERROROUT(), so don't here.
981 		 */
982 		if (error == ENOENT)
983 			nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
984 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
985 					     NFS_LATTR_NOSHRINK);
986 		if (tmp_error) {
987 			error = tmp_error;
988 			goto nfsmout;
989 		}
990 		m_freem(info.mrep);
991 		info.mrep = NULL;
992 		goto nfsmout;
993 	}
994 
995 	/*
996 	 * Success, get the file handle, do various checks, and load
997 	 * post-operation data from the reply packet.  Theoretically
998 	 * we should never be looking up "." so, theoretically, we
999 	 * should never get the same file handle as our directory.  But
1000 	 * we check anyway. XXX
1001 	 *
1002 	 * Note that no timeout is set for the positive cache hit.  We
1003 	 * assume, theoretically, that ESTALE returns will be dealt with
1004 	 * properly to handle NFS races and in anycase we cannot depend
1005 	 * on a timeout to deal with NFS open/create/excl issues so instead
1006 	 * of a bad hack here the rest of the NFS client code needs to do
1007 	 * the right thing.
1008 	 */
1009 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1010 
1011 	np = VTONFS(dvp);
1012 	if (NFS_CMPFH(np, fhp, fhsize)) {
1013 		vref(dvp);
1014 		nvp = dvp;
1015 	} else {
1016 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1017 		if (error) {
1018 			m_freem(info.mrep);
1019 			info.mrep = NULL;
1020 			vput(dvp);
1021 			lwkt_reltoken(&nmp->nm_token);
1022 			return (error);
1023 		}
1024 		nvp = NFSTOV(np);
1025 	}
1026 	if (info.v3) {
1027 		ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1028 					  NFS_LATTR_NOSHRINK));
1029 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1030 					  NFS_LATTR_NOSHRINK));
1031 	} else {
1032 		ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1033 	}
1034 	nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1035 	m_freem(info.mrep);
1036 	info.mrep = NULL;
1037 nfsmout:
1038 	lwkt_reltoken(&nmp->nm_token);
1039 	vput(dvp);
1040 	if (nvp) {
1041 		if (nvp == dvp)
1042 			vrele(nvp);
1043 		else
1044 			vput(nvp);
1045 	}
1046 	return (error);
1047 }
1048 
1049 /*
1050  * 'cached' nfs directory lookup
1051  *
1052  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1053  *
1054  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1055  *	      struct componentname *a_cnp)
1056  */
1057 static int
1058 nfs_lookup(struct vop_old_lookup_args *ap)
1059 {
1060 	struct componentname *cnp = ap->a_cnp;
1061 	struct vnode *dvp = ap->a_dvp;
1062 	struct vnode **vpp = ap->a_vpp;
1063 	int flags = cnp->cn_flags;
1064 	struct vnode *newvp;
1065 	struct nfsmount *nmp;
1066 	long len;
1067 	nfsfh_t *fhp;
1068 	struct nfsnode *np;
1069 	int lockparent, wantparent, attrflag, fhsize;
1070 	int error;
1071 	int tmp_error;
1072 	struct nfsm_info info;
1073 
1074 	info.mrep = NULL;
1075 	info.v3 = NFS_ISV3(dvp);
1076 	error = 0;
1077 
1078 	/*
1079 	 * Read-only mount check and directory check.
1080 	 */
1081 	*vpp = NULLVP;
1082 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1083 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1084 		return (EROFS);
1085 
1086 	if (dvp->v_type != VDIR)
1087 		return (ENOTDIR);
1088 
1089 	/*
1090 	 * Look it up in the cache.  Note that ENOENT is only returned if we
1091 	 * previously entered a negative hit (see later on).  The additional
1092 	 * nfsneg_cache_timeout check causes previously cached results to
1093 	 * be instantly ignored if the negative caching is turned off.
1094 	 */
1095 	lockparent = flags & CNP_LOCKPARENT;
1096 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1097 	nmp = VFSTONFS(dvp->v_mount);
1098 	np = VTONFS(dvp);
1099 
1100 	lwkt_gettoken(&nmp->nm_token);
1101 
1102 	/*
1103 	 * Go to the wire.
1104 	 */
1105 	error = 0;
1106 	newvp = NULLVP;
1107 	nfsstats.lookupcache_misses++;
1108 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1109 	len = cnp->cn_namelen;
1110 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1111 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1112 	ERROROUT(nfsm_fhtom(&info, dvp));
1113 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1114 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1115 				cnp->cn_cred, &error));
1116 	if (error) {
1117 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1118 					     NFS_LATTR_NOSHRINK);
1119 		if (tmp_error) {
1120 			error = tmp_error;
1121 			goto nfsmout;
1122 		}
1123 
1124 		m_freem(info.mrep);
1125 		info.mrep = NULL;
1126 		goto nfsmout;
1127 	}
1128 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1129 
1130 	/*
1131 	 * Handle RENAME case...
1132 	 */
1133 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1134 		if (NFS_CMPFH(np, fhp, fhsize)) {
1135 			m_freem(info.mrep);
1136 			info.mrep = NULL;
1137 			lwkt_reltoken(&nmp->nm_token);
1138 			return (EISDIR);
1139 		}
1140 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1141 		if (error) {
1142 			m_freem(info.mrep);
1143 			info.mrep = NULL;
1144 			lwkt_reltoken(&nmp->nm_token);
1145 			return (error);
1146 		}
1147 		newvp = NFSTOV(np);
1148 		if (info.v3) {
1149 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1150 						  NFS_LATTR_NOSHRINK));
1151 			ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1152 						  NFS_LATTR_NOSHRINK));
1153 		} else {
1154 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1155 		}
1156 		*vpp = newvp;
1157 		m_freem(info.mrep);
1158 		info.mrep = NULL;
1159 		if (!lockparent) {
1160 			vn_unlock(dvp);
1161 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1162 		}
1163 		lwkt_reltoken(&nmp->nm_token);
1164 		return (0);
1165 	}
1166 
1167 	if (flags & CNP_ISDOTDOT) {
1168 		vn_unlock(dvp);
1169 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1170 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1171 		if (error) {
1172 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1173 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1174 			lwkt_reltoken(&nmp->nm_token);
1175 			return (error); /* NOTE: return error from nget */
1176 		}
1177 		newvp = NFSTOV(np);
1178 		if (lockparent) {
1179 			error = vn_lock(dvp, LK_EXCLUSIVE);
1180 			if (error) {
1181 				vput(newvp);
1182 				lwkt_reltoken(&nmp->nm_token);
1183 				return (error);
1184 			}
1185 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1186 		}
1187 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1188 		vref(dvp);
1189 		newvp = dvp;
1190 	} else {
1191 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1192 		if (error) {
1193 			m_freem(info.mrep);
1194 			info.mrep = NULL;
1195 			lwkt_reltoken(&nmp->nm_token);
1196 			return (error);
1197 		}
1198 		if (!lockparent) {
1199 			vn_unlock(dvp);
1200 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1201 		}
1202 		newvp = NFSTOV(np);
1203 	}
1204 	if (info.v3) {
1205 		ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1206 					  NFS_LATTR_NOSHRINK));
1207 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1208 					  NFS_LATTR_NOSHRINK));
1209 	} else {
1210 		ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1211 	}
1212 #if 0
1213 	/* XXX MOVE TO nfs_nremove() */
1214 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1215 	    cnp->cn_nameiop != NAMEI_DELETE) {
1216 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1217 	}
1218 #endif
1219 	*vpp = newvp;
1220 	m_freem(info.mrep);
1221 	info.mrep = NULL;
1222 nfsmout:
1223 	if (error) {
1224 		if (newvp != NULLVP) {
1225 			vrele(newvp);
1226 			*vpp = NULLVP;
1227 		}
1228 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1229 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1230 		    error == ENOENT) {
1231 			if (!lockparent) {
1232 				vn_unlock(dvp);
1233 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1234 			}
1235 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1236 				error = EROFS;
1237 			else
1238 				error = EJUSTRETURN;
1239 		}
1240 	}
1241 	lwkt_reltoken(&nmp->nm_token);
1242 	return (error);
1243 }
1244 
1245 /*
1246  * nfs read call.
1247  * Just call nfs_bioread() to do the work.
1248  *
1249  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1250  *	    struct ucred *a_cred)
1251  */
1252 static int
1253 nfs_read(struct vop_read_args *ap)
1254 {
1255 	struct vnode *vp = ap->a_vp;
1256 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1257 	int error;
1258 
1259 	lwkt_gettoken(&nmp->nm_token);
1260 	error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1261 	lwkt_reltoken(&nmp->nm_token);
1262 
1263 	return error;
1264 }
1265 
1266 /*
1267  * nfs readlink call
1268  *
1269  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1270  */
1271 static int
1272 nfs_readlink(struct vop_readlink_args *ap)
1273 {
1274 	struct vnode *vp = ap->a_vp;
1275 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1276 	int error;
1277 
1278 	if (vp->v_type != VLNK)
1279 		return (EINVAL);
1280 
1281 	lwkt_gettoken(&nmp->nm_token);
1282 	error = nfs_bioread(vp, ap->a_uio, 0);
1283 	lwkt_reltoken(&nmp->nm_token);
1284 
1285 	return error;
1286 }
1287 
1288 /*
1289  * Do a readlink rpc.
1290  * Called by nfs_doio() from below the buffer cache.
1291  */
1292 int
1293 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1294 {
1295 	int error = 0, len, attrflag;
1296 	struct nfsm_info info;
1297 
1298 	info.mrep = NULL;
1299 	info.v3 = NFS_ISV3(vp);
1300 
1301 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1302 	nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1303 	ERROROUT(nfsm_fhtom(&info, vp));
1304 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1305 				nfs_vpcred(vp, ND_CHECK), &error));
1306 	if (info.v3) {
1307 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1308 					  NFS_LATTR_NOSHRINK));
1309 	}
1310 	if (!error) {
1311 		NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1312 		if (len == NFS_MAXPATHLEN) {
1313 			struct nfsnode *np = VTONFS(vp);
1314 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1315 				len = np->n_size;
1316 		}
1317 		ERROROUT(nfsm_mtouio(&info, uiop, len));
1318 	}
1319 	m_freem(info.mrep);
1320 	info.mrep = NULL;
1321 nfsmout:
1322 	return (error);
1323 }
1324 
1325 /*
1326  * nfs synchronous read rpc using UIO
1327  */
1328 int
1329 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1330 {
1331 	u_int32_t *tl;
1332 	struct nfsmount *nmp;
1333 	int error = 0, len, retlen, tsiz, eof, attrflag;
1334 	struct nfsm_info info;
1335 	off_t tmp_off;
1336 
1337 	info.mrep = NULL;
1338 	info.v3 = NFS_ISV3(vp);
1339 
1340 #ifndef nolint
1341 	eof = 0;
1342 #endif
1343 	nmp = VFSTONFS(vp->v_mount);
1344 
1345 	tsiz = uiop->uio_resid;
1346 	tmp_off = uiop->uio_offset + tsiz;
1347 	if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1348 		return (EFBIG);
1349 	tmp_off = uiop->uio_offset;
1350 	while (tsiz > 0) {
1351 		nfsstats.rpccnt[NFSPROC_READ]++;
1352 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1353 		nfsm_reqhead(&info, vp, NFSPROC_READ,
1354 			     NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1355 		ERROROUT(nfsm_fhtom(&info, vp));
1356 		tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1357 		if (info.v3) {
1358 			txdr_hyper(uiop->uio_offset, tl);
1359 			*(tl + 2) = txdr_unsigned(len);
1360 		} else {
1361 			*tl++ = txdr_unsigned(uiop->uio_offset);
1362 			*tl++ = txdr_unsigned(len);
1363 			*tl = 0;
1364 		}
1365 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1366 					nfs_vpcred(vp, ND_READ), &error));
1367 		if (info.v3) {
1368 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1369 						 NFS_LATTR_NOSHRINK));
1370 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1371 			eof = fxdr_unsigned(int, *(tl + 1));
1372 		} else {
1373 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1374 		}
1375 		NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1376 		ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1377 		m_freem(info.mrep);
1378 		info.mrep = NULL;
1379 
1380 		/*
1381 		 * Handle short-read from server (NFSv3).  If EOF is not
1382 		 * flagged (and no error occurred), but retlen is less
1383 		 * then the request size, we must zero-fill the remainder.
1384 		 */
1385 		if (retlen < len && info.v3 && eof == 0) {
1386 			ERROROUT(uiomovez(len - retlen, uiop));
1387 			retlen = len;
1388 		}
1389 		tsiz -= retlen;
1390 
1391 		/*
1392 		 * Terminate loop on EOF or zero-length read.
1393 		 *
1394 		 * For NFSv2 a short-read indicates EOF, not zero-fill,
1395 		 * and also terminates the loop.
1396 		 */
1397 		if (info.v3) {
1398 			if (eof || retlen == 0)
1399 				tsiz = 0;
1400 		} else if (retlen < len) {
1401 			tsiz = 0;
1402 		}
1403 	}
1404 nfsmout:
1405 	return (error);
1406 }
1407 
1408 /*
1409  * nfs write call
1410  */
1411 int
1412 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1413 		 int *iomode, int *must_commit)
1414 {
1415 	u_int32_t *tl;
1416 	int32_t backup;
1417 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1418 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1419 	int  committed = NFSV3WRITE_FILESYNC;
1420 	struct nfsm_info info;
1421 
1422 	info.mrep = NULL;
1423 	info.v3 = NFS_ISV3(vp);
1424 
1425 #ifndef DIAGNOSTIC
1426 	if (uiop->uio_iovcnt != 1)
1427 		panic("nfs: writerpc iovcnt > 1");
1428 #endif
1429 	*must_commit = 0;
1430 	tsiz = uiop->uio_resid;
1431 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1432 		return (EFBIG);
1433 	while (tsiz > 0) {
1434 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1435 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1436 		nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1437 			     NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1438 		ERROROUT(nfsm_fhtom(&info, vp));
1439 		if (info.v3) {
1440 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1441 			txdr_hyper(uiop->uio_offset, tl);
1442 			tl += 2;
1443 			*tl++ = txdr_unsigned(len);
1444 			*tl++ = txdr_unsigned(*iomode);
1445 			*tl = txdr_unsigned(len);
1446 		} else {
1447 			u_int32_t x;
1448 
1449 			tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1450 			/* Set both "begin" and "current" to non-garbage. */
1451 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1452 			*tl++ = x;	/* "begin offset" */
1453 			*tl++ = x;	/* "current offset" */
1454 			x = txdr_unsigned(len);
1455 			*tl++ = x;	/* total to this offset */
1456 			*tl = x;	/* size of this write */
1457 		}
1458 		ERROROUT(nfsm_uiotom(&info, uiop, len));
1459 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1460 					nfs_vpcred(vp, ND_WRITE), &error));
1461 		if (info.v3) {
1462 			/*
1463 			 * The write RPC returns a before and after mtime.  The
1464 			 * nfsm_wcc_data() macro checks the before n_mtime
1465 			 * against the before time and stores the after time
1466 			 * in the nfsnode's cached vattr and n_mtime field.
1467 			 * The NRMODIFIED bit will be set if the before
1468 			 * time did not match the original mtime.
1469 			 */
1470 			wccflag = NFSV3_WCCCHK;
1471 			ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1472 			if (error == 0) {
1473 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1474 				rlen = fxdr_unsigned(int, *tl++);
1475 				if (rlen == 0) {
1476 					error = NFSERR_IO;
1477 					m_freem(info.mrep);
1478 					info.mrep = NULL;
1479 					break;
1480 				} else if (rlen < len) {
1481 					backup = len - rlen;
1482 					uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1483 					uiop->uio_iov->iov_len += backup;
1484 					uiop->uio_offset -= backup;
1485 					uiop->uio_resid += backup;
1486 					len = rlen;
1487 				}
1488 				commit = fxdr_unsigned(int, *tl++);
1489 
1490 				/*
1491 				 * Return the lowest committment level
1492 				 * obtained by any of the RPCs.
1493 				 */
1494 				if (committed == NFSV3WRITE_FILESYNC)
1495 					committed = commit;
1496 				else if (committed == NFSV3WRITE_DATASYNC &&
1497 					commit == NFSV3WRITE_UNSTABLE)
1498 					committed = commit;
1499 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1500 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1501 					NFSX_V3WRITEVERF);
1502 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1503 				} else if (bcmp((caddr_t)tl,
1504 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1505 				    *must_commit = 1;
1506 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1507 					NFSX_V3WRITEVERF);
1508 				}
1509 			}
1510 		} else {
1511 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1512 		}
1513 		m_freem(info.mrep);
1514 		info.mrep = NULL;
1515 		if (error)
1516 			break;
1517 		tsiz -= len;
1518 	}
1519 nfsmout:
1520 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1521 		committed = NFSV3WRITE_FILESYNC;
1522 	*iomode = committed;
1523 	if (error)
1524 		uiop->uio_resid = tsiz;
1525 	return (error);
1526 }
1527 
1528 /*
1529  * nfs mknod rpc
1530  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1531  * mode set to specify the file type and the size field for rdev.
1532  */
1533 static int
1534 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1535 	     struct vattr *vap)
1536 {
1537 	struct nfsv2_sattr *sp;
1538 	u_int32_t *tl;
1539 	struct vnode *newvp = NULL;
1540 	struct nfsnode *np = NULL;
1541 	struct vattr vattr;
1542 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1543 	int rmajor, rminor;
1544 	struct nfsm_info info;
1545 
1546 	info.mrep = NULL;
1547 	info.v3 = NFS_ISV3(dvp);
1548 
1549 	if (vap->va_type == VCHR || vap->va_type == VBLK) {
1550 		rmajor = txdr_unsigned(vap->va_rmajor);
1551 		rminor = txdr_unsigned(vap->va_rminor);
1552 	} else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1553 		rmajor = nfs_xdrneg1;
1554 		rminor = nfs_xdrneg1;
1555 	} else {
1556 		return (EOPNOTSUPP);
1557 	}
1558 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1559 		return (error);
1560 	}
1561 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1562 	nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1563 		     NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1564 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1565 	ERROROUT(nfsm_fhtom(&info, dvp));
1566 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1567 			     NFS_MAXNAMLEN));
1568 	if (info.v3) {
1569 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1570 		*tl++ = vtonfsv3_type(vap->va_type);
1571 		nfsm_v3attrbuild(&info, vap, FALSE);
1572 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1573 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1574 			*tl++ = txdr_unsigned(vap->va_rmajor);
1575 			*tl = txdr_unsigned(vap->va_rminor);
1576 		}
1577 	} else {
1578 		sp = nfsm_build(&info, NFSX_V2SATTR);
1579 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1580 		sp->sa_uid = nfs_xdrneg1;
1581 		sp->sa_gid = nfs_xdrneg1;
1582 		sp->sa_size = makeudev(rmajor, rminor);
1583 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1584 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1585 	}
1586 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1587 				cnp->cn_cred, &error));
1588 	if (!error) {
1589 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1590 		if (!gotvp) {
1591 			if (newvp) {
1592 				vput(newvp);
1593 				newvp = NULL;
1594 			}
1595 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1596 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1597 			if (!error)
1598 				newvp = NFSTOV(np);
1599 		}
1600 	}
1601 	if (info.v3) {
1602 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1603 	}
1604 	m_freem(info.mrep);
1605 	info.mrep = NULL;
1606 nfsmout:
1607 	if (error) {
1608 		if (newvp)
1609 			vput(newvp);
1610 	} else {
1611 		*vpp = newvp;
1612 	}
1613 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1614 	if (!wccflag)
1615 		VTONFS(dvp)->n_attrstamp = 0;
1616 	return (error);
1617 }
1618 
1619 /*
1620  * nfs mknod vop
1621  * just call nfs_mknodrpc() to do the work.
1622  *
1623  * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1624  *	     struct componentname *a_cnp, struct vattr *a_vap)
1625  */
1626 /* ARGSUSED */
1627 static int
1628 nfs_mknod(struct vop_old_mknod_args *ap)
1629 {
1630 	struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1631 	int error;
1632 
1633 	lwkt_gettoken(&nmp->nm_token);
1634 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1635 	lwkt_reltoken(&nmp->nm_token);
1636 
1637 	return error;
1638 }
1639 
1640 static u_long create_verf;
1641 /*
1642  * nfs file create call
1643  *
1644  * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1645  *	      struct componentname *a_cnp, struct vattr *a_vap)
1646  */
1647 static int
1648 nfs_create(struct vop_old_create_args *ap)
1649 {
1650 	struct vnode *dvp = ap->a_dvp;
1651 	struct vattr *vap = ap->a_vap;
1652 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1653 	struct componentname *cnp = ap->a_cnp;
1654 	struct nfsv2_sattr *sp;
1655 	u_int32_t *tl;
1656 	struct nfsnode *np = NULL;
1657 	struct vnode *newvp = NULL;
1658 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1659 	struct vattr vattr;
1660 	struct nfsm_info info;
1661 
1662 	info.mrep = NULL;
1663 	info.v3 = NFS_ISV3(dvp);
1664 	lwkt_gettoken(&nmp->nm_token);
1665 
1666 	/*
1667 	 * Oops, not for me..
1668 	 */
1669 	if (vap->va_type == VSOCK) {
1670 		error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1671 		lwkt_reltoken(&nmp->nm_token);
1672 		return error;
1673 	}
1674 
1675 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1676 		lwkt_reltoken(&nmp->nm_token);
1677 		return (error);
1678 	}
1679 	if (vap->va_vaflags & VA_EXCLUSIVE)
1680 		fmode |= O_EXCL;
1681 again:
1682 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1683 	nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1684 		     NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1685 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1686 	ERROROUT(nfsm_fhtom(&info, dvp));
1687 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1688 			     NFS_MAXNAMLEN));
1689 	if (info.v3) {
1690 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1691 		if (fmode & O_EXCL) {
1692 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1693 			tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1694 #ifdef INET
1695 			if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1696 				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1697 			else
1698 #endif
1699 				*tl++ = create_verf;
1700 			*tl = ++create_verf;
1701 		} else {
1702 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1703 			nfsm_v3attrbuild(&info, vap, FALSE);
1704 		}
1705 	} else {
1706 		sp = nfsm_build(&info, NFSX_V2SATTR);
1707 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1708 		sp->sa_uid = nfs_xdrneg1;
1709 		sp->sa_gid = nfs_xdrneg1;
1710 		sp->sa_size = 0;
1711 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1712 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1713 	}
1714 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1715 				cnp->cn_cred, &error));
1716 	if (error == 0) {
1717 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1718 		if (!gotvp) {
1719 			if (newvp) {
1720 				vput(newvp);
1721 				newvp = NULL;
1722 			}
1723 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1724 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1725 			if (!error)
1726 				newvp = NFSTOV(np);
1727 		}
1728 	}
1729 	if (info.v3) {
1730 		if (error == 0)
1731 			error = nfsm_wcc_data(&info, dvp, &wccflag);
1732 		else
1733 			(void)nfsm_wcc_data(&info, dvp, &wccflag);
1734 	}
1735 	m_freem(info.mrep);
1736 	info.mrep = NULL;
1737 nfsmout:
1738 	if (error) {
1739 		if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1740 			KKASSERT(newvp == NULL);
1741 			fmode &= ~O_EXCL;
1742 			goto again;
1743 		}
1744 	} else if (info.v3 && (fmode & O_EXCL)) {
1745 		/*
1746 		 * We are normally called with only a partially initialized
1747 		 * VAP.  Since the NFSv3 spec says that server may use the
1748 		 * file attributes to store the verifier, the spec requires
1749 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1750 		 * in atime, but we can't really assume that all servers will
1751 		 * so we ensure that our SETATTR sets both atime and mtime.
1752 		 */
1753 		if (vap->va_mtime.tv_sec == VNOVAL)
1754 			vfs_timestamp(&vap->va_mtime);
1755 		if (vap->va_atime.tv_sec == VNOVAL)
1756 			vap->va_atime = vap->va_mtime;
1757 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1758 	}
1759 	if (error == 0) {
1760 		/*
1761 		 * The new np may have enough info for access
1762 		 * checks, make sure rucred and wucred are
1763 		 * initialized for read and write rpc's.
1764 		 */
1765 		np = VTONFS(newvp);
1766 		if (np->n_rucred == NULL)
1767 			np->n_rucred = crhold(cnp->cn_cred);
1768 		if (np->n_wucred == NULL)
1769 			np->n_wucred = crhold(cnp->cn_cred);
1770 		*ap->a_vpp = newvp;
1771 	} else if (newvp) {
1772 		vput(newvp);
1773 	}
1774 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1775 	if (!wccflag)
1776 		VTONFS(dvp)->n_attrstamp = 0;
1777 	lwkt_reltoken(&nmp->nm_token);
1778 	return (error);
1779 }
1780 
1781 /*
1782  * nfs file remove call
1783  * To try and make nfs semantics closer to ufs semantics, a file that has
1784  * other processes using the vnode is renamed instead of removed and then
1785  * removed later on the last close.
1786  * - If v_sysref.refcnt > 1
1787  *	  If a rename is not already in the works
1788  *	     call nfs_sillyrename() to set it up
1789  *     else
1790  *	  do the remove rpc
1791  *
1792  * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1793  *	      struct componentname *a_cnp)
1794  */
1795 static int
1796 nfs_remove(struct vop_old_remove_args *ap)
1797 {
1798 	struct vnode *vp = ap->a_vp;
1799 	struct vnode *dvp = ap->a_dvp;
1800 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1801 	struct componentname *cnp = ap->a_cnp;
1802 	struct nfsnode *np = VTONFS(vp);
1803 	int error = 0;
1804 	struct vattr vattr;
1805 
1806 	lwkt_gettoken(&nmp->nm_token);
1807 #ifndef DIAGNOSTIC
1808 	if (vp->v_sysref.refcnt < 1)
1809 		panic("nfs_remove: bad v_sysref.refcnt");
1810 #endif
1811 	if (vp->v_type == VDIR) {
1812 		error = EPERM;
1813 	} else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1814 		   VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1815 		/*
1816 		 * throw away biocache buffers, mainly to avoid
1817 		 * unnecessary delayed writes later.
1818 		 */
1819 		error = nfs_vinvalbuf(vp, 0, 1);
1820 		/* Do the rpc */
1821 		if (error != EINTR)
1822 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1823 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1824 		/*
1825 		 * Kludge City: If the first reply to the remove rpc is lost..
1826 		 *   the reply to the retransmitted request will be ENOENT
1827 		 *   since the file was in fact removed
1828 		 *   Therefore, we cheat and return success.
1829 		 */
1830 		if (error == ENOENT)
1831 			error = 0;
1832 	} else if (!np->n_sillyrename) {
1833 		error = nfs_sillyrename(dvp, vp, cnp);
1834 	}
1835 	np->n_attrstamp = 0;
1836 	lwkt_reltoken(&nmp->nm_token);
1837 
1838 	return (error);
1839 }
1840 
1841 /*
1842  * nfs file remove rpc called from nfs_inactive
1843  */
1844 int
1845 nfs_removeit(struct sillyrename *sp)
1846 {
1847 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1848 		sp->s_cred, NULL));
1849 }
1850 
1851 /*
1852  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1853  */
1854 static int
1855 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1856 	      struct ucred *cred, struct thread *td)
1857 {
1858 	int error = 0, wccflag = NFSV3_WCCRATTR;
1859 	struct nfsm_info info;
1860 
1861 	info.mrep = NULL;
1862 	info.v3 = NFS_ISV3(dvp);
1863 
1864 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1865 	nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1866 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1867 	ERROROUT(nfsm_fhtom(&info, dvp));
1868 	ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1869 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1870 	if (info.v3) {
1871 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1872 	}
1873 	m_freem(info.mrep);
1874 	info.mrep = NULL;
1875 nfsmout:
1876 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1877 	if (!wccflag)
1878 		VTONFS(dvp)->n_attrstamp = 0;
1879 	return (error);
1880 }
1881 
1882 /*
1883  * nfs file rename call
1884  *
1885  * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1886  *	      struct componentname *a_fcnp, struct vnode *a_tdvp,
1887  *	      struct vnode *a_tvp, struct componentname *a_tcnp)
1888  */
1889 static int
1890 nfs_rename(struct vop_old_rename_args *ap)
1891 {
1892 	struct vnode *fvp = ap->a_fvp;
1893 	struct vnode *tvp = ap->a_tvp;
1894 	struct vnode *fdvp = ap->a_fdvp;
1895 	struct vnode *tdvp = ap->a_tdvp;
1896 	struct componentname *tcnp = ap->a_tcnp;
1897 	struct componentname *fcnp = ap->a_fcnp;
1898 	struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1899 	int error;
1900 
1901 	lwkt_gettoken(&nmp->nm_token);
1902 
1903 	/* Check for cross-device rename */
1904 	if ((fvp->v_mount != tdvp->v_mount) ||
1905 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1906 		error = EXDEV;
1907 		goto out;
1908 	}
1909 
1910 	/*
1911 	 * We shouldn't have to flush fvp on rename for most server-side
1912 	 * filesystems as the file handle should not change.  Unfortunately
1913 	 * the inode for some filesystems (msdosfs) might be tied to the
1914 	 * file name or directory position so to be completely safe
1915 	 * vfs.nfs.flush_on_rename is set by default.  Clear to improve
1916 	 * performance.
1917 	 *
1918 	 * We must flush tvp on rename because it might become stale on the
1919 	 * server after the rename.
1920 	 */
1921 	if (nfs_flush_on_rename)
1922 	    VOP_FSYNC(fvp, MNT_WAIT, 0);
1923 	if (tvp)
1924 	    VOP_FSYNC(tvp, MNT_WAIT, 0);
1925 
1926 	/*
1927 	 * If the tvp exists and is in use, sillyrename it before doing the
1928 	 * rename of the new file over it.
1929 	 *
1930 	 * XXX Can't sillyrename a directory.
1931 	 *
1932 	 * We do not attempt to do any namecache purges in this old API
1933 	 * routine.  The new API compat functions have access to the actual
1934 	 * namecache structures and will do it for us.
1935 	 */
1936 	if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1937 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1938 		vput(tvp);
1939 		tvp = NULL;
1940 	} else if (tvp) {
1941 		;
1942 	}
1943 
1944 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1945 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1946 		tcnp->cn_td);
1947 
1948 out:
1949 	lwkt_reltoken(&nmp->nm_token);
1950 	if (tdvp == tvp)
1951 		vrele(tdvp);
1952 	else
1953 		vput(tdvp);
1954 	if (tvp)
1955 		vput(tvp);
1956 	vrele(fdvp);
1957 	vrele(fvp);
1958 	/*
1959 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1960 	 */
1961 	if (error == ENOENT)
1962 		error = 0;
1963 	return (error);
1964 }
1965 
1966 /*
1967  * nfs file rename rpc called from nfs_remove() above
1968  */
1969 static int
1970 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1971 	     struct sillyrename *sp)
1972 {
1973 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1974 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1975 }
1976 
1977 /*
1978  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1979  */
1980 static int
1981 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1982 	      struct vnode *tdvp, const char *tnameptr, int tnamelen,
1983 	      struct ucred *cred, struct thread *td)
1984 {
1985 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1986 	struct nfsm_info info;
1987 
1988 	info.mrep = NULL;
1989 	info.v3 = NFS_ISV3(fdvp);
1990 
1991 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1992 	nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1993 		    (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1994 		    nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1995 	ERROROUT(nfsm_fhtom(&info, fdvp));
1996 	ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1997 	ERROROUT(nfsm_fhtom(&info, tdvp));
1998 	ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1999 	NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
2000 	if (info.v3) {
2001 		ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
2002 		ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
2003 	}
2004 	m_freem(info.mrep);
2005 	info.mrep = NULL;
2006 nfsmout:
2007 	VTONFS(fdvp)->n_flag |= NLMODIFIED;
2008 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2009 	if (!fwccflag)
2010 		VTONFS(fdvp)->n_attrstamp = 0;
2011 	if (!twccflag)
2012 		VTONFS(tdvp)->n_attrstamp = 0;
2013 	return (error);
2014 }
2015 
2016 /*
2017  * nfs hard link create call
2018  *
2019  * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2020  *	    struct componentname *a_cnp)
2021  */
2022 static int
2023 nfs_link(struct vop_old_link_args *ap)
2024 {
2025 	struct vnode *vp = ap->a_vp;
2026 	struct vnode *tdvp = ap->a_tdvp;
2027 	struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2028 	struct componentname *cnp = ap->a_cnp;
2029 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2030 	struct nfsm_info info;
2031 
2032 	if (vp->v_mount != tdvp->v_mount) {
2033 		return (EXDEV);
2034 	}
2035 	lwkt_gettoken(&nmp->nm_token);
2036 
2037 	/*
2038 	 * The attribute cache may get out of sync with the server on link.
2039 	 * Pushing writes to the server before handle was inherited from
2040 	 * long long ago and it is unclear if we still need to do this.
2041 	 * Defaults to off.
2042 	 */
2043 	if (nfs_flush_on_hlink)
2044 		VOP_FSYNC(vp, MNT_WAIT, 0);
2045 
2046 	info.mrep = NULL;
2047 	info.v3 = NFS_ISV3(vp);
2048 
2049 	nfsstats.rpccnt[NFSPROC_LINK]++;
2050 	nfsm_reqhead(&info, vp, NFSPROC_LINK,
2051 		     NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2052 		     nfsm_rndup(cnp->cn_namelen));
2053 	ERROROUT(nfsm_fhtom(&info, vp));
2054 	ERROROUT(nfsm_fhtom(&info, tdvp));
2055 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2056 			     NFS_MAXNAMLEN));
2057 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2058 				cnp->cn_cred, &error));
2059 	if (info.v3) {
2060 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2061 					 NFS_LATTR_NOSHRINK));
2062 		ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2063 	}
2064 	m_freem(info.mrep);
2065 	info.mrep = NULL;
2066 nfsmout:
2067 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2068 	if (!attrflag)
2069 		VTONFS(vp)->n_attrstamp = 0;
2070 	if (!wccflag)
2071 		VTONFS(tdvp)->n_attrstamp = 0;
2072 	/*
2073 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2074 	 */
2075 	if (error == EEXIST)
2076 		error = 0;
2077 	lwkt_reltoken(&nmp->nm_token);
2078 	return (error);
2079 }
2080 
2081 /*
2082  * nfs symbolic link create call
2083  *
2084  * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2085  *		struct componentname *a_cnp, struct vattr *a_vap,
2086  *		char *a_target)
2087  */
2088 static int
2089 nfs_symlink(struct vop_old_symlink_args *ap)
2090 {
2091 	struct vnode *dvp = ap->a_dvp;
2092 	struct vattr *vap = ap->a_vap;
2093 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2094 	struct componentname *cnp = ap->a_cnp;
2095 	struct nfsv2_sattr *sp;
2096 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2097 	struct vnode *newvp = NULL;
2098 	struct nfsm_info info;
2099 
2100 	info.mrep = NULL;
2101 	info.v3 = NFS_ISV3(dvp);
2102 	lwkt_gettoken(&nmp->nm_token);
2103 
2104 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2105 	slen = strlen(ap->a_target);
2106 	nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2107 		     NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2108 		     nfsm_rndup(cnp->cn_namelen) +
2109 		     nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2110 	ERROROUT(nfsm_fhtom(&info, dvp));
2111 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2112 			     NFS_MAXNAMLEN));
2113 	if (info.v3) {
2114 		nfsm_v3attrbuild(&info, vap, FALSE);
2115 	}
2116 	ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2117 	if (info.v3 == 0) {
2118 		sp = nfsm_build(&info, NFSX_V2SATTR);
2119 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2120 		sp->sa_uid = nfs_xdrneg1;
2121 		sp->sa_gid = nfs_xdrneg1;
2122 		sp->sa_size = nfs_xdrneg1;
2123 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2124 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2125 	}
2126 
2127 	/*
2128 	 * Issue the NFS request and get the rpc response.
2129 	 *
2130 	 * Only NFSv3 responses returning an error of 0 actually return
2131 	 * a file handle that can be converted into newvp without having
2132 	 * to do an extra lookup rpc.
2133 	 */
2134 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2135 				cnp->cn_cred, &error));
2136 	if (info.v3) {
2137 		if (error == 0) {
2138 		       ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2139 		}
2140 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2141 	}
2142 
2143 	/*
2144 	 * out code jumps -> here, mrep is also freed.
2145 	 */
2146 
2147 	m_freem(info.mrep);
2148 	info.mrep = NULL;
2149 nfsmout:
2150 
2151 	/*
2152 	 * If we get an EEXIST error, silently convert it to no-error
2153 	 * in case of an NFS retry.
2154 	 */
2155 	if (error == EEXIST)
2156 		error = 0;
2157 
2158 	/*
2159 	 * If we do not have (or no longer have) an error, and we could
2160 	 * not extract the newvp from the response due to the request being
2161 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
2162 	 * to obtain a newvp to return.
2163 	 */
2164 	if (error == 0 && newvp == NULL) {
2165 		struct nfsnode *np = NULL;
2166 
2167 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2168 				     cnp->cn_cred, cnp->cn_td, &np);
2169 		if (!error)
2170 			newvp = NFSTOV(np);
2171 	}
2172 	if (error) {
2173 		if (newvp)
2174 			vput(newvp);
2175 	} else {
2176 		*ap->a_vpp = newvp;
2177 	}
2178 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2179 	if (!wccflag)
2180 		VTONFS(dvp)->n_attrstamp = 0;
2181 	lwkt_reltoken(&nmp->nm_token);
2182 
2183 	return (error);
2184 }
2185 
2186 /*
2187  * nfs make dir call
2188  *
2189  * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2190  *	     struct componentname *a_cnp, struct vattr *a_vap)
2191  */
2192 static int
2193 nfs_mkdir(struct vop_old_mkdir_args *ap)
2194 {
2195 	struct vnode *dvp = ap->a_dvp;
2196 	struct vattr *vap = ap->a_vap;
2197 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2198 	struct componentname *cnp = ap->a_cnp;
2199 	struct nfsv2_sattr *sp;
2200 	struct nfsnode *np = NULL;
2201 	struct vnode *newvp = NULL;
2202 	struct vattr vattr;
2203 	int error = 0, wccflag = NFSV3_WCCRATTR;
2204 	int gotvp = 0;
2205 	int len;
2206 	struct nfsm_info info;
2207 
2208 	info.mrep = NULL;
2209 	info.v3 = NFS_ISV3(dvp);
2210 	lwkt_gettoken(&nmp->nm_token);
2211 
2212 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2213 		lwkt_reltoken(&nmp->nm_token);
2214 		return (error);
2215 	}
2216 	len = cnp->cn_namelen;
2217 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2218 	nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2219 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2220 		     nfsm_rndup(len) + NFSX_SATTR(info.v3));
2221 	ERROROUT(nfsm_fhtom(&info, dvp));
2222 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2223 	if (info.v3) {
2224 		nfsm_v3attrbuild(&info, vap, FALSE);
2225 	} else {
2226 		sp = nfsm_build(&info, NFSX_V2SATTR);
2227 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2228 		sp->sa_uid = nfs_xdrneg1;
2229 		sp->sa_gid = nfs_xdrneg1;
2230 		sp->sa_size = nfs_xdrneg1;
2231 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2232 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2233 	}
2234 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2235 		    cnp->cn_cred, &error));
2236 	if (error == 0) {
2237 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2238 	}
2239 	if (info.v3) {
2240 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2241 	}
2242 	m_freem(info.mrep);
2243 	info.mrep = NULL;
2244 nfsmout:
2245 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2246 	if (!wccflag)
2247 		VTONFS(dvp)->n_attrstamp = 0;
2248 	/*
2249 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2250 	 * if we can succeed in looking up the directory.
2251 	 */
2252 	if (error == EEXIST || (!error && !gotvp)) {
2253 		if (newvp) {
2254 			vrele(newvp);
2255 			newvp = NULL;
2256 		}
2257 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2258 			cnp->cn_td, &np);
2259 		if (!error) {
2260 			newvp = NFSTOV(np);
2261 			if (newvp->v_type != VDIR)
2262 				error = EEXIST;
2263 		}
2264 	}
2265 	if (error) {
2266 		if (newvp)
2267 			vrele(newvp);
2268 	} else {
2269 		*ap->a_vpp = newvp;
2270 	}
2271 	lwkt_reltoken(&nmp->nm_token);
2272 	return (error);
2273 }
2274 
2275 /*
2276  * nfs remove directory call
2277  *
2278  * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2279  *	     struct componentname *a_cnp)
2280  */
2281 static int
2282 nfs_rmdir(struct vop_old_rmdir_args *ap)
2283 {
2284 	struct vnode *vp = ap->a_vp;
2285 	struct vnode *dvp = ap->a_dvp;
2286 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2287 	struct componentname *cnp = ap->a_cnp;
2288 	int error = 0, wccflag = NFSV3_WCCRATTR;
2289 	struct nfsm_info info;
2290 
2291 	info.mrep = NULL;
2292 	info.v3 = NFS_ISV3(dvp);
2293 
2294 	if (dvp == vp)
2295 		return (EINVAL);
2296 
2297 	lwkt_gettoken(&nmp->nm_token);
2298 
2299 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2300 	nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2301 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2302 		     nfsm_rndup(cnp->cn_namelen));
2303 	ERROROUT(nfsm_fhtom(&info, dvp));
2304 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2305 		 NFS_MAXNAMLEN));
2306 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2307 				cnp->cn_cred, &error));
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 ENOENT => 0 assuming that you have a reply to a retry.
2319 	 */
2320 	if (error == ENOENT)
2321 		error = 0;
2322 	lwkt_reltoken(&nmp->nm_token);
2323 
2324 	return (error);
2325 }
2326 
2327 /*
2328  * nfs readdir call
2329  *
2330  * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2331  */
2332 static int
2333 nfs_readdir(struct vop_readdir_args *ap)
2334 {
2335 	struct vnode *vp = ap->a_vp;
2336 	struct nfsnode *np = VTONFS(vp);
2337 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2338 	struct uio *uio = ap->a_uio;
2339 	int tresid, error;
2340 	struct vattr vattr;
2341 
2342 	if (vp->v_type != VDIR)
2343 		return (EPERM);
2344 
2345 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2346 		return (error);
2347 
2348 	lwkt_gettoken(&nmp->nm_token);
2349 
2350 	/*
2351 	 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2352 	 * and then check that is still valid, or if this is an NQNFS mount
2353 	 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR().  Note that
2354 	 * VOP_GETATTR() does not necessarily go to the wire.
2355 	 */
2356 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2357 	    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2358 		if (VOP_GETATTR(vp, &vattr) == 0 &&
2359 		    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2360 		) {
2361 			nfsstats.direofcache_hits++;
2362 			goto done;
2363 		}
2364 	}
2365 
2366 	/*
2367 	 * Call nfs_bioread() to do the real work.  nfs_bioread() does its
2368 	 * own cache coherency checks so we do not have to.
2369 	 */
2370 	tresid = uio->uio_resid;
2371 	error = nfs_bioread(vp, uio, 0);
2372 
2373 	if (!error && uio->uio_resid == tresid)
2374 		nfsstats.direofcache_misses++;
2375 done:
2376 	lwkt_reltoken(&nmp->nm_token);
2377 	vn_unlock(vp);
2378 
2379 	return (error);
2380 }
2381 
2382 /*
2383  * Readdir rpc call.  nfs_bioread->nfs_doio->nfs_readdirrpc.
2384  *
2385  * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2386  * offset/block and converts the nfs formatted directory entries for userland
2387  * consumption as well as deals with offsets into the middle of blocks.
2388  * nfs_doio only deals with logical blocks.  In particular, uio_offset will
2389  * be block-bounded.  It must convert to cookies for the actual RPC.
2390  */
2391 int
2392 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2393 {
2394 	int len, left;
2395 	struct nfs_dirent *dp = NULL;
2396 	u_int32_t *tl;
2397 	nfsuint64 *cookiep;
2398 	caddr_t cp;
2399 	nfsuint64 cookie;
2400 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2401 	struct nfsnode *dnp = VTONFS(vp);
2402 	u_quad_t fileno;
2403 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2404 	int attrflag;
2405 	struct nfsm_info info;
2406 
2407 	info.mrep = NULL;
2408 	info.v3 = NFS_ISV3(vp);
2409 
2410 #ifndef DIAGNOSTIC
2411 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2412 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2413 		panic("nfs readdirrpc bad uio");
2414 #endif
2415 
2416 	/*
2417 	 * If there is no cookie, assume directory was stale.
2418 	 */
2419 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2420 	if (cookiep)
2421 		cookie = *cookiep;
2422 	else
2423 		return (NFSERR_BAD_COOKIE);
2424 	/*
2425 	 * Loop around doing readdir rpc's of size nm_readdirsize
2426 	 * truncated to a multiple of DIRBLKSIZ.
2427 	 * The stopping criteria is EOF or buffer full.
2428 	 */
2429 	while (more_dirs && bigenough) {
2430 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2431 		nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2432 			     NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2433 		ERROROUT(nfsm_fhtom(&info, vp));
2434 		if (info.v3) {
2435 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2436 			*tl++ = cookie.nfsuquad[0];
2437 			*tl++ = cookie.nfsuquad[1];
2438 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2439 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2440 		} else {
2441 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2442 			*tl++ = cookie.nfsuquad[0];
2443 		}
2444 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2445 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2446 					uiop->uio_td,
2447 					nfs_vpcred(vp, ND_READ), &error));
2448 		if (info.v3) {
2449 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2450 						  NFS_LATTR_NOSHRINK));
2451 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2452 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2453 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2454 		}
2455 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2456 		more_dirs = fxdr_unsigned(int, *tl);
2457 
2458 		/* loop thru the dir entries, converting them to std form */
2459 		while (more_dirs && bigenough) {
2460 			if (info.v3) {
2461 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2462 				fileno = fxdr_hyper(tl);
2463 				len = fxdr_unsigned(int, *(tl + 2));
2464 			} else {
2465 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2466 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2467 				len = fxdr_unsigned(int, *tl);
2468 			}
2469 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2470 				error = EBADRPC;
2471 				m_freem(info.mrep);
2472 				info.mrep = NULL;
2473 				goto nfsmout;
2474 			}
2475 
2476 			/*
2477 			 * len is the number of bytes in the path element
2478 			 * name, not including the \0 termination.
2479 			 *
2480 			 * tlen is the number of bytes w have to reserve for
2481 			 * the path element name.
2482 			 */
2483 			tlen = nfsm_rndup(len);
2484 			if (tlen == len)
2485 				tlen += 4;	/* To ensure null termination */
2486 
2487 			/*
2488 			 * If the entry would cross a DIRBLKSIZ boundary,
2489 			 * extend the previous nfs_dirent to cover the
2490 			 * remaining space.
2491 			 */
2492 			left = DIRBLKSIZ - blksiz;
2493 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2494 				dp->nfs_reclen += left;
2495 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2496 				uiop->uio_iov->iov_len -= left;
2497 				uiop->uio_offset += left;
2498 				uiop->uio_resid -= left;
2499 				blksiz = 0;
2500 			}
2501 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2502 				bigenough = 0;
2503 			if (bigenough) {
2504 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2505 				dp->nfs_ino = fileno;
2506 				dp->nfs_namlen = len;
2507 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2508 				dp->nfs_type = DT_UNKNOWN;
2509 				blksiz += dp->nfs_reclen;
2510 				if (blksiz == DIRBLKSIZ)
2511 					blksiz = 0;
2512 				uiop->uio_offset += sizeof(struct nfs_dirent);
2513 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2514 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2515 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2516 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2517 
2518 				/*
2519 				 * The uiop has advanced by nfs_dirent + len
2520 				 * but really needs to advance by
2521 				 * nfs_dirent + tlen
2522 				 */
2523 				cp = uiop->uio_iov->iov_base;
2524 				tlen -= len;
2525 				*cp = '\0';	/* null terminate */
2526 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2527 				uiop->uio_iov->iov_len -= tlen;
2528 				uiop->uio_offset += tlen;
2529 				uiop->uio_resid -= tlen;
2530 			} else {
2531 				/*
2532 				 * NFS strings must be rounded up (nfsm_myouio
2533 				 * handled that in the bigenough case).
2534 				 */
2535 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2536 			}
2537 			if (info.v3) {
2538 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2539 			} else {
2540 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2541 			}
2542 
2543 			/*
2544 			 * If we were able to accomodate the last entry,
2545 			 * get the cookie for the next one.  Otherwise
2546 			 * hold-over the cookie for the one we were not
2547 			 * able to accomodate.
2548 			 */
2549 			if (bigenough) {
2550 				cookie.nfsuquad[0] = *tl++;
2551 				if (info.v3)
2552 					cookie.nfsuquad[1] = *tl++;
2553 			} else if (info.v3) {
2554 				tl += 2;
2555 			} else {
2556 				tl++;
2557 			}
2558 			more_dirs = fxdr_unsigned(int, *tl);
2559 		}
2560 		/*
2561 		 * If at end of rpc data, get the eof boolean
2562 		 */
2563 		if (!more_dirs) {
2564 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2565 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2566 		}
2567 		m_freem(info.mrep);
2568 		info.mrep = NULL;
2569 	}
2570 	/*
2571 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2572 	 * by increasing d_reclen for the last record.
2573 	 */
2574 	if (blksiz > 0) {
2575 		left = DIRBLKSIZ - blksiz;
2576 		dp->nfs_reclen += left;
2577 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2578 		uiop->uio_iov->iov_len -= left;
2579 		uiop->uio_offset += left;
2580 		uiop->uio_resid -= left;
2581 	}
2582 
2583 	if (bigenough) {
2584 		/*
2585 		 * We hit the end of the directory, update direofoffset.
2586 		 */
2587 		dnp->n_direofoffset = uiop->uio_offset;
2588 	} else {
2589 		/*
2590 		 * There is more to go, insert the link cookie so the
2591 		 * next block can be read.
2592 		 */
2593 		if (uiop->uio_resid > 0)
2594 			kprintf("EEK! readdirrpc resid > 0\n");
2595 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2596 		*cookiep = cookie;
2597 	}
2598 nfsmout:
2599 	return (error);
2600 }
2601 
2602 /*
2603  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2604  */
2605 int
2606 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2607 {
2608 	int len, left;
2609 	struct nfs_dirent *dp;
2610 	u_int32_t *tl;
2611 	struct vnode *newvp;
2612 	nfsuint64 *cookiep;
2613 	caddr_t dpossav1, dpossav2;
2614 	caddr_t cp;
2615 	struct mbuf *mdsav1, *mdsav2;
2616 	nfsuint64 cookie;
2617 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2618 	struct nfsnode *dnp = VTONFS(vp), *np;
2619 	nfsfh_t *fhp;
2620 	u_quad_t fileno;
2621 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2622 	int attrflag, fhsize;
2623 	struct nchandle nch;
2624 	struct nchandle dnch;
2625 	struct nlcomponent nlc;
2626 	struct nfsm_info info;
2627 
2628 	info.mrep = NULL;
2629 	info.v3 = 1;
2630 
2631 #ifndef nolint
2632 	dp = NULL;
2633 #endif
2634 #ifndef DIAGNOSTIC
2635 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2636 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2637 		panic("nfs readdirplusrpc bad uio");
2638 #endif
2639 	/*
2640 	 * Obtain the namecache record for the directory so we have something
2641 	 * to use as a basis for creating the entries.  This function will
2642 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2643 	 * from the tree and cannot be used for upward traversals, and the
2644 	 * ncp may be unnamed.  Note that other unrelated operations may
2645 	 * cause the ncp to be named at any time.
2646 	 *
2647 	 * We have to lock the ncp to prevent a lock order reversal when
2648 	 * rdirplus does nlookups of the children, because the vnode is
2649 	 * locked and has to stay that way.
2650 	 */
2651 	cache_fromdvp(vp, NULL, 0, &dnch);
2652 	bzero(&nlc, sizeof(nlc));
2653 	newvp = NULLVP;
2654 
2655 	/*
2656 	 * If there is no cookie, assume directory was stale.
2657 	 */
2658 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2659 	if (cookiep) {
2660 		cookie = *cookiep;
2661 	} else {
2662 		if (dnch.ncp)
2663 			cache_drop(&dnch);
2664 		return (NFSERR_BAD_COOKIE);
2665 	}
2666 
2667 	/*
2668 	 * Loop around doing readdir rpc's of size nm_readdirsize
2669 	 * truncated to a multiple of DIRBLKSIZ.
2670 	 * The stopping criteria is EOF or buffer full.
2671 	 */
2672 	while (more_dirs && bigenough) {
2673 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2674 		nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2675 			     NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2676 		ERROROUT(nfsm_fhtom(&info, vp));
2677 		tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2678 		*tl++ = cookie.nfsuquad[0];
2679 		*tl++ = cookie.nfsuquad[1];
2680 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2681 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2682 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2683 		*tl = txdr_unsigned(nmp->nm_rsize);
2684 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2685 					uiop->uio_td,
2686 					nfs_vpcred(vp, ND_READ), &error));
2687 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2688 					  NFS_LATTR_NOSHRINK));
2689 		NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2690 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2691 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2692 		more_dirs = fxdr_unsigned(int, *tl);
2693 
2694 		/* loop thru the dir entries, doctoring them to 4bsd form */
2695 		while (more_dirs && bigenough) {
2696 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2697 			fileno = fxdr_hyper(tl);
2698 			len = fxdr_unsigned(int, *(tl + 2));
2699 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2700 				error = EBADRPC;
2701 				m_freem(info.mrep);
2702 				info.mrep = NULL;
2703 				goto nfsmout;
2704 			}
2705 			tlen = nfsm_rndup(len);
2706 			if (tlen == len)
2707 				tlen += 4;	/* To ensure null termination*/
2708 			left = DIRBLKSIZ - blksiz;
2709 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2710 				dp->nfs_reclen += left;
2711 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2712 				uiop->uio_iov->iov_len -= left;
2713 				uiop->uio_offset += left;
2714 				uiop->uio_resid -= left;
2715 				blksiz = 0;
2716 			}
2717 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2718 				bigenough = 0;
2719 			if (bigenough) {
2720 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2721 				dp->nfs_ino = fileno;
2722 				dp->nfs_namlen = len;
2723 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2724 				dp->nfs_type = DT_UNKNOWN;
2725 				blksiz += dp->nfs_reclen;
2726 				if (blksiz == DIRBLKSIZ)
2727 					blksiz = 0;
2728 				uiop->uio_offset += sizeof(struct nfs_dirent);
2729 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2730 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2731 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2732 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2733 				nlc.nlc_namelen = len;
2734 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2735 				cp = uiop->uio_iov->iov_base;
2736 				tlen -= len;
2737 				*cp = '\0';
2738 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2739 				uiop->uio_iov->iov_len -= tlen;
2740 				uiop->uio_offset += tlen;
2741 				uiop->uio_resid -= tlen;
2742 			} else {
2743 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2744 			}
2745 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2746 			if (bigenough) {
2747 				cookie.nfsuquad[0] = *tl++;
2748 				cookie.nfsuquad[1] = *tl++;
2749 			} else {
2750 				tl += 2;
2751 			}
2752 
2753 			/*
2754 			 * Since the attributes are before the file handle
2755 			 * (sigh), we must skip over the attributes and then
2756 			 * come back and get them.
2757 			 */
2758 			attrflag = fxdr_unsigned(int, *tl);
2759 			if (attrflag) {
2760 			    dpossav1 = info.dpos;
2761 			    mdsav1 = info.md;
2762 			    ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2763 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2764 			    doit = fxdr_unsigned(int, *tl);
2765 			    if (doit) {
2766 				NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2767 			    }
2768 			    if (doit && bigenough && !nlcdegenerate(&nlc) &&
2769 				!NFS_CMPFH(dnp, fhp, fhsize)
2770 			    ) {
2771 				if (dnch.ncp) {
2772 #if 0
2773 				    kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2774 					nlc.nlc_namelen, nlc.nlc_namelen,
2775 					nlc.nlc_nameptr);
2776 #endif
2777 				    /*
2778 				     * This is a bit hokey but there isn't
2779 				     * much we can do about it.  We can't
2780 				     * hold the directory vp locked while
2781 				     * doing lookups and gets.
2782 				     */
2783 				    nch = cache_nlookup_nonblock(&dnch, &nlc);
2784 				    if (nch.ncp == NULL)
2785 					goto rdfail;
2786 				    cache_setunresolved(&nch);
2787 				    error = nfs_nget_nonblock(vp->v_mount, fhp,
2788 							      fhsize, &np);
2789 				    if (error) {
2790 					cache_put(&nch);
2791 					goto rdfail;
2792 				    }
2793 				    newvp = NFSTOV(np);
2794 				    dpossav2 = info.dpos;
2795 				    info.dpos = dpossav1;
2796 				    mdsav2 = info.md;
2797 				    info.md = mdsav1;
2798 				    ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2799 				    info.dpos = dpossav2;
2800 				    info.md = mdsav2;
2801 				    dp->nfs_type =
2802 					    IFTODT(VTTOIF(np->n_vattr.va_type));
2803 				    nfs_cache_setvp(&nch, newvp,
2804 						    nfspos_cache_timeout);
2805 				    vput(newvp);
2806 				    newvp = NULLVP;
2807 				    cache_put(&nch);
2808 				} else {
2809 rdfail:
2810 				    ;
2811 #if 0
2812 				    kprintf("Warning: NFS/rddirplus, "
2813 					    "UNABLE TO ENTER %*.*s\n",
2814 					nlc.nlc_namelen, nlc.nlc_namelen,
2815 					nlc.nlc_nameptr);
2816 #endif
2817 				}
2818 			    }
2819 			} else {
2820 			    /* Just skip over the file handle */
2821 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2822 			    i = fxdr_unsigned(int, *tl);
2823 			    ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2824 			}
2825 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2826 			more_dirs = fxdr_unsigned(int, *tl);
2827 		}
2828 		/*
2829 		 * If at end of rpc data, get the eof boolean
2830 		 */
2831 		if (!more_dirs) {
2832 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2833 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2834 		}
2835 		m_freem(info.mrep);
2836 		info.mrep = NULL;
2837 	}
2838 	/*
2839 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2840 	 * by increasing d_reclen for the last record.
2841 	 */
2842 	if (blksiz > 0) {
2843 		left = DIRBLKSIZ - blksiz;
2844 		dp->nfs_reclen += left;
2845 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2846 		uiop->uio_iov->iov_len -= left;
2847 		uiop->uio_offset += left;
2848 		uiop->uio_resid -= left;
2849 	}
2850 
2851 	/*
2852 	 * We are now either at the end of the directory or have filled the
2853 	 * block.
2854 	 */
2855 	if (bigenough) {
2856 		dnp->n_direofoffset = uiop->uio_offset;
2857 	} else {
2858 		if (uiop->uio_resid > 0)
2859 			kprintf("EEK! readdirplusrpc resid > 0\n");
2860 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2861 		*cookiep = cookie;
2862 	}
2863 nfsmout:
2864 	if (newvp != NULLVP) {
2865 	        if (newvp == vp)
2866 			vrele(newvp);
2867 		else
2868 			vput(newvp);
2869 		newvp = NULLVP;
2870 	}
2871 	if (dnch.ncp)
2872 		cache_drop(&dnch);
2873 	return (error);
2874 }
2875 
2876 /*
2877  * Silly rename. To make the NFS filesystem that is stateless look a little
2878  * more like the "ufs" a remove of an active vnode is translated to a rename
2879  * to a funny looking filename that is removed by nfs_inactive on the
2880  * nfsnode. There is the potential for another process on a different client
2881  * to create the same funny name between the nfs_lookitup() fails and the
2882  * nfs_rename() completes, but...
2883  */
2884 static int
2885 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2886 {
2887 	struct sillyrename *sp;
2888 	struct nfsnode *np;
2889 	int error;
2890 
2891 	/*
2892 	 * We previously purged dvp instead of vp.  I don't know why, it
2893 	 * completely destroys performance.  We can't do it anyway with the
2894 	 * new VFS API since we would be breaking the namecache topology.
2895 	 */
2896 	cache_purge(vp);	/* XXX */
2897 	np = VTONFS(vp);
2898 #ifndef DIAGNOSTIC
2899 	if (vp->v_type == VDIR)
2900 		panic("nfs: sillyrename dir");
2901 #endif
2902 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2903 		M_NFSREQ, M_WAITOK);
2904 	sp->s_cred = crdup(cnp->cn_cred);
2905 	sp->s_dvp = dvp;
2906 	vref(dvp);
2907 
2908 	/* Fudge together a funny name */
2909 	sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2910 				(int)(intptr_t)cnp->cn_td);
2911 
2912 	/* Try lookitups until we get one that isn't there */
2913 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2914 		cnp->cn_td, NULL) == 0) {
2915 		sp->s_name[4]++;
2916 		if (sp->s_name[4] > 'z') {
2917 			error = EINVAL;
2918 			goto bad;
2919 		}
2920 	}
2921 	error = nfs_renameit(dvp, cnp, sp);
2922 	if (error)
2923 		goto bad;
2924 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2925 		cnp->cn_td, &np);
2926 	np->n_sillyrename = sp;
2927 	return (0);
2928 bad:
2929 	vrele(sp->s_dvp);
2930 	crfree(sp->s_cred);
2931 	kfree((caddr_t)sp, M_NFSREQ);
2932 	return (error);
2933 }
2934 
2935 /*
2936  * Look up a file name and optionally either update the file handle or
2937  * allocate an nfsnode, depending on the value of npp.
2938  * npp == NULL	--> just do the lookup
2939  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2940  *			handled too
2941  * *npp != NULL --> update the file handle in the vnode
2942  */
2943 static int
2944 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2945 	     struct thread *td, struct nfsnode **npp)
2946 {
2947 	struct vnode *newvp = NULL;
2948 	struct nfsnode *np, *dnp = VTONFS(dvp);
2949 	int error = 0, fhlen, attrflag;
2950 	nfsfh_t *nfhp;
2951 	struct nfsm_info info;
2952 
2953 	info.mrep = NULL;
2954 	info.v3 = NFS_ISV3(dvp);
2955 
2956 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2957 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2958 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2959 	ERROROUT(nfsm_fhtom(&info, dvp));
2960 	ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2961 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2962 	if (npp && !error) {
2963 		NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2964 		if (*npp) {
2965 		    np = *npp;
2966 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2967 			kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2968 			np->n_fhp = &np->n_fh;
2969 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2970 			np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2971 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2972 		    np->n_fhsize = fhlen;
2973 		    newvp = NFSTOV(np);
2974 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2975 		    vref(dvp);
2976 		    newvp = dvp;
2977 		} else {
2978 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2979 		    if (error) {
2980 			m_freem(info.mrep);
2981 			info.mrep = NULL;
2982 			return (error);
2983 		    }
2984 		    newvp = NFSTOV(np);
2985 		}
2986 		if (info.v3) {
2987 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2988 						  NFS_LATTR_NOSHRINK));
2989 			if (!attrflag && *npp == NULL) {
2990 				m_freem(info.mrep);
2991 				info.mrep = NULL;
2992 				if (newvp == dvp)
2993 					vrele(newvp);
2994 				else
2995 					vput(newvp);
2996 				return (ENOENT);
2997 			}
2998 		} else {
2999 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3000 		}
3001 	}
3002 	m_freem(info.mrep);
3003 	info.mrep = NULL;
3004 nfsmout:
3005 	if (npp && *npp == NULL) {
3006 		if (error) {
3007 			if (newvp) {
3008 				if (newvp == dvp)
3009 					vrele(newvp);
3010 				else
3011 					vput(newvp);
3012 			}
3013 		} else
3014 			*npp = np;
3015 	}
3016 	return (error);
3017 }
3018 
3019 /*
3020  * Nfs Version 3 commit rpc
3021  *
3022  * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3023  * involved.
3024  */
3025 int
3026 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3027 {
3028 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3029 	int error = 0, wccflag = NFSV3_WCCRATTR;
3030 	struct nfsm_info info;
3031 	u_int32_t *tl;
3032 
3033 	info.mrep = NULL;
3034 	info.v3 = 1;
3035 
3036 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3037 		return (0);
3038 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3039 	nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3040 	ERROROUT(nfsm_fhtom(&info, vp));
3041 	tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3042 	txdr_hyper(offset, tl);
3043 	tl += 2;
3044 	*tl = txdr_unsigned(cnt);
3045 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3046 				nfs_vpcred(vp, ND_WRITE), &error));
3047 	ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3048 	if (!error) {
3049 		NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3050 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3051 			NFSX_V3WRITEVERF)) {
3052 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3053 				NFSX_V3WRITEVERF);
3054 			error = NFSERR_STALEWRITEVERF;
3055 		}
3056 	}
3057 	m_freem(info.mrep);
3058 	info.mrep = NULL;
3059 nfsmout:
3060 	return (error);
3061 }
3062 
3063 /*
3064  * Kludge City..
3065  * - make nfs_bmap() essentially a no-op that does no translation
3066  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3067  *   (Maybe I could use the process's page mapping, but I was concerned that
3068  *    Kernel Write might not be enabled and also figured copyout() would do
3069  *    a lot more work than bcopy() and also it currently happens in the
3070  *    context of the swapper process (2).
3071  *
3072  * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3073  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
3074  */
3075 static int
3076 nfs_bmap(struct vop_bmap_args *ap)
3077 {
3078 	/* no token lock required */
3079 	if (ap->a_doffsetp != NULL)
3080 		*ap->a_doffsetp = ap->a_loffset;
3081 	if (ap->a_runp != NULL)
3082 		*ap->a_runp = 0;
3083 	if (ap->a_runb != NULL)
3084 		*ap->a_runb = 0;
3085 	return (0);
3086 }
3087 
3088 /*
3089  * Strategy routine.
3090  */
3091 static int
3092 nfs_strategy(struct vop_strategy_args *ap)
3093 {
3094 	struct bio *bio = ap->a_bio;
3095 	struct bio *nbio;
3096 	struct buf *bp __debugvar = bio->bio_buf;
3097 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3098 	struct thread *td;
3099 	int error;
3100 
3101 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
3102 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
3103 	KASSERT(BUF_REFCNT(bp) > 0,
3104 		("nfs_strategy: buffer %p not locked", bp));
3105 
3106 	if (bio->bio_flags & BIO_SYNC)
3107 		td = curthread;	/* XXX */
3108 	else
3109 		td = NULL;
3110 
3111 	lwkt_gettoken(&nmp->nm_token);
3112 
3113         /*
3114 	 * We probably don't need to push an nbio any more since no
3115 	 * block conversion is required due to the use of 64 bit byte
3116 	 * offsets, but do it anyway.
3117 	 *
3118 	 * NOTE: When NFS callers itself via this strategy routines and
3119 	 *	 sets up a synchronous I/O, it expects the I/O to run
3120 	 *	 synchronously (its bio_done routine just assumes it),
3121 	 *	 so for now we have to honor the bit.
3122          */
3123 	nbio = push_bio(bio);
3124 	nbio->bio_offset = bio->bio_offset;
3125 	nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3126 
3127 	/*
3128 	 * If the op is asynchronous and an i/o daemon is waiting
3129 	 * queue the request, wake it up and wait for completion
3130 	 * otherwise just do it ourselves.
3131 	 */
3132 	if (bio->bio_flags & BIO_SYNC) {
3133 		error = nfs_doio(ap->a_vp, nbio, td);
3134 	} else {
3135 		nfs_asyncio(ap->a_vp, nbio);
3136 		error = 0;
3137 	}
3138 	lwkt_reltoken(&nmp->nm_token);
3139 
3140 	return (error);
3141 }
3142 
3143 /*
3144  * Mmap a file
3145  *
3146  * NB Currently unsupported.
3147  *
3148  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3149  */
3150 /* ARGSUSED */
3151 static int
3152 nfs_mmap(struct vop_mmap_args *ap)
3153 {
3154 	/* no token lock required */
3155 	return (EINVAL);
3156 }
3157 
3158 /*
3159  * fsync vnode op. Just call nfs_flush() with commit == 1.
3160  *
3161  * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3162  */
3163 /* ARGSUSED */
3164 static int
3165 nfs_fsync(struct vop_fsync_args *ap)
3166 {
3167 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3168 	int error;
3169 
3170 	lwkt_gettoken(&nmp->nm_token);
3171 	error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3172 	lwkt_reltoken(&nmp->nm_token);
3173 
3174 	return error;
3175 }
3176 
3177 /*
3178  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
3179  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
3180  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
3181  * set the buffer contains data that has already been written to the server
3182  * and which now needs a commit RPC.
3183  *
3184  * If commit is 0 we only take one pass and only flush buffers containing new
3185  * dirty data.
3186  *
3187  * If commit is 1 we take two passes, issuing a commit RPC in the second
3188  * pass.
3189  *
3190  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3191  * to completely flush all pending data.
3192  *
3193  * Note that the RB_SCAN code properly handles the case where the
3194  * callback might block and directly or indirectly (another thread) cause
3195  * the RB tree to change.
3196  */
3197 
3198 #ifndef NFS_COMMITBVECSIZ
3199 #define NFS_COMMITBVECSIZ	16
3200 #endif
3201 
3202 struct nfs_flush_info {
3203 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3204 	struct thread *td;
3205 	struct vnode *vp;
3206 	int waitfor;
3207 	int slpflag;
3208 	int slptimeo;
3209 	int loops;
3210 	struct buf *bvary[NFS_COMMITBVECSIZ];
3211 	int bvsize;
3212 	off_t beg_off;
3213 	off_t end_off;
3214 };
3215 
3216 static int nfs_flush_bp(struct buf *bp, void *data);
3217 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3218 
3219 int
3220 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3221 {
3222 	struct nfsnode *np = VTONFS(vp);
3223 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3224 	struct nfs_flush_info info;
3225 	int error;
3226 
3227 	bzero(&info, sizeof(info));
3228 	info.td = td;
3229 	info.vp = vp;
3230 	info.waitfor = waitfor;
3231 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3232 	info.loops = 0;
3233 	lwkt_gettoken(&vp->v_token);
3234 
3235 	do {
3236 		/*
3237 		 * Flush mode
3238 		 */
3239 		info.mode = NFI_FLUSHNEW;
3240 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3241 				nfs_flush_bp, &info);
3242 
3243 		/*
3244 		 * Take a second pass if committing and no error occured.
3245 		 * Clean up any left over collection (whether an error
3246 		 * occurs or not).
3247 		 */
3248 		if (commit && error == 0) {
3249 			info.mode = NFI_COMMIT;
3250 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3251 					nfs_flush_bp, &info);
3252 			if (info.bvsize)
3253 				error = nfs_flush_docommit(&info, error);
3254 		}
3255 
3256 		/*
3257 		 * Wait for pending I/O to complete before checking whether
3258 		 * any further dirty buffers exist.
3259 		 */
3260 		while (waitfor == MNT_WAIT &&
3261 		       bio_track_active(&vp->v_track_write)) {
3262 			error = bio_track_wait(&vp->v_track_write,
3263 					       info.slpflag, info.slptimeo);
3264 			if (error) {
3265 				/*
3266 				 * We have to be able to break out if this
3267 				 * is an 'intr' mount.
3268 				 */
3269 				if (nfs_sigintr(nmp, NULL, td)) {
3270 					error = -EINTR;
3271 					break;
3272 				}
3273 
3274 				/*
3275 				 * Since we do not process pending signals,
3276 				 * once we get a PCATCH our tsleep() will no
3277 				 * longer sleep, switch to a fixed timeout
3278 				 * instead.
3279 				 */
3280 				if (info.slpflag == PCATCH) {
3281 					info.slpflag = 0;
3282 					info.slptimeo = 2 * hz;
3283 				}
3284 				error = 0;
3285 			}
3286 		}
3287 		++info.loops;
3288 		/*
3289 		 * Loop if we are flushing synchronous as well as committing,
3290 		 * and dirty buffers are still present.  Otherwise we might livelock.
3291 		 */
3292 	} while (waitfor == MNT_WAIT && commit &&
3293 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3294 
3295 	/*
3296 	 * The callbacks have to return a negative error to terminate the
3297 	 * RB scan.
3298 	 */
3299 	if (error < 0)
3300 		error = -error;
3301 
3302 	/*
3303 	 * Deal with any error collection
3304 	 */
3305 	if (np->n_flag & NWRITEERR) {
3306 		error = np->n_error;
3307 		np->n_flag &= ~NWRITEERR;
3308 	}
3309 	lwkt_reltoken(&vp->v_token);
3310 	return (error);
3311 }
3312 
3313 static
3314 int
3315 nfs_flush_bp(struct buf *bp, void *data)
3316 {
3317 	struct nfs_flush_info *info = data;
3318 	int lkflags;
3319 	int error;
3320 	off_t toff;
3321 
3322 	error = 0;
3323 	switch(info->mode) {
3324 	case NFI_FLUSHNEW:
3325 		error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3326 		if (error && info->loops && info->waitfor == MNT_WAIT) {
3327 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3328 			if (error) {
3329 				lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3330 				if (info->slpflag & PCATCH)
3331 					lkflags |= LK_PCATCH;
3332 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3333 						     info->slptimeo);
3334 			}
3335 		}
3336 
3337 		/*
3338 		 * Ignore locking errors
3339 		 */
3340 		if (error) {
3341 			error = 0;
3342 			break;
3343 		}
3344 
3345 		/*
3346 		 * The buffer may have changed out from under us, even if
3347 		 * we did not block (MPSAFE).  Check again now that it is
3348 		 * locked.
3349 		 */
3350 		if (bp->b_vp == info->vp &&
3351 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3352 			bremfree(bp);
3353 			bawrite(bp);
3354 		} else {
3355 			BUF_UNLOCK(bp);
3356 		}
3357 		break;
3358 	case NFI_COMMIT:
3359 		/*
3360 		 * Only process buffers in need of a commit which we can
3361 		 * immediately lock.  This may prevent a buffer from being
3362 		 * committed, but the normal flush loop will block on the
3363 		 * same buffer so we shouldn't get into an endless loop.
3364 		 */
3365 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3366 		    (B_DELWRI | B_NEEDCOMMIT)) {
3367 			break;
3368 		}
3369 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3370 			break;
3371 
3372 		/*
3373 		 * We must recheck after successfully locking the buffer.
3374 		 */
3375 		if (bp->b_vp != info->vp ||
3376 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3377 		    (B_DELWRI | B_NEEDCOMMIT)) {
3378 			BUF_UNLOCK(bp);
3379 			break;
3380 		}
3381 
3382 		/*
3383 		 * NOTE: storing the bp in the bvary[] basically sets
3384 		 * it up for a commit operation.
3385 		 *
3386 		 * We must call vfs_busy_pages() now so the commit operation
3387 		 * is interlocked with user modifications to memory mapped
3388 		 * pages.  The b_dirtyoff/b_dirtyend range is not correct
3389 		 * until after the pages have been busied.
3390 		 *
3391 		 * Note: to avoid loopback deadlocks, we do not
3392 		 * assign b_runningbufspace.
3393 		 */
3394 		bremfree(bp);
3395 		bp->b_cmd = BUF_CMD_WRITE;
3396 		vfs_busy_pages(bp->b_vp, bp);
3397 		info->bvary[info->bvsize] = bp;
3398 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3399 		if (info->bvsize == 0 || toff < info->beg_off)
3400 			info->beg_off = toff;
3401 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3402 		if (info->bvsize == 0 || toff > info->end_off)
3403 			info->end_off = toff;
3404 		++info->bvsize;
3405 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3406 			error = nfs_flush_docommit(info, 0);
3407 			KKASSERT(info->bvsize == 0);
3408 		}
3409 	}
3410 	return (error);
3411 }
3412 
3413 static
3414 int
3415 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3416 {
3417 	struct vnode *vp;
3418 	struct buf *bp;
3419 	off_t bytes;
3420 	int retv;
3421 	int i;
3422 
3423 	vp = info->vp;
3424 
3425 	if (info->bvsize > 0) {
3426 		/*
3427 		 * Commit data on the server, as required.  Note that
3428 		 * nfs_commit will use the vnode's cred for the commit.
3429 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3430 		 */
3431 		bytes = info->end_off - info->beg_off;
3432 		if (bytes > 0x40000000)
3433 			bytes = 0x40000000;
3434 		if (error) {
3435 			retv = -error;
3436 		} else {
3437 			retv = nfs_commitrpc_uio(vp, info->beg_off,
3438 						 (int)bytes, info->td);
3439 			if (retv == NFSERR_STALEWRITEVERF)
3440 				nfs_clearcommit(vp->v_mount);
3441 		}
3442 
3443 		/*
3444 		 * Now, either mark the blocks I/O done or mark the
3445 		 * blocks dirty, depending on whether the commit
3446 		 * succeeded.
3447 		 */
3448 		for (i = 0; i < info->bvsize; ++i) {
3449 			bp = info->bvary[i];
3450 			if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3451 				/*
3452 				 * Either an error or the original
3453 				 * vfs_busy_pages() cleared B_NEEDCOMMIT
3454 				 * due to finding new dirty VM pages in
3455 				 * the buffer.
3456 				 *
3457 				 * Leave B_DELWRI intact.
3458 				 */
3459 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3460 				vfs_unbusy_pages(bp);
3461 				bp->b_cmd = BUF_CMD_DONE;
3462 				bqrelse(bp);
3463 			} else {
3464 				/*
3465 				 * Success, remove B_DELWRI ( bundirty() ).
3466 				 *
3467 				 * b_dirtyoff/b_dirtyend seem to be NFS
3468 				 * specific.  We should probably move that
3469 				 * into bundirty(). XXX
3470 				 *
3471 				 * We are faking an I/O write, we have to
3472 				 * start the transaction in order to
3473 				 * immediately biodone() it.
3474 				 */
3475 				bundirty(bp);
3476 				bp->b_flags &= ~B_ERROR;
3477 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3478 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3479 				biodone(&bp->b_bio1);
3480 			}
3481 		}
3482 		info->bvsize = 0;
3483 	}
3484 	return (error);
3485 }
3486 
3487 /*
3488  * NFS advisory byte-level locks.
3489  * Currently unsupported.
3490  *
3491  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3492  *		int a_flags)
3493  */
3494 static int
3495 nfs_advlock(struct vop_advlock_args *ap)
3496 {
3497 	struct nfsnode *np = VTONFS(ap->a_vp);
3498 
3499 	/* no token lock currently required */
3500 	/*
3501 	 * The following kludge is to allow diskless support to work
3502 	 * until a real NFS lockd is implemented. Basically, just pretend
3503 	 * that this is a local lock.
3504 	 */
3505 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3506 }
3507 
3508 /*
3509  * Print out the contents of an nfsnode.
3510  *
3511  * nfs_print(struct vnode *a_vp)
3512  */
3513 static int
3514 nfs_print(struct vop_print_args *ap)
3515 {
3516 	struct vnode *vp = ap->a_vp;
3517 	struct nfsnode *np = VTONFS(vp);
3518 
3519 	kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3520 		(long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3521 	if (vp->v_type == VFIFO)
3522 		fifo_printinfo(vp);
3523 	kprintf("\n");
3524 	return (0);
3525 }
3526 
3527 /*
3528  * nfs special file access vnode op.
3529  *
3530  * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3531  */
3532 static int
3533 nfs_laccess(struct vop_access_args *ap)
3534 {
3535 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3536 	struct vattr vattr;
3537 	int error;
3538 
3539 	lwkt_gettoken(&nmp->nm_token);
3540 	error = VOP_GETATTR(ap->a_vp, &vattr);
3541 	if (error == 0) {
3542 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3543 					  vattr.va_mode, 0);
3544 	}
3545 	lwkt_reltoken(&nmp->nm_token);
3546 
3547 	return (error);
3548 }
3549 
3550 /*
3551  * Read wrapper for fifos.
3552  *
3553  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3554  *		struct ucred *a_cred)
3555  */
3556 static int
3557 nfsfifo_read(struct vop_read_args *ap)
3558 {
3559 	struct nfsnode *np = VTONFS(ap->a_vp);
3560 
3561 	/* no token access required */
3562 	/*
3563 	 * Set access flag.
3564 	 */
3565 	np->n_flag |= NACC;
3566 	getnanotime(&np->n_atim);
3567 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3568 }
3569 
3570 /*
3571  * Write wrapper for fifos.
3572  *
3573  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3574  *		 struct ucred *a_cred)
3575  */
3576 static int
3577 nfsfifo_write(struct vop_write_args *ap)
3578 {
3579 	struct nfsnode *np = VTONFS(ap->a_vp);
3580 
3581 	/* no token access required */
3582 	/*
3583 	 * Set update flag.
3584 	 */
3585 	np->n_flag |= NUPD;
3586 	getnanotime(&np->n_mtim);
3587 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3588 }
3589 
3590 /*
3591  * Close wrapper for fifos.
3592  *
3593  * Update the times on the nfsnode then do fifo close.
3594  *
3595  * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3596  */
3597 static int
3598 nfsfifo_close(struct vop_close_args *ap)
3599 {
3600 	struct vnode *vp = ap->a_vp;
3601 	struct nfsnode *np = VTONFS(vp);
3602 	struct vattr vattr;
3603 	struct timespec ts;
3604 
3605 	/* no token access required */
3606 
3607 	if (np->n_flag & (NACC | NUPD)) {
3608 		getnanotime(&ts);
3609 		if (np->n_flag & NACC)
3610 			np->n_atim = ts;
3611 		if (np->n_flag & NUPD)
3612 			np->n_mtim = ts;
3613 		np->n_flag |= NCHG;
3614 		if (vp->v_sysref.refcnt == 1 &&
3615 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3616 			VATTR_NULL(&vattr);
3617 			if (np->n_flag & NACC)
3618 				vattr.va_atime = np->n_atim;
3619 			if (np->n_flag & NUPD)
3620 				vattr.va_mtime = np->n_mtim;
3621 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3622 		}
3623 	}
3624 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3625 }
3626 
3627