xref: /dflybsd-src/sys/vfs/nfs/nfs_vnops.c (revision a563ca70e68142ccf7f50a6f129665fd8cb66d98)
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 
71 #include <sys/buf2.h>
72 
73 #include <vfs/fifofs/fifo.h>
74 #include <vfs/ufs/dir.h>
75 
76 #undef DIRBLKSIZ
77 
78 #include "rpcv2.h"
79 #include "nfsproto.h"
80 #include "nfs.h"
81 #include "nfsmount.h"
82 #include "nfsnode.h"
83 #include "xdr_subs.h"
84 #include "nfsm_subs.h"
85 
86 #include <net/if.h>
87 #include <netinet/in.h>
88 #include <netinet/in_var.h>
89 
90 #include <sys/thread2.h>
91 
92 /* Defs */
93 #define	TRUE	1
94 #define	FALSE	0
95 
96 static int	nfsfifo_read (struct vop_read_args *);
97 static int	nfsfifo_write (struct vop_write_args *);
98 static int	nfsfifo_close (struct vop_close_args *);
99 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
100 static	int	nfs_lookup (struct vop_old_lookup_args *);
101 static	int	nfs_create (struct vop_old_create_args *);
102 static	int	nfs_mknod (struct vop_old_mknod_args *);
103 static	int	nfs_open (struct vop_open_args *);
104 static	int	nfs_close (struct vop_close_args *);
105 static	int	nfs_access (struct vop_access_args *);
106 static	int	nfs_getattr (struct vop_getattr_args *);
107 static	int	nfs_setattr (struct vop_setattr_args *);
108 static	int	nfs_read (struct vop_read_args *);
109 static	int	nfs_mmap (struct vop_mmap_args *);
110 static	int	nfs_fsync (struct vop_fsync_args *);
111 static	int	nfs_remove (struct vop_old_remove_args *);
112 static	int	nfs_link (struct vop_old_link_args *);
113 static	int	nfs_rename (struct vop_old_rename_args *);
114 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
115 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
116 static	int	nfs_symlink (struct vop_old_symlink_args *);
117 static	int	nfs_readdir (struct vop_readdir_args *);
118 static	int	nfs_bmap (struct vop_bmap_args *);
119 static	int	nfs_strategy (struct vop_strategy_args *);
120 static	int	nfs_lookitup (struct vnode *, const char *, int,
121 			struct ucred *, struct thread *, struct nfsnode **);
122 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
123 static int	nfs_laccess (struct vop_access_args *);
124 static int	nfs_readlink (struct vop_readlink_args *);
125 static int	nfs_print (struct vop_print_args *);
126 static int	nfs_advlock (struct vop_advlock_args *);
127 
128 static	int	nfs_nresolve (struct vop_nresolve_args *);
129 /*
130  * Global vfs data structures for nfs
131  */
132 struct vop_ops nfsv2_vnode_vops = {
133 	.vop_default =		vop_defaultop,
134 	.vop_access =		nfs_access,
135 	.vop_advlock =		nfs_advlock,
136 	.vop_bmap =		nfs_bmap,
137 	.vop_close =		nfs_close,
138 	.vop_old_create =	nfs_create,
139 	.vop_fsync =		nfs_fsync,
140 	.vop_getattr =		nfs_getattr,
141 	.vop_getpages =		vop_stdgetpages,
142 	.vop_putpages =		vop_stdputpages,
143 	.vop_inactive =		nfs_inactive,
144 	.vop_old_link =		nfs_link,
145 	.vop_old_lookup =	nfs_lookup,
146 	.vop_old_mkdir =	nfs_mkdir,
147 	.vop_old_mknod =	nfs_mknod,
148 	.vop_mmap =		nfs_mmap,
149 	.vop_open =		nfs_open,
150 	.vop_print =		nfs_print,
151 	.vop_read =		nfs_read,
152 	.vop_readdir =		nfs_readdir,
153 	.vop_readlink =		nfs_readlink,
154 	.vop_reclaim =		nfs_reclaim,
155 	.vop_old_remove =	nfs_remove,
156 	.vop_old_rename =	nfs_rename,
157 	.vop_old_rmdir =	nfs_rmdir,
158 	.vop_setattr =		nfs_setattr,
159 	.vop_strategy =		nfs_strategy,
160 	.vop_old_symlink =	nfs_symlink,
161 	.vop_write =		nfs_write,
162 	.vop_nresolve =		nfs_nresolve
163 };
164 
165 /*
166  * Special device vnode ops
167  */
168 struct vop_ops nfsv2_spec_vops = {
169 	.vop_default =		vop_defaultop,
170 	.vop_access =		nfs_laccess,
171 	.vop_close =		nfs_close,
172 	.vop_fsync =		nfs_fsync,
173 	.vop_getattr =		nfs_getattr,
174 	.vop_inactive =		nfs_inactive,
175 	.vop_print =		nfs_print,
176 	.vop_read =		vop_stdnoread,
177 	.vop_reclaim =		nfs_reclaim,
178 	.vop_setattr =		nfs_setattr,
179 	.vop_write =		vop_stdnowrite
180 };
181 
182 struct vop_ops nfsv2_fifo_vops = {
183 	.vop_default =		fifo_vnoperate,
184 	.vop_access =		nfs_laccess,
185 	.vop_close =		nfsfifo_close,
186 	.vop_fsync =		nfs_fsync,
187 	.vop_getattr =		nfs_getattr,
188 	.vop_inactive =		nfs_inactive,
189 	.vop_print =		nfs_print,
190 	.vop_read =		nfsfifo_read,
191 	.vop_reclaim =		nfs_reclaim,
192 	.vop_setattr =		nfs_setattr,
193 	.vop_write =		nfsfifo_write
194 };
195 
196 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
197 				  struct componentname *cnp,
198 				  struct vattr *vap);
199 static int	nfs_removerpc (struct vnode *dvp, const char *name,
200 				   int namelen,
201 				   struct ucred *cred, struct thread *td);
202 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
203 				   int fnamelen, struct vnode *tdvp,
204 				   const char *tnameptr, int tnamelen,
205 				   struct ucred *cred, struct thread *td);
206 static int	nfs_renameit (struct vnode *sdvp,
207 				  struct componentname *scnp,
208 				  struct sillyrename *sp);
209 
210 SYSCTL_DECL(_vfs_nfs);
211 
212 static int nfs_flush_on_rename = 1;
213 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
214 	   &nfs_flush_on_rename, 0, "flush fvp prior to rename");
215 static int nfs_flush_on_hlink = 0;
216 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
217 	   &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
218 
219 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
220 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
221 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
222 
223 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
224 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
225 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
226 
227 static int	nfspos_cache_timeout = NFS_MINATTRTIMO;
228 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
229 	   &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
230 
231 static int	nfsv3_commit_on_close = 0;
232 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
233 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
234 #if 0
235 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
236 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
237 
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
239 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
240 #endif
241 
242 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
243 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
244 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
245 
246 /*
247  * Returns whether a name component is a degenerate '.' or '..'.
248  */
249 static __inline
250 int
251 nlcdegenerate(struct nlcomponent *nlc)
252 {
253 	if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
254 		return(1);
255 	if (nlc->nlc_namelen == 2 &&
256 	    nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
257 		return(1);
258 	return(0);
259 }
260 
261 static int
262 nfs3_access_otw(struct vnode *vp, int wmode,
263 		struct thread *td, struct ucred *cred)
264 {
265 	struct nfsnode *np = VTONFS(vp);
266 	int attrflag;
267 	int error = 0;
268 	u_int32_t *tl;
269 	u_int32_t rmode;
270 	struct nfsm_info info;
271 
272 	info.mrep = NULL;
273 	info.v3 = 1;
274 
275 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
276 	nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
277 		     NFSX_FH(info.v3) + NFSX_UNSIGNED);
278 	ERROROUT(nfsm_fhtom(&info, vp));
279 	tl = nfsm_build(&info, NFSX_UNSIGNED);
280 	*tl = txdr_unsigned(wmode);
281 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
282 	ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
283 	if (error == 0) {
284 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
285 		rmode = fxdr_unsigned(u_int32_t, *tl);
286 		np->n_mode = rmode;
287 		np->n_modeuid = cred->cr_uid;
288 		np->n_modestamp = mycpu->gd_time_seconds;
289 	}
290 	m_freem(info.mrep);
291 	info.mrep = NULL;
292 nfsmout:
293 	return error;
294 }
295 
296 /*
297  * nfs access vnode op.
298  * For nfs version 2, just return ok. File accesses may fail later.
299  * For nfs version 3, use the access rpc to check accessibility. If file modes
300  * are changed on the server, accesses might still fail later.
301  *
302  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
303  */
304 static int
305 nfs_access(struct vop_access_args *ap)
306 {
307 	struct ucred *cred;
308 	struct vnode *vp = ap->a_vp;
309 	thread_t td = curthread;
310 	int error = 0;
311 	u_int32_t mode, wmode;
312 	struct nfsnode *np = VTONFS(vp);
313 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
314 	int v3 = NFS_ISV3(vp);
315 
316 	lwkt_gettoken(&nmp->nm_token);
317 
318 	/*
319 	 * Disallow write attempts on filesystems mounted read-only;
320 	 * unless the file is a socket, fifo, or a block or character
321 	 * device resident on the filesystem.
322 	 */
323 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
324 		switch (vp->v_type) {
325 		case VREG:
326 		case VDIR:
327 		case VLNK:
328 			lwkt_reltoken(&nmp->nm_token);
329 			return (EROFS);
330 		default:
331 			break;
332 		}
333 	}
334 
335 	/*
336 	 * The NFS protocol passes only the effective uid/gid over the wire but
337 	 * we need to check access against real ids if AT_EACCESS not set.
338 	 * Handle this case by cloning the credentials and setting the
339 	 * effective ids to the real ones.
340 	 */
341 	if (ap->a_flags & AT_EACCESS) {
342 		cred = crhold(ap->a_cred);
343 	} else {
344 		cred = crdup(ap->a_cred);
345 		cred->cr_uid = cred->cr_ruid;
346 		cred->cr_gid = cred->cr_rgid;
347 	}
348 
349 	/*
350 	 * For nfs v3, check to see if we have done this recently, and if
351 	 * so return our cached result instead of making an ACCESS call.
352 	 * If not, do an access rpc, otherwise you are stuck emulating
353 	 * ufs_access() locally using the vattr. This may not be correct,
354 	 * since the server may apply other access criteria such as
355 	 * client uid-->server uid mapping that we do not know about.
356 	 */
357 	if (v3) {
358 		if (ap->a_mode & VREAD)
359 			mode = NFSV3ACCESS_READ;
360 		else
361 			mode = 0;
362 		if (vp->v_type != VDIR) {
363 			if (ap->a_mode & VWRITE)
364 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
365 			if (ap->a_mode & VEXEC)
366 				mode |= NFSV3ACCESS_EXECUTE;
367 		} else {
368 			if (ap->a_mode & VWRITE)
369 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
370 					 NFSV3ACCESS_DELETE);
371 			if (ap->a_mode & VEXEC)
372 				mode |= NFSV3ACCESS_LOOKUP;
373 		}
374 		/* XXX safety belt, only make blanket request if caching */
375 		if (nfsaccess_cache_timeout > 0) {
376 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
377 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
378 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
379 		} else {
380 			wmode = mode;
381 		}
382 
383 		/*
384 		 * Does our cached result allow us to give a definite yes to
385 		 * this request?
386 		 */
387 		if (np->n_modestamp &&
388 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
389 		   (cred->cr_uid == np->n_modeuid) &&
390 		   ((np->n_mode & mode) == mode)) {
391 			nfsstats.accesscache_hits++;
392 		} else {
393 			/*
394 			 * Either a no, or a don't know.  Go to the wire.
395 			 */
396 			nfsstats.accesscache_misses++;
397 		        error = nfs3_access_otw(vp, wmode, td, cred);
398 			if (!error) {
399 				if ((np->n_mode & mode) != mode) {
400 					error = EACCES;
401 				}
402 			}
403 		}
404 	} else {
405 		if ((error = nfs_laccess(ap)) != 0) {
406 			crfree(cred);
407 			lwkt_reltoken(&nmp->nm_token);
408 			return (error);
409 		}
410 
411 		/*
412 		 * Attempt to prevent a mapped root from accessing a file
413 		 * which it shouldn't.  We try to read a byte from the file
414 		 * if the user is root and the file is not zero length.
415 		 * After calling nfs_laccess, we should have the correct
416 		 * file size cached.
417 		 */
418 		if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
419 		    && VTONFS(vp)->n_size > 0) {
420 			struct iovec aiov;
421 			struct uio auio;
422 			char buf[1];
423 
424 			aiov.iov_base = buf;
425 			aiov.iov_len = 1;
426 			auio.uio_iov = &aiov;
427 			auio.uio_iovcnt = 1;
428 			auio.uio_offset = 0;
429 			auio.uio_resid = 1;
430 			auio.uio_segflg = UIO_SYSSPACE;
431 			auio.uio_rw = UIO_READ;
432 			auio.uio_td = td;
433 
434 			if (vp->v_type == VREG) {
435 				error = nfs_readrpc_uio(vp, &auio);
436 			} else if (vp->v_type == VDIR) {
437 				char* bp;
438 				bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
439 				aiov.iov_base = bp;
440 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
441 				error = nfs_readdirrpc_uio(vp, &auio);
442 				kfree(bp, M_TEMP);
443 			} else if (vp->v_type == VLNK) {
444 				error = nfs_readlinkrpc_uio(vp, &auio);
445 			} else {
446 				error = EACCES;
447 			}
448 		}
449 	}
450 	/*
451 	 * [re]record creds for reading and/or writing if access
452 	 * was granted.  Assume the NFS server will grant read access
453 	 * for execute requests.
454 	 */
455 	if (error == 0) {
456 		if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
457 			crhold(cred);
458 			if (np->n_rucred)
459 				crfree(np->n_rucred);
460 			np->n_rucred = cred;
461 		}
462 		if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
463 			crhold(cred);
464 			if (np->n_wucred)
465 				crfree(np->n_wucred);
466 			np->n_wucred = cred;
467 		}
468 	}
469 	lwkt_reltoken(&nmp->nm_token);
470 	crfree(cred);
471 	return(error);
472 }
473 
474 /*
475  * nfs open vnode op
476  * Check to see if the type is ok
477  * and that deletion is not in progress.
478  * For paged in text files, you will need to flush the page cache
479  * if consistency is lost.
480  *
481  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
482  *	    struct file *a_fp)
483  */
484 /* ARGSUSED */
485 static int
486 nfs_open(struct vop_open_args *ap)
487 {
488 	struct vnode *vp = ap->a_vp;
489 	struct nfsnode *np = VTONFS(vp);
490 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
491 	struct vattr vattr;
492 	int error;
493 
494 	lwkt_gettoken(&nmp->nm_token);
495 
496 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
497 #ifdef DIAGNOSTIC
498 		kprintf("open eacces vtyp=%d\n",vp->v_type);
499 #endif
500 		lwkt_reltoken(&nmp->nm_token);
501 		return (EOPNOTSUPP);
502 	}
503 
504 	/*
505 	 * Save valid creds for reading and writing for later RPCs.
506 	 */
507 	if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
508 		crhold(ap->a_cred);
509 		if (np->n_rucred)
510 			crfree(np->n_rucred);
511 		np->n_rucred = ap->a_cred;
512 	}
513 	if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
514 		crhold(ap->a_cred);
515 		if (np->n_wucred)
516 			crfree(np->n_wucred);
517 		np->n_wucred = ap->a_cred;
518 	}
519 
520 	/*
521 	 * Clear the attribute cache only if opening with write access.  It
522 	 * is unclear if we should do this at all here, but we certainly
523 	 * should not clear the cache unconditionally simply because a file
524 	 * is being opened.
525 	 */
526 	if (ap->a_mode & FWRITE)
527 		np->n_attrstamp = 0;
528 
529 	/*
530 	 * For normal NFS, reconcile changes made locally verses
531 	 * changes made remotely.  Note that VOP_GETATTR only goes
532 	 * to the wire if the cached attribute has timed out or been
533 	 * cleared.
534 	 *
535 	 * If local modifications have been made clear the attribute
536 	 * cache to force an attribute and modified time check.  If
537 	 * GETATTR detects that the file has been changed by someone
538 	 * other then us it will set NRMODIFIED.
539 	 *
540 	 * If we are opening a directory and local changes have been
541 	 * made we have to invalidate the cache in order to ensure
542 	 * that we get the most up-to-date information from the
543 	 * server.  XXX
544 	 */
545 	if (np->n_flag & NLMODIFIED) {
546 		np->n_attrstamp = 0;
547 		if (vp->v_type == VDIR) {
548 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
549 			if (error == EINTR)
550 				return (error);
551 			nfs_invaldir(vp);
552 		}
553 	}
554 	error = VOP_GETATTR(vp, &vattr);
555 	if (error) {
556 		lwkt_reltoken(&nmp->nm_token);
557 		return (error);
558 	}
559 	if (np->n_flag & NRMODIFIED) {
560 		if (vp->v_type == VDIR)
561 			nfs_invaldir(vp);
562 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
563 		if (error == EINTR) {
564 			lwkt_reltoken(&nmp->nm_token);
565 			return (error);
566 		}
567 		np->n_flag &= ~NRMODIFIED;
568 	}
569 	error = vop_stdopen(ap);
570 	lwkt_reltoken(&nmp->nm_token);
571 
572 	return error;
573 }
574 
575 /*
576  * nfs close vnode op
577  * What an NFS client should do upon close after writing is a debatable issue.
578  * Most NFS clients push delayed writes to the server upon close, basically for
579  * two reasons:
580  * 1 - So that any write errors may be reported back to the client process
581  *     doing the close system call. By far the two most likely errors are
582  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
583  * 2 - To put a worst case upper bound on cache inconsistency between
584  *     multiple clients for the file.
585  * There is also a consistency problem for Version 2 of the protocol w.r.t.
586  * not being able to tell if other clients are writing a file concurrently,
587  * since there is no way of knowing if the changed modify time in the reply
588  * is only due to the write for this client.
589  * (NFS Version 3 provides weak cache consistency data in the reply that
590  *  should be sufficient to detect and handle this case.)
591  *
592  * The current code does the following:
593  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
594  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
595  *                     or commit them (this satisfies 1 and 2 except for the
596  *                     case where the server crashes after this close but
597  *                     before the commit RPC, which is felt to be "good
598  *                     enough". Changing the last argument to nfs_flush() to
599  *                     a 1 would force a commit operation, if it is felt a
600  *                     commit is necessary now.
601  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
602  *                     1 should be dealt with via an fsync() system call for
603  *                     cases where write errors are important.
604  *
605  * nfs_close(struct vnode *a_vp, int a_fflag)
606  */
607 /* ARGSUSED */
608 static int
609 nfs_close(struct vop_close_args *ap)
610 {
611 	struct vnode *vp = ap->a_vp;
612 	struct nfsnode *np = VTONFS(vp);
613 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
614 	int error = 0;
615 	thread_t td = curthread;
616 
617 	lwkt_gettoken(&nmp->nm_token);
618 
619 	if (vp->v_type == VREG) {
620 	    if (np->n_flag & NLMODIFIED) {
621 		if (NFS_ISV3(vp)) {
622 		    /*
623 		     * Under NFSv3 we have dirty buffers to dispose of.  We
624 		     * must flush them to the NFS server.  We have the option
625 		     * of waiting all the way through the commit rpc or just
626 		     * waiting for the initial write.  The default is to only
627 		     * wait through the initial write so the data is in the
628 		     * server's cache, which is roughly similar to the state
629 		     * a standard disk subsystem leaves the file in on close().
630 		     *
631 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
632 		     * potential races with other processes, and certainly
633 		     * cannot clear it if we don't commit.
634 		     */
635 		    int cm = nfsv3_commit_on_close ? 1 : 0;
636 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
637 		    /* np->n_flag &= ~NLMODIFIED; */
638 		} else {
639 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
640 		}
641 		np->n_attrstamp = 0;
642 	    }
643 	    if (np->n_flag & NWRITEERR) {
644 		np->n_flag &= ~NWRITEERR;
645 		error = np->n_error;
646 	    }
647 	}
648 	vop_stdclose(ap);
649 	lwkt_reltoken(&nmp->nm_token);
650 
651 	return (error);
652 }
653 
654 /*
655  * nfs getattr call from vfs.
656  *
657  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
658  */
659 static int
660 nfs_getattr(struct vop_getattr_args *ap)
661 {
662 	struct vnode *vp = ap->a_vp;
663 	struct nfsnode *np = VTONFS(vp);
664 	struct nfsmount *nmp;
665 	int error = 0;
666 	thread_t td = curthread;
667 	struct nfsm_info info;
668 
669 	info.mrep = NULL;
670 	info.v3 = NFS_ISV3(vp);
671 	nmp = VFSTONFS(vp->v_mount);
672 
673 	lwkt_gettoken(&nmp->nm_token);
674 
675 	/*
676 	 * Update local times for special files.
677 	 */
678 	if (np->n_flag & (NACC | NUPD))
679 		np->n_flag |= NCHG;
680 	/*
681 	 * First look in the cache.
682 	 */
683 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
684 		goto done;
685 
686 	if (info.v3 && nfsaccess_cache_timeout > 0) {
687 		nfsstats.accesscache_misses++;
688 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
689 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
690 			goto done;
691 	}
692 
693 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
694 	nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
695 	ERROROUT(nfsm_fhtom(&info, vp));
696 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
697 				nfs_vpcred(vp, ND_CHECK), &error));
698 	if (error == 0) {
699 		ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
700 	}
701 	m_freem(info.mrep);
702 	info.mrep = NULL;
703 done:
704 	/*
705 	 * NFS doesn't support chflags flags.  If the nfs mount was
706 	 * made -o cache set the UF_CACHE bit for swapcache.
707 	 */
708 	if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
709 		ap->a_vap->va_flags |= UF_CACHE;
710 nfsmout:
711 	lwkt_reltoken(&nmp->nm_token);
712 	return (error);
713 }
714 
715 /*
716  * nfs setattr call.
717  *
718  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
719  */
720 static int
721 nfs_setattr(struct vop_setattr_args *ap)
722 {
723 	struct vnode *vp = ap->a_vp;
724 	struct nfsnode *np = VTONFS(vp);
725 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
726 	struct vattr *vap = ap->a_vap;
727 	int biosize = vp->v_mount->mnt_stat.f_iosize;
728 	int error = 0;
729 	int boff;
730 	off_t tsize;
731 	thread_t td = curthread;
732 
733 #ifndef nolint
734 	tsize = (off_t)0;
735 #endif
736 	/*
737 	 * Setting of flags is not supported.
738 	 */
739 	if (vap->va_flags != VNOVAL)
740 		return (EOPNOTSUPP);
741 
742 	/*
743 	 * Disallow write attempts if the filesystem is mounted read-only.
744 	 */
745   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
746 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
747 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
748 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
749 		return (EROFS);
750 
751 	lwkt_gettoken(&nmp->nm_token);
752 
753 	if (vap->va_size != VNOVAL) {
754 		/*
755 		 * truncation requested
756 		 */
757  		switch (vp->v_type) {
758  		case VDIR:
759 			lwkt_reltoken(&nmp->nm_token);
760  			return (EISDIR);
761  		case VCHR:
762  		case VBLK:
763  		case VSOCK:
764  		case VFIFO:
765 			if (vap->va_mtime.tv_sec == VNOVAL &&
766 			    vap->va_atime.tv_sec == VNOVAL &&
767 			    vap->va_mode == (mode_t)VNOVAL &&
768 			    vap->va_uid == (uid_t)VNOVAL &&
769 			    vap->va_gid == (gid_t)VNOVAL) {
770 				lwkt_reltoken(&nmp->nm_token);
771 				return (0);
772 			}
773  			vap->va_size = VNOVAL;
774  			break;
775  		default:
776 			/*
777 			 * Disallow write attempts if the filesystem is
778 			 * mounted read-only.
779 			 */
780 			if (vp->v_mount->mnt_flag & MNT_RDONLY) {
781 				lwkt_reltoken(&nmp->nm_token);
782 				return (EROFS);
783 			}
784 
785 			tsize = np->n_size;
786 again:
787 			boff = (int)vap->va_size & (biosize - 1);
788 			error = nfs_meta_setsize(vp, td, vap->va_size, 0);
789 
790 #if 0
791  			if (np->n_flag & NLMODIFIED) {
792  			    if (vap->va_size == 0)
793  				error = nfs_vinvalbuf(vp, 0, 1);
794  			    else
795  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
796  			}
797 #endif
798 			/*
799 			 * note: this loop case almost always happens at
800 			 * least once per truncation.
801 			 */
802 			if (error == 0 && np->n_size != vap->va_size)
803 				goto again;
804 			np->n_vattr.va_size = vap->va_size;
805 			break;
806 		}
807 	} else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
808 		/*
809 		 * What to do.  If we are modifying the mtime we lose
810 		 * mtime detection of changes made by the server or other
811 		 * clients.  But programs like rsync/rdist/cpdup are going
812 		 * to call utimes a lot.  We don't want to piecemeal sync.
813 		 *
814 		 * For now sync if any prior remote changes were detected,
815 		 * but allow us to lose track of remote changes made during
816 		 * the utimes operation.
817 		 */
818 		if (np->n_flag & NRMODIFIED)
819 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
820 		if (error == EINTR)
821 			return (error);
822 		if (error == 0) {
823 			if (vap->va_mtime.tv_sec != VNOVAL) {
824 				np->n_mtime = vap->va_mtime.tv_sec;
825 			}
826 		}
827 	}
828 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
829 
830 	/*
831 	 * Sanity check if a truncation was issued.  This should only occur
832 	 * if multiple processes are racing on the same file.
833 	 */
834 	if (error == 0 && vap->va_size != VNOVAL &&
835 	    np->n_size != vap->va_size) {
836 		kprintf("NFS ftruncate: server disagrees on the file size: "
837 			"%jd/%jd/%jd\n",
838 			(intmax_t)tsize,
839 			(intmax_t)vap->va_size,
840 			(intmax_t)np->n_size);
841 		goto again;
842 	}
843 	if (error && vap->va_size != VNOVAL) {
844 		np->n_size = np->n_vattr.va_size = tsize;
845 		nfs_meta_setsize(vp, td, np->n_size, 0);
846 	}
847 	lwkt_reltoken(&nmp->nm_token);
848 
849 	return (error);
850 }
851 
852 /*
853  * Do an nfs setattr rpc.
854  */
855 static int
856 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
857 	       struct ucred *cred, struct thread *td)
858 {
859 	struct nfsv2_sattr *sp;
860 	struct nfsnode *np = VTONFS(vp);
861 	u_int32_t *tl;
862 	int error = 0, wccflag = NFSV3_WCCRATTR;
863 	struct nfsm_info info;
864 
865 	info.mrep = NULL;
866 	info.v3 = NFS_ISV3(vp);
867 
868 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
869 	nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
870 		     NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
871 	ERROROUT(nfsm_fhtom(&info, vp));
872 	if (info.v3) {
873 		nfsm_v3attrbuild(&info, vap, TRUE);
874 		tl = nfsm_build(&info, NFSX_UNSIGNED);
875 		*tl = nfs_false;
876 	} else {
877 		sp = nfsm_build(&info, NFSX_V2SATTR);
878 		if (vap->va_mode == (mode_t)VNOVAL)
879 			sp->sa_mode = nfs_xdrneg1;
880 		else
881 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
882 		if (vap->va_uid == (uid_t)VNOVAL)
883 			sp->sa_uid = nfs_xdrneg1;
884 		else
885 			sp->sa_uid = txdr_unsigned(vap->va_uid);
886 		if (vap->va_gid == (gid_t)VNOVAL)
887 			sp->sa_gid = nfs_xdrneg1;
888 		else
889 			sp->sa_gid = txdr_unsigned(vap->va_gid);
890 		sp->sa_size = txdr_unsigned(vap->va_size);
891 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
892 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
893 	}
894 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
895 	if (info.v3) {
896 		np->n_modestamp = 0;
897 		ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
898 	} else {
899 		ERROROUT(nfsm_loadattr(&info, vp, NULL));
900 	}
901 	m_freem(info.mrep);
902 	info.mrep = NULL;
903 nfsmout:
904 	return (error);
905 }
906 
907 static
908 void
909 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
910 {
911 	if (nctimeout == 0)
912 		nctimeout = 1;
913 	else
914 		nctimeout *= hz;
915 	cache_setvp(nch, vp);
916 	cache_settimeout(nch, nctimeout);
917 }
918 
919 /*
920  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
921  * nfs_lookup() until all remaining new api calls are implemented.
922  *
923  * Resolve a namecache entry.  This function is passed a locked ncp and
924  * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
925  */
926 static int
927 nfs_nresolve(struct vop_nresolve_args *ap)
928 {
929 	struct thread *td = curthread;
930 	struct namecache *ncp;
931 	struct nfsmount *nmp;
932 	struct ucred *cred;
933 	struct nfsnode *np;
934 	struct vnode *dvp;
935 	struct vnode *nvp;
936 	nfsfh_t *fhp;
937 	int attrflag;
938 	int fhsize;
939 	int error;
940 	int tmp_error;
941 	int len;
942 	struct nfsm_info info;
943 
944 	cred = ap->a_cred;
945 	dvp = ap->a_dvp;
946 	nmp = VFSTONFS(dvp->v_mount);
947 
948 	lwkt_gettoken(&nmp->nm_token);
949 
950 	if ((error = vget(dvp, LK_SHARED)) != 0) {
951 		lwkt_reltoken(&nmp->nm_token);
952 		return (error);
953 	}
954 
955 	info.mrep = NULL;
956 	info.v3 = NFS_ISV3(dvp);
957 
958 	nvp = NULL;
959 	nfsstats.lookupcache_misses++;
960 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
961 	ncp = ap->a_nch->ncp;
962 	len = ncp->nc_nlen;
963 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
964 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
965 	ERROROUT(nfsm_fhtom(&info, dvp));
966 	ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
967 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
968 				ap->a_cred, &error));
969 	if (error) {
970 		/*
971 		 * Cache negatve lookups to reduce NFS traffic, but use
972 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
973 		 * XXX we should add a namecache flag for no-caching
974 		 * to uncache the negative hit as soon as possible, but
975 		 * we cannot simply destroy the entry because it is used
976 		 * as a placeholder by the caller.
977 		 *
978 		 * The refactored nfs code will overwrite a non-zero error
979 		 * with 0 when we use ERROROUT(), so don't here.
980 		 */
981 		if (error == ENOENT)
982 			nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
983 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
984 					     NFS_LATTR_NOSHRINK);
985 		if (tmp_error) {
986 			error = tmp_error;
987 			goto nfsmout;
988 		}
989 		m_freem(info.mrep);
990 		info.mrep = NULL;
991 		goto nfsmout;
992 	}
993 
994 	/*
995 	 * Success, get the file handle, do various checks, and load
996 	 * post-operation data from the reply packet.  Theoretically
997 	 * we should never be looking up "." so, theoretically, we
998 	 * should never get the same file handle as our directory.  But
999 	 * we check anyway. XXX
1000 	 *
1001 	 * Note that no timeout is set for the positive cache hit.  We
1002 	 * assume, theoretically, that ESTALE returns will be dealt with
1003 	 * properly to handle NFS races and in anycase we cannot depend
1004 	 * on a timeout to deal with NFS open/create/excl issues so instead
1005 	 * of a bad hack here the rest of the NFS client code needs to do
1006 	 * the right thing.
1007 	 */
1008 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1009 
1010 	np = VTONFS(dvp);
1011 	if (NFS_CMPFH(np, fhp, fhsize)) {
1012 		vref(dvp);
1013 		nvp = dvp;
1014 	} else {
1015 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1016 		if (error) {
1017 			m_freem(info.mrep);
1018 			info.mrep = NULL;
1019 			vput(dvp);
1020 			lwkt_reltoken(&nmp->nm_token);
1021 			return (error);
1022 		}
1023 		nvp = NFSTOV(np);
1024 	}
1025 	if (info.v3) {
1026 		ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1027 					  NFS_LATTR_NOSHRINK));
1028 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1029 					  NFS_LATTR_NOSHRINK));
1030 	} else {
1031 		ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1032 	}
1033 	nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1034 	m_freem(info.mrep);
1035 	info.mrep = NULL;
1036 nfsmout:
1037 	lwkt_reltoken(&nmp->nm_token);
1038 	vput(dvp);
1039 	if (nvp) {
1040 		if (nvp == dvp)
1041 			vrele(nvp);
1042 		else
1043 			vput(nvp);
1044 	}
1045 	return (error);
1046 }
1047 
1048 /*
1049  * 'cached' nfs directory lookup
1050  *
1051  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1052  *
1053  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1054  *	      struct componentname *a_cnp)
1055  */
1056 static int
1057 nfs_lookup(struct vop_old_lookup_args *ap)
1058 {
1059 	struct componentname *cnp = ap->a_cnp;
1060 	struct vnode *dvp = ap->a_dvp;
1061 	struct vnode **vpp = ap->a_vpp;
1062 	int flags = cnp->cn_flags;
1063 	struct vnode *newvp;
1064 	struct nfsmount *nmp;
1065 	long len;
1066 	nfsfh_t *fhp;
1067 	struct nfsnode *np;
1068 	int lockparent, wantparent, attrflag, fhsize;
1069 	int error;
1070 	int tmp_error;
1071 	struct nfsm_info info;
1072 
1073 	info.mrep = NULL;
1074 	info.v3 = NFS_ISV3(dvp);
1075 	error = 0;
1076 
1077 	/*
1078 	 * Read-only mount check and directory check.
1079 	 */
1080 	*vpp = NULLVP;
1081 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1082 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1083 		return (EROFS);
1084 
1085 	if (dvp->v_type != VDIR)
1086 		return (ENOTDIR);
1087 
1088 	/*
1089 	 * Look it up in the cache.  Note that ENOENT is only returned if we
1090 	 * previously entered a negative hit (see later on).  The additional
1091 	 * nfsneg_cache_timeout check causes previously cached results to
1092 	 * be instantly ignored if the negative caching is turned off.
1093 	 */
1094 	lockparent = flags & CNP_LOCKPARENT;
1095 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1096 	nmp = VFSTONFS(dvp->v_mount);
1097 	np = VTONFS(dvp);
1098 
1099 	lwkt_gettoken(&nmp->nm_token);
1100 
1101 	/*
1102 	 * Go to the wire.
1103 	 */
1104 	error = 0;
1105 	newvp = NULLVP;
1106 	nfsstats.lookupcache_misses++;
1107 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1108 	len = cnp->cn_namelen;
1109 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1110 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1111 	ERROROUT(nfsm_fhtom(&info, dvp));
1112 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1113 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1114 				cnp->cn_cred, &error));
1115 	if (error) {
1116 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1117 					     NFS_LATTR_NOSHRINK);
1118 		if (tmp_error) {
1119 			error = tmp_error;
1120 			goto nfsmout;
1121 		}
1122 
1123 		m_freem(info.mrep);
1124 		info.mrep = NULL;
1125 		goto nfsmout;
1126 	}
1127 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1128 
1129 	/*
1130 	 * Handle RENAME case...
1131 	 */
1132 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1133 		if (NFS_CMPFH(np, fhp, fhsize)) {
1134 			m_freem(info.mrep);
1135 			info.mrep = NULL;
1136 			lwkt_reltoken(&nmp->nm_token);
1137 			return (EISDIR);
1138 		}
1139 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1140 		if (error) {
1141 			m_freem(info.mrep);
1142 			info.mrep = NULL;
1143 			lwkt_reltoken(&nmp->nm_token);
1144 			return (error);
1145 		}
1146 		newvp = NFSTOV(np);
1147 		if (info.v3) {
1148 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1149 						  NFS_LATTR_NOSHRINK));
1150 			ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1151 						  NFS_LATTR_NOSHRINK));
1152 		} else {
1153 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1154 		}
1155 		*vpp = newvp;
1156 		m_freem(info.mrep);
1157 		info.mrep = NULL;
1158 		if (!lockparent) {
1159 			vn_unlock(dvp);
1160 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1161 		}
1162 		lwkt_reltoken(&nmp->nm_token);
1163 		return (0);
1164 	}
1165 
1166 	if (flags & CNP_ISDOTDOT) {
1167 		vn_unlock(dvp);
1168 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1169 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1170 		if (error) {
1171 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1172 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1173 			lwkt_reltoken(&nmp->nm_token);
1174 			return (error); /* NOTE: return error from nget */
1175 		}
1176 		newvp = NFSTOV(np);
1177 		if (lockparent) {
1178 			error = vn_lock(dvp, LK_EXCLUSIVE);
1179 			if (error) {
1180 				vput(newvp);
1181 				lwkt_reltoken(&nmp->nm_token);
1182 				return (error);
1183 			}
1184 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1185 		}
1186 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1187 		vref(dvp);
1188 		newvp = dvp;
1189 	} else {
1190 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1191 		if (error) {
1192 			m_freem(info.mrep);
1193 			info.mrep = NULL;
1194 			lwkt_reltoken(&nmp->nm_token);
1195 			return (error);
1196 		}
1197 		if (!lockparent) {
1198 			vn_unlock(dvp);
1199 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1200 		}
1201 		newvp = NFSTOV(np);
1202 	}
1203 	if (info.v3) {
1204 		ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1205 					  NFS_LATTR_NOSHRINK));
1206 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1207 					  NFS_LATTR_NOSHRINK));
1208 	} else {
1209 		ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1210 	}
1211 #if 0
1212 	/* XXX MOVE TO nfs_nremove() */
1213 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1214 	    cnp->cn_nameiop != NAMEI_DELETE) {
1215 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1216 	}
1217 #endif
1218 	*vpp = newvp;
1219 	m_freem(info.mrep);
1220 	info.mrep = NULL;
1221 nfsmout:
1222 	if (error) {
1223 		if (newvp != NULLVP) {
1224 			vrele(newvp);
1225 			*vpp = NULLVP;
1226 		}
1227 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1228 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1229 		    error == ENOENT) {
1230 			if (!lockparent) {
1231 				vn_unlock(dvp);
1232 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1233 			}
1234 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1235 				error = EROFS;
1236 			else
1237 				error = EJUSTRETURN;
1238 		}
1239 	}
1240 	lwkt_reltoken(&nmp->nm_token);
1241 	return (error);
1242 }
1243 
1244 /*
1245  * nfs read call.
1246  * Just call nfs_bioread() to do the work.
1247  *
1248  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1249  *	    struct ucred *a_cred)
1250  */
1251 static int
1252 nfs_read(struct vop_read_args *ap)
1253 {
1254 	struct vnode *vp = ap->a_vp;
1255 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1256 	int error;
1257 
1258 	lwkt_gettoken(&nmp->nm_token);
1259 	error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1260 	lwkt_reltoken(&nmp->nm_token);
1261 
1262 	return error;
1263 }
1264 
1265 /*
1266  * nfs readlink call
1267  *
1268  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1269  */
1270 static int
1271 nfs_readlink(struct vop_readlink_args *ap)
1272 {
1273 	struct vnode *vp = ap->a_vp;
1274 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1275 	int error;
1276 
1277 	if (vp->v_type != VLNK)
1278 		return (EINVAL);
1279 
1280 	lwkt_gettoken(&nmp->nm_token);
1281 	error = nfs_bioread(vp, ap->a_uio, 0);
1282 	lwkt_reltoken(&nmp->nm_token);
1283 
1284 	return error;
1285 }
1286 
1287 /*
1288  * Do a readlink rpc.
1289  * Called by nfs_doio() from below the buffer cache.
1290  */
1291 int
1292 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1293 {
1294 	int error = 0, len, attrflag;
1295 	struct nfsm_info info;
1296 
1297 	info.mrep = NULL;
1298 	info.v3 = NFS_ISV3(vp);
1299 
1300 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1301 	nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1302 	ERROROUT(nfsm_fhtom(&info, vp));
1303 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1304 				nfs_vpcred(vp, ND_CHECK), &error));
1305 	if (info.v3) {
1306 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1307 					  NFS_LATTR_NOSHRINK));
1308 	}
1309 	if (!error) {
1310 		NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1311 		if (len == NFS_MAXPATHLEN) {
1312 			struct nfsnode *np = VTONFS(vp);
1313 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1314 				len = np->n_size;
1315 		}
1316 		ERROROUT(nfsm_mtouio(&info, uiop, len));
1317 	}
1318 	m_freem(info.mrep);
1319 	info.mrep = NULL;
1320 nfsmout:
1321 	return (error);
1322 }
1323 
1324 /*
1325  * nfs synchronous read rpc using UIO
1326  */
1327 int
1328 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1329 {
1330 	u_int32_t *tl;
1331 	struct nfsmount *nmp;
1332 	int error = 0, len, retlen, tsiz, eof, attrflag;
1333 	struct nfsm_info info;
1334 	off_t tmp_off;
1335 
1336 	info.mrep = NULL;
1337 	info.v3 = NFS_ISV3(vp);
1338 
1339 #ifndef nolint
1340 	eof = 0;
1341 #endif
1342 	nmp = VFSTONFS(vp->v_mount);
1343 
1344 	tsiz = uiop->uio_resid;
1345 	tmp_off = uiop->uio_offset + tsiz;
1346 	if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1347 		return (EFBIG);
1348 	tmp_off = uiop->uio_offset;
1349 	while (tsiz > 0) {
1350 		nfsstats.rpccnt[NFSPROC_READ]++;
1351 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1352 		nfsm_reqhead(&info, vp, NFSPROC_READ,
1353 			     NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1354 		ERROROUT(nfsm_fhtom(&info, vp));
1355 		tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1356 		if (info.v3) {
1357 			txdr_hyper(uiop->uio_offset, tl);
1358 			*(tl + 2) = txdr_unsigned(len);
1359 		} else {
1360 			*tl++ = txdr_unsigned(uiop->uio_offset);
1361 			*tl++ = txdr_unsigned(len);
1362 			*tl = 0;
1363 		}
1364 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1365 					nfs_vpcred(vp, ND_READ), &error));
1366 		if (info.v3) {
1367 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1368 						 NFS_LATTR_NOSHRINK));
1369 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1370 			eof = fxdr_unsigned(int, *(tl + 1));
1371 		} else {
1372 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1373 		}
1374 		NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1375 		ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1376 		m_freem(info.mrep);
1377 		info.mrep = NULL;
1378 
1379 		/*
1380 		 * Handle short-read from server (NFSv3).  If EOF is not
1381 		 * flagged (and no error occurred), but retlen is less
1382 		 * then the request size, we must zero-fill the remainder.
1383 		 */
1384 		if (retlen < len && info.v3 && eof == 0) {
1385 			ERROROUT(uiomovez(len - retlen, uiop));
1386 			retlen = len;
1387 		}
1388 		tsiz -= retlen;
1389 
1390 		/*
1391 		 * Terminate loop on EOF or zero-length read.
1392 		 *
1393 		 * For NFSv2 a short-read indicates EOF, not zero-fill,
1394 		 * and also terminates the loop.
1395 		 */
1396 		if (info.v3) {
1397 			if (eof || retlen == 0)
1398 				tsiz = 0;
1399 		} else if (retlen < len) {
1400 			tsiz = 0;
1401 		}
1402 	}
1403 nfsmout:
1404 	return (error);
1405 }
1406 
1407 /*
1408  * nfs write call
1409  */
1410 int
1411 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1412 		 int *iomode, int *must_commit)
1413 {
1414 	u_int32_t *tl;
1415 	int32_t backup;
1416 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1417 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1418 	int  committed = NFSV3WRITE_FILESYNC;
1419 	struct nfsm_info info;
1420 
1421 	info.mrep = NULL;
1422 	info.v3 = NFS_ISV3(vp);
1423 
1424 #ifndef DIAGNOSTIC
1425 	if (uiop->uio_iovcnt != 1)
1426 		panic("nfs: writerpc iovcnt > 1");
1427 #endif
1428 	*must_commit = 0;
1429 	tsiz = uiop->uio_resid;
1430 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1431 		return (EFBIG);
1432 	while (tsiz > 0) {
1433 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1434 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1435 		nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1436 			     NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1437 			     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 			/*
2442 			 * WARNING!  HAMMER DIRECTORIES WILL NOT WORK WELL
2443 			 * WITH NFSv2!!!  There's nothing I can really do
2444 			 * about it other than to hope the server supports
2445 			 * rdirplus w/NFSv2.
2446 			 */
2447 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2448 			*tl++ = cookie.nfsuquad[0];
2449 		}
2450 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2451 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2452 					uiop->uio_td,
2453 					nfs_vpcred(vp, ND_READ), &error));
2454 		if (info.v3) {
2455 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2456 						  NFS_LATTR_NOSHRINK));
2457 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2458 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2459 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2460 		}
2461 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2462 		more_dirs = fxdr_unsigned(int, *tl);
2463 
2464 		/* loop thru the dir entries, converting them to std form */
2465 		while (more_dirs && bigenough) {
2466 			if (info.v3) {
2467 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2468 				fileno = fxdr_hyper(tl);
2469 				len = fxdr_unsigned(int, *(tl + 2));
2470 			} else {
2471 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2472 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2473 				len = fxdr_unsigned(int, *tl);
2474 			}
2475 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2476 				error = EBADRPC;
2477 				m_freem(info.mrep);
2478 				info.mrep = NULL;
2479 				goto nfsmout;
2480 			}
2481 
2482 			/*
2483 			 * len is the number of bytes in the path element
2484 			 * name, not including the \0 termination.
2485 			 *
2486 			 * tlen is the number of bytes w have to reserve for
2487 			 * the path element name.
2488 			 */
2489 			tlen = nfsm_rndup(len);
2490 			if (tlen == len)
2491 				tlen += 4;	/* To ensure null termination */
2492 
2493 			/*
2494 			 * If the entry would cross a DIRBLKSIZ boundary,
2495 			 * extend the previous nfs_dirent to cover the
2496 			 * remaining space.
2497 			 */
2498 			left = DIRBLKSIZ - blksiz;
2499 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2500 				dp->nfs_reclen += left;
2501 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2502 				uiop->uio_iov->iov_len -= left;
2503 				uiop->uio_offset += left;
2504 				uiop->uio_resid -= left;
2505 				blksiz = 0;
2506 			}
2507 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2508 				bigenough = 0;
2509 			if (bigenough) {
2510 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2511 				dp->nfs_ino = fileno;
2512 				dp->nfs_namlen = len;
2513 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2514 				dp->nfs_type = DT_UNKNOWN;
2515 				blksiz += dp->nfs_reclen;
2516 				if (blksiz == DIRBLKSIZ)
2517 					blksiz = 0;
2518 				uiop->uio_offset += sizeof(struct nfs_dirent);
2519 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2520 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2521 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2522 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2523 
2524 				/*
2525 				 * The uiop has advanced by nfs_dirent + len
2526 				 * but really needs to advance by
2527 				 * nfs_dirent + tlen
2528 				 */
2529 				cp = uiop->uio_iov->iov_base;
2530 				tlen -= len;
2531 				*cp = '\0';	/* null terminate */
2532 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2533 				uiop->uio_iov->iov_len -= tlen;
2534 				uiop->uio_offset += tlen;
2535 				uiop->uio_resid -= tlen;
2536 			} else {
2537 				/*
2538 				 * NFS strings must be rounded up (nfsm_myouio
2539 				 * handled that in the bigenough case).
2540 				 */
2541 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2542 			}
2543 			if (info.v3) {
2544 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2545 			} else {
2546 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2547 			}
2548 
2549 			/*
2550 			 * If we were able to accomodate the last entry,
2551 			 * get the cookie for the next one.  Otherwise
2552 			 * hold-over the cookie for the one we were not
2553 			 * able to accomodate.
2554 			 */
2555 			if (bigenough) {
2556 				cookie.nfsuquad[0] = *tl++;
2557 				if (info.v3)
2558 					cookie.nfsuquad[1] = *tl++;
2559 			} else if (info.v3) {
2560 				tl += 2;
2561 			} else {
2562 				tl++;
2563 			}
2564 			more_dirs = fxdr_unsigned(int, *tl);
2565 		}
2566 		/*
2567 		 * If at end of rpc data, get the eof boolean
2568 		 */
2569 		if (!more_dirs) {
2570 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2571 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2572 		}
2573 		m_freem(info.mrep);
2574 		info.mrep = NULL;
2575 	}
2576 	/*
2577 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2578 	 * by increasing d_reclen for the last record.
2579 	 */
2580 	if (blksiz > 0) {
2581 		left = DIRBLKSIZ - blksiz;
2582 		dp->nfs_reclen += left;
2583 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2584 		uiop->uio_iov->iov_len -= left;
2585 		uiop->uio_offset += left;
2586 		uiop->uio_resid -= left;
2587 	}
2588 
2589 	if (bigenough) {
2590 		/*
2591 		 * We hit the end of the directory, update direofoffset.
2592 		 */
2593 		dnp->n_direofoffset = uiop->uio_offset;
2594 	} else {
2595 		/*
2596 		 * There is more to go, insert the link cookie so the
2597 		 * next block can be read.
2598 		 */
2599 		if (uiop->uio_resid > 0)
2600 			kprintf("EEK! readdirrpc resid > 0\n");
2601 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2602 		*cookiep = cookie;
2603 	}
2604 nfsmout:
2605 	return (error);
2606 }
2607 
2608 /*
2609  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2610  */
2611 int
2612 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2613 {
2614 	int len, left;
2615 	struct nfs_dirent *dp;
2616 	u_int32_t *tl;
2617 	struct vnode *newvp;
2618 	nfsuint64 *cookiep;
2619 	caddr_t dpossav1, dpossav2;
2620 	caddr_t cp;
2621 	struct mbuf *mdsav1, *mdsav2;
2622 	nfsuint64 cookie;
2623 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2624 	struct nfsnode *dnp = VTONFS(vp), *np;
2625 	nfsfh_t *fhp;
2626 	u_quad_t fileno;
2627 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2628 	int attrflag, fhsize;
2629 	struct nchandle nch;
2630 	struct nchandle dnch;
2631 	struct nlcomponent nlc;
2632 	struct nfsm_info info;
2633 
2634 	info.mrep = NULL;
2635 	info.v3 = 1;
2636 
2637 #ifndef nolint
2638 	dp = NULL;
2639 #endif
2640 #ifndef DIAGNOSTIC
2641 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2642 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2643 		panic("nfs readdirplusrpc bad uio");
2644 #endif
2645 	/*
2646 	 * Obtain the namecache record for the directory so we have something
2647 	 * to use as a basis for creating the entries.  This function will
2648 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2649 	 * from the tree and cannot be used for upward traversals, and the
2650 	 * ncp may be unnamed.  Note that other unrelated operations may
2651 	 * cause the ncp to be named at any time.
2652 	 *
2653 	 * We have to lock the ncp to prevent a lock order reversal when
2654 	 * rdirplus does nlookups of the children, because the vnode is
2655 	 * locked and has to stay that way.
2656 	 */
2657 	cache_fromdvp(vp, NULL, 0, &dnch);
2658 	bzero(&nlc, sizeof(nlc));
2659 	newvp = NULLVP;
2660 
2661 	/*
2662 	 * If there is no cookie, assume directory was stale.
2663 	 */
2664 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2665 	if (cookiep) {
2666 		cookie = *cookiep;
2667 	} else {
2668 		if (dnch.ncp)
2669 			cache_drop(&dnch);
2670 		return (NFSERR_BAD_COOKIE);
2671 	}
2672 
2673 	/*
2674 	 * Loop around doing readdir rpc's of size nm_readdirsize
2675 	 * truncated to a multiple of DIRBLKSIZ.
2676 	 * The stopping criteria is EOF or buffer full.
2677 	 */
2678 	while (more_dirs && bigenough) {
2679 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2680 		nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2681 			     NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2682 		ERROROUT(nfsm_fhtom(&info, vp));
2683 		tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2684 		*tl++ = cookie.nfsuquad[0];
2685 		*tl++ = cookie.nfsuquad[1];
2686 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2687 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2688 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2689 		*tl = txdr_unsigned(nmp->nm_rsize);
2690 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2691 					uiop->uio_td,
2692 					nfs_vpcred(vp, ND_READ), &error));
2693 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2694 					  NFS_LATTR_NOSHRINK));
2695 		NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2696 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2697 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2698 		more_dirs = fxdr_unsigned(int, *tl);
2699 
2700 		/* loop thru the dir entries, doctoring them to 4bsd form */
2701 		while (more_dirs && bigenough) {
2702 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2703 			fileno = fxdr_hyper(tl);
2704 			len = fxdr_unsigned(int, *(tl + 2));
2705 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2706 				error = EBADRPC;
2707 				m_freem(info.mrep);
2708 				info.mrep = NULL;
2709 				goto nfsmout;
2710 			}
2711 			tlen = nfsm_rndup(len);
2712 			if (tlen == len)
2713 				tlen += 4;	/* To ensure null termination*/
2714 			left = DIRBLKSIZ - blksiz;
2715 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2716 				dp->nfs_reclen += left;
2717 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2718 				uiop->uio_iov->iov_len -= left;
2719 				uiop->uio_offset += left;
2720 				uiop->uio_resid -= left;
2721 				blksiz = 0;
2722 			}
2723 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2724 				bigenough = 0;
2725 			if (bigenough) {
2726 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2727 				dp->nfs_ino = fileno;
2728 				dp->nfs_namlen = len;
2729 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2730 				dp->nfs_type = DT_UNKNOWN;
2731 				blksiz += dp->nfs_reclen;
2732 				if (blksiz == DIRBLKSIZ)
2733 					blksiz = 0;
2734 				uiop->uio_offset += sizeof(struct nfs_dirent);
2735 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2736 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2737 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2738 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2739 				nlc.nlc_namelen = len;
2740 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2741 				cp = uiop->uio_iov->iov_base;
2742 				tlen -= len;
2743 				*cp = '\0';
2744 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2745 				uiop->uio_iov->iov_len -= tlen;
2746 				uiop->uio_offset += tlen;
2747 				uiop->uio_resid -= tlen;
2748 			} else {
2749 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2750 			}
2751 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2752 			if (bigenough) {
2753 				cookie.nfsuquad[0] = *tl++;
2754 				cookie.nfsuquad[1] = *tl++;
2755 			} else {
2756 				tl += 2;
2757 			}
2758 
2759 			/*
2760 			 * Since the attributes are before the file handle
2761 			 * (sigh), we must skip over the attributes and then
2762 			 * come back and get them.
2763 			 */
2764 			attrflag = fxdr_unsigned(int, *tl);
2765 			if (attrflag) {
2766 			    dpossav1 = info.dpos;
2767 			    mdsav1 = info.md;
2768 			    ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2769 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2770 			    doit = fxdr_unsigned(int, *tl);
2771 			    if (doit) {
2772 				NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2773 			    }
2774 			    if (doit && bigenough && !nlcdegenerate(&nlc) &&
2775 				!NFS_CMPFH(dnp, fhp, fhsize)
2776 			    ) {
2777 				if (dnch.ncp) {
2778 #if 0
2779 				    kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2780 					nlc.nlc_namelen, nlc.nlc_namelen,
2781 					nlc.nlc_nameptr);
2782 #endif
2783 				    /*
2784 				     * This is a bit hokey but there isn't
2785 				     * much we can do about it.  We can't
2786 				     * hold the directory vp locked while
2787 				     * doing lookups and gets.
2788 				     */
2789 				    nch = cache_nlookup_nonblock(&dnch, &nlc);
2790 				    if (nch.ncp == NULL)
2791 					goto rdfail;
2792 				    cache_setunresolved(&nch);
2793 				    error = nfs_nget_nonblock(vp->v_mount, fhp,
2794 							      fhsize, &np);
2795 				    if (error) {
2796 					cache_put(&nch);
2797 					goto rdfail;
2798 				    }
2799 				    newvp = NFSTOV(np);
2800 				    dpossav2 = info.dpos;
2801 				    info.dpos = dpossav1;
2802 				    mdsav2 = info.md;
2803 				    info.md = mdsav1;
2804 				    ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2805 				    info.dpos = dpossav2;
2806 				    info.md = mdsav2;
2807 				    dp->nfs_type =
2808 					    IFTODT(VTTOIF(np->n_vattr.va_type));
2809 				    nfs_cache_setvp(&nch, newvp,
2810 						    nfspos_cache_timeout);
2811 				    vput(newvp);
2812 				    newvp = NULLVP;
2813 				    cache_put(&nch);
2814 				} else {
2815 rdfail:
2816 				    ;
2817 #if 0
2818 				    kprintf("Warning: NFS/rddirplus, "
2819 					    "UNABLE TO ENTER %*.*s\n",
2820 					nlc.nlc_namelen, nlc.nlc_namelen,
2821 					nlc.nlc_nameptr);
2822 #endif
2823 				}
2824 			    }
2825 			} else {
2826 			    /* Just skip over the file handle */
2827 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2828 			    i = fxdr_unsigned(int, *tl);
2829 			    ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2830 			}
2831 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2832 			more_dirs = fxdr_unsigned(int, *tl);
2833 		}
2834 		/*
2835 		 * If at end of rpc data, get the eof boolean
2836 		 */
2837 		if (!more_dirs) {
2838 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2839 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2840 		}
2841 		m_freem(info.mrep);
2842 		info.mrep = NULL;
2843 	}
2844 	/*
2845 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2846 	 * by increasing d_reclen for the last record.
2847 	 */
2848 	if (blksiz > 0) {
2849 		left = DIRBLKSIZ - blksiz;
2850 		dp->nfs_reclen += left;
2851 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2852 		uiop->uio_iov->iov_len -= left;
2853 		uiop->uio_offset += left;
2854 		uiop->uio_resid -= left;
2855 	}
2856 
2857 	/*
2858 	 * We are now either at the end of the directory or have filled the
2859 	 * block.
2860 	 */
2861 	if (bigenough) {
2862 		dnp->n_direofoffset = uiop->uio_offset;
2863 	} else {
2864 		if (uiop->uio_resid > 0)
2865 			kprintf("EEK! readdirplusrpc resid > 0\n");
2866 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2867 		*cookiep = cookie;
2868 	}
2869 nfsmout:
2870 	if (newvp != NULLVP) {
2871 	        if (newvp == vp)
2872 			vrele(newvp);
2873 		else
2874 			vput(newvp);
2875 		newvp = NULLVP;
2876 	}
2877 	if (dnch.ncp)
2878 		cache_drop(&dnch);
2879 	return (error);
2880 }
2881 
2882 /*
2883  * Silly rename. To make the NFS filesystem that is stateless look a little
2884  * more like the "ufs" a remove of an active vnode is translated to a rename
2885  * to a funny looking filename that is removed by nfs_inactive on the
2886  * nfsnode. There is the potential for another process on a different client
2887  * to create the same funny name between the nfs_lookitup() fails and the
2888  * nfs_rename() completes, but...
2889  */
2890 static int
2891 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2892 {
2893 	struct sillyrename *sp;
2894 	struct nfsnode *np;
2895 	int error;
2896 
2897 	/*
2898 	 * We previously purged dvp instead of vp.  I don't know why, it
2899 	 * completely destroys performance.  We can't do it anyway with the
2900 	 * new VFS API since we would be breaking the namecache topology.
2901 	 */
2902 	cache_purge(vp);	/* XXX */
2903 	np = VTONFS(vp);
2904 #ifndef DIAGNOSTIC
2905 	if (vp->v_type == VDIR)
2906 		panic("nfs: sillyrename dir");
2907 #endif
2908 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2909 		M_NFSREQ, M_WAITOK);
2910 	sp->s_cred = crdup(cnp->cn_cred);
2911 	sp->s_dvp = dvp;
2912 	vref(dvp);
2913 
2914 	/* Fudge together a funny name */
2915 	sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2916 				(int)(intptr_t)cnp->cn_td);
2917 
2918 	/* Try lookitups until we get one that isn't there */
2919 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2920 		cnp->cn_td, NULL) == 0) {
2921 		sp->s_name[4]++;
2922 		if (sp->s_name[4] > 'z') {
2923 			error = EINVAL;
2924 			goto bad;
2925 		}
2926 	}
2927 	error = nfs_renameit(dvp, cnp, sp);
2928 	if (error)
2929 		goto bad;
2930 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2931 		cnp->cn_td, &np);
2932 	np->n_sillyrename = sp;
2933 	return (0);
2934 bad:
2935 	vrele(sp->s_dvp);
2936 	crfree(sp->s_cred);
2937 	kfree((caddr_t)sp, M_NFSREQ);
2938 	return (error);
2939 }
2940 
2941 /*
2942  * Look up a file name and optionally either update the file handle or
2943  * allocate an nfsnode, depending on the value of npp.
2944  * npp == NULL	--> just do the lookup
2945  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2946  *			handled too
2947  * *npp != NULL --> update the file handle in the vnode
2948  */
2949 static int
2950 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2951 	     struct thread *td, struct nfsnode **npp)
2952 {
2953 	struct vnode *newvp = NULL;
2954 	struct nfsnode *np, *dnp = VTONFS(dvp);
2955 	int error = 0, fhlen, attrflag;
2956 	nfsfh_t *nfhp;
2957 	struct nfsm_info info;
2958 
2959 	info.mrep = NULL;
2960 	info.v3 = NFS_ISV3(dvp);
2961 
2962 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2963 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2964 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2965 	ERROROUT(nfsm_fhtom(&info, dvp));
2966 	ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2967 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2968 	if (npp && !error) {
2969 		NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2970 		if (*npp) {
2971 		    np = *npp;
2972 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2973 			kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2974 			np->n_fhp = &np->n_fh;
2975 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2976 			np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2977 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2978 		    np->n_fhsize = fhlen;
2979 		    newvp = NFSTOV(np);
2980 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2981 		    vref(dvp);
2982 		    newvp = dvp;
2983 		} else {
2984 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2985 		    if (error) {
2986 			m_freem(info.mrep);
2987 			info.mrep = NULL;
2988 			return (error);
2989 		    }
2990 		    newvp = NFSTOV(np);
2991 		}
2992 		if (info.v3) {
2993 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2994 						  NFS_LATTR_NOSHRINK));
2995 			if (!attrflag && *npp == NULL) {
2996 				m_freem(info.mrep);
2997 				info.mrep = NULL;
2998 				if (newvp == dvp)
2999 					vrele(newvp);
3000 				else
3001 					vput(newvp);
3002 				return (ENOENT);
3003 			}
3004 		} else {
3005 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3006 		}
3007 	}
3008 	m_freem(info.mrep);
3009 	info.mrep = NULL;
3010 nfsmout:
3011 	if (npp && *npp == NULL) {
3012 		if (error) {
3013 			if (newvp) {
3014 				if (newvp == dvp)
3015 					vrele(newvp);
3016 				else
3017 					vput(newvp);
3018 			}
3019 		} else
3020 			*npp = np;
3021 	}
3022 	return (error);
3023 }
3024 
3025 /*
3026  * Nfs Version 3 commit rpc
3027  *
3028  * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3029  * involved.
3030  */
3031 int
3032 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3033 {
3034 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3035 	int error = 0, wccflag = NFSV3_WCCRATTR;
3036 	struct nfsm_info info;
3037 	u_int32_t *tl;
3038 
3039 	info.mrep = NULL;
3040 	info.v3 = 1;
3041 
3042 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3043 		return (0);
3044 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3045 	nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3046 	ERROROUT(nfsm_fhtom(&info, vp));
3047 	tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3048 	txdr_hyper(offset, tl);
3049 	tl += 2;
3050 	*tl = txdr_unsigned(cnt);
3051 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3052 				nfs_vpcred(vp, ND_WRITE), &error));
3053 	ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3054 	if (!error) {
3055 		NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3056 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3057 			NFSX_V3WRITEVERF)) {
3058 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3059 				NFSX_V3WRITEVERF);
3060 			error = NFSERR_STALEWRITEVERF;
3061 		}
3062 	}
3063 	m_freem(info.mrep);
3064 	info.mrep = NULL;
3065 nfsmout:
3066 	return (error);
3067 }
3068 
3069 /*
3070  * Kludge City..
3071  * - make nfs_bmap() essentially a no-op that does no translation
3072  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3073  *   (Maybe I could use the process's page mapping, but I was concerned that
3074  *    Kernel Write might not be enabled and also figured copyout() would do
3075  *    a lot more work than bcopy() and also it currently happens in the
3076  *    context of the swapper process (2).
3077  *
3078  * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3079  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
3080  */
3081 static int
3082 nfs_bmap(struct vop_bmap_args *ap)
3083 {
3084 	/* no token lock required */
3085 	if (ap->a_doffsetp != NULL)
3086 		*ap->a_doffsetp = ap->a_loffset;
3087 	if (ap->a_runp != NULL)
3088 		*ap->a_runp = 0;
3089 	if (ap->a_runb != NULL)
3090 		*ap->a_runb = 0;
3091 	return (0);
3092 }
3093 
3094 /*
3095  * Strategy routine.
3096  */
3097 static int
3098 nfs_strategy(struct vop_strategy_args *ap)
3099 {
3100 	struct bio *bio = ap->a_bio;
3101 	struct bio *nbio;
3102 	struct buf *bp __debugvar = bio->bio_buf;
3103 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3104 	struct thread *td;
3105 	int error;
3106 
3107 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
3108 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
3109 	KASSERT(BUF_REFCNT(bp) > 0,
3110 		("nfs_strategy: buffer %p not locked", bp));
3111 
3112 	if (bio->bio_flags & BIO_SYNC)
3113 		td = curthread;	/* XXX */
3114 	else
3115 		td = NULL;
3116 
3117 	lwkt_gettoken(&nmp->nm_token);
3118 
3119         /*
3120 	 * We probably don't need to push an nbio any more since no
3121 	 * block conversion is required due to the use of 64 bit byte
3122 	 * offsets, but do it anyway.
3123 	 *
3124 	 * NOTE: When NFS callers itself via this strategy routines and
3125 	 *	 sets up a synchronous I/O, it expects the I/O to run
3126 	 *	 synchronously (its bio_done routine just assumes it),
3127 	 *	 so for now we have to honor the bit.
3128          */
3129 	nbio = push_bio(bio);
3130 	nbio->bio_offset = bio->bio_offset;
3131 	nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3132 
3133 	/*
3134 	 * If the op is asynchronous and an i/o daemon is waiting
3135 	 * queue the request, wake it up and wait for completion
3136 	 * otherwise just do it ourselves.
3137 	 */
3138 	if (bio->bio_flags & BIO_SYNC) {
3139 		error = nfs_doio(ap->a_vp, nbio, td);
3140 	} else {
3141 		nfs_asyncio(ap->a_vp, nbio);
3142 		error = 0;
3143 	}
3144 	lwkt_reltoken(&nmp->nm_token);
3145 
3146 	return (error);
3147 }
3148 
3149 /*
3150  * Mmap a file
3151  *
3152  * NB Currently unsupported.
3153  *
3154  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3155  */
3156 /* ARGSUSED */
3157 static int
3158 nfs_mmap(struct vop_mmap_args *ap)
3159 {
3160 	/* no token lock required */
3161 	return (EINVAL);
3162 }
3163 
3164 /*
3165  * fsync vnode op. Just call nfs_flush() with commit == 1.
3166  *
3167  * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3168  */
3169 /* ARGSUSED */
3170 static int
3171 nfs_fsync(struct vop_fsync_args *ap)
3172 {
3173 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3174 	int error;
3175 
3176 	lwkt_gettoken(&nmp->nm_token);
3177 	error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3178 	lwkt_reltoken(&nmp->nm_token);
3179 
3180 	return error;
3181 }
3182 
3183 /*
3184  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
3185  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
3186  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
3187  * set the buffer contains data that has already been written to the server
3188  * and which now needs a commit RPC.
3189  *
3190  * If commit is 0 we only take one pass and only flush buffers containing new
3191  * dirty data.
3192  *
3193  * If commit is 1 we take two passes, issuing a commit RPC in the second
3194  * pass.
3195  *
3196  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3197  * to completely flush all pending data.
3198  *
3199  * Note that the RB_SCAN code properly handles the case where the
3200  * callback might block and directly or indirectly (another thread) cause
3201  * the RB tree to change.
3202  */
3203 
3204 #ifndef NFS_COMMITBVECSIZ
3205 #define NFS_COMMITBVECSIZ	16
3206 #endif
3207 
3208 struct nfs_flush_info {
3209 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3210 	struct thread *td;
3211 	struct vnode *vp;
3212 	int waitfor;
3213 	int slpflag;
3214 	int slptimeo;
3215 	int loops;
3216 	struct buf *bvary[NFS_COMMITBVECSIZ];
3217 	int bvsize;
3218 	off_t beg_off;
3219 	off_t end_off;
3220 };
3221 
3222 static int nfs_flush_bp(struct buf *bp, void *data);
3223 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3224 
3225 int
3226 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3227 {
3228 	struct nfsnode *np = VTONFS(vp);
3229 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3230 	struct nfs_flush_info info;
3231 	int error;
3232 
3233 	bzero(&info, sizeof(info));
3234 	info.td = td;
3235 	info.vp = vp;
3236 	info.waitfor = waitfor;
3237 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3238 	info.loops = 0;
3239 	lwkt_gettoken(&vp->v_token);
3240 
3241 	do {
3242 		/*
3243 		 * Flush mode
3244 		 */
3245 		info.mode = NFI_FLUSHNEW;
3246 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3247 				nfs_flush_bp, &info);
3248 
3249 		/*
3250 		 * Take a second pass if committing and no error occured.
3251 		 * Clean up any left over collection (whether an error
3252 		 * occurs or not).
3253 		 */
3254 		if (commit && error == 0) {
3255 			info.mode = NFI_COMMIT;
3256 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3257 					nfs_flush_bp, &info);
3258 			if (info.bvsize)
3259 				error = nfs_flush_docommit(&info, error);
3260 		}
3261 
3262 		/*
3263 		 * Wait for pending I/O to complete before checking whether
3264 		 * any further dirty buffers exist.
3265 		 */
3266 		while (waitfor == MNT_WAIT &&
3267 		       bio_track_active(&vp->v_track_write)) {
3268 			error = bio_track_wait(&vp->v_track_write,
3269 					       info.slpflag, info.slptimeo);
3270 			if (error) {
3271 				/*
3272 				 * We have to be able to break out if this
3273 				 * is an 'intr' mount.
3274 				 */
3275 				if (nfs_sigintr(nmp, NULL, td)) {
3276 					error = -EINTR;
3277 					break;
3278 				}
3279 
3280 				/*
3281 				 * Since we do not process pending signals,
3282 				 * once we get a PCATCH our tsleep() will no
3283 				 * longer sleep, switch to a fixed timeout
3284 				 * instead.
3285 				 */
3286 				if (info.slpflag == PCATCH) {
3287 					info.slpflag = 0;
3288 					info.slptimeo = 2 * hz;
3289 				}
3290 				error = 0;
3291 			}
3292 		}
3293 		++info.loops;
3294 		/*
3295 		 * Loop if we are flushing synchronous as well as committing,
3296 		 * and dirty buffers are still present.  Otherwise we might livelock.
3297 		 */
3298 	} while (waitfor == MNT_WAIT && commit &&
3299 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3300 
3301 	/*
3302 	 * The callbacks have to return a negative error to terminate the
3303 	 * RB scan.
3304 	 */
3305 	if (error < 0)
3306 		error = -error;
3307 
3308 	/*
3309 	 * Deal with any error collection
3310 	 */
3311 	if (np->n_flag & NWRITEERR) {
3312 		error = np->n_error;
3313 		np->n_flag &= ~NWRITEERR;
3314 	}
3315 	lwkt_reltoken(&vp->v_token);
3316 	return (error);
3317 }
3318 
3319 static
3320 int
3321 nfs_flush_bp(struct buf *bp, void *data)
3322 {
3323 	struct nfs_flush_info *info = data;
3324 	int lkflags;
3325 	int error;
3326 	off_t toff;
3327 
3328 	error = 0;
3329 	switch(info->mode) {
3330 	case NFI_FLUSHNEW:
3331 		error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3332 		if (error && info->loops && info->waitfor == MNT_WAIT) {
3333 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3334 			if (error) {
3335 				lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3336 				if (info->slpflag & PCATCH)
3337 					lkflags |= LK_PCATCH;
3338 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3339 						     info->slptimeo);
3340 			}
3341 		}
3342 
3343 		/*
3344 		 * Ignore locking errors
3345 		 */
3346 		if (error) {
3347 			error = 0;
3348 			break;
3349 		}
3350 
3351 		/*
3352 		 * The buffer may have changed out from under us, even if
3353 		 * we did not block (MPSAFE).  Check again now that it is
3354 		 * locked.
3355 		 */
3356 		if (bp->b_vp == info->vp &&
3357 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3358 			bremfree(bp);
3359 			bawrite(bp);
3360 		} else {
3361 			BUF_UNLOCK(bp);
3362 		}
3363 		break;
3364 	case NFI_COMMIT:
3365 		/*
3366 		 * Only process buffers in need of a commit which we can
3367 		 * immediately lock.  This may prevent a buffer from being
3368 		 * committed, but the normal flush loop will block on the
3369 		 * same buffer so we shouldn't get into an endless loop.
3370 		 */
3371 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3372 		    (B_DELWRI | B_NEEDCOMMIT)) {
3373 			break;
3374 		}
3375 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3376 			break;
3377 
3378 		/*
3379 		 * We must recheck after successfully locking the buffer.
3380 		 */
3381 		if (bp->b_vp != info->vp ||
3382 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3383 		    (B_DELWRI | B_NEEDCOMMIT)) {
3384 			BUF_UNLOCK(bp);
3385 			break;
3386 		}
3387 
3388 		/*
3389 		 * NOTE: storing the bp in the bvary[] basically sets
3390 		 * it up for a commit operation.
3391 		 *
3392 		 * We must call vfs_busy_pages() now so the commit operation
3393 		 * is interlocked with user modifications to memory mapped
3394 		 * pages.  The b_dirtyoff/b_dirtyend range is not correct
3395 		 * until after the pages have been busied.
3396 		 *
3397 		 * Note: to avoid loopback deadlocks, we do not
3398 		 * assign b_runningbufspace.
3399 		 */
3400 		bremfree(bp);
3401 		bp->b_cmd = BUF_CMD_WRITE;
3402 		vfs_busy_pages(bp->b_vp, bp);
3403 		info->bvary[info->bvsize] = bp;
3404 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3405 		if (info->bvsize == 0 || toff < info->beg_off)
3406 			info->beg_off = toff;
3407 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3408 		if (info->bvsize == 0 || toff > info->end_off)
3409 			info->end_off = toff;
3410 		++info->bvsize;
3411 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3412 			error = nfs_flush_docommit(info, 0);
3413 			KKASSERT(info->bvsize == 0);
3414 		}
3415 	}
3416 	return (error);
3417 }
3418 
3419 static
3420 int
3421 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3422 {
3423 	struct vnode *vp;
3424 	struct buf *bp;
3425 	off_t bytes;
3426 	int retv;
3427 	int i;
3428 
3429 	vp = info->vp;
3430 
3431 	if (info->bvsize > 0) {
3432 		/*
3433 		 * Commit data on the server, as required.  Note that
3434 		 * nfs_commit will use the vnode's cred for the commit.
3435 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3436 		 */
3437 		bytes = info->end_off - info->beg_off;
3438 		if (bytes > 0x40000000)
3439 			bytes = 0x40000000;
3440 		if (error) {
3441 			retv = -error;
3442 		} else {
3443 			retv = nfs_commitrpc_uio(vp, info->beg_off,
3444 						 (int)bytes, info->td);
3445 			if (retv == NFSERR_STALEWRITEVERF)
3446 				nfs_clearcommit(vp->v_mount);
3447 		}
3448 
3449 		/*
3450 		 * Now, either mark the blocks I/O done or mark the
3451 		 * blocks dirty, depending on whether the commit
3452 		 * succeeded.
3453 		 */
3454 		for (i = 0; i < info->bvsize; ++i) {
3455 			bp = info->bvary[i];
3456 			if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3457 				/*
3458 				 * Either an error or the original
3459 				 * vfs_busy_pages() cleared B_NEEDCOMMIT
3460 				 * due to finding new dirty VM pages in
3461 				 * the buffer.
3462 				 *
3463 				 * Leave B_DELWRI intact.
3464 				 */
3465 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3466 				vfs_unbusy_pages(bp);
3467 				bp->b_cmd = BUF_CMD_DONE;
3468 				bqrelse(bp);
3469 			} else {
3470 				/*
3471 				 * Success, remove B_DELWRI ( bundirty() ).
3472 				 *
3473 				 * b_dirtyoff/b_dirtyend seem to be NFS
3474 				 * specific.  We should probably move that
3475 				 * into bundirty(). XXX
3476 				 *
3477 				 * We are faking an I/O write, we have to
3478 				 * start the transaction in order to
3479 				 * immediately biodone() it.
3480 				 */
3481 				bundirty(bp);
3482 				bp->b_flags &= ~B_ERROR;
3483 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3484 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3485 				biodone(&bp->b_bio1);
3486 			}
3487 		}
3488 		info->bvsize = 0;
3489 	}
3490 	return (error);
3491 }
3492 
3493 /*
3494  * NFS advisory byte-level locks.
3495  * Currently unsupported.
3496  *
3497  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3498  *		int a_flags)
3499  */
3500 static int
3501 nfs_advlock(struct vop_advlock_args *ap)
3502 {
3503 	struct nfsnode *np = VTONFS(ap->a_vp);
3504 
3505 	/* no token lock currently required */
3506 	/*
3507 	 * The following kludge is to allow diskless support to work
3508 	 * until a real NFS lockd is implemented. Basically, just pretend
3509 	 * that this is a local lock.
3510 	 */
3511 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3512 }
3513 
3514 /*
3515  * Print out the contents of an nfsnode.
3516  *
3517  * nfs_print(struct vnode *a_vp)
3518  */
3519 static int
3520 nfs_print(struct vop_print_args *ap)
3521 {
3522 	struct vnode *vp = ap->a_vp;
3523 	struct nfsnode *np = VTONFS(vp);
3524 
3525 	kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3526 		(long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3527 	if (vp->v_type == VFIFO)
3528 		fifo_printinfo(vp);
3529 	kprintf("\n");
3530 	return (0);
3531 }
3532 
3533 /*
3534  * nfs special file access vnode op.
3535  *
3536  * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3537  */
3538 static int
3539 nfs_laccess(struct vop_access_args *ap)
3540 {
3541 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3542 	struct vattr vattr;
3543 	int error;
3544 
3545 	lwkt_gettoken(&nmp->nm_token);
3546 	error = VOP_GETATTR(ap->a_vp, &vattr);
3547 	if (error == 0) {
3548 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3549 					  vattr.va_mode, 0);
3550 	}
3551 	lwkt_reltoken(&nmp->nm_token);
3552 
3553 	return (error);
3554 }
3555 
3556 /*
3557  * Read wrapper for fifos.
3558  *
3559  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3560  *		struct ucred *a_cred)
3561  */
3562 static int
3563 nfsfifo_read(struct vop_read_args *ap)
3564 {
3565 	struct nfsnode *np = VTONFS(ap->a_vp);
3566 
3567 	/* no token access required */
3568 	/*
3569 	 * Set access flag.
3570 	 */
3571 	np->n_flag |= NACC;
3572 	getnanotime(&np->n_atim);
3573 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3574 }
3575 
3576 /*
3577  * Write wrapper for fifos.
3578  *
3579  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3580  *		 struct ucred *a_cred)
3581  */
3582 static int
3583 nfsfifo_write(struct vop_write_args *ap)
3584 {
3585 	struct nfsnode *np = VTONFS(ap->a_vp);
3586 
3587 	/* no token access required */
3588 	/*
3589 	 * Set update flag.
3590 	 */
3591 	np->n_flag |= NUPD;
3592 	getnanotime(&np->n_mtim);
3593 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3594 }
3595 
3596 /*
3597  * Close wrapper for fifos.
3598  *
3599  * Update the times on the nfsnode then do fifo close.
3600  *
3601  * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3602  */
3603 static int
3604 nfsfifo_close(struct vop_close_args *ap)
3605 {
3606 	struct vnode *vp = ap->a_vp;
3607 	struct nfsnode *np = VTONFS(vp);
3608 	struct vattr vattr;
3609 	struct timespec ts;
3610 
3611 	/* no token access required */
3612 
3613 	if (np->n_flag & (NACC | NUPD)) {
3614 		getnanotime(&ts);
3615 		if (np->n_flag & NACC)
3616 			np->n_atim = ts;
3617 		if (np->n_flag & NUPD)
3618 			np->n_mtim = ts;
3619 		np->n_flag |= NCHG;
3620 		if (vp->v_sysref.refcnt == 1 &&
3621 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3622 			VATTR_NULL(&vattr);
3623 			if (np->n_flag & NACC)
3624 				vattr.va_atime = np->n_atim;
3625 			if (np->n_flag & NUPD)
3626 				vattr.va_mtime = np->n_mtim;
3627 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3628 		}
3629 	}
3630 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3631 }
3632 
3633