xref: /dflybsd-src/sys/vfs/nfs/nfs_vnops.c (revision 0ea552d68c3a0ee380451241d06e39c6dd43d66a)
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 1.64 2006/08/12 00:26:21 dillon Exp $
39  */
40 
41 
42 /*
43  * vnode op calls for Sun NFS version 2 and 3
44  */
45 
46 #include "opt_inet.h"
47 
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/nlookup.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
64 #include <sys/stat.h>
65 #include <sys/sysctl.h>
66 #include <sys/conf.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
71 
72 #include <sys/buf2.h>
73 
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
76 
77 #undef DIRBLKSIZ
78 
79 #include "rpcv2.h"
80 #include "nfsproto.h"
81 #include "nfs.h"
82 #include "nfsmount.h"
83 #include "nfsnode.h"
84 #include "xdr_subs.h"
85 #include "nfsm_subs.h"
86 
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 
91 #include <sys/thread2.h>
92 
93 /* Defs */
94 #define	TRUE	1
95 #define	FALSE	0
96 
97 static int	nfsspec_read (struct vop_read_args *);
98 static int	nfsspec_write (struct vop_write_args *);
99 static int	nfsfifo_read (struct vop_read_args *);
100 static int	nfsfifo_write (struct vop_write_args *);
101 static int	nfsspec_close (struct vop_close_args *);
102 static int	nfsfifo_close (struct vop_close_args *);
103 #define nfs_poll vop_nopoll
104 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
105 static	int	nfs_lookup (struct vop_old_lookup_args *);
106 static	int	nfs_create (struct vop_old_create_args *);
107 static	int	nfs_mknod (struct vop_old_mknod_args *);
108 static	int	nfs_open (struct vop_open_args *);
109 static	int	nfs_close (struct vop_close_args *);
110 static	int	nfs_access (struct vop_access_args *);
111 static	int	nfs_getattr (struct vop_getattr_args *);
112 static	int	nfs_setattr (struct vop_setattr_args *);
113 static	int	nfs_read (struct vop_read_args *);
114 static	int	nfs_mmap (struct vop_mmap_args *);
115 static	int	nfs_fsync (struct vop_fsync_args *);
116 static	int	nfs_remove (struct vop_old_remove_args *);
117 static	int	nfs_link (struct vop_old_link_args *);
118 static	int	nfs_rename (struct vop_old_rename_args *);
119 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
120 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
121 static	int	nfs_symlink (struct vop_old_symlink_args *);
122 static	int	nfs_readdir (struct vop_readdir_args *);
123 static	int	nfs_bmap (struct vop_bmap_args *);
124 static	int	nfs_strategy (struct vop_strategy_args *);
125 static	int	nfs_lookitup (struct vnode *, const char *, int,
126 			struct ucred *, struct thread *, struct nfsnode **);
127 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
128 static int	nfsspec_access (struct vop_access_args *);
129 static int	nfs_readlink (struct vop_readlink_args *);
130 static int	nfs_print (struct vop_print_args *);
131 static int	nfs_advlock (struct vop_advlock_args *);
132 
133 static	int	nfs_nresolve (struct vop_nresolve_args *);
134 /*
135  * Global vfs data structures for nfs
136  */
137 struct vop_ops nfsv2_vnode_vops = {
138 	.vop_default =		vop_defaultop,
139 	.vop_access =		nfs_access,
140 	.vop_advlock =		nfs_advlock,
141 	.vop_bmap =		nfs_bmap,
142 	.vop_close =		nfs_close,
143 	.vop_old_create =	nfs_create,
144 	.vop_fsync =		nfs_fsync,
145 	.vop_getattr =		nfs_getattr,
146 	.vop_getpages =		nfs_getpages,
147 	.vop_putpages =		nfs_putpages,
148 	.vop_inactive =		nfs_inactive,
149 	.vop_old_link =		nfs_link,
150 	.vop_old_lookup =	nfs_lookup,
151 	.vop_old_mkdir =	nfs_mkdir,
152 	.vop_old_mknod =	nfs_mknod,
153 	.vop_mmap =		nfs_mmap,
154 	.vop_open =		nfs_open,
155 	.vop_poll =		nfs_poll,
156 	.vop_print =		nfs_print,
157 	.vop_read =		nfs_read,
158 	.vop_readdir =		nfs_readdir,
159 	.vop_readlink =		nfs_readlink,
160 	.vop_reclaim =		nfs_reclaim,
161 	.vop_old_remove =	nfs_remove,
162 	.vop_old_rename =	nfs_rename,
163 	.vop_old_rmdir =	nfs_rmdir,
164 	.vop_setattr =		nfs_setattr,
165 	.vop_strategy =		nfs_strategy,
166 	.vop_old_symlink =	nfs_symlink,
167 	.vop_write =		nfs_write,
168 	.vop_nresolve =		nfs_nresolve
169 };
170 
171 /*
172  * Special device vnode ops
173  */
174 struct vop_ops nfsv2_spec_vops = {
175 	.vop_default =		spec_vnoperate,
176 	.vop_access =		nfsspec_access,
177 	.vop_close =		nfsspec_close,
178 	.vop_fsync =		nfs_fsync,
179 	.vop_getattr =		nfs_getattr,
180 	.vop_inactive =		nfs_inactive,
181 	.vop_print =		nfs_print,
182 	.vop_read =		nfsspec_read,
183 	.vop_reclaim =		nfs_reclaim,
184 	.vop_setattr =		nfs_setattr,
185 	.vop_write =		nfsspec_write
186 };
187 
188 struct vop_ops nfsv2_fifo_vops = {
189 	.vop_default =		fifo_vnoperate,
190 	.vop_access =		nfsspec_access,
191 	.vop_close =		nfsfifo_close,
192 	.vop_fsync =		nfs_fsync,
193 	.vop_getattr =		nfs_getattr,
194 	.vop_inactive =		nfs_inactive,
195 	.vop_print =		nfs_print,
196 	.vop_read =		nfsfifo_read,
197 	.vop_reclaim =		nfs_reclaim,
198 	.vop_setattr =		nfs_setattr,
199 	.vop_write =		nfsfifo_write
200 };
201 
202 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
203 				  struct componentname *cnp,
204 				  struct vattr *vap);
205 static int	nfs_removerpc (struct vnode *dvp, const char *name,
206 				   int namelen,
207 				   struct ucred *cred, struct thread *td);
208 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
209 				   int fnamelen, struct vnode *tdvp,
210 				   const char *tnameptr, int tnamelen,
211 				   struct ucred *cred, struct thread *td);
212 static int	nfs_renameit (struct vnode *sdvp,
213 				  struct componentname *scnp,
214 				  struct sillyrename *sp);
215 
216 /*
217  * Global variables
218  */
219 extern u_int32_t nfs_true, nfs_false;
220 extern u_int32_t nfs_xdrneg1;
221 extern struct nfsstats nfsstats;
222 extern nfstype nfsv3_type[9];
223 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
224 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
225 int nfs_numasync = 0;
226 
227 SYSCTL_DECL(_vfs_nfs);
228 
229 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
231 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
232 
233 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
235 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout");
236 
237 static int	nfsv3_commit_on_close = 0;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
239 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
240 #if 0
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
242 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
243 
244 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
245 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
246 #endif
247 
248 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
249 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
250 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
251 static int
252 nfs3_access_otw(struct vnode *vp, int wmode,
253 		struct thread *td, struct ucred *cred)
254 {
255 	const int v3 = 1;
256 	u_int32_t *tl;
257 	int error = 0, attrflag;
258 
259 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
260 	caddr_t bpos, dpos, cp2;
261 	int32_t t1, t2;
262 	caddr_t cp;
263 	u_int32_t rmode;
264 	struct nfsnode *np = VTONFS(vp);
265 
266 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
267 	nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
268 	nfsm_fhtom(vp, v3);
269 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
270 	*tl = txdr_unsigned(wmode);
271 	nfsm_request(vp, NFSPROC_ACCESS, td, cred);
272 	nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
273 	if (!error) {
274 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
275 		rmode = fxdr_unsigned(u_int32_t, *tl);
276 		np->n_mode = rmode;
277 		np->n_modeuid = cred->cr_uid;
278 		np->n_modestamp = mycpu->gd_time_seconds;
279 	}
280 	m_freem(mrep);
281 nfsmout:
282 	return error;
283 }
284 
285 /*
286  * nfs access vnode op.
287  * For nfs version 2, just return ok. File accesses may fail later.
288  * For nfs version 3, use the access rpc to check accessibility. If file modes
289  * are changed on the server, accesses might still fail later.
290  *
291  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
292  *	      struct thread *a_td)
293  */
294 static int
295 nfs_access(struct vop_access_args *ap)
296 {
297 	struct vnode *vp = ap->a_vp;
298 	thread_t td = curthread;
299 	int error = 0;
300 	u_int32_t mode, wmode;
301 	int v3 = NFS_ISV3(vp);
302 	struct nfsnode *np = VTONFS(vp);
303 
304 	/*
305 	 * Disallow write attempts on filesystems mounted read-only;
306 	 * unless the file is a socket, fifo, or a block or character
307 	 * device resident on the filesystem.
308 	 */
309 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
310 		switch (vp->v_type) {
311 		case VREG:
312 		case VDIR:
313 		case VLNK:
314 			return (EROFS);
315 		default:
316 			break;
317 		}
318 	}
319 	/*
320 	 * For nfs v3, check to see if we have done this recently, and if
321 	 * so return our cached result instead of making an ACCESS call.
322 	 * If not, do an access rpc, otherwise you are stuck emulating
323 	 * ufs_access() locally using the vattr. This may not be correct,
324 	 * since the server may apply other access criteria such as
325 	 * client uid-->server uid mapping that we do not know about.
326 	 */
327 	if (v3) {
328 		if (ap->a_mode & VREAD)
329 			mode = NFSV3ACCESS_READ;
330 		else
331 			mode = 0;
332 		if (vp->v_type != VDIR) {
333 			if (ap->a_mode & VWRITE)
334 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
335 			if (ap->a_mode & VEXEC)
336 				mode |= NFSV3ACCESS_EXECUTE;
337 		} else {
338 			if (ap->a_mode & VWRITE)
339 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
340 					 NFSV3ACCESS_DELETE);
341 			if (ap->a_mode & VEXEC)
342 				mode |= NFSV3ACCESS_LOOKUP;
343 		}
344 		/* XXX safety belt, only make blanket request if caching */
345 		if (nfsaccess_cache_timeout > 0) {
346 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
347 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
348 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
349 		} else {
350 			wmode = mode;
351 		}
352 
353 		/*
354 		 * Does our cached result allow us to give a definite yes to
355 		 * this request?
356 		 */
357 		if (np->n_modestamp &&
358 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
359 		   (ap->a_cred->cr_uid == np->n_modeuid) &&
360 		   ((np->n_mode & mode) == mode)) {
361 			nfsstats.accesscache_hits++;
362 		} else {
363 			/*
364 			 * Either a no, or a don't know.  Go to the wire.
365 			 */
366 			nfsstats.accesscache_misses++;
367 		        error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
368 			if (!error) {
369 				if ((np->n_mode & mode) != mode) {
370 					error = EACCES;
371 				}
372 			}
373 		}
374 	} else {
375 		if ((error = nfsspec_access(ap)) != 0)
376 			return (error);
377 
378 		/*
379 		 * Attempt to prevent a mapped root from accessing a file
380 		 * which it shouldn't.  We try to read a byte from the file
381 		 * if the user is root and the file is not zero length.
382 		 * After calling nfsspec_access, we should have the correct
383 		 * file size cached.
384 		 */
385 		if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
386 		    && VTONFS(vp)->n_size > 0) {
387 			struct iovec aiov;
388 			struct uio auio;
389 			char buf[1];
390 
391 			aiov.iov_base = buf;
392 			aiov.iov_len = 1;
393 			auio.uio_iov = &aiov;
394 			auio.uio_iovcnt = 1;
395 			auio.uio_offset = 0;
396 			auio.uio_resid = 1;
397 			auio.uio_segflg = UIO_SYSSPACE;
398 			auio.uio_rw = UIO_READ;
399 			auio.uio_td = td;
400 
401 			if (vp->v_type == VREG) {
402 				error = nfs_readrpc(vp, &auio);
403 			} else if (vp->v_type == VDIR) {
404 				char* bp;
405 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
406 				aiov.iov_base = bp;
407 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
408 				error = nfs_readdirrpc(vp, &auio);
409 				free(bp, M_TEMP);
410 			} else if (vp->v_type == VLNK) {
411 				error = nfs_readlinkrpc(vp, &auio);
412 			} else {
413 				error = EACCES;
414 			}
415 		}
416 	}
417 	/*
418 	 * [re]record creds for reading and/or writing if access
419 	 * was granted.  Assume the NFS server will grant read access
420 	 * for execute requests.
421 	 */
422 	if (error == 0) {
423 		if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
424 			crhold(ap->a_cred);
425 			if (np->n_rucred)
426 				crfree(np->n_rucred);
427 			np->n_rucred = ap->a_cred;
428 		}
429 		if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
430 			crhold(ap->a_cred);
431 			if (np->n_wucred)
432 				crfree(np->n_wucred);
433 			np->n_wucred = ap->a_cred;
434 		}
435 	}
436 	return(error);
437 }
438 
439 /*
440  * nfs open vnode op
441  * Check to see if the type is ok
442  * and that deletion is not in progress.
443  * For paged in text files, you will need to flush the page cache
444  * if consistency is lost.
445  *
446  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
447  */
448 /* ARGSUSED */
449 static int
450 nfs_open(struct vop_open_args *ap)
451 {
452 	struct vnode *vp = ap->a_vp;
453 	struct nfsnode *np = VTONFS(vp);
454 	struct vattr vattr;
455 	int error;
456 
457 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
458 #ifdef DIAGNOSTIC
459 		printf("open eacces vtyp=%d\n",vp->v_type);
460 #endif
461 		return (EOPNOTSUPP);
462 	}
463 
464 	/*
465 	 * Clear the attribute cache only if opening with write access.  It
466 	 * is unclear if we should do this at all here, but we certainly
467 	 * should not clear the cache unconditionally simply because a file
468 	 * is being opened.
469 	 */
470 	if (ap->a_mode & FWRITE)
471 		np->n_attrstamp = 0;
472 
473 	/*
474 	 * For normal NFS, reconcile changes made locally verses
475 	 * changes made remotely.  Note that VOP_GETATTR only goes
476 	 * to the wire if the cached attribute has timed out or been
477 	 * cleared.
478 	 *
479 	 * If local modifications have been made clear the attribute
480 	 * cache to force an attribute and modified time check.  If
481 	 * GETATTR detects that the file has been changed by someone
482 	 * other then us it will set NRMODIFIED.
483 	 *
484 	 * If we are opening a directory and local changes have been
485 	 * made we have to invalidate the cache in order to ensure
486 	 * that we get the most up-to-date information from the
487 	 * server.  XXX
488 	 */
489 	if (np->n_flag & NLMODIFIED) {
490 		np->n_attrstamp = 0;
491 		if (vp->v_type == VDIR) {
492 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
493 			if (error == EINTR)
494 				return (error);
495 			nfs_invaldir(vp);
496 		}
497 	}
498 	error = VOP_GETATTR(vp, &vattr);
499 	if (error)
500 		return (error);
501 	if (np->n_flag & NRMODIFIED) {
502 		if (vp->v_type == VDIR)
503 			nfs_invaldir(vp);
504 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
505 		if (error == EINTR)
506 			return (error);
507 		np->n_flag &= ~NRMODIFIED;
508 	}
509 
510 	return (vop_stdopen(ap));
511 }
512 
513 /*
514  * nfs close vnode op
515  * What an NFS client should do upon close after writing is a debatable issue.
516  * Most NFS clients push delayed writes to the server upon close, basically for
517  * two reasons:
518  * 1 - So that any write errors may be reported back to the client process
519  *     doing the close system call. By far the two most likely errors are
520  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
521  * 2 - To put a worst case upper bound on cache inconsistency between
522  *     multiple clients for the file.
523  * There is also a consistency problem for Version 2 of the protocol w.r.t.
524  * not being able to tell if other clients are writing a file concurrently,
525  * since there is no way of knowing if the changed modify time in the reply
526  * is only due to the write for this client.
527  * (NFS Version 3 provides weak cache consistency data in the reply that
528  *  should be sufficient to detect and handle this case.)
529  *
530  * The current code does the following:
531  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
532  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
533  *                     or commit them (this satisfies 1 and 2 except for the
534  *                     case where the server crashes after this close but
535  *                     before the commit RPC, which is felt to be "good
536  *                     enough". Changing the last argument to nfs_flush() to
537  *                     a 1 would force a commit operation, if it is felt a
538  *                     commit is necessary now.
539  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
540  *                     1 should be dealt with via an fsync() system call for
541  *                     cases where write errors are important.
542  *
543  * nfs_close(struct vnode *a_vp, int a_fflag,
544  *	     struct ucred *a_cred, struct thread *a_td)
545  */
546 /* ARGSUSED */
547 static int
548 nfs_close(struct vop_close_args *ap)
549 {
550 	struct vnode *vp = ap->a_vp;
551 	struct nfsnode *np = VTONFS(vp);
552 	int error = 0;
553 	thread_t td = curthread;
554 
555 	if (vp->v_type == VREG) {
556 	    if (np->n_flag & NLMODIFIED) {
557 		if (NFS_ISV3(vp)) {
558 		    /*
559 		     * Under NFSv3 we have dirty buffers to dispose of.  We
560 		     * must flush them to the NFS server.  We have the option
561 		     * of waiting all the way through the commit rpc or just
562 		     * waiting for the initial write.  The default is to only
563 		     * wait through the initial write so the data is in the
564 		     * server's cache, which is roughly similar to the state
565 		     * a standard disk subsystem leaves the file in on close().
566 		     *
567 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
568 		     * potential races with other processes, and certainly
569 		     * cannot clear it if we don't commit.
570 		     */
571 		    int cm = nfsv3_commit_on_close ? 1 : 0;
572 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
573 		    /* np->n_flag &= ~NLMODIFIED; */
574 		} else {
575 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
576 		}
577 		np->n_attrstamp = 0;
578 	    }
579 	    if (np->n_flag & NWRITEERR) {
580 		np->n_flag &= ~NWRITEERR;
581 		error = np->n_error;
582 	    }
583 	}
584 	vop_stdclose(ap);
585 	return (error);
586 }
587 
588 /*
589  * nfs getattr call from vfs.
590  *
591  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
592  *		struct thread *a_td)
593  */
594 static int
595 nfs_getattr(struct vop_getattr_args *ap)
596 {
597 	struct vnode *vp = ap->a_vp;
598 	struct nfsnode *np = VTONFS(vp);
599 	caddr_t cp;
600 	u_int32_t *tl;
601 	int32_t t1, t2;
602 	caddr_t bpos, dpos;
603 	int error = 0;
604 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
605 	int v3 = NFS_ISV3(vp);
606 	thread_t td = curthread;
607 
608 	/*
609 	 * Update local times for special files.
610 	 */
611 	if (np->n_flag & (NACC | NUPD))
612 		np->n_flag |= NCHG;
613 	/*
614 	 * First look in the cache.
615 	 */
616 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
617 		return (0);
618 
619 	if (v3 && nfsaccess_cache_timeout > 0) {
620 		nfsstats.accesscache_misses++;
621 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
622 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
623 			return (0);
624 	}
625 
626 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
627 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
628 	nfsm_fhtom(vp, v3);
629 	nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK));
630 	if (!error) {
631 		nfsm_loadattr(vp, ap->a_vap);
632 	}
633 	m_freem(mrep);
634 nfsmout:
635 	return (error);
636 }
637 
638 /*
639  * nfs setattr call.
640  *
641  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
642  */
643 static int
644 nfs_setattr(struct vop_setattr_args *ap)
645 {
646 	struct vnode *vp = ap->a_vp;
647 	struct nfsnode *np = VTONFS(vp);
648 	struct vattr *vap = ap->a_vap;
649 	int error = 0;
650 	u_quad_t tsize;
651 	thread_t td = curthread;
652 
653 #ifndef nolint
654 	tsize = (u_quad_t)0;
655 #endif
656 
657 	/*
658 	 * Setting of flags is not supported.
659 	 */
660 	if (vap->va_flags != VNOVAL)
661 		return (EOPNOTSUPP);
662 
663 	/*
664 	 * Disallow write attempts if the filesystem is mounted read-only.
665 	 */
666   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
667 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
668 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
669 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
670 		return (EROFS);
671 	if (vap->va_size != VNOVAL) {
672  		switch (vp->v_type) {
673  		case VDIR:
674  			return (EISDIR);
675  		case VCHR:
676  		case VBLK:
677  		case VSOCK:
678  		case VFIFO:
679 			if (vap->va_mtime.tv_sec == VNOVAL &&
680 			    vap->va_atime.tv_sec == VNOVAL &&
681 			    vap->va_mode == (mode_t)VNOVAL &&
682 			    vap->va_uid == (uid_t)VNOVAL &&
683 			    vap->va_gid == (gid_t)VNOVAL)
684 				return (0);
685  			vap->va_size = VNOVAL;
686  			break;
687  		default:
688 			/*
689 			 * Disallow write attempts if the filesystem is
690 			 * mounted read-only.
691 			 */
692 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
693 				return (EROFS);
694 
695 			/*
696 			 * This is nasty.  The RPCs we send to flush pending
697 			 * data often return attribute information which is
698 			 * cached via a callback to nfs_loadattrcache(), which
699 			 * has the effect of changing our notion of the file
700 			 * size.  Due to flushed appends and other operations
701 			 * the file size can be set to virtually anything,
702 			 * including values that do not match either the old
703 			 * or intended file size.
704 			 *
705 			 * When this condition is detected we must loop to
706 			 * try the operation again.  Hopefully no more
707 			 * flushing is required on the loop so it works the
708 			 * second time around.  THIS CASE ALMOST ALWAYS
709 			 * HAPPENS!
710 			 */
711 			tsize = np->n_size;
712 again:
713 			error = nfs_meta_setsize(vp, td, vap->va_size);
714 
715  			if (np->n_flag & NLMODIFIED) {
716  			    if (vap->va_size == 0)
717  				error = nfs_vinvalbuf(vp, 0, 1);
718  			    else
719  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
720  			}
721 			/*
722 			 * note: this loop case almost always happens at
723 			 * least once per truncation.
724 			 */
725 			if (error == 0 && np->n_size != vap->va_size)
726 				goto again;
727 			np->n_vattr.va_size = vap->va_size;
728 			break;
729 		}
730   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
731 		vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
732 		vp->v_type == VREG &&
733   		(error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR
734 	) {
735 		return (error);
736 	}
737 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
738 
739 	/*
740 	 * Sanity check if a truncation was issued.  This should only occur
741 	 * if multiple processes are racing on the same file.
742 	 */
743 	if (error == 0 && vap->va_size != VNOVAL &&
744 	    np->n_size != vap->va_size) {
745 		printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
746 		goto again;
747 	}
748 	if (error && vap->va_size != VNOVAL) {
749 		np->n_size = np->n_vattr.va_size = tsize;
750 		vnode_pager_setsize(vp, np->n_size);
751 	}
752 	return (error);
753 }
754 
755 /*
756  * Do an nfs setattr rpc.
757  */
758 static int
759 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
760 	       struct ucred *cred, struct thread *td)
761 {
762 	struct nfsv2_sattr *sp;
763 	struct nfsnode *np = VTONFS(vp);
764 	caddr_t cp;
765 	int32_t t1, t2;
766 	caddr_t bpos, dpos, cp2;
767 	u_int32_t *tl;
768 	int error = 0, wccflag = NFSV3_WCCRATTR;
769 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
770 	int v3 = NFS_ISV3(vp);
771 
772 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
773 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
774 	nfsm_fhtom(vp, v3);
775 	if (v3) {
776 		nfsm_v3attrbuild(vap, TRUE);
777 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
778 		*tl = nfs_false;
779 	} else {
780 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
781 		if (vap->va_mode == (mode_t)VNOVAL)
782 			sp->sa_mode = nfs_xdrneg1;
783 		else
784 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
785 		if (vap->va_uid == (uid_t)VNOVAL)
786 			sp->sa_uid = nfs_xdrneg1;
787 		else
788 			sp->sa_uid = txdr_unsigned(vap->va_uid);
789 		if (vap->va_gid == (gid_t)VNOVAL)
790 			sp->sa_gid = nfs_xdrneg1;
791 		else
792 			sp->sa_gid = txdr_unsigned(vap->va_gid);
793 		sp->sa_size = txdr_unsigned(vap->va_size);
794 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
795 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
796 	}
797 	nfsm_request(vp, NFSPROC_SETATTR, td, cred);
798 	if (v3) {
799 		np->n_modestamp = 0;
800 		nfsm_wcc_data(vp, wccflag);
801 	} else
802 		nfsm_loadattr(vp, (struct vattr *)0);
803 	m_freem(mrep);
804 nfsmout:
805 	return (error);
806 }
807 
808 /*
809  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
810  * nfs_lookup() until all remaining new api calls are implemented.
811  *
812  * Resolve a namecache entry.  This function is passed a locked ncp and
813  * must call cache_setvp() on it as appropriate to resolve the entry.
814  */
815 static int
816 nfs_nresolve(struct vop_nresolve_args *ap)
817 {
818 	struct thread *td = curthread;
819 	struct namecache *ncp;
820 	struct ucred *cred;
821 	struct nfsnode *np;
822 	struct vnode *dvp;
823 	struct vnode *nvp;
824 	nfsfh_t *fhp;
825 	int attrflag;
826 	int fhsize;
827 	int error;
828 	int len;
829 	int v3;
830 	/******NFSM MACROS********/
831 	struct mbuf *mb, *mrep, *mreq, *mb2, *md;
832 	caddr_t bpos, dpos, cp, cp2;
833 	u_int32_t *tl;
834 	int32_t t1, t2;
835 
836 	cred = ap->a_cred;
837 	ncp = ap->a_ncp;
838 
839 	KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
840 	dvp = ncp->nc_parent->nc_vp;
841 	if ((error = vget(dvp, LK_SHARED)) != 0)
842 		return (error);
843 
844 	nvp = NULL;
845 	v3 = NFS_ISV3(dvp);
846 	nfsstats.lookupcache_misses++;
847 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
848 	len = ncp->nc_nlen;
849 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
850 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
851 	nfsm_fhtom(dvp, v3);
852 	nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
853 	nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
854 	if (error) {
855 		/*
856 		 * Cache negatve lookups to reduce NFS traffic, but use
857 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
858 		 * XXX we should add a namecache flag for no-caching
859 		 * to uncache the negative hit as soon as possible, but
860 		 * we cannot simply destroy the entry because it is used
861 		 * as a placeholder by the caller.
862 		 */
863 		if (error == ENOENT) {
864 			int nticks;
865 
866 			if (nfsneg_cache_timeout)
867 				nticks = nfsneg_cache_timeout * hz;
868 			else
869 				nticks = 1;
870 			cache_setvp(ncp, NULL);
871 			cache_settimeout(ncp, nticks);
872 		}
873 		nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
874 		m_freem(mrep);
875 		goto nfsmout;
876 	}
877 
878 	/*
879 	 * Success, get the file handle, do various checks, and load
880 	 * post-operation data from the reply packet.  Theoretically
881 	 * we should never be looking up "." so, theoretically, we
882 	 * should never get the same file handle as our directory.  But
883 	 * we check anyway. XXX
884 	 *
885 	 * Note that no timeout is set for the positive cache hit.  We
886 	 * assume, theoretically, that ESTALE returns will be dealt with
887 	 * properly to handle NFS races and in anycase we cannot depend
888 	 * on a timeout to deal with NFS open/create/excl issues so instead
889 	 * of a bad hack here the rest of the NFS client code needs to do
890 	 * the right thing.
891 	 */
892 	nfsm_getfh(fhp, fhsize, v3);
893 
894 	np = VTONFS(dvp);
895 	if (NFS_CMPFH(np, fhp, fhsize)) {
896 		vref(dvp);
897 		nvp = dvp;
898 	} else {
899 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
900 		if (error) {
901 			m_freem(mrep);
902 			vput(dvp);
903 			return (error);
904 		}
905 		nvp = NFSTOV(np);
906 	}
907 	if (v3) {
908 		nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
909 		nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
910 	} else {
911 		nfsm_loadattr(nvp, NULL);
912 	}
913 	cache_setvp(ncp, nvp);
914 	m_freem(mrep);
915 nfsmout:
916 	vput(dvp);
917 	if (nvp) {
918 		if (nvp == dvp)
919 			vrele(nvp);
920 		else
921 			vput(nvp);
922 	}
923 	return (error);
924 }
925 
926 /*
927  * 'cached' nfs directory lookup
928  *
929  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
930  *
931  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
932  *	      struct componentname *a_cnp)
933  */
934 static int
935 nfs_lookup(struct vop_old_lookup_args *ap)
936 {
937 	struct componentname *cnp = ap->a_cnp;
938 	struct vnode *dvp = ap->a_dvp;
939 	struct vnode **vpp = ap->a_vpp;
940 	int flags = cnp->cn_flags;
941 	struct vnode *newvp;
942 	u_int32_t *tl;
943 	caddr_t cp;
944 	int32_t t1, t2;
945 	struct nfsmount *nmp;
946 	caddr_t bpos, dpos, cp2;
947 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
948 	long len;
949 	nfsfh_t *fhp;
950 	struct nfsnode *np;
951 	int lockparent, wantparent, error = 0, attrflag, fhsize;
952 	int v3 = NFS_ISV3(dvp);
953 
954 	/*
955 	 * Read-only mount check and directory check.
956 	 */
957 	*vpp = NULLVP;
958 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
959 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
960 		return (EROFS);
961 
962 	if (dvp->v_type != VDIR)
963 		return (ENOTDIR);
964 
965 	/*
966 	 * Look it up in the cache.  Note that ENOENT is only returned if we
967 	 * previously entered a negative hit (see later on).  The additional
968 	 * nfsneg_cache_timeout check causes previously cached results to
969 	 * be instantly ignored if the negative caching is turned off.
970 	 */
971 	lockparent = flags & CNP_LOCKPARENT;
972 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
973 	nmp = VFSTONFS(dvp->v_mount);
974 	np = VTONFS(dvp);
975 
976 	/*
977 	 * Go to the wire.
978 	 */
979 	error = 0;
980 	newvp = NULLVP;
981 	nfsstats.lookupcache_misses++;
982 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
983 	len = cnp->cn_namelen;
984 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
985 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
986 	nfsm_fhtom(dvp, v3);
987 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
988 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
989 	if (error) {
990 		nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
991 		m_freem(mrep);
992 		goto nfsmout;
993 	}
994 	nfsm_getfh(fhp, fhsize, v3);
995 
996 	/*
997 	 * Handle RENAME case...
998 	 */
999 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1000 		if (NFS_CMPFH(np, fhp, fhsize)) {
1001 			m_freem(mrep);
1002 			return (EISDIR);
1003 		}
1004 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1005 		if (error) {
1006 			m_freem(mrep);
1007 			return (error);
1008 		}
1009 		newvp = NFSTOV(np);
1010 		if (v3) {
1011 			nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1012 			nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1013 		} else
1014 			nfsm_loadattr(newvp, (struct vattr *)0);
1015 		*vpp = newvp;
1016 		m_freem(mrep);
1017 		if (!lockparent) {
1018 			vn_unlock(dvp);
1019 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1020 		}
1021 		return (0);
1022 	}
1023 
1024 	if (flags & CNP_ISDOTDOT) {
1025 		vn_unlock(dvp);
1026 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1027 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1028 		if (error) {
1029 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1030 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1031 			return (error); /* NOTE: return error from nget */
1032 		}
1033 		newvp = NFSTOV(np);
1034 		if (lockparent) {
1035 			error = vn_lock(dvp, LK_EXCLUSIVE);
1036 			if (error) {
1037 				vput(newvp);
1038 				return (error);
1039 			}
1040 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1041 		}
1042 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1043 		vref(dvp);
1044 		newvp = dvp;
1045 	} else {
1046 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1047 		if (error) {
1048 			m_freem(mrep);
1049 			return (error);
1050 		}
1051 		if (!lockparent) {
1052 			vn_unlock(dvp);
1053 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1054 		}
1055 		newvp = NFSTOV(np);
1056 	}
1057 	if (v3) {
1058 		nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1059 		nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1060 	} else
1061 		nfsm_loadattr(newvp, (struct vattr *)0);
1062 #if 0
1063 	/* XXX MOVE TO nfs_nremove() */
1064 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1065 	    cnp->cn_nameiop != NAMEI_DELETE) {
1066 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1067 	}
1068 #endif
1069 	*vpp = newvp;
1070 	m_freem(mrep);
1071 nfsmout:
1072 	if (error) {
1073 		if (newvp != NULLVP) {
1074 			vrele(newvp);
1075 			*vpp = NULLVP;
1076 		}
1077 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1078 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1079 		    error == ENOENT) {
1080 			if (!lockparent) {
1081 				vn_unlock(dvp);
1082 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1083 			}
1084 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1085 				error = EROFS;
1086 			else
1087 				error = EJUSTRETURN;
1088 		}
1089 	}
1090 	return (error);
1091 }
1092 
1093 /*
1094  * nfs read call.
1095  * Just call nfs_bioread() to do the work.
1096  *
1097  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1098  *	    struct ucred *a_cred)
1099  */
1100 static int
1101 nfs_read(struct vop_read_args *ap)
1102 {
1103 	struct vnode *vp = ap->a_vp;
1104 
1105 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1106 	switch (vp->v_type) {
1107 	case VREG:
1108 		return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1109 	case VDIR:
1110 		return (EISDIR);
1111 	default:
1112 		return EOPNOTSUPP;
1113 	}
1114 }
1115 
1116 /*
1117  * nfs readlink call
1118  *
1119  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1120  */
1121 static int
1122 nfs_readlink(struct vop_readlink_args *ap)
1123 {
1124 	struct vnode *vp = ap->a_vp;
1125 
1126 	if (vp->v_type != VLNK)
1127 		return (EINVAL);
1128 	return (nfs_bioread(vp, ap->a_uio, 0));
1129 }
1130 
1131 /*
1132  * Do a readlink rpc.
1133  * Called by nfs_doio() from below the buffer cache.
1134  */
1135 int
1136 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1137 {
1138 	u_int32_t *tl;
1139 	caddr_t cp;
1140 	int32_t t1, t2;
1141 	caddr_t bpos, dpos, cp2;
1142 	int error = 0, len, attrflag;
1143 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1144 	int v3 = NFS_ISV3(vp);
1145 
1146 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1147 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1148 	nfsm_fhtom(vp, v3);
1149 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1150 	if (v3)
1151 		nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1152 	if (!error) {
1153 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1154 		if (len == NFS_MAXPATHLEN) {
1155 			struct nfsnode *np = VTONFS(vp);
1156 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1157 				len = np->n_size;
1158 		}
1159 		nfsm_mtouio(uiop, len);
1160 	}
1161 	m_freem(mrep);
1162 nfsmout:
1163 	return (error);
1164 }
1165 
1166 /*
1167  * nfs read rpc call
1168  * Ditto above
1169  */
1170 int
1171 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1172 {
1173 	u_int32_t *tl;
1174 	caddr_t cp;
1175 	int32_t t1, t2;
1176 	caddr_t bpos, dpos, cp2;
1177 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1178 	struct nfsmount *nmp;
1179 	int error = 0, len, retlen, tsiz, eof, attrflag;
1180 	int v3 = NFS_ISV3(vp);
1181 
1182 #ifndef nolint
1183 	eof = 0;
1184 #endif
1185 	nmp = VFSTONFS(vp->v_mount);
1186 	tsiz = uiop->uio_resid;
1187 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1188 		return (EFBIG);
1189 	while (tsiz > 0) {
1190 		nfsstats.rpccnt[NFSPROC_READ]++;
1191 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1192 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1193 		nfsm_fhtom(vp, v3);
1194 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1195 		if (v3) {
1196 			txdr_hyper(uiop->uio_offset, tl);
1197 			*(tl + 2) = txdr_unsigned(len);
1198 		} else {
1199 			*tl++ = txdr_unsigned(uiop->uio_offset);
1200 			*tl++ = txdr_unsigned(len);
1201 			*tl = 0;
1202 		}
1203 		nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1204 		if (v3) {
1205 			nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1206 			if (error) {
1207 				m_freem(mrep);
1208 				goto nfsmout;
1209 			}
1210 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1211 			eof = fxdr_unsigned(int, *(tl + 1));
1212 		} else
1213 			nfsm_loadattr(vp, (struct vattr *)0);
1214 		nfsm_strsiz(retlen, nmp->nm_rsize);
1215 		nfsm_mtouio(uiop, retlen);
1216 		m_freem(mrep);
1217 		tsiz -= retlen;
1218 		if (v3) {
1219 			if (eof || retlen == 0) {
1220 				tsiz = 0;
1221 			}
1222 		} else if (retlen < len) {
1223 			tsiz = 0;
1224 		}
1225 	}
1226 nfsmout:
1227 	return (error);
1228 }
1229 
1230 /*
1231  * nfs write call
1232  */
1233 int
1234 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1235 {
1236 	u_int32_t *tl;
1237 	caddr_t cp;
1238 	int32_t t1, t2, backup;
1239 	caddr_t bpos, dpos, cp2;
1240 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1241 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1242 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1243 	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1244 
1245 #ifndef DIAGNOSTIC
1246 	if (uiop->uio_iovcnt != 1)
1247 		panic("nfs: writerpc iovcnt > 1");
1248 #endif
1249 	*must_commit = 0;
1250 	tsiz = uiop->uio_resid;
1251 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1252 		return (EFBIG);
1253 	while (tsiz > 0) {
1254 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1255 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1256 		nfsm_reqhead(vp, NFSPROC_WRITE,
1257 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1258 		nfsm_fhtom(vp, v3);
1259 		if (v3) {
1260 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1261 			txdr_hyper(uiop->uio_offset, tl);
1262 			tl += 2;
1263 			*tl++ = txdr_unsigned(len);
1264 			*tl++ = txdr_unsigned(*iomode);
1265 			*tl = txdr_unsigned(len);
1266 		} else {
1267 			u_int32_t x;
1268 
1269 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1270 			/* Set both "begin" and "current" to non-garbage. */
1271 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1272 			*tl++ = x;	/* "begin offset" */
1273 			*tl++ = x;	/* "current offset" */
1274 			x = txdr_unsigned(len);
1275 			*tl++ = x;	/* total to this offset */
1276 			*tl = x;	/* size of this write */
1277 		}
1278 		nfsm_uiotom(uiop, len);
1279 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1280 		if (v3) {
1281 			/*
1282 			 * The write RPC returns a before and after mtime.  The
1283 			 * nfsm_wcc_data() macro checks the before n_mtime
1284 			 * against the before time and stores the after time
1285 			 * in the nfsnode's cached vattr and n_mtime field.
1286 			 * The NRMODIFIED bit will be set if the before
1287 			 * time did not match the original mtime.
1288 			 */
1289 			wccflag = NFSV3_WCCCHK;
1290 			nfsm_wcc_data(vp, wccflag);
1291 			if (!error) {
1292 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1293 					+ NFSX_V3WRITEVERF);
1294 				rlen = fxdr_unsigned(int, *tl++);
1295 				if (rlen == 0) {
1296 					error = NFSERR_IO;
1297 					m_freem(mrep);
1298 					break;
1299 				} else if (rlen < len) {
1300 					backup = len - rlen;
1301 					uiop->uio_iov->iov_base -= backup;
1302 					uiop->uio_iov->iov_len += backup;
1303 					uiop->uio_offset -= backup;
1304 					uiop->uio_resid += backup;
1305 					len = rlen;
1306 				}
1307 				commit = fxdr_unsigned(int, *tl++);
1308 
1309 				/*
1310 				 * Return the lowest committment level
1311 				 * obtained by any of the RPCs.
1312 				 */
1313 				if (committed == NFSV3WRITE_FILESYNC)
1314 					committed = commit;
1315 				else if (committed == NFSV3WRITE_DATASYNC &&
1316 					commit == NFSV3WRITE_UNSTABLE)
1317 					committed = commit;
1318 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1319 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1320 					NFSX_V3WRITEVERF);
1321 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1322 				} else if (bcmp((caddr_t)tl,
1323 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1324 				    *must_commit = 1;
1325 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1326 					NFSX_V3WRITEVERF);
1327 				}
1328 			}
1329 		} else {
1330 			nfsm_loadattr(vp, (struct vattr *)0);
1331 		}
1332 		m_freem(mrep);
1333 		if (error)
1334 			break;
1335 		tsiz -= len;
1336 	}
1337 nfsmout:
1338 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1339 		committed = NFSV3WRITE_FILESYNC;
1340 	*iomode = committed;
1341 	if (error)
1342 		uiop->uio_resid = tsiz;
1343 	return (error);
1344 }
1345 
1346 /*
1347  * nfs mknod rpc
1348  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1349  * mode set to specify the file type and the size field for rdev.
1350  */
1351 static int
1352 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1353 	     struct vattr *vap)
1354 {
1355 	struct nfsv2_sattr *sp;
1356 	u_int32_t *tl;
1357 	caddr_t cp;
1358 	int32_t t1, t2;
1359 	struct vnode *newvp = (struct vnode *)0;
1360 	struct nfsnode *np = (struct nfsnode *)0;
1361 	struct vattr vattr;
1362 	char *cp2;
1363 	caddr_t bpos, dpos;
1364 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1365 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1366 	u_int32_t rdev;
1367 	int v3 = NFS_ISV3(dvp);
1368 
1369 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1370 		rdev = txdr_unsigned(vap->va_rdev);
1371 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1372 		rdev = nfs_xdrneg1;
1373 	else {
1374 		return (EOPNOTSUPP);
1375 	}
1376 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1377 		return (error);
1378 	}
1379 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1380 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1381 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1382 	nfsm_fhtom(dvp, v3);
1383 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1384 	if (v3) {
1385 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1386 		*tl++ = vtonfsv3_type(vap->va_type);
1387 		nfsm_v3attrbuild(vap, FALSE);
1388 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1389 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1390 			*tl++ = txdr_unsigned(umajor(vap->va_rdev));
1391 			*tl = txdr_unsigned(uminor(vap->va_rdev));
1392 		}
1393 	} else {
1394 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1395 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1396 		sp->sa_uid = nfs_xdrneg1;
1397 		sp->sa_gid = nfs_xdrneg1;
1398 		sp->sa_size = rdev;
1399 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1400 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1401 	}
1402 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1403 	if (!error) {
1404 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1405 		if (!gotvp) {
1406 			if (newvp) {
1407 				vput(newvp);
1408 				newvp = (struct vnode *)0;
1409 			}
1410 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1411 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1412 			if (!error)
1413 				newvp = NFSTOV(np);
1414 		}
1415 	}
1416 	if (v3)
1417 		nfsm_wcc_data(dvp, wccflag);
1418 	m_freem(mrep);
1419 nfsmout:
1420 	if (error) {
1421 		if (newvp)
1422 			vput(newvp);
1423 	} else {
1424 		*vpp = newvp;
1425 	}
1426 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1427 	if (!wccflag)
1428 		VTONFS(dvp)->n_attrstamp = 0;
1429 	return (error);
1430 }
1431 
1432 /*
1433  * nfs mknod vop
1434  * just call nfs_mknodrpc() to do the work.
1435  *
1436  * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1437  *	     struct componentname *a_cnp, struct vattr *a_vap)
1438  */
1439 /* ARGSUSED */
1440 static int
1441 nfs_mknod(struct vop_old_mknod_args *ap)
1442 {
1443 	return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1444 }
1445 
1446 static u_long create_verf;
1447 /*
1448  * nfs file create call
1449  *
1450  * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1451  *	      struct componentname *a_cnp, struct vattr *a_vap)
1452  */
1453 static int
1454 nfs_create(struct vop_old_create_args *ap)
1455 {
1456 	struct vnode *dvp = ap->a_dvp;
1457 	struct vattr *vap = ap->a_vap;
1458 	struct componentname *cnp = ap->a_cnp;
1459 	struct nfsv2_sattr *sp;
1460 	u_int32_t *tl;
1461 	caddr_t cp;
1462 	int32_t t1, t2;
1463 	struct nfsnode *np = (struct nfsnode *)0;
1464 	struct vnode *newvp = (struct vnode *)0;
1465 	caddr_t bpos, dpos, cp2;
1466 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1467 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1468 	struct vattr vattr;
1469 	int v3 = NFS_ISV3(dvp);
1470 
1471 	/*
1472 	 * Oops, not for me..
1473 	 */
1474 	if (vap->va_type == VSOCK)
1475 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1476 
1477 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1478 		return (error);
1479 	}
1480 	if (vap->va_vaflags & VA_EXCLUSIVE)
1481 		fmode |= O_EXCL;
1482 again:
1483 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1484 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1485 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1486 	nfsm_fhtom(dvp, v3);
1487 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1488 	if (v3) {
1489 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1490 		if (fmode & O_EXCL) {
1491 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1492 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1493 #ifdef INET
1494 			if (!TAILQ_EMPTY(&in_ifaddrhead))
1495 				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1496 			else
1497 #endif
1498 				*tl++ = create_verf;
1499 			*tl = ++create_verf;
1500 		} else {
1501 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1502 			nfsm_v3attrbuild(vap, FALSE);
1503 		}
1504 	} else {
1505 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1506 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1507 		sp->sa_uid = nfs_xdrneg1;
1508 		sp->sa_gid = nfs_xdrneg1;
1509 		sp->sa_size = 0;
1510 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1511 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1512 	}
1513 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1514 	if (!error) {
1515 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1516 		if (!gotvp) {
1517 			if (newvp) {
1518 				vput(newvp);
1519 				newvp = (struct vnode *)0;
1520 			}
1521 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1522 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1523 			if (!error)
1524 				newvp = NFSTOV(np);
1525 		}
1526 	}
1527 	if (v3)
1528 		nfsm_wcc_data(dvp, wccflag);
1529 	m_freem(mrep);
1530 nfsmout:
1531 	if (error) {
1532 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1533 			fmode &= ~O_EXCL;
1534 			goto again;
1535 		}
1536 		if (newvp)
1537 			vput(newvp);
1538 	} else if (v3 && (fmode & O_EXCL)) {
1539 		/*
1540 		 * We are normally called with only a partially initialized
1541 		 * VAP.  Since the NFSv3 spec says that server may use the
1542 		 * file attributes to store the verifier, the spec requires
1543 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1544 		 * in atime, but we can't really assume that all servers will
1545 		 * so we ensure that our SETATTR sets both atime and mtime.
1546 		 */
1547 		if (vap->va_mtime.tv_sec == VNOVAL)
1548 			vfs_timestamp(&vap->va_mtime);
1549 		if (vap->va_atime.tv_sec == VNOVAL)
1550 			vap->va_atime = vap->va_mtime;
1551 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1552 	}
1553 	if (!error) {
1554 		/*
1555 		 * The new np may have enough info for access
1556 		 * checks, make sure rucred and wucred are
1557 		 * initialized for read and write rpc's.
1558 		 */
1559 		np = VTONFS(newvp);
1560 		if (np->n_rucred == NULL)
1561 			np->n_rucred = crhold(cnp->cn_cred);
1562 		if (np->n_wucred == NULL)
1563 			np->n_wucred = crhold(cnp->cn_cred);
1564 		*ap->a_vpp = newvp;
1565 	}
1566 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1567 	if (!wccflag)
1568 		VTONFS(dvp)->n_attrstamp = 0;
1569 	return (error);
1570 }
1571 
1572 /*
1573  * nfs file remove call
1574  * To try and make nfs semantics closer to ufs semantics, a file that has
1575  * other processes using the vnode is renamed instead of removed and then
1576  * removed later on the last close.
1577  * - If v_usecount > 1
1578  *	  If a rename is not already in the works
1579  *	     call nfs_sillyrename() to set it up
1580  *     else
1581  *	  do the remove rpc
1582  *
1583  * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1584  *	      struct componentname *a_cnp)
1585  */
1586 static int
1587 nfs_remove(struct vop_old_remove_args *ap)
1588 {
1589 	struct vnode *vp = ap->a_vp;
1590 	struct vnode *dvp = ap->a_dvp;
1591 	struct componentname *cnp = ap->a_cnp;
1592 	struct nfsnode *np = VTONFS(vp);
1593 	int error = 0;
1594 	struct vattr vattr;
1595 
1596 #ifndef DIAGNOSTIC
1597 	if (vp->v_usecount < 1)
1598 		panic("nfs_remove: bad v_usecount");
1599 #endif
1600 	if (vp->v_type == VDIR)
1601 		error = EPERM;
1602 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1603 	    VOP_GETATTR(vp, &vattr) == 0 &&
1604 	    vattr.va_nlink > 1)) {
1605 		/*
1606 		 * throw away biocache buffers, mainly to avoid
1607 		 * unnecessary delayed writes later.
1608 		 */
1609 		error = nfs_vinvalbuf(vp, 0, 1);
1610 		/* Do the rpc */
1611 		if (error != EINTR)
1612 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1613 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1614 		/*
1615 		 * Kludge City: If the first reply to the remove rpc is lost..
1616 		 *   the reply to the retransmitted request will be ENOENT
1617 		 *   since the file was in fact removed
1618 		 *   Therefore, we cheat and return success.
1619 		 */
1620 		if (error == ENOENT)
1621 			error = 0;
1622 	} else if (!np->n_sillyrename) {
1623 		error = nfs_sillyrename(dvp, vp, cnp);
1624 	}
1625 	np->n_attrstamp = 0;
1626 	return (error);
1627 }
1628 
1629 /*
1630  * nfs file remove rpc called from nfs_inactive
1631  */
1632 int
1633 nfs_removeit(struct sillyrename *sp)
1634 {
1635 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1636 		sp->s_cred, NULL));
1637 }
1638 
1639 /*
1640  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1641  */
1642 static int
1643 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1644 	      struct ucred *cred, struct thread *td)
1645 {
1646 	u_int32_t *tl;
1647 	caddr_t cp;
1648 	int32_t t1, t2;
1649 	caddr_t bpos, dpos, cp2;
1650 	int error = 0, wccflag = NFSV3_WCCRATTR;
1651 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1652 	int v3 = NFS_ISV3(dvp);
1653 
1654 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1655 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1656 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1657 	nfsm_fhtom(dvp, v3);
1658 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1659 	nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1660 	if (v3)
1661 		nfsm_wcc_data(dvp, wccflag);
1662 	m_freem(mrep);
1663 nfsmout:
1664 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1665 	if (!wccflag)
1666 		VTONFS(dvp)->n_attrstamp = 0;
1667 	return (error);
1668 }
1669 
1670 /*
1671  * nfs file rename call
1672  *
1673  * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1674  *	      struct componentname *a_fcnp, struct vnode *a_tdvp,
1675  *	      struct vnode *a_tvp, struct componentname *a_tcnp)
1676  */
1677 static int
1678 nfs_rename(struct vop_old_rename_args *ap)
1679 {
1680 	struct vnode *fvp = ap->a_fvp;
1681 	struct vnode *tvp = ap->a_tvp;
1682 	struct vnode *fdvp = ap->a_fdvp;
1683 	struct vnode *tdvp = ap->a_tdvp;
1684 	struct componentname *tcnp = ap->a_tcnp;
1685 	struct componentname *fcnp = ap->a_fcnp;
1686 	int error;
1687 
1688 	/* Check for cross-device rename */
1689 	if ((fvp->v_mount != tdvp->v_mount) ||
1690 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1691 		error = EXDEV;
1692 		goto out;
1693 	}
1694 
1695 	/*
1696 	 * We have to flush B_DELWRI data prior to renaming
1697 	 * the file.  If we don't, the delayed-write buffers
1698 	 * can be flushed out later after the file has gone stale
1699 	 * under NFSV3.  NFSV2 does not have this problem because
1700 	 * ( as far as I can tell ) it flushes dirty buffers more
1701 	 * often.
1702 	 */
1703 
1704 	VOP_FSYNC(fvp, MNT_WAIT);
1705 	if (tvp)
1706 	    VOP_FSYNC(tvp, MNT_WAIT);
1707 
1708 	/*
1709 	 * If the tvp exists and is in use, sillyrename it before doing the
1710 	 * rename of the new file over it.
1711 	 *
1712 	 * XXX Can't sillyrename a directory.
1713 	 *
1714 	 * We do not attempt to do any namecache purges in this old API
1715 	 * routine.  The new API compat functions have access to the actual
1716 	 * namecache structures and will do it for us.
1717 	 */
1718 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1719 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1720 		vput(tvp);
1721 		tvp = NULL;
1722 	} else if (tvp) {
1723 		;
1724 	}
1725 
1726 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1727 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1728 		tcnp->cn_td);
1729 
1730 out:
1731 	if (tdvp == tvp)
1732 		vrele(tdvp);
1733 	else
1734 		vput(tdvp);
1735 	if (tvp)
1736 		vput(tvp);
1737 	vrele(fdvp);
1738 	vrele(fvp);
1739 	/*
1740 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1741 	 */
1742 	if (error == ENOENT)
1743 		error = 0;
1744 	return (error);
1745 }
1746 
1747 /*
1748  * nfs file rename rpc called from nfs_remove() above
1749  */
1750 static int
1751 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1752 	     struct sillyrename *sp)
1753 {
1754 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1755 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1756 }
1757 
1758 /*
1759  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1760  */
1761 static int
1762 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1763 	      struct vnode *tdvp, const char *tnameptr, int tnamelen,
1764 	      struct ucred *cred, struct thread *td)
1765 {
1766 	u_int32_t *tl;
1767 	caddr_t cp;
1768 	int32_t t1, t2;
1769 	caddr_t bpos, dpos, cp2;
1770 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1771 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1772 	int v3 = NFS_ISV3(fdvp);
1773 
1774 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1775 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1776 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1777 		nfsm_rndup(tnamelen));
1778 	nfsm_fhtom(fdvp, v3);
1779 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1780 	nfsm_fhtom(tdvp, v3);
1781 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1782 	nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1783 	if (v3) {
1784 		nfsm_wcc_data(fdvp, fwccflag);
1785 		nfsm_wcc_data(tdvp, twccflag);
1786 	}
1787 	m_freem(mrep);
1788 nfsmout:
1789 	VTONFS(fdvp)->n_flag |= NLMODIFIED;
1790 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
1791 	if (!fwccflag)
1792 		VTONFS(fdvp)->n_attrstamp = 0;
1793 	if (!twccflag)
1794 		VTONFS(tdvp)->n_attrstamp = 0;
1795 	return (error);
1796 }
1797 
1798 /*
1799  * nfs hard link create call
1800  *
1801  * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1802  *	    struct componentname *a_cnp)
1803  */
1804 static int
1805 nfs_link(struct vop_old_link_args *ap)
1806 {
1807 	struct vnode *vp = ap->a_vp;
1808 	struct vnode *tdvp = ap->a_tdvp;
1809 	struct componentname *cnp = ap->a_cnp;
1810 	u_int32_t *tl;
1811 	caddr_t cp;
1812 	int32_t t1, t2;
1813 	caddr_t bpos, dpos, cp2;
1814 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1815 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1816 	int v3;
1817 
1818 	if (vp->v_mount != tdvp->v_mount) {
1819 		return (EXDEV);
1820 	}
1821 
1822 	/*
1823 	 * Push all writes to the server, so that the attribute cache
1824 	 * doesn't get "out of sync" with the server.
1825 	 * XXX There should be a better way!
1826 	 */
1827 	VOP_FSYNC(vp, MNT_WAIT);
1828 
1829 	v3 = NFS_ISV3(vp);
1830 	nfsstats.rpccnt[NFSPROC_LINK]++;
1831 	nfsm_reqhead(vp, NFSPROC_LINK,
1832 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1833 	nfsm_fhtom(vp, v3);
1834 	nfsm_fhtom(tdvp, v3);
1835 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1836 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1837 	if (v3) {
1838 		nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1839 		nfsm_wcc_data(tdvp, wccflag);
1840 	}
1841 	m_freem(mrep);
1842 nfsmout:
1843 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
1844 	if (!attrflag)
1845 		VTONFS(vp)->n_attrstamp = 0;
1846 	if (!wccflag)
1847 		VTONFS(tdvp)->n_attrstamp = 0;
1848 	/*
1849 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1850 	 */
1851 	if (error == EEXIST)
1852 		error = 0;
1853 	return (error);
1854 }
1855 
1856 /*
1857  * nfs symbolic link create call
1858  *
1859  * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1860  *		struct componentname *a_cnp, struct vattr *a_vap,
1861  *		char *a_target)
1862  */
1863 static int
1864 nfs_symlink(struct vop_old_symlink_args *ap)
1865 {
1866 	struct vnode *dvp = ap->a_dvp;
1867 	struct vattr *vap = ap->a_vap;
1868 	struct componentname *cnp = ap->a_cnp;
1869 	struct nfsv2_sattr *sp;
1870 	u_int32_t *tl;
1871 	caddr_t cp;
1872 	int32_t t1, t2;
1873 	caddr_t bpos, dpos, cp2;
1874 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1875 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1876 	struct vnode *newvp = (struct vnode *)0;
1877 	int v3 = NFS_ISV3(dvp);
1878 
1879 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1880 	slen = strlen(ap->a_target);
1881 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1882 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1883 	nfsm_fhtom(dvp, v3);
1884 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1885 	if (v3) {
1886 		nfsm_v3attrbuild(vap, FALSE);
1887 	}
1888 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1889 	if (!v3) {
1890 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1891 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1892 		sp->sa_uid = nfs_xdrneg1;
1893 		sp->sa_gid = nfs_xdrneg1;
1894 		sp->sa_size = nfs_xdrneg1;
1895 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1896 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1897 	}
1898 
1899 	/*
1900 	 * Issue the NFS request and get the rpc response.
1901 	 *
1902 	 * Only NFSv3 responses returning an error of 0 actually return
1903 	 * a file handle that can be converted into newvp without having
1904 	 * to do an extra lookup rpc.
1905 	 */
1906 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1907 	if (v3) {
1908 		if (error == 0)
1909 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1910 		nfsm_wcc_data(dvp, wccflag);
1911 	}
1912 
1913 	/*
1914 	 * out code jumps -> here, mrep is also freed.
1915 	 */
1916 
1917 	m_freem(mrep);
1918 nfsmout:
1919 
1920 	/*
1921 	 * If we get an EEXIST error, silently convert it to no-error
1922 	 * in case of an NFS retry.
1923 	 */
1924 	if (error == EEXIST)
1925 		error = 0;
1926 
1927 	/*
1928 	 * If we do not have (or no longer have) an error, and we could
1929 	 * not extract the newvp from the response due to the request being
1930 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
1931 	 * to obtain a newvp to return.
1932 	 */
1933 	if (error == 0 && newvp == NULL) {
1934 		struct nfsnode *np = NULL;
1935 
1936 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1937 		    cnp->cn_cred, cnp->cn_td, &np);
1938 		if (!error)
1939 			newvp = NFSTOV(np);
1940 	}
1941 	if (error) {
1942 		if (newvp)
1943 			vput(newvp);
1944 	} else {
1945 		*ap->a_vpp = newvp;
1946 	}
1947 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1948 	if (!wccflag)
1949 		VTONFS(dvp)->n_attrstamp = 0;
1950 	return (error);
1951 }
1952 
1953 /*
1954  * nfs make dir call
1955  *
1956  * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1957  *	     struct componentname *a_cnp, struct vattr *a_vap)
1958  */
1959 static int
1960 nfs_mkdir(struct vop_old_mkdir_args *ap)
1961 {
1962 	struct vnode *dvp = ap->a_dvp;
1963 	struct vattr *vap = ap->a_vap;
1964 	struct componentname *cnp = ap->a_cnp;
1965 	struct nfsv2_sattr *sp;
1966 	u_int32_t *tl;
1967 	caddr_t cp;
1968 	int32_t t1, t2;
1969 	int len;
1970 	struct nfsnode *np = (struct nfsnode *)0;
1971 	struct vnode *newvp = (struct vnode *)0;
1972 	caddr_t bpos, dpos, cp2;
1973 	int error = 0, wccflag = NFSV3_WCCRATTR;
1974 	int gotvp = 0;
1975 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1976 	struct vattr vattr;
1977 	int v3 = NFS_ISV3(dvp);
1978 
1979 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1980 		return (error);
1981 	}
1982 	len = cnp->cn_namelen;
1983 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1984 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1985 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1986 	nfsm_fhtom(dvp, v3);
1987 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1988 	if (v3) {
1989 		nfsm_v3attrbuild(vap, FALSE);
1990 	} else {
1991 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1992 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1993 		sp->sa_uid = nfs_xdrneg1;
1994 		sp->sa_gid = nfs_xdrneg1;
1995 		sp->sa_size = nfs_xdrneg1;
1996 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1997 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1998 	}
1999 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2000 	if (!error)
2001 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2002 	if (v3)
2003 		nfsm_wcc_data(dvp, wccflag);
2004 	m_freem(mrep);
2005 nfsmout:
2006 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2007 	if (!wccflag)
2008 		VTONFS(dvp)->n_attrstamp = 0;
2009 	/*
2010 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2011 	 * if we can succeed in looking up the directory.
2012 	 */
2013 	if (error == EEXIST || (!error && !gotvp)) {
2014 		if (newvp) {
2015 			vrele(newvp);
2016 			newvp = (struct vnode *)0;
2017 		}
2018 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2019 			cnp->cn_td, &np);
2020 		if (!error) {
2021 			newvp = NFSTOV(np);
2022 			if (newvp->v_type != VDIR)
2023 				error = EEXIST;
2024 		}
2025 	}
2026 	if (error) {
2027 		if (newvp)
2028 			vrele(newvp);
2029 	} else
2030 		*ap->a_vpp = newvp;
2031 	return (error);
2032 }
2033 
2034 /*
2035  * nfs remove directory call
2036  *
2037  * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2038  *	     struct componentname *a_cnp)
2039  */
2040 static int
2041 nfs_rmdir(struct vop_old_rmdir_args *ap)
2042 {
2043 	struct vnode *vp = ap->a_vp;
2044 	struct vnode *dvp = ap->a_dvp;
2045 	struct componentname *cnp = ap->a_cnp;
2046 	u_int32_t *tl;
2047 	caddr_t cp;
2048 	int32_t t1, t2;
2049 	caddr_t bpos, dpos, cp2;
2050 	int error = 0, wccflag = NFSV3_WCCRATTR;
2051 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2052 	int v3 = NFS_ISV3(dvp);
2053 
2054 	if (dvp == vp)
2055 		return (EINVAL);
2056 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2057 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
2058 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2059 	nfsm_fhtom(dvp, v3);
2060 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2061 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2062 	if (v3)
2063 		nfsm_wcc_data(dvp, wccflag);
2064 	m_freem(mrep);
2065 nfsmout:
2066 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2067 	if (!wccflag)
2068 		VTONFS(dvp)->n_attrstamp = 0;
2069 	/*
2070 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2071 	 */
2072 	if (error == ENOENT)
2073 		error = 0;
2074 	return (error);
2075 }
2076 
2077 /*
2078  * nfs readdir call
2079  *
2080  * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2081  */
2082 static int
2083 nfs_readdir(struct vop_readdir_args *ap)
2084 {
2085 	struct vnode *vp = ap->a_vp;
2086 	struct nfsnode *np = VTONFS(vp);
2087 	struct uio *uio = ap->a_uio;
2088 	int tresid, error;
2089 	struct vattr vattr;
2090 
2091 	if (vp->v_type != VDIR)
2092 		return (EPERM);
2093 
2094 	/*
2095 	 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2096 	 * and then check that is still valid, or if this is an NQNFS mount
2097 	 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR().  Note that
2098 	 * VOP_GETATTR() does not necessarily go to the wire.
2099 	 */
2100 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2101 	    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2102 		if (VOP_GETATTR(vp, &vattr) == 0 &&
2103 		    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2104 		) {
2105 			nfsstats.direofcache_hits++;
2106 			return (0);
2107 		}
2108 	}
2109 
2110 	/*
2111 	 * Call nfs_bioread() to do the real work.  nfs_bioread() does its
2112 	 * own cache coherency checks so we do not have to.
2113 	 */
2114 	tresid = uio->uio_resid;
2115 	error = nfs_bioread(vp, uio, 0);
2116 
2117 	if (!error && uio->uio_resid == tresid)
2118 		nfsstats.direofcache_misses++;
2119 	return (error);
2120 }
2121 
2122 /*
2123  * Readdir rpc call.  nfs_bioread->nfs_doio->nfs_readdirrpc.
2124  *
2125  * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2126  * offset/block and converts the nfs formatted directory entries for userland
2127  * consumption as well as deals with offsets into the middle of blocks.
2128  * nfs_doio only deals with logical blocks.  In particular, uio_offset will
2129  * be block-bounded.  It must convert to cookies for the actual RPC.
2130  */
2131 int
2132 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2133 {
2134 	int len, left;
2135 	struct nfs_dirent *dp = NULL;
2136 	u_int32_t *tl;
2137 	caddr_t cp;
2138 	int32_t t1, t2;
2139 	nfsuint64 *cookiep;
2140 	caddr_t bpos, dpos, cp2;
2141 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2142 	nfsuint64 cookie;
2143 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2144 	struct nfsnode *dnp = VTONFS(vp);
2145 	u_quad_t fileno;
2146 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2147 	int attrflag;
2148 	int v3 = NFS_ISV3(vp);
2149 
2150 #ifndef DIAGNOSTIC
2151 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2152 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2153 		panic("nfs readdirrpc bad uio");
2154 #endif
2155 
2156 	/*
2157 	 * If there is no cookie, assume directory was stale.
2158 	 */
2159 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2160 	if (cookiep)
2161 		cookie = *cookiep;
2162 	else
2163 		return (NFSERR_BAD_COOKIE);
2164 	/*
2165 	 * Loop around doing readdir rpc's of size nm_readdirsize
2166 	 * truncated to a multiple of DIRBLKSIZ.
2167 	 * The stopping criteria is EOF or buffer full.
2168 	 */
2169 	while (more_dirs && bigenough) {
2170 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2171 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2172 			NFSX_READDIR(v3));
2173 		nfsm_fhtom(vp, v3);
2174 		if (v3) {
2175 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2176 			*tl++ = cookie.nfsuquad[0];
2177 			*tl++ = cookie.nfsuquad[1];
2178 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2179 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2180 		} else {
2181 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2182 			*tl++ = cookie.nfsuquad[0];
2183 		}
2184 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2185 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2186 		if (v3) {
2187 			nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2188 			if (!error) {
2189 				nfsm_dissect(tl, u_int32_t *,
2190 				    2 * NFSX_UNSIGNED);
2191 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2192 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2193 			} else {
2194 				m_freem(mrep);
2195 				goto nfsmout;
2196 			}
2197 		}
2198 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2199 		more_dirs = fxdr_unsigned(int, *tl);
2200 
2201 		/* loop thru the dir entries, converting them to std form */
2202 		while (more_dirs && bigenough) {
2203 			if (v3) {
2204 				nfsm_dissect(tl, u_int32_t *,
2205 				    3 * NFSX_UNSIGNED);
2206 				fileno = fxdr_hyper(tl);
2207 				len = fxdr_unsigned(int, *(tl + 2));
2208 			} else {
2209 				nfsm_dissect(tl, u_int32_t *,
2210 				    2 * NFSX_UNSIGNED);
2211 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2212 				len = fxdr_unsigned(int, *tl);
2213 			}
2214 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2215 				error = EBADRPC;
2216 				m_freem(mrep);
2217 				goto nfsmout;
2218 			}
2219 
2220 			/*
2221 			 * len is the number of bytes in the path element
2222 			 * name, not including the \0 termination.
2223 			 *
2224 			 * tlen is the number of bytes w have to reserve for
2225 			 * the path element name.
2226 			 */
2227 			tlen = nfsm_rndup(len);
2228 			if (tlen == len)
2229 				tlen += 4;	/* To ensure null termination */
2230 
2231 			/*
2232 			 * If the entry would cross a DIRBLKSIZ boundary,
2233 			 * extend the previous nfs_dirent to cover the
2234 			 * remaining space.
2235 			 */
2236 			left = DIRBLKSIZ - blksiz;
2237 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2238 				dp->nfs_reclen += left;
2239 				uiop->uio_iov->iov_base += left;
2240 				uiop->uio_iov->iov_len -= left;
2241 				uiop->uio_offset += left;
2242 				uiop->uio_resid -= left;
2243 				blksiz = 0;
2244 			}
2245 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2246 				bigenough = 0;
2247 			if (bigenough) {
2248 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2249 				dp->nfs_ino = fileno;
2250 				dp->nfs_namlen = len;
2251 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2252 				dp->nfs_type = DT_UNKNOWN;
2253 				blksiz += dp->nfs_reclen;
2254 				if (blksiz == DIRBLKSIZ)
2255 					blksiz = 0;
2256 				uiop->uio_offset += sizeof(struct nfs_dirent);
2257 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2258 				uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2259 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2260 				nfsm_mtouio(uiop, len);
2261 
2262 				/*
2263 				 * The uiop has advanced by nfs_dirent + len
2264 				 * but really needs to advance by
2265 				 * nfs_dirent + tlen
2266 				 */
2267 				cp = uiop->uio_iov->iov_base;
2268 				tlen -= len;
2269 				*cp = '\0';	/* null terminate */
2270 				uiop->uio_iov->iov_base += tlen;
2271 				uiop->uio_iov->iov_len -= tlen;
2272 				uiop->uio_offset += tlen;
2273 				uiop->uio_resid -= tlen;
2274 			} else {
2275 				/*
2276 				 * NFS strings must be rounded up (nfsm_myouio
2277 				 * handled that in the bigenough case).
2278 				 */
2279 				nfsm_adv(nfsm_rndup(len));
2280 			}
2281 			if (v3) {
2282 				nfsm_dissect(tl, u_int32_t *,
2283 				    3 * NFSX_UNSIGNED);
2284 			} else {
2285 				nfsm_dissect(tl, u_int32_t *,
2286 				    2 * NFSX_UNSIGNED);
2287 			}
2288 
2289 			/*
2290 			 * If we were able to accomodate the last entry,
2291 			 * get the cookie for the next one.  Otherwise
2292 			 * hold-over the cookie for the one we were not
2293 			 * able to accomodate.
2294 			 */
2295 			if (bigenough) {
2296 				cookie.nfsuquad[0] = *tl++;
2297 				if (v3)
2298 					cookie.nfsuquad[1] = *tl++;
2299 			} else if (v3) {
2300 				tl += 2;
2301 			} else {
2302 				tl++;
2303 			}
2304 			more_dirs = fxdr_unsigned(int, *tl);
2305 		}
2306 		/*
2307 		 * If at end of rpc data, get the eof boolean
2308 		 */
2309 		if (!more_dirs) {
2310 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2311 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2312 		}
2313 		m_freem(mrep);
2314 	}
2315 	/*
2316 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2317 	 * by increasing d_reclen for the last record.
2318 	 */
2319 	if (blksiz > 0) {
2320 		left = DIRBLKSIZ - blksiz;
2321 		dp->nfs_reclen += left;
2322 		uiop->uio_iov->iov_base += left;
2323 		uiop->uio_iov->iov_len -= left;
2324 		uiop->uio_offset += left;
2325 		uiop->uio_resid -= left;
2326 	}
2327 
2328 	if (bigenough) {
2329 		/*
2330 		 * We hit the end of the directory, update direofoffset.
2331 		 */
2332 		dnp->n_direofoffset = uiop->uio_offset;
2333 	} else {
2334 		/*
2335 		 * There is more to go, insert the link cookie so the
2336 		 * next block can be read.
2337 		 */
2338 		if (uiop->uio_resid > 0)
2339 			printf("EEK! readdirrpc resid > 0\n");
2340 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2341 		*cookiep = cookie;
2342 	}
2343 nfsmout:
2344 	return (error);
2345 }
2346 
2347 /*
2348  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2349  */
2350 int
2351 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2352 {
2353 	int len, left;
2354 	struct nfs_dirent *dp;
2355 	u_int32_t *tl;
2356 	caddr_t cp;
2357 	int32_t t1, t2;
2358 	struct vnode *newvp;
2359 	nfsuint64 *cookiep;
2360 	caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2361 	struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2362 	nfsuint64 cookie;
2363 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2364 	struct nfsnode *dnp = VTONFS(vp), *np;
2365 	nfsfh_t *fhp;
2366 	u_quad_t fileno;
2367 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2368 	int attrflag, fhsize;
2369 	struct namecache *ncp;
2370 	struct namecache *dncp;
2371 	struct nlcomponent nlc;
2372 
2373 #ifndef nolint
2374 	dp = NULL;
2375 #endif
2376 #ifndef DIAGNOSTIC
2377 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2378 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2379 		panic("nfs readdirplusrpc bad uio");
2380 #endif
2381 	/*
2382 	 * Obtain the namecache record for the directory so we have something
2383 	 * to use as a basis for creating the entries.  This function will
2384 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2385 	 * from the tree and cannot be used for upward traversals, and the
2386 	 * ncp may be unnamed.  Note that other unrelated operations may
2387 	 * cause the ncp to be named at any time.
2388 	 */
2389 	dncp = cache_fromdvp(vp, NULL, 0);
2390 	bzero(&nlc, sizeof(nlc));
2391 	newvp = NULLVP;
2392 
2393 	/*
2394 	 * If there is no cookie, assume directory was stale.
2395 	 */
2396 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2397 	if (cookiep)
2398 		cookie = *cookiep;
2399 	else
2400 		return (NFSERR_BAD_COOKIE);
2401 	/*
2402 	 * Loop around doing readdir rpc's of size nm_readdirsize
2403 	 * truncated to a multiple of DIRBLKSIZ.
2404 	 * The stopping criteria is EOF or buffer full.
2405 	 */
2406 	while (more_dirs && bigenough) {
2407 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2408 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2409 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2410 		nfsm_fhtom(vp, 1);
2411  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2412 		*tl++ = cookie.nfsuquad[0];
2413 		*tl++ = cookie.nfsuquad[1];
2414 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2415 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2416 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2417 		*tl = txdr_unsigned(nmp->nm_rsize);
2418 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2419 		nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2420 		if (error) {
2421 			m_freem(mrep);
2422 			goto nfsmout;
2423 		}
2424 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2425 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2426 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2427 		more_dirs = fxdr_unsigned(int, *tl);
2428 
2429 		/* loop thru the dir entries, doctoring them to 4bsd form */
2430 		while (more_dirs && bigenough) {
2431 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2432 			fileno = fxdr_hyper(tl);
2433 			len = fxdr_unsigned(int, *(tl + 2));
2434 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2435 				error = EBADRPC;
2436 				m_freem(mrep);
2437 				goto nfsmout;
2438 			}
2439 			tlen = nfsm_rndup(len);
2440 			if (tlen == len)
2441 				tlen += 4;	/* To ensure null termination*/
2442 			left = DIRBLKSIZ - blksiz;
2443 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2444 				dp->nfs_reclen += left;
2445 				uiop->uio_iov->iov_base += left;
2446 				uiop->uio_iov->iov_len -= left;
2447 				uiop->uio_offset += left;
2448 				uiop->uio_resid -= left;
2449 				blksiz = 0;
2450 			}
2451 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2452 				bigenough = 0;
2453 			if (bigenough) {
2454 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2455 				dp->nfs_ino = fileno;
2456 				dp->nfs_namlen = len;
2457 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2458 				dp->nfs_type = DT_UNKNOWN;
2459 				blksiz += dp->nfs_reclen;
2460 				if (blksiz == DIRBLKSIZ)
2461 					blksiz = 0;
2462 				uiop->uio_offset += sizeof(struct nfs_dirent);
2463 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2464 				uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2465 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2466 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2467 				nlc.nlc_namelen = len;
2468 				nfsm_mtouio(uiop, len);
2469 				cp = uiop->uio_iov->iov_base;
2470 				tlen -= len;
2471 				*cp = '\0';
2472 				uiop->uio_iov->iov_base += tlen;
2473 				uiop->uio_iov->iov_len -= tlen;
2474 				uiop->uio_offset += tlen;
2475 				uiop->uio_resid -= tlen;
2476 			} else
2477 				nfsm_adv(nfsm_rndup(len));
2478 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2479 			if (bigenough) {
2480 				cookie.nfsuquad[0] = *tl++;
2481 				cookie.nfsuquad[1] = *tl++;
2482 			} else
2483 				tl += 2;
2484 
2485 			/*
2486 			 * Since the attributes are before the file handle
2487 			 * (sigh), we must skip over the attributes and then
2488 			 * come back and get them.
2489 			 */
2490 			attrflag = fxdr_unsigned(int, *tl);
2491 			if (attrflag) {
2492 			    dpossav1 = dpos;
2493 			    mdsav1 = md;
2494 			    nfsm_adv(NFSX_V3FATTR);
2495 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2496 			    doit = fxdr_unsigned(int, *tl);
2497 			    if (doit) {
2498 				nfsm_getfh(fhp, fhsize, 1);
2499 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2500 				    vref(vp);
2501 				    newvp = vp;
2502 				    np = dnp;
2503 				} else {
2504 				    error = nfs_nget(vp->v_mount, fhp,
2505 					fhsize, &np);
2506 				    if (error)
2507 					doit = 0;
2508 				    else
2509 					newvp = NFSTOV(np);
2510 				}
2511 			    }
2512 			    if (doit && bigenough) {
2513 				dpossav2 = dpos;
2514 				dpos = dpossav1;
2515 				mdsav2 = md;
2516 				md = mdsav1;
2517 				nfsm_loadattr(newvp, (struct vattr *)0);
2518 				dpos = dpossav2;
2519 				md = mdsav2;
2520 				dp->nfs_type =
2521 				    IFTODT(VTTOIF(np->n_vattr.va_type));
2522 				if (dncp) {
2523 				    printf("NFS/READDIRPLUS, ENTER %*.*s\n",
2524 					nlc.nlc_namelen, nlc.nlc_namelen,
2525 					nlc.nlc_nameptr);
2526 				    ncp = cache_nlookup(dncp, &nlc);
2527 				    cache_setunresolved(ncp);
2528 				    cache_setvp(ncp, newvp);
2529 				    cache_put(ncp);
2530 				} else {
2531 				    printf("NFS/READDIRPLUS, UNABLE TO ENTER"
2532 					" %*.*s\n",
2533 					nlc.nlc_namelen, nlc.nlc_namelen,
2534 					nlc.nlc_nameptr);
2535 				}
2536 			    }
2537 			} else {
2538 			    /* Just skip over the file handle */
2539 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2540 			    i = fxdr_unsigned(int, *tl);
2541 			    nfsm_adv(nfsm_rndup(i));
2542 			}
2543 			if (newvp != NULLVP) {
2544 			    if (newvp == vp)
2545 				vrele(newvp);
2546 			    else
2547 				vput(newvp);
2548 			    newvp = NULLVP;
2549 			}
2550 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2551 			more_dirs = fxdr_unsigned(int, *tl);
2552 		}
2553 		/*
2554 		 * If at end of rpc data, get the eof boolean
2555 		 */
2556 		if (!more_dirs) {
2557 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2558 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2559 		}
2560 		m_freem(mrep);
2561 	}
2562 	/*
2563 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2564 	 * by increasing d_reclen for the last record.
2565 	 */
2566 	if (blksiz > 0) {
2567 		left = DIRBLKSIZ - blksiz;
2568 		dp->nfs_reclen += left;
2569 		uiop->uio_iov->iov_base += left;
2570 		uiop->uio_iov->iov_len -= left;
2571 		uiop->uio_offset += left;
2572 		uiop->uio_resid -= left;
2573 	}
2574 
2575 	/*
2576 	 * We are now either at the end of the directory or have filled the
2577 	 * block.
2578 	 */
2579 	if (bigenough)
2580 		dnp->n_direofoffset = uiop->uio_offset;
2581 	else {
2582 		if (uiop->uio_resid > 0)
2583 			printf("EEK! readdirplusrpc resid > 0\n");
2584 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2585 		*cookiep = cookie;
2586 	}
2587 nfsmout:
2588 	if (newvp != NULLVP) {
2589 	        if (newvp == vp)
2590 			vrele(newvp);
2591 		else
2592 			vput(newvp);
2593 		newvp = NULLVP;
2594 	}
2595 	if (dncp)
2596 		cache_drop(dncp);
2597 	return (error);
2598 }
2599 
2600 /*
2601  * Silly rename. To make the NFS filesystem that is stateless look a little
2602  * more like the "ufs" a remove of an active vnode is translated to a rename
2603  * to a funny looking filename that is removed by nfs_inactive on the
2604  * nfsnode. There is the potential for another process on a different client
2605  * to create the same funny name between the nfs_lookitup() fails and the
2606  * nfs_rename() completes, but...
2607  */
2608 static int
2609 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2610 {
2611 	struct sillyrename *sp;
2612 	struct nfsnode *np;
2613 	int error;
2614 
2615 	/*
2616 	 * We previously purged dvp instead of vp.  I don't know why, it
2617 	 * completely destroys performance.  We can't do it anyway with the
2618 	 * new VFS API since we would be breaking the namecache topology.
2619 	 */
2620 	cache_purge(vp);	/* XXX */
2621 	np = VTONFS(vp);
2622 #ifndef DIAGNOSTIC
2623 	if (vp->v_type == VDIR)
2624 		panic("nfs: sillyrename dir");
2625 #endif
2626 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2627 		M_NFSREQ, M_WAITOK);
2628 	sp->s_cred = crdup(cnp->cn_cred);
2629 	sp->s_dvp = dvp;
2630 	vref(dvp);
2631 
2632 	/* Fudge together a funny name */
2633 	sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2634 
2635 	/* Try lookitups until we get one that isn't there */
2636 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2637 		cnp->cn_td, (struct nfsnode **)0) == 0) {
2638 		sp->s_name[4]++;
2639 		if (sp->s_name[4] > 'z') {
2640 			error = EINVAL;
2641 			goto bad;
2642 		}
2643 	}
2644 	error = nfs_renameit(dvp, cnp, sp);
2645 	if (error)
2646 		goto bad;
2647 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2648 		cnp->cn_td, &np);
2649 	np->n_sillyrename = sp;
2650 	return (0);
2651 bad:
2652 	vrele(sp->s_dvp);
2653 	crfree(sp->s_cred);
2654 	free((caddr_t)sp, M_NFSREQ);
2655 	return (error);
2656 }
2657 
2658 /*
2659  * Look up a file name and optionally either update the file handle or
2660  * allocate an nfsnode, depending on the value of npp.
2661  * npp == NULL	--> just do the lookup
2662  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2663  *			handled too
2664  * *npp != NULL --> update the file handle in the vnode
2665  */
2666 static int
2667 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2668 	     struct thread *td, struct nfsnode **npp)
2669 {
2670 	u_int32_t *tl;
2671 	caddr_t cp;
2672 	int32_t t1, t2;
2673 	struct vnode *newvp = (struct vnode *)0;
2674 	struct nfsnode *np, *dnp = VTONFS(dvp);
2675 	caddr_t bpos, dpos, cp2;
2676 	int error = 0, fhlen, attrflag;
2677 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2678 	nfsfh_t *nfhp;
2679 	int v3 = NFS_ISV3(dvp);
2680 
2681 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2682 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2683 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2684 	nfsm_fhtom(dvp, v3);
2685 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2686 	nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2687 	if (npp && !error) {
2688 		nfsm_getfh(nfhp, fhlen, v3);
2689 		if (*npp) {
2690 		    np = *npp;
2691 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2692 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2693 			np->n_fhp = &np->n_fh;
2694 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2695 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2696 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2697 		    np->n_fhsize = fhlen;
2698 		    newvp = NFSTOV(np);
2699 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2700 		    vref(dvp);
2701 		    newvp = dvp;
2702 		} else {
2703 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2704 		    if (error) {
2705 			m_freem(mrep);
2706 			return (error);
2707 		    }
2708 		    newvp = NFSTOV(np);
2709 		}
2710 		if (v3) {
2711 			nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2712 			if (!attrflag && *npp == NULL) {
2713 				m_freem(mrep);
2714 				if (newvp == dvp)
2715 					vrele(newvp);
2716 				else
2717 					vput(newvp);
2718 				return (ENOENT);
2719 			}
2720 		} else
2721 			nfsm_loadattr(newvp, (struct vattr *)0);
2722 	}
2723 	m_freem(mrep);
2724 nfsmout:
2725 	if (npp && *npp == NULL) {
2726 		if (error) {
2727 			if (newvp) {
2728 				if (newvp == dvp)
2729 					vrele(newvp);
2730 				else
2731 					vput(newvp);
2732 			}
2733 		} else
2734 			*npp = np;
2735 	}
2736 	return (error);
2737 }
2738 
2739 /*
2740  * Nfs Version 3 commit rpc
2741  */
2742 int
2743 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2744 {
2745 	caddr_t cp;
2746 	u_int32_t *tl;
2747 	int32_t t1, t2;
2748 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2749 	caddr_t bpos, dpos, cp2;
2750 	int error = 0, wccflag = NFSV3_WCCRATTR;
2751 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2752 
2753 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2754 		return (0);
2755 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2756 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2757 	nfsm_fhtom(vp, 1);
2758 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2759 	txdr_hyper(offset, tl);
2760 	tl += 2;
2761 	*tl = txdr_unsigned(cnt);
2762 	nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2763 	nfsm_wcc_data(vp, wccflag);
2764 	if (!error) {
2765 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2766 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2767 			NFSX_V3WRITEVERF)) {
2768 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2769 				NFSX_V3WRITEVERF);
2770 			error = NFSERR_STALEWRITEVERF;
2771 		}
2772 	}
2773 	m_freem(mrep);
2774 nfsmout:
2775 	return (error);
2776 }
2777 
2778 /*
2779  * Kludge City..
2780  * - make nfs_bmap() essentially a no-op that does no translation
2781  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2782  *   (Maybe I could use the process's page mapping, but I was concerned that
2783  *    Kernel Write might not be enabled and also figured copyout() would do
2784  *    a lot more work than bcopy() and also it currently happens in the
2785  *    context of the swapper process (2).
2786  *
2787  * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
2788  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
2789  */
2790 static int
2791 nfs_bmap(struct vop_bmap_args *ap)
2792 {
2793 	struct vnode *vp = ap->a_vp;
2794 
2795 	if (ap->a_vpp != NULL)
2796 		*ap->a_vpp = vp;
2797 	if (ap->a_doffsetp != NULL)
2798 		*ap->a_doffsetp = ap->a_loffset;
2799 	if (ap->a_runp != NULL)
2800 		*ap->a_runp = 0;
2801 	if (ap->a_runb != NULL)
2802 		*ap->a_runb = 0;
2803 	return (0);
2804 }
2805 
2806 /*
2807  * Strategy routine.
2808  *
2809  * For async requests when nfsiod(s) are running, queue the request by
2810  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2811  * request.
2812  */
2813 static int
2814 nfs_strategy(struct vop_strategy_args *ap)
2815 {
2816 	struct bio *bio = ap->a_bio;
2817 	struct bio *nbio;
2818 	struct buf *bp = bio->bio_buf;
2819 	struct thread *td;
2820 	int error = 0;
2821 
2822 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
2823 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
2824 	KASSERT(BUF_REFCNT(bp) > 0,
2825 		("nfs_strategy: buffer %p not locked", bp));
2826 
2827 	if (bp->b_flags & B_ASYNC)
2828 		td = NULL;
2829 	else
2830 		td = curthread;	/* XXX */
2831 
2832         /*
2833 	 * We probably don't need to push an nbio any more since no
2834 	 * block conversion is required due to the use of 64 bit byte
2835 	 * offsets, but do it anyway.
2836          */
2837 	nbio = push_bio(bio);
2838 	nbio->bio_offset = bio->bio_offset;
2839 
2840 	/*
2841 	 * If the op is asynchronous and an i/o daemon is waiting
2842 	 * queue the request, wake it up and wait for completion
2843 	 * otherwise just do it ourselves.
2844 	 */
2845 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2846 		error = nfs_doio(ap->a_vp, nbio, td);
2847 	return (error);
2848 }
2849 
2850 /*
2851  * Mmap a file
2852  *
2853  * NB Currently unsupported.
2854  *
2855  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2856  *	    struct thread *a_td)
2857  */
2858 /* ARGSUSED */
2859 static int
2860 nfs_mmap(struct vop_mmap_args *ap)
2861 {
2862 	return (EINVAL);
2863 }
2864 
2865 /*
2866  * fsync vnode op. Just call nfs_flush() with commit == 1.
2867  *
2868  * nfs_fsync(struct vnode *a_vp, struct ucred * a_cred, int a_waitfor,
2869  *	     struct thread *a_td)
2870  */
2871 /* ARGSUSED */
2872 static int
2873 nfs_fsync(struct vop_fsync_args *ap)
2874 {
2875 	return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2876 }
2877 
2878 /*
2879  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
2880  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
2881  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
2882  * set the buffer contains data that has already been written to the server
2883  * and which now needs a commit RPC.
2884  *
2885  * If commit is 0 we only take one pass and only flush buffers containing new
2886  * dirty data.
2887  *
2888  * If commit is 1 we take two passes, issuing a commit RPC in the second
2889  * pass.
2890  *
2891  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2892  * to completely flush all pending data.
2893  *
2894  * Note that the RB_SCAN code properly handles the case where the
2895  * callback might block and directly or indirectly (another thread) cause
2896  * the RB tree to change.
2897  */
2898 
2899 #ifndef NFS_COMMITBVECSIZ
2900 #define NFS_COMMITBVECSIZ	16
2901 #endif
2902 
2903 struct nfs_flush_info {
2904 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2905 	struct thread *td;
2906 	struct vnode *vp;
2907 	int waitfor;
2908 	int slpflag;
2909 	int slptimeo;
2910 	int loops;
2911 	struct buf *bvary[NFS_COMMITBVECSIZ];
2912 	int bvsize;
2913 	off_t beg_off;
2914 	off_t end_off;
2915 };
2916 
2917 static int nfs_flush_bp(struct buf *bp, void *data);
2918 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2919 
2920 int
2921 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2922 {
2923 	struct nfsnode *np = VTONFS(vp);
2924 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2925 	struct nfs_flush_info info;
2926 	int error;
2927 
2928 	bzero(&info, sizeof(info));
2929 	info.td = td;
2930 	info.vp = vp;
2931 	info.waitfor = waitfor;
2932 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2933 	info.loops = 0;
2934 
2935 	do {
2936 		/*
2937 		 * Flush mode
2938 		 */
2939 		info.mode = NFI_FLUSHNEW;
2940 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2941 				nfs_flush_bp, &info);
2942 
2943 		/*
2944 		 * Take a second pass if committing and no error occured.
2945 		 * Clean up any left over collection (whether an error
2946 		 * occurs or not).
2947 		 */
2948 		if (commit && error == 0) {
2949 			info.mode = NFI_COMMIT;
2950 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2951 					nfs_flush_bp, &info);
2952 			if (info.bvsize)
2953 				error = nfs_flush_docommit(&info, error);
2954 		}
2955 
2956 		/*
2957 		 * Wait for pending I/O to complete before checking whether
2958 		 * any further dirty buffers exist.
2959 		 */
2960 		while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
2961 			vp->v_track_write.bk_waitflag = 1;
2962 			error = tsleep(&vp->v_track_write,
2963 				info.slpflag, "nfsfsync", info.slptimeo);
2964 			if (error) {
2965 				/*
2966 				 * We have to be able to break out if this
2967 				 * is an 'intr' mount.
2968 				 */
2969 				if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2970 					error = -EINTR;
2971 					break;
2972 				}
2973 
2974 				/*
2975 				 * Since we do not process pending signals,
2976 				 * once we get a PCATCH our tsleep() will no
2977 				 * longer sleep, switch to a fixed timeout
2978 				 * instead.
2979 				 */
2980 				if (info.slpflag == PCATCH) {
2981 					info.slpflag = 0;
2982 					info.slptimeo = 2 * hz;
2983 				}
2984 				error = 0;
2985 			}
2986 		}
2987 		++info.loops;
2988 		/*
2989 		 * Loop if we are flushing synchronous as well as committing,
2990 		 * and dirty buffers are still present.  Otherwise we might livelock.
2991 		 */
2992 	} while (waitfor == MNT_WAIT && commit &&
2993 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
2994 
2995 	/*
2996 	 * The callbacks have to return a negative error to terminate the
2997 	 * RB scan.
2998 	 */
2999 	if (error < 0)
3000 		error = -error;
3001 
3002 	/*
3003 	 * Deal with any error collection
3004 	 */
3005 	if (np->n_flag & NWRITEERR) {
3006 		error = np->n_error;
3007 		np->n_flag &= ~NWRITEERR;
3008 	}
3009 	return (error);
3010 }
3011 
3012 
3013 static
3014 int
3015 nfs_flush_bp(struct buf *bp, void *data)
3016 {
3017 	struct nfs_flush_info *info = data;
3018 	off_t toff;
3019 	int error;
3020 
3021 	error = 0;
3022 	switch(info->mode) {
3023 	case NFI_FLUSHNEW:
3024 		crit_enter();
3025 		if (info->loops && info->waitfor == MNT_WAIT) {
3026 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3027 			if (error) {
3028 				int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3029 				if (info->slpflag & PCATCH)
3030 					lkflags |= LK_PCATCH;
3031 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3032 						     info->slptimeo);
3033 			}
3034 		} else {
3035 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3036 		}
3037 		if (error == 0) {
3038 			KKASSERT(bp->b_vp == info->vp);
3039 
3040 			if ((bp->b_flags & B_DELWRI) == 0)
3041 				panic("nfs_fsync: not dirty");
3042 			if (bp->b_flags & B_NEEDCOMMIT) {
3043 				BUF_UNLOCK(bp);
3044 				crit_exit();
3045 				break;
3046 			}
3047 			bremfree(bp);
3048 
3049 			bp->b_flags |= B_ASYNC;
3050 			crit_exit();
3051 			bwrite(bp);
3052 		} else {
3053 			crit_exit();
3054 			error = 0;
3055 		}
3056 		break;
3057 	case NFI_COMMIT:
3058 		/*
3059 		 * Only process buffers in need of a commit which we can
3060 		 * immediately lock.  This may prevent a buffer from being
3061 		 * committed, but the normal flush loop will block on the
3062 		 * same buffer so we shouldn't get into an endless loop.
3063 		 */
3064 		crit_enter();
3065 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3066 		    (B_DELWRI | B_NEEDCOMMIT) ||
3067 		    BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3068 			crit_exit();
3069 			break;
3070 		}
3071 
3072 		KKASSERT(bp->b_vp == info->vp);
3073 		bremfree(bp);
3074 
3075 		/*
3076 		 * NOTE: storing the bp in the bvary[] basically sets
3077 		 * it up for a commit operation.
3078 		 *
3079 		 * We must call vfs_busy_pages() now so the commit operation
3080 		 * is interlocked with user modifications to memory mapped
3081 		 * pages.
3082 		 *
3083 		 * Note: to avoid loopback deadlocks, we do not
3084 		 * assign b_runningbufspace.
3085 		 */
3086 		bp->b_cmd = BUF_CMD_WRITE;
3087 		vfs_busy_pages(bp->b_vp, bp);
3088 		info->bvary[info->bvsize] = bp;
3089 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3090 		if (info->bvsize == 0 || toff < info->beg_off)
3091 			info->beg_off = toff;
3092 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3093 		if (info->bvsize == 0 || toff > info->end_off)
3094 			info->end_off = toff;
3095 		++info->bvsize;
3096 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3097 			error = nfs_flush_docommit(info, 0);
3098 			KKASSERT(info->bvsize == 0);
3099 		}
3100 		crit_exit();
3101 	}
3102 	return (error);
3103 }
3104 
3105 static
3106 int
3107 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3108 {
3109 	struct vnode *vp;
3110 	struct buf *bp;
3111 	off_t bytes;
3112 	int retv;
3113 	int i;
3114 
3115 	vp = info->vp;
3116 
3117 	if (info->bvsize > 0) {
3118 		/*
3119 		 * Commit data on the server, as required.  Note that
3120 		 * nfs_commit will use the vnode's cred for the commit.
3121 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3122 		 */
3123 		bytes = info->end_off - info->beg_off;
3124 		if (bytes > 0x40000000)
3125 			bytes = 0x40000000;
3126 		if (error) {
3127 			retv = -error;
3128 		} else {
3129 			retv = nfs_commit(vp, info->beg_off,
3130 					    (int)bytes, info->td);
3131 			if (retv == NFSERR_STALEWRITEVERF)
3132 				nfs_clearcommit(vp->v_mount);
3133 		}
3134 
3135 		/*
3136 		 * Now, either mark the blocks I/O done or mark the
3137 		 * blocks dirty, depending on whether the commit
3138 		 * succeeded.
3139 		 */
3140 		for (i = 0; i < info->bvsize; ++i) {
3141 			bp = info->bvary[i];
3142 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3143 			if (retv) {
3144 				/*
3145 				 * Error, leave B_DELWRI intact
3146 				 */
3147 				vfs_unbusy_pages(bp);
3148 				bp->b_cmd = BUF_CMD_DONE;
3149 				brelse(bp);
3150 			} else {
3151 				/*
3152 				 * Success, remove B_DELWRI ( bundirty() ).
3153 				 *
3154 				 * b_dirtyoff/b_dirtyend seem to be NFS
3155 				 * specific.  We should probably move that
3156 				 * into bundirty(). XXX
3157 				 *
3158 				 * We are faking an I/O write, we have to
3159 				 * start the transaction in order to
3160 				 * immediately biodone() it.
3161 				 */
3162 				crit_enter();
3163 				bp->b_flags |= B_ASYNC;
3164 				bundirty(bp);
3165 				bp->b_flags &= ~B_ERROR;
3166 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3167 				crit_exit();
3168 				biodone(&bp->b_bio1);
3169 			}
3170 		}
3171 		info->bvsize = 0;
3172 	}
3173 	return (error);
3174 }
3175 
3176 /*
3177  * NFS advisory byte-level locks.
3178  * Currently unsupported.
3179  *
3180  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3181  *		int a_flags)
3182  */
3183 static int
3184 nfs_advlock(struct vop_advlock_args *ap)
3185 {
3186 	struct nfsnode *np = VTONFS(ap->a_vp);
3187 
3188 	/*
3189 	 * The following kludge is to allow diskless support to work
3190 	 * until a real NFS lockd is implemented. Basically, just pretend
3191 	 * that this is a local lock.
3192 	 */
3193 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3194 }
3195 
3196 /*
3197  * Print out the contents of an nfsnode.
3198  *
3199  * nfs_print(struct vnode *a_vp)
3200  */
3201 static int
3202 nfs_print(struct vop_print_args *ap)
3203 {
3204 	struct vnode *vp = ap->a_vp;
3205 	struct nfsnode *np = VTONFS(vp);
3206 
3207 	printf("tag VT_NFS, fileid %ld fsid 0x%x",
3208 		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3209 	if (vp->v_type == VFIFO)
3210 		fifo_printinfo(vp);
3211 	printf("\n");
3212 	return (0);
3213 }
3214 
3215 /*
3216  * nfs special file access vnode op.
3217  * Essentially just get vattr and then imitate iaccess() since the device is
3218  * local to the client.
3219  *
3220  * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3221  *		  struct thread *a_td)
3222  */
3223 static int
3224 nfsspec_access(struct vop_access_args *ap)
3225 {
3226 	struct vattr *vap;
3227 	gid_t *gp;
3228 	struct ucred *cred = ap->a_cred;
3229 	struct vnode *vp = ap->a_vp;
3230 	mode_t mode = ap->a_mode;
3231 	struct vattr vattr;
3232 	int i;
3233 	int error;
3234 
3235 	/*
3236 	 * Disallow write attempts on filesystems mounted read-only;
3237 	 * unless the file is a socket, fifo, or a block or character
3238 	 * device resident on the filesystem.
3239 	 */
3240 	if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3241 		switch (vp->v_type) {
3242 		case VREG:
3243 		case VDIR:
3244 		case VLNK:
3245 			return (EROFS);
3246 		default:
3247 			break;
3248 		}
3249 	}
3250 	/*
3251 	 * If you're the super-user,
3252 	 * you always get access.
3253 	 */
3254 	if (cred->cr_uid == 0)
3255 		return (0);
3256 	vap = &vattr;
3257 	error = VOP_GETATTR(vp, vap);
3258 	if (error)
3259 		return (error);
3260 	/*
3261 	 * Access check is based on only one of owner, group, public.
3262 	 * If not owner, then check group. If not a member of the
3263 	 * group, then check public access.
3264 	 */
3265 	if (cred->cr_uid != vap->va_uid) {
3266 		mode >>= 3;
3267 		gp = cred->cr_groups;
3268 		for (i = 0; i < cred->cr_ngroups; i++, gp++)
3269 			if (vap->va_gid == *gp)
3270 				goto found;
3271 		mode >>= 3;
3272 found:
3273 		;
3274 	}
3275 	error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3276 	return (error);
3277 }
3278 
3279 /*
3280  * Read wrapper for special devices.
3281  *
3282  * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3283  *		struct ucred *a_cred)
3284  */
3285 static int
3286 nfsspec_read(struct vop_read_args *ap)
3287 {
3288 	struct nfsnode *np = VTONFS(ap->a_vp);
3289 
3290 	/*
3291 	 * Set access flag.
3292 	 */
3293 	np->n_flag |= NACC;
3294 	getnanotime(&np->n_atim);
3295 	return (VOCALL(&spec_vnode_vops, &ap->a_head));
3296 }
3297 
3298 /*
3299  * Write wrapper for special devices.
3300  *
3301  * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3302  *		 struct ucred *a_cred)
3303  */
3304 static int
3305 nfsspec_write(struct vop_write_args *ap)
3306 {
3307 	struct nfsnode *np = VTONFS(ap->a_vp);
3308 
3309 	/*
3310 	 * Set update flag.
3311 	 */
3312 	np->n_flag |= NUPD;
3313 	getnanotime(&np->n_mtim);
3314 	return (VOCALL(&spec_vnode_vops, &ap->a_head));
3315 }
3316 
3317 /*
3318  * Close wrapper for special devices.
3319  *
3320  * Update the times on the nfsnode then do device close.
3321  *
3322  * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3323  *		 struct thread *a_td)
3324  */
3325 static int
3326 nfsspec_close(struct vop_close_args *ap)
3327 {
3328 	struct vnode *vp = ap->a_vp;
3329 	struct nfsnode *np = VTONFS(vp);
3330 	struct vattr vattr;
3331 
3332 	if (np->n_flag & (NACC | NUPD)) {
3333 		np->n_flag |= NCHG;
3334 		if (vp->v_usecount == 1 &&
3335 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3336 			VATTR_NULL(&vattr);
3337 			if (np->n_flag & NACC)
3338 				vattr.va_atime = np->n_atim;
3339 			if (np->n_flag & NUPD)
3340 				vattr.va_mtime = np->n_mtim;
3341 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3342 		}
3343 	}
3344 	return (VOCALL(&spec_vnode_vops, &ap->a_head));
3345 }
3346 
3347 /*
3348  * Read wrapper for fifos.
3349  *
3350  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3351  *		struct ucred *a_cred)
3352  */
3353 static int
3354 nfsfifo_read(struct vop_read_args *ap)
3355 {
3356 	struct nfsnode *np = VTONFS(ap->a_vp);
3357 
3358 	/*
3359 	 * Set access flag.
3360 	 */
3361 	np->n_flag |= NACC;
3362 	getnanotime(&np->n_atim);
3363 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3364 }
3365 
3366 /*
3367  * Write wrapper for fifos.
3368  *
3369  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3370  *		 struct ucred *a_cred)
3371  */
3372 static int
3373 nfsfifo_write(struct vop_write_args *ap)
3374 {
3375 	struct nfsnode *np = VTONFS(ap->a_vp);
3376 
3377 	/*
3378 	 * Set update flag.
3379 	 */
3380 	np->n_flag |= NUPD;
3381 	getnanotime(&np->n_mtim);
3382 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3383 }
3384 
3385 /*
3386  * Close wrapper for fifos.
3387  *
3388  * Update the times on the nfsnode then do fifo close.
3389  *
3390  * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3391  */
3392 static int
3393 nfsfifo_close(struct vop_close_args *ap)
3394 {
3395 	struct vnode *vp = ap->a_vp;
3396 	struct nfsnode *np = VTONFS(vp);
3397 	struct vattr vattr;
3398 	struct timespec ts;
3399 
3400 	if (np->n_flag & (NACC | NUPD)) {
3401 		getnanotime(&ts);
3402 		if (np->n_flag & NACC)
3403 			np->n_atim = ts;
3404 		if (np->n_flag & NUPD)
3405 			np->n_mtim = ts;
3406 		np->n_flag |= NCHG;
3407 		if (vp->v_usecount == 1 &&
3408 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3409 			VATTR_NULL(&vattr);
3410 			if (np->n_flag & NACC)
3411 				vattr.va_atime = np->n_atim;
3412 			if (np->n_flag & NUPD)
3413 				vattr.va_mtime = np->n_mtim;
3414 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3415 		}
3416 	}
3417 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3418 }
3419 
3420