xref: /dragonfly/sys/vfs/nfs/nfs_vnops.c (revision 984263bc)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
37  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38  */
39 
40 
41 /*
42  * vnode op calls for Sun NFS version 2 and 3
43  */
44 
45 #include "opt_inet.h"
46 
47 #include <sys/param.h>
48 #include <sys/kernel.h>
49 #include <sys/systm.h>
50 #include <sys/resourcevar.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/buf.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/conf.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_zone.h>
69 
70 #include <miscfs/fifofs/fifo.h>
71 
72 #include <nfs/rpcv2.h>
73 #include <nfs/nfsproto.h>
74 #include <nfs/nfs.h>
75 #include <nfs/nfsnode.h>
76 #include <nfs/nfsmount.h>
77 #include <nfs/xdr_subs.h>
78 #include <nfs/nfsm_subs.h>
79 #include <nfs/nqnfs.h>
80 
81 #include <net/if.h>
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 
85 /* Defs */
86 #define	TRUE	1
87 #define	FALSE	0
88 
89 /*
90  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91  * calls are not in getblk() and brelse() so that they would not be necessary
92  * here.
93  */
94 #ifndef B_VMIO
95 #define vfs_busy_pages(bp, f)
96 #endif
97 
98 static int	nfsspec_read __P((struct vop_read_args *));
99 static int	nfsspec_write __P((struct vop_write_args *));
100 static int	nfsfifo_read __P((struct vop_read_args *));
101 static int	nfsfifo_write __P((struct vop_write_args *));
102 static int	nfsspec_close __P((struct vop_close_args *));
103 static int	nfsfifo_close __P((struct vop_close_args *));
104 #define nfs_poll vop_nopoll
105 static int	nfs_flush __P((struct vnode *,struct ucred *,int,struct proc *,int));
106 static int	nfs_setattrrpc __P((struct vnode *,struct vattr *,struct ucred *,struct proc *));
107 static	int	nfs_lookup __P((struct vop_lookup_args *));
108 static	int	nfs_create __P((struct vop_create_args *));
109 static	int	nfs_mknod __P((struct vop_mknod_args *));
110 static	int	nfs_open __P((struct vop_open_args *));
111 static	int	nfs_close __P((struct vop_close_args *));
112 static	int	nfs_access __P((struct vop_access_args *));
113 static	int	nfs_getattr __P((struct vop_getattr_args *));
114 static	int	nfs_setattr __P((struct vop_setattr_args *));
115 static	int	nfs_read __P((struct vop_read_args *));
116 static	int	nfs_mmap __P((struct vop_mmap_args *));
117 static	int	nfs_fsync __P((struct vop_fsync_args *));
118 static	int	nfs_remove __P((struct vop_remove_args *));
119 static	int	nfs_link __P((struct vop_link_args *));
120 static	int	nfs_rename __P((struct vop_rename_args *));
121 static	int	nfs_mkdir __P((struct vop_mkdir_args *));
122 static	int	nfs_rmdir __P((struct vop_rmdir_args *));
123 static	int	nfs_symlink __P((struct vop_symlink_args *));
124 static	int	nfs_readdir __P((struct vop_readdir_args *));
125 static	int	nfs_bmap __P((struct vop_bmap_args *));
126 static	int	nfs_strategy __P((struct vop_strategy_args *));
127 static	int	nfs_lookitup __P((struct vnode *, const char *, int,
128 			struct ucred *, struct proc *, struct nfsnode **));
129 static	int	nfs_sillyrename __P((struct vnode *,struct vnode *,struct componentname *));
130 static int	nfsspec_access __P((struct vop_access_args *));
131 static int	nfs_readlink __P((struct vop_readlink_args *));
132 static int	nfs_print __P((struct vop_print_args *));
133 static int	nfs_advlock __P((struct vop_advlock_args *));
134 static int	nfs_bwrite __P((struct vop_bwrite_args *));
135 /*
136  * Global vfs data structures for nfs
137  */
138 vop_t **nfsv2_vnodeop_p;
139 static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
140 	{ &vop_default_desc,		(vop_t *) vop_defaultop },
141 	{ &vop_access_desc,		(vop_t *) nfs_access },
142 	{ &vop_advlock_desc,		(vop_t *) nfs_advlock },
143 	{ &vop_bmap_desc,		(vop_t *) nfs_bmap },
144 	{ &vop_bwrite_desc,		(vop_t *) nfs_bwrite },
145 	{ &vop_close_desc,		(vop_t *) nfs_close },
146 	{ &vop_create_desc,		(vop_t *) nfs_create },
147 	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
148 	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
149 	{ &vop_getpages_desc,		(vop_t *) nfs_getpages },
150 	{ &vop_putpages_desc,		(vop_t *) nfs_putpages },
151 	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
152 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
153 	{ &vop_lease_desc,		(vop_t *) vop_null },
154 	{ &vop_link_desc,		(vop_t *) nfs_link },
155 	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
156 	{ &vop_lookup_desc,		(vop_t *) nfs_lookup },
157 	{ &vop_mkdir_desc,		(vop_t *) nfs_mkdir },
158 	{ &vop_mknod_desc,		(vop_t *) nfs_mknod },
159 	{ &vop_mmap_desc,		(vop_t *) nfs_mmap },
160 	{ &vop_open_desc,		(vop_t *) nfs_open },
161 	{ &vop_poll_desc,		(vop_t *) nfs_poll },
162 	{ &vop_print_desc,		(vop_t *) nfs_print },
163 	{ &vop_read_desc,		(vop_t *) nfs_read },
164 	{ &vop_readdir_desc,		(vop_t *) nfs_readdir },
165 	{ &vop_readlink_desc,		(vop_t *) nfs_readlink },
166 	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
167 	{ &vop_remove_desc,		(vop_t *) nfs_remove },
168 	{ &vop_rename_desc,		(vop_t *) nfs_rename },
169 	{ &vop_rmdir_desc,		(vop_t *) nfs_rmdir },
170 	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
171 	{ &vop_strategy_desc,		(vop_t *) nfs_strategy },
172 	{ &vop_symlink_desc,		(vop_t *) nfs_symlink },
173 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
174 	{ &vop_write_desc,		(vop_t *) nfs_write },
175 	{ NULL, NULL }
176 };
177 static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
178 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
179 VNODEOP_SET(nfsv2_vnodeop_opv_desc);
180 
181 /*
182  * Special device vnode ops
183  */
184 vop_t **spec_nfsv2nodeop_p;
185 static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
186 	{ &vop_default_desc,		(vop_t *) spec_vnoperate },
187 	{ &vop_access_desc,		(vop_t *) nfsspec_access },
188 	{ &vop_close_desc,		(vop_t *) nfsspec_close },
189 	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
190 	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
191 	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
192 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
193 	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
194 	{ &vop_print_desc,		(vop_t *) nfs_print },
195 	{ &vop_read_desc,		(vop_t *) nfsspec_read },
196 	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
197 	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
198 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
199 	{ &vop_write_desc,		(vop_t *) nfsspec_write },
200 	{ NULL, NULL }
201 };
202 static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
203 	{ &spec_nfsv2nodeop_p, nfsv2_specop_entries };
204 VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
205 
206 vop_t **fifo_nfsv2nodeop_p;
207 static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
208 	{ &vop_default_desc,		(vop_t *) fifo_vnoperate },
209 	{ &vop_access_desc,		(vop_t *) nfsspec_access },
210 	{ &vop_close_desc,		(vop_t *) nfsfifo_close },
211 	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
212 	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
213 	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
214 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
215 	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
216 	{ &vop_print_desc,		(vop_t *) nfs_print },
217 	{ &vop_read_desc,		(vop_t *) nfsfifo_read },
218 	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
219 	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
220 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
221 	{ &vop_write_desc,		(vop_t *) nfsfifo_write },
222 	{ NULL, NULL }
223 };
224 static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
225 	{ &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
226 VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
227 
228 static int	nfs_mknodrpc __P((struct vnode *dvp, struct vnode **vpp,
229 				  struct componentname *cnp,
230 				  struct vattr *vap));
231 static int	nfs_removerpc __P((struct vnode *dvp, const char *name,
232 				   int namelen,
233 				   struct ucred *cred, struct proc *proc));
234 static int	nfs_renamerpc __P((struct vnode *fdvp, const char *fnameptr,
235 				   int fnamelen, struct vnode *tdvp,
236 				   const char *tnameptr, int tnamelen,
237 				   struct ucred *cred, struct proc *proc));
238 static int	nfs_renameit __P((struct vnode *sdvp,
239 				  struct componentname *scnp,
240 				  struct sillyrename *sp));
241 
242 /*
243  * Global variables
244  */
245 extern u_int32_t nfs_true, nfs_false;
246 extern u_int32_t nfs_xdrneg1;
247 extern struct nfsstats nfsstats;
248 extern nfstype nfsv3_type[9];
249 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
250 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
251 int nfs_numasync = 0;
252 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
253 
254 SYSCTL_DECL(_vfs_nfs);
255 
256 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
257 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
258 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
259 
260 static int	nfsv3_commit_on_close = 0;
261 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
262 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
263 #if 0
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
265 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
266 
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
268 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
269 #endif
270 
271 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
272 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
273 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
274 static int
275 nfs3_access_otw(struct vnode *vp,
276 		int wmode,
277 		struct proc *p,
278 		struct ucred *cred)
279 {
280 	const int v3 = 1;
281 	u_int32_t *tl;
282 	int error = 0, attrflag;
283 
284 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
285 	caddr_t bpos, dpos, cp2;
286 	register int32_t t1, t2;
287 	register caddr_t cp;
288 	u_int32_t rmode;
289 	struct nfsnode *np = VTONFS(vp);
290 
291 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
292 	nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
293 	nfsm_fhtom(vp, v3);
294 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
295 	*tl = txdr_unsigned(wmode);
296 	nfsm_request(vp, NFSPROC_ACCESS, p, cred);
297 	nfsm_postop_attr(vp, attrflag);
298 	if (!error) {
299 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
300 		rmode = fxdr_unsigned(u_int32_t, *tl);
301 		np->n_mode = rmode;
302 		np->n_modeuid = cred->cr_uid;
303 		np->n_modestamp = time_second;
304 	}
305 	nfsm_reqdone;
306 	return error;
307 }
308 
309 /*
310  * nfs access vnode op.
311  * For nfs version 2, just return ok. File accesses may fail later.
312  * For nfs version 3, use the access rpc to check accessibility. If file modes
313  * are changed on the server, accesses might still fail later.
314  */
315 static int
316 nfs_access(ap)
317 	struct vop_access_args /* {
318 		struct vnode *a_vp;
319 		int  a_mode;
320 		struct ucred *a_cred;
321 		struct proc *a_p;
322 	} */ *ap;
323 {
324 	register struct vnode *vp = ap->a_vp;
325 	int error = 0;
326 	u_int32_t mode, wmode;
327 	int v3 = NFS_ISV3(vp);
328 	struct nfsnode *np = VTONFS(vp);
329 
330 	/*
331 	 * Disallow write attempts on filesystems mounted read-only;
332 	 * unless the file is a socket, fifo, or a block or character
333 	 * device resident on the filesystem.
334 	 */
335 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
336 		switch (vp->v_type) {
337 		case VREG:
338 		case VDIR:
339 		case VLNK:
340 			return (EROFS);
341 		default:
342 			break;
343 		}
344 	}
345 	/*
346 	 * For nfs v3, check to see if we have done this recently, and if
347 	 * so return our cached result instead of making an ACCESS call.
348 	 * If not, do an access rpc, otherwise you are stuck emulating
349 	 * ufs_access() locally using the vattr. This may not be correct,
350 	 * since the server may apply other access criteria such as
351 	 * client uid-->server uid mapping that we do not know about.
352 	 */
353 	if (v3) {
354 		if (ap->a_mode & VREAD)
355 			mode = NFSV3ACCESS_READ;
356 		else
357 			mode = 0;
358 		if (vp->v_type != VDIR) {
359 			if (ap->a_mode & VWRITE)
360 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
361 			if (ap->a_mode & VEXEC)
362 				mode |= NFSV3ACCESS_EXECUTE;
363 		} else {
364 			if (ap->a_mode & VWRITE)
365 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
366 					 NFSV3ACCESS_DELETE);
367 			if (ap->a_mode & VEXEC)
368 				mode |= NFSV3ACCESS_LOOKUP;
369 		}
370 		/* XXX safety belt, only make blanket request if caching */
371 		if (nfsaccess_cache_timeout > 0) {
372 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
373 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
374 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
375 		} else {
376 			wmode = mode;
377 		}
378 
379 		/*
380 		 * Does our cached result allow us to give a definite yes to
381 		 * this request?
382 		 */
383 		if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
384 		    (ap->a_cred->cr_uid == np->n_modeuid) &&
385 		    ((np->n_mode & mode) == mode)) {
386 			nfsstats.accesscache_hits++;
387 		} else {
388 			/*
389 			 * Either a no, or a don't know.  Go to the wire.
390 			 */
391 			nfsstats.accesscache_misses++;
392 		        error = nfs3_access_otw(vp, wmode, ap->a_p,ap->a_cred);
393 			if (!error) {
394 				if ((np->n_mode & mode) != mode) {
395 					error = EACCES;
396 				}
397 			}
398 		}
399 		return (error);
400 	} else {
401 		if ((error = nfsspec_access(ap)) != 0)
402 			return (error);
403 
404 		/*
405 		 * Attempt to prevent a mapped root from accessing a file
406 		 * which it shouldn't.  We try to read a byte from the file
407 		 * if the user is root and the file is not zero length.
408 		 * After calling nfsspec_access, we should have the correct
409 		 * file size cached.
410 		 */
411 		if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
412 		    && VTONFS(vp)->n_size > 0) {
413 			struct iovec aiov;
414 			struct uio auio;
415 			char buf[1];
416 
417 			aiov.iov_base = buf;
418 			aiov.iov_len = 1;
419 			auio.uio_iov = &aiov;
420 			auio.uio_iovcnt = 1;
421 			auio.uio_offset = 0;
422 			auio.uio_resid = 1;
423 			auio.uio_segflg = UIO_SYSSPACE;
424 			auio.uio_rw = UIO_READ;
425 			auio.uio_procp = ap->a_p;
426 
427 			if (vp->v_type == VREG)
428 				error = nfs_readrpc(vp, &auio, ap->a_cred);
429 			else if (vp->v_type == VDIR) {
430 				char* bp;
431 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
432 				aiov.iov_base = bp;
433 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
434 				error = nfs_readdirrpc(vp, &auio, ap->a_cred);
435 				free(bp, M_TEMP);
436 			} else if (vp->v_type == VLNK)
437 				error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
438 			else
439 				error = EACCES;
440 		}
441 		return (error);
442 	}
443 }
444 
445 /*
446  * nfs open vnode op
447  * Check to see if the type is ok
448  * and that deletion is not in progress.
449  * For paged in text files, you will need to flush the page cache
450  * if consistency is lost.
451  */
452 /* ARGSUSED */
453 static int
454 nfs_open(ap)
455 	struct vop_open_args /* {
456 		struct vnode *a_vp;
457 		int  a_mode;
458 		struct ucred *a_cred;
459 		struct proc *a_p;
460 	} */ *ap;
461 {
462 	register struct vnode *vp = ap->a_vp;
463 	struct nfsnode *np = VTONFS(vp);
464 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
465 	struct vattr vattr;
466 	int error;
467 
468 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
469 #ifdef DIAGNOSTIC
470 		printf("open eacces vtyp=%d\n",vp->v_type);
471 #endif
472 		return (EACCES);
473 	}
474 	/*
475 	 * Get a valid lease. If cached data is stale, flush it.
476 	 */
477 	if (nmp->nm_flag & NFSMNT_NQNFS) {
478 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
479 		    do {
480 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
481 			    ap->a_p);
482 		    } while (error == NQNFS_EXPIRED);
483 		    if (error)
484 			return (error);
485 		    if (np->n_lrev != np->n_brev ||
486 			(np->n_flag & NQNFSNONCACHE)) {
487 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
488 				ap->a_p, 1)) == EINTR)
489 				return (error);
490 			np->n_brev = np->n_lrev;
491 		    }
492 		}
493 	} else {
494 		if (np->n_flag & NMODIFIED) {
495 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
496 				ap->a_p, 1)) == EINTR)
497 				return (error);
498 			np->n_attrstamp = 0;
499 			if (vp->v_type == VDIR)
500 				np->n_direofoffset = 0;
501 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
502 			if (error)
503 				return (error);
504 			np->n_mtime = vattr.va_mtime.tv_sec;
505 		} else {
506 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
507 			if (error)
508 				return (error);
509 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
510 				if (vp->v_type == VDIR)
511 					np->n_direofoffset = 0;
512 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
513 					ap->a_cred, ap->a_p, 1)) == EINTR)
514 					return (error);
515 				np->n_mtime = vattr.va_mtime.tv_sec;
516 			}
517 		}
518 	}
519 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
520 		np->n_attrstamp = 0; /* For Open/Close consistency */
521 	return (0);
522 }
523 
524 /*
525  * nfs close vnode op
526  * What an NFS client should do upon close after writing is a debatable issue.
527  * Most NFS clients push delayed writes to the server upon close, basically for
528  * two reasons:
529  * 1 - So that any write errors may be reported back to the client process
530  *     doing the close system call. By far the two most likely errors are
531  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
532  * 2 - To put a worst case upper bound on cache inconsistency between
533  *     multiple clients for the file.
534  * There is also a consistency problem for Version 2 of the protocol w.r.t.
535  * not being able to tell if other clients are writing a file concurrently,
536  * since there is no way of knowing if the changed modify time in the reply
537  * is only due to the write for this client.
538  * (NFS Version 3 provides weak cache consistency data in the reply that
539  *  should be sufficient to detect and handle this case.)
540  *
541  * The current code does the following:
542  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
543  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
544  *                     or commit them (this satisfies 1 and 2 except for the
545  *                     case where the server crashes after this close but
546  *                     before the commit RPC, which is felt to be "good
547  *                     enough". Changing the last argument to nfs_flush() to
548  *                     a 1 would force a commit operation, if it is felt a
549  *                     commit is necessary now.
550  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
551  *                     1 should be dealt with via an fsync() system call for
552  *                     cases where write errors are important.
553  */
554 /* ARGSUSED */
555 static int
556 nfs_close(ap)
557 	struct vop_close_args /* {
558 		struct vnodeop_desc *a_desc;
559 		struct vnode *a_vp;
560 		int  a_fflag;
561 		struct ucred *a_cred;
562 		struct proc *a_p;
563 	} */ *ap;
564 {
565 	register struct vnode *vp = ap->a_vp;
566 	register struct nfsnode *np = VTONFS(vp);
567 	int error = 0;
568 
569 	if (vp->v_type == VREG) {
570 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
571 		(np->n_flag & NMODIFIED)) {
572 		if (NFS_ISV3(vp)) {
573 		    /*
574 		     * Under NFSv3 we have dirty buffers to dispose of.  We
575 		     * must flush them to the NFS server.  We have the option
576 		     * of waiting all the way through the commit rpc or just
577 		     * waiting for the initial write.  The default is to only
578 		     * wait through the initial write so the data is in the
579 		     * server's cache, which is roughly similar to the state
580 		     * a standard disk subsystem leaves the file in on close().
581 		     *
582 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
583 		     * potential races with other processes, and certainly
584 		     * cannot clear it if we don't commit.
585 		     */
586 		    int cm = nfsv3_commit_on_close ? 1 : 0;
587 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, cm);
588 		    /* np->n_flag &= ~NMODIFIED; */
589 		} else {
590 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
591 		}
592 		np->n_attrstamp = 0;
593 	    }
594 	    if (np->n_flag & NWRITEERR) {
595 		np->n_flag &= ~NWRITEERR;
596 		error = np->n_error;
597 	    }
598 	}
599 	return (error);
600 }
601 
602 /*
603  * nfs getattr call from vfs.
604  */
605 static int
606 nfs_getattr(ap)
607 	struct vop_getattr_args /* {
608 		struct vnode *a_vp;
609 		struct vattr *a_vap;
610 		struct ucred *a_cred;
611 		struct proc *a_p;
612 	} */ *ap;
613 {
614 	register struct vnode *vp = ap->a_vp;
615 	register struct nfsnode *np = VTONFS(vp);
616 	register caddr_t cp;
617 	register u_int32_t *tl;
618 	register int32_t t1, t2;
619 	caddr_t bpos, dpos;
620 	int error = 0;
621 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
622 	int v3 = NFS_ISV3(vp);
623 
624 	/*
625 	 * Update local times for special files.
626 	 */
627 	if (np->n_flag & (NACC | NUPD))
628 		np->n_flag |= NCHG;
629 	/*
630 	 * First look in the cache.
631 	 */
632 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
633 		return (0);
634 
635 	if (v3 && nfsaccess_cache_timeout > 0) {
636 		nfsstats.accesscache_misses++;
637 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_p, ap->a_cred);
638 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
639 			return (0);
640 	}
641 
642 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
643 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
644 	nfsm_fhtom(vp, v3);
645 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
646 	if (!error) {
647 		nfsm_loadattr(vp, ap->a_vap);
648 	}
649 	nfsm_reqdone;
650 	return (error);
651 }
652 
653 /*
654  * nfs setattr call.
655  */
656 static int
657 nfs_setattr(ap)
658 	struct vop_setattr_args /* {
659 		struct vnodeop_desc *a_desc;
660 		struct vnode *a_vp;
661 		struct vattr *a_vap;
662 		struct ucred *a_cred;
663 		struct proc *a_p;
664 	} */ *ap;
665 {
666 	register struct vnode *vp = ap->a_vp;
667 	register struct nfsnode *np = VTONFS(vp);
668 	register struct vattr *vap = ap->a_vap;
669 	int error = 0;
670 	u_quad_t tsize;
671 
672 #ifndef nolint
673 	tsize = (u_quad_t)0;
674 #endif
675 
676 	/*
677 	 * Setting of flags is not supported.
678 	 */
679 	if (vap->va_flags != VNOVAL)
680 		return (EOPNOTSUPP);
681 
682 	/*
683 	 * Disallow write attempts if the filesystem is mounted read-only.
684 	 */
685   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
686 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
687 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
688 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
689 		return (EROFS);
690 	if (vap->va_size != VNOVAL) {
691  		switch (vp->v_type) {
692  		case VDIR:
693  			return (EISDIR);
694  		case VCHR:
695  		case VBLK:
696  		case VSOCK:
697  		case VFIFO:
698 			if (vap->va_mtime.tv_sec == VNOVAL &&
699 			    vap->va_atime.tv_sec == VNOVAL &&
700 			    vap->va_mode == (mode_t)VNOVAL &&
701 			    vap->va_uid == (uid_t)VNOVAL &&
702 			    vap->va_gid == (gid_t)VNOVAL)
703 				return (0);
704  			vap->va_size = VNOVAL;
705  			break;
706  		default:
707 			/*
708 			 * Disallow write attempts if the filesystem is
709 			 * mounted read-only.
710 			 */
711 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
712 				return (EROFS);
713 
714 			/*
715 			 * We run vnode_pager_setsize() early (why?),
716 			 * we must set np->n_size now to avoid vinvalbuf
717 			 * V_SAVE races that might setsize a lower
718 			 * value.
719 			 */
720 
721 			tsize = np->n_size;
722 			error = nfs_meta_setsize(vp, ap->a_cred,
723 						ap->a_p, vap->va_size);
724 
725  			if (np->n_flag & NMODIFIED) {
726  			    if (vap->va_size == 0)
727  				error = nfs_vinvalbuf(vp, 0,
728  					ap->a_cred, ap->a_p, 1);
729  			    else
730  				error = nfs_vinvalbuf(vp, V_SAVE,
731  					ap->a_cred, ap->a_p, 1);
732  			    if (error) {
733 				np->n_size = tsize;
734 				vnode_pager_setsize(vp, np->n_size);
735  				return (error);
736 			    }
737  			}
738 			np->n_vattr.va_size = vap->va_size;
739   		};
740   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
741 		vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
742 		vp->v_type == VREG &&
743   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
744 		 ap->a_p, 1)) == EINTR)
745 		return (error);
746 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
747 	if (error && vap->va_size != VNOVAL) {
748 		np->n_size = np->n_vattr.va_size = tsize;
749 		vnode_pager_setsize(vp, np->n_size);
750 	}
751 	return (error);
752 }
753 
754 /*
755  * Do an nfs setattr rpc.
756  */
757 static int
758 nfs_setattrrpc(vp, vap, cred, procp)
759 	register struct vnode *vp;
760 	register struct vattr *vap;
761 	struct ucred *cred;
762 	struct proc *procp;
763 {
764 	register struct nfsv2_sattr *sp;
765 	register caddr_t cp;
766 	register int32_t t1, t2;
767 	caddr_t bpos, dpos, cp2;
768 	u_int32_t *tl;
769 	int error = 0, wccflag = NFSV3_WCCRATTR;
770 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
771 	int v3 = NFS_ISV3(vp);
772 
773 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
774 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
775 	nfsm_fhtom(vp, v3);
776 	if (v3) {
777 		nfsm_v3attrbuild(vap, TRUE);
778 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
779 		*tl = nfs_false;
780 	} else {
781 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
782 		if (vap->va_mode == (mode_t)VNOVAL)
783 			sp->sa_mode = nfs_xdrneg1;
784 		else
785 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
786 		if (vap->va_uid == (uid_t)VNOVAL)
787 			sp->sa_uid = nfs_xdrneg1;
788 		else
789 			sp->sa_uid = txdr_unsigned(vap->va_uid);
790 		if (vap->va_gid == (gid_t)VNOVAL)
791 			sp->sa_gid = nfs_xdrneg1;
792 		else
793 			sp->sa_gid = txdr_unsigned(vap->va_gid);
794 		sp->sa_size = txdr_unsigned(vap->va_size);
795 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
796 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
797 	}
798 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
799 	if (v3) {
800 		nfsm_wcc_data(vp, wccflag);
801 	} else
802 		nfsm_loadattr(vp, (struct vattr *)0);
803 	nfsm_reqdone;
804 	return (error);
805 }
806 
807 /*
808  * nfs lookup call, one step at a time...
809  * First look in cache
810  * If not found, unlock the directory nfsnode and do the rpc
811  */
812 static int
813 nfs_lookup(ap)
814 	struct vop_lookup_args /* {
815 		struct vnodeop_desc *a_desc;
816 		struct vnode *a_dvp;
817 		struct vnode **a_vpp;
818 		struct componentname *a_cnp;
819 	} */ *ap;
820 {
821 	struct componentname *cnp = ap->a_cnp;
822 	struct vnode *dvp = ap->a_dvp;
823 	struct vnode **vpp = ap->a_vpp;
824 	int flags = cnp->cn_flags;
825 	struct vnode *newvp;
826 	u_int32_t *tl;
827 	caddr_t cp;
828 	int32_t t1, t2;
829 	struct nfsmount *nmp;
830 	caddr_t bpos, dpos, cp2;
831 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
832 	long len;
833 	nfsfh_t *fhp;
834 	struct nfsnode *np;
835 	int lockparent, wantparent, error = 0, attrflag, fhsize;
836 	int v3 = NFS_ISV3(dvp);
837 	struct proc *p = cnp->cn_proc;
838 
839 	*vpp = NULLVP;
840 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
841 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
842 		return (EROFS);
843 	if (dvp->v_type != VDIR)
844 		return (ENOTDIR);
845 	lockparent = flags & LOCKPARENT;
846 	wantparent = flags & (LOCKPARENT|WANTPARENT);
847 	nmp = VFSTONFS(dvp->v_mount);
848 	np = VTONFS(dvp);
849 	if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
850 		struct vattr vattr;
851 		int vpid;
852 
853 		if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, p)) != 0) {
854 			*vpp = NULLVP;
855 			return (error);
856 		}
857 
858 		newvp = *vpp;
859 		vpid = newvp->v_id;
860 		/*
861 		 * See the comment starting `Step through' in ufs/ufs_lookup.c
862 		 * for an explanation of the locking protocol
863 		 */
864 		if (dvp == newvp) {
865 			VREF(newvp);
866 			error = 0;
867 		} else if (flags & ISDOTDOT) {
868 			VOP_UNLOCK(dvp, 0, p);
869 			error = vget(newvp, LK_EXCLUSIVE, p);
870 			if (!error && lockparent && (flags & ISLASTCN))
871 				error = vn_lock(dvp, LK_EXCLUSIVE, p);
872 		} else {
873 			error = vget(newvp, LK_EXCLUSIVE, p);
874 			if (!lockparent || error || !(flags & ISLASTCN))
875 				VOP_UNLOCK(dvp, 0, p);
876 		}
877 		if (!error) {
878 			if (vpid == newvp->v_id) {
879 			   if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, p)
880 			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
881 				nfsstats.lookupcache_hits++;
882 				if (cnp->cn_nameiop != LOOKUP &&
883 				    (flags & ISLASTCN))
884 					cnp->cn_flags |= SAVENAME;
885 				return (0);
886 			   }
887 			   cache_purge(newvp);
888 			}
889 			vput(newvp);
890 			if (lockparent && dvp != newvp && (flags & ISLASTCN))
891 				VOP_UNLOCK(dvp, 0, p);
892 		}
893 		error = vn_lock(dvp, LK_EXCLUSIVE, p);
894 		*vpp = NULLVP;
895 		if (error)
896 			return (error);
897 	}
898 	error = 0;
899 	newvp = NULLVP;
900 	nfsstats.lookupcache_misses++;
901 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
902 	len = cnp->cn_namelen;
903 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
904 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
905 	nfsm_fhtom(dvp, v3);
906 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
907 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
908 	if (error) {
909 		nfsm_postop_attr(dvp, attrflag);
910 		m_freem(mrep);
911 		goto nfsmout;
912 	}
913 	nfsm_getfh(fhp, fhsize, v3);
914 
915 	/*
916 	 * Handle RENAME case...
917 	 */
918 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
919 		if (NFS_CMPFH(np, fhp, fhsize)) {
920 			m_freem(mrep);
921 			return (EISDIR);
922 		}
923 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
924 		if (error) {
925 			m_freem(mrep);
926 			return (error);
927 		}
928 		newvp = NFSTOV(np);
929 		if (v3) {
930 			nfsm_postop_attr(newvp, attrflag);
931 			nfsm_postop_attr(dvp, attrflag);
932 		} else
933 			nfsm_loadattr(newvp, (struct vattr *)0);
934 		*vpp = newvp;
935 		m_freem(mrep);
936 		cnp->cn_flags |= SAVENAME;
937 		if (!lockparent)
938 			VOP_UNLOCK(dvp, 0, p);
939 		return (0);
940 	}
941 
942 	if (flags & ISDOTDOT) {
943 		VOP_UNLOCK(dvp, 0, p);
944 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
945 		if (error) {
946 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
947 			return (error);
948 		}
949 		newvp = NFSTOV(np);
950 		if (lockparent && (flags & ISLASTCN) &&
951 		    (error = vn_lock(dvp, LK_EXCLUSIVE, p))) {
952 		    	vput(newvp);
953 			return (error);
954 		}
955 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
956 		VREF(dvp);
957 		newvp = dvp;
958 	} else {
959 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
960 		if (error) {
961 			m_freem(mrep);
962 			return (error);
963 		}
964 		if (!lockparent || !(flags & ISLASTCN))
965 			VOP_UNLOCK(dvp, 0, p);
966 		newvp = NFSTOV(np);
967 	}
968 	if (v3) {
969 		nfsm_postop_attr(newvp, attrflag);
970 		nfsm_postop_attr(dvp, attrflag);
971 	} else
972 		nfsm_loadattr(newvp, (struct vattr *)0);
973 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
974 		cnp->cn_flags |= SAVENAME;
975 	if ((cnp->cn_flags & MAKEENTRY) &&
976 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
977 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
978 		cache_enter(dvp, newvp, cnp);
979 	}
980 	*vpp = newvp;
981 	nfsm_reqdone;
982 	if (error) {
983 		if (newvp != NULLVP) {
984 			vrele(newvp);
985 			*vpp = NULLVP;
986 		}
987 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
988 		    (flags & ISLASTCN) && error == ENOENT) {
989 			if (!lockparent)
990 				VOP_UNLOCK(dvp, 0, p);
991 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
992 				error = EROFS;
993 			else
994 				error = EJUSTRETURN;
995 		}
996 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
997 			cnp->cn_flags |= SAVENAME;
998 	}
999 	return (error);
1000 }
1001 
1002 /*
1003  * nfs read call.
1004  * Just call nfs_bioread() to do the work.
1005  */
1006 static int
1007 nfs_read(ap)
1008 	struct vop_read_args /* {
1009 		struct vnode *a_vp;
1010 		struct uio *a_uio;
1011 		int  a_ioflag;
1012 		struct ucred *a_cred;
1013 	} */ *ap;
1014 {
1015 	register struct vnode *vp = ap->a_vp;
1016 
1017 	if (vp->v_type != VREG)
1018 		return (EPERM);
1019 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1020 }
1021 
1022 /*
1023  * nfs readlink call
1024  */
1025 static int
1026 nfs_readlink(ap)
1027 	struct vop_readlink_args /* {
1028 		struct vnode *a_vp;
1029 		struct uio *a_uio;
1030 		struct ucred *a_cred;
1031 	} */ *ap;
1032 {
1033 	register struct vnode *vp = ap->a_vp;
1034 
1035 	if (vp->v_type != VLNK)
1036 		return (EINVAL);
1037 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1038 }
1039 
1040 /*
1041  * Do a readlink rpc.
1042  * Called by nfs_doio() from below the buffer cache.
1043  */
1044 int
1045 nfs_readlinkrpc(vp, uiop, cred)
1046 	register struct vnode *vp;
1047 	struct uio *uiop;
1048 	struct ucred *cred;
1049 {
1050 	register u_int32_t *tl;
1051 	register caddr_t cp;
1052 	register int32_t t1, t2;
1053 	caddr_t bpos, dpos, cp2;
1054 	int error = 0, len, attrflag;
1055 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1056 	int v3 = NFS_ISV3(vp);
1057 
1058 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1059 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1060 	nfsm_fhtom(vp, v3);
1061 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
1062 	if (v3)
1063 		nfsm_postop_attr(vp, attrflag);
1064 	if (!error) {
1065 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1066 		if (len == NFS_MAXPATHLEN) {
1067 			struct nfsnode *np = VTONFS(vp);
1068 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1069 				len = np->n_size;
1070 		}
1071 		nfsm_mtouio(uiop, len);
1072 	}
1073 	nfsm_reqdone;
1074 	return (error);
1075 }
1076 
1077 /*
1078  * nfs read rpc call
1079  * Ditto above
1080  */
1081 int
1082 nfs_readrpc(vp, uiop, cred)
1083 	register struct vnode *vp;
1084 	struct uio *uiop;
1085 	struct ucred *cred;
1086 {
1087 	register u_int32_t *tl;
1088 	register caddr_t cp;
1089 	register int32_t t1, t2;
1090 	caddr_t bpos, dpos, cp2;
1091 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1092 	struct nfsmount *nmp;
1093 	int error = 0, len, retlen, tsiz, eof, attrflag;
1094 	int v3 = NFS_ISV3(vp);
1095 
1096 #ifndef nolint
1097 	eof = 0;
1098 #endif
1099 	nmp = VFSTONFS(vp->v_mount);
1100 	tsiz = uiop->uio_resid;
1101 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1102 		return (EFBIG);
1103 	while (tsiz > 0) {
1104 		nfsstats.rpccnt[NFSPROC_READ]++;
1105 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1106 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1107 		nfsm_fhtom(vp, v3);
1108 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1109 		if (v3) {
1110 			txdr_hyper(uiop->uio_offset, tl);
1111 			*(tl + 2) = txdr_unsigned(len);
1112 		} else {
1113 			*tl++ = txdr_unsigned(uiop->uio_offset);
1114 			*tl++ = txdr_unsigned(len);
1115 			*tl = 0;
1116 		}
1117 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp, cred);
1118 		if (v3) {
1119 			nfsm_postop_attr(vp, attrflag);
1120 			if (error) {
1121 				m_freem(mrep);
1122 				goto nfsmout;
1123 			}
1124 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1125 			eof = fxdr_unsigned(int, *(tl + 1));
1126 		} else
1127 			nfsm_loadattr(vp, (struct vattr *)0);
1128 		nfsm_strsiz(retlen, nmp->nm_rsize);
1129 		nfsm_mtouio(uiop, retlen);
1130 		m_freem(mrep);
1131 		tsiz -= retlen;
1132 		if (v3) {
1133 			if (eof || retlen == 0) {
1134 				tsiz = 0;
1135 			}
1136 		} else if (retlen < len) {
1137 			tsiz = 0;
1138 		}
1139 	}
1140 nfsmout:
1141 	return (error);
1142 }
1143 
1144 /*
1145  * nfs write call
1146  */
1147 int
1148 nfs_writerpc(vp, uiop, cred, iomode, must_commit)
1149 	register struct vnode *vp;
1150 	register struct uio *uiop;
1151 	struct ucred *cred;
1152 	int *iomode, *must_commit;
1153 {
1154 	register u_int32_t *tl;
1155 	register caddr_t cp;
1156 	register int32_t t1, t2, backup;
1157 	caddr_t bpos, dpos, cp2;
1158 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1159 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1160 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1161 	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1162 
1163 #ifndef DIAGNOSTIC
1164 	if (uiop->uio_iovcnt != 1)
1165 		panic("nfs: writerpc iovcnt > 1");
1166 #endif
1167 	*must_commit = 0;
1168 	tsiz = uiop->uio_resid;
1169 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1170 		return (EFBIG);
1171 	while (tsiz > 0) {
1172 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1173 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1174 		nfsm_reqhead(vp, NFSPROC_WRITE,
1175 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1176 		nfsm_fhtom(vp, v3);
1177 		if (v3) {
1178 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1179 			txdr_hyper(uiop->uio_offset, tl);
1180 			tl += 2;
1181 			*tl++ = txdr_unsigned(len);
1182 			*tl++ = txdr_unsigned(*iomode);
1183 			*tl = txdr_unsigned(len);
1184 		} else {
1185 			register u_int32_t x;
1186 
1187 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1188 			/* Set both "begin" and "current" to non-garbage. */
1189 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1190 			*tl++ = x;	/* "begin offset" */
1191 			*tl++ = x;	/* "current offset" */
1192 			x = txdr_unsigned(len);
1193 			*tl++ = x;	/* total to this offset */
1194 			*tl = x;	/* size of this write */
1195 		}
1196 		nfsm_uiotom(uiop, len);
1197 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp, cred);
1198 		if (v3) {
1199 			wccflag = NFSV3_WCCCHK;
1200 			nfsm_wcc_data(vp, wccflag);
1201 			if (!error) {
1202 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1203 					+ NFSX_V3WRITEVERF);
1204 				rlen = fxdr_unsigned(int, *tl++);
1205 				if (rlen == 0) {
1206 					error = NFSERR_IO;
1207 					m_freem(mrep);
1208 					break;
1209 				} else if (rlen < len) {
1210 					backup = len - rlen;
1211 					uiop->uio_iov->iov_base -= backup;
1212 					uiop->uio_iov->iov_len += backup;
1213 					uiop->uio_offset -= backup;
1214 					uiop->uio_resid += backup;
1215 					len = rlen;
1216 				}
1217 				commit = fxdr_unsigned(int, *tl++);
1218 
1219 				/*
1220 				 * Return the lowest committment level
1221 				 * obtained by any of the RPCs.
1222 				 */
1223 				if (committed == NFSV3WRITE_FILESYNC)
1224 					committed = commit;
1225 				else if (committed == NFSV3WRITE_DATASYNC &&
1226 					commit == NFSV3WRITE_UNSTABLE)
1227 					committed = commit;
1228 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1229 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1230 					NFSX_V3WRITEVERF);
1231 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1232 				} else if (bcmp((caddr_t)tl,
1233 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1234 				    *must_commit = 1;
1235 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1236 					NFSX_V3WRITEVERF);
1237 				}
1238 			}
1239 		} else
1240 		    nfsm_loadattr(vp, (struct vattr *)0);
1241 		if (wccflag)
1242 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1243 		m_freem(mrep);
1244 		if (error)
1245 			break;
1246 		tsiz -= len;
1247 	}
1248 nfsmout:
1249 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1250 		committed = NFSV3WRITE_FILESYNC;
1251 	*iomode = committed;
1252 	if (error)
1253 		uiop->uio_resid = tsiz;
1254 	return (error);
1255 }
1256 
1257 /*
1258  * nfs mknod rpc
1259  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1260  * mode set to specify the file type and the size field for rdev.
1261  */
1262 static int
1263 nfs_mknodrpc(dvp, vpp, cnp, vap)
1264 	register struct vnode *dvp;
1265 	register struct vnode **vpp;
1266 	register struct componentname *cnp;
1267 	register struct vattr *vap;
1268 {
1269 	register struct nfsv2_sattr *sp;
1270 	register u_int32_t *tl;
1271 	register caddr_t cp;
1272 	register int32_t t1, t2;
1273 	struct vnode *newvp = (struct vnode *)0;
1274 	struct nfsnode *np = (struct nfsnode *)0;
1275 	struct vattr vattr;
1276 	char *cp2;
1277 	caddr_t bpos, dpos;
1278 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1279 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1280 	u_int32_t rdev;
1281 	int v3 = NFS_ISV3(dvp);
1282 
1283 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1284 		rdev = txdr_unsigned(vap->va_rdev);
1285 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1286 		rdev = nfs_xdrneg1;
1287 	else {
1288 		return (EOPNOTSUPP);
1289 	}
1290 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_proc)) != 0) {
1291 		return (error);
1292 	}
1293 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1294 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1295 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1296 	nfsm_fhtom(dvp, v3);
1297 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1298 	if (v3) {
1299 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1300 		*tl++ = vtonfsv3_type(vap->va_type);
1301 		nfsm_v3attrbuild(vap, FALSE);
1302 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1303 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1304 			*tl++ = txdr_unsigned(umajor(vap->va_rdev));
1305 			*tl = txdr_unsigned(uminor(vap->va_rdev));
1306 		}
1307 	} else {
1308 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1309 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1310 		sp->sa_uid = nfs_xdrneg1;
1311 		sp->sa_gid = nfs_xdrneg1;
1312 		sp->sa_size = rdev;
1313 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1314 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1315 	}
1316 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1317 	if (!error) {
1318 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1319 		if (!gotvp) {
1320 			if (newvp) {
1321 				vput(newvp);
1322 				newvp = (struct vnode *)0;
1323 			}
1324 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1325 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1326 			if (!error)
1327 				newvp = NFSTOV(np);
1328 		}
1329 	}
1330 	if (v3)
1331 		nfsm_wcc_data(dvp, wccflag);
1332 	nfsm_reqdone;
1333 	if (error) {
1334 		if (newvp)
1335 			vput(newvp);
1336 	} else {
1337 		if (cnp->cn_flags & MAKEENTRY)
1338 			cache_enter(dvp, newvp, cnp);
1339 		*vpp = newvp;
1340 	}
1341 	VTONFS(dvp)->n_flag |= NMODIFIED;
1342 	if (!wccflag)
1343 		VTONFS(dvp)->n_attrstamp = 0;
1344 	return (error);
1345 }
1346 
1347 /*
1348  * nfs mknod vop
1349  * just call nfs_mknodrpc() to do the work.
1350  */
1351 /* ARGSUSED */
1352 static int
1353 nfs_mknod(ap)
1354 	struct vop_mknod_args /* {
1355 		struct vnode *a_dvp;
1356 		struct vnode **a_vpp;
1357 		struct componentname *a_cnp;
1358 		struct vattr *a_vap;
1359 	} */ *ap;
1360 {
1361 	return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1362 }
1363 
1364 static u_long create_verf;
1365 /*
1366  * nfs file create call
1367  */
1368 static int
1369 nfs_create(ap)
1370 	struct vop_create_args /* {
1371 		struct vnode *a_dvp;
1372 		struct vnode **a_vpp;
1373 		struct componentname *a_cnp;
1374 		struct vattr *a_vap;
1375 	} */ *ap;
1376 {
1377 	register struct vnode *dvp = ap->a_dvp;
1378 	register struct vattr *vap = ap->a_vap;
1379 	register struct componentname *cnp = ap->a_cnp;
1380 	register struct nfsv2_sattr *sp;
1381 	register u_int32_t *tl;
1382 	register caddr_t cp;
1383 	register int32_t t1, t2;
1384 	struct nfsnode *np = (struct nfsnode *)0;
1385 	struct vnode *newvp = (struct vnode *)0;
1386 	caddr_t bpos, dpos, cp2;
1387 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1388 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1389 	struct vattr vattr;
1390 	int v3 = NFS_ISV3(dvp);
1391 
1392 	/*
1393 	 * Oops, not for me..
1394 	 */
1395 	if (vap->va_type == VSOCK)
1396 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1397 
1398 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_proc)) != 0) {
1399 		return (error);
1400 	}
1401 	if (vap->va_vaflags & VA_EXCLUSIVE)
1402 		fmode |= O_EXCL;
1403 again:
1404 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1405 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1406 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1407 	nfsm_fhtom(dvp, v3);
1408 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1409 	if (v3) {
1410 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1411 		if (fmode & O_EXCL) {
1412 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1413 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1414 #ifdef INET
1415 			if (!TAILQ_EMPTY(&in_ifaddrhead))
1416 				*tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
1417 			else
1418 #endif
1419 				*tl++ = create_verf;
1420 			*tl = ++create_verf;
1421 		} else {
1422 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1423 			nfsm_v3attrbuild(vap, FALSE);
1424 		}
1425 	} else {
1426 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1427 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1428 		sp->sa_uid = nfs_xdrneg1;
1429 		sp->sa_gid = nfs_xdrneg1;
1430 		sp->sa_size = 0;
1431 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1432 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1433 	}
1434 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1435 	if (!error) {
1436 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1437 		if (!gotvp) {
1438 			if (newvp) {
1439 				vput(newvp);
1440 				newvp = (struct vnode *)0;
1441 			}
1442 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1443 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1444 			if (!error)
1445 				newvp = NFSTOV(np);
1446 		}
1447 	}
1448 	if (v3)
1449 		nfsm_wcc_data(dvp, wccflag);
1450 	nfsm_reqdone;
1451 	if (error) {
1452 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1453 			fmode &= ~O_EXCL;
1454 			goto again;
1455 		}
1456 		if (newvp)
1457 			vput(newvp);
1458 	} else if (v3 && (fmode & O_EXCL)) {
1459 		/*
1460 		 * We are normally called with only a partially initialized
1461 		 * VAP.  Since the NFSv3 spec says that server may use the
1462 		 * file attributes to store the verifier, the spec requires
1463 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1464 		 * in atime, but we can't really assume that all servers will
1465 		 * so we ensure that our SETATTR sets both atime and mtime.
1466 		 */
1467 		if (vap->va_mtime.tv_sec == VNOVAL)
1468 			vfs_timestamp(&vap->va_mtime);
1469 		if (vap->va_atime.tv_sec == VNOVAL)
1470 			vap->va_atime = vap->va_mtime;
1471 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1472 	}
1473 	if (!error) {
1474 		if (cnp->cn_flags & MAKEENTRY)
1475 			cache_enter(dvp, newvp, cnp);
1476 		*ap->a_vpp = newvp;
1477 	}
1478 	VTONFS(dvp)->n_flag |= NMODIFIED;
1479 	if (!wccflag)
1480 		VTONFS(dvp)->n_attrstamp = 0;
1481 	return (error);
1482 }
1483 
1484 /*
1485  * nfs file remove call
1486  * To try and make nfs semantics closer to ufs semantics, a file that has
1487  * other processes using the vnode is renamed instead of removed and then
1488  * removed later on the last close.
1489  * - If v_usecount > 1
1490  *	  If a rename is not already in the works
1491  *	     call nfs_sillyrename() to set it up
1492  *     else
1493  *	  do the remove rpc
1494  */
1495 static int
1496 nfs_remove(ap)
1497 	struct vop_remove_args /* {
1498 		struct vnodeop_desc *a_desc;
1499 		struct vnode * a_dvp;
1500 		struct vnode * a_vp;
1501 		struct componentname * a_cnp;
1502 	} */ *ap;
1503 {
1504 	register struct vnode *vp = ap->a_vp;
1505 	register struct vnode *dvp = ap->a_dvp;
1506 	register struct componentname *cnp = ap->a_cnp;
1507 	register struct nfsnode *np = VTONFS(vp);
1508 	int error = 0;
1509 	struct vattr vattr;
1510 
1511 #ifndef DIAGNOSTIC
1512 	if ((cnp->cn_flags & HASBUF) == 0)
1513 		panic("nfs_remove: no name");
1514 	if (vp->v_usecount < 1)
1515 		panic("nfs_remove: bad v_usecount");
1516 #endif
1517 	if (vp->v_type == VDIR)
1518 		error = EPERM;
1519 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1520 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1521 	    vattr.va_nlink > 1)) {
1522 		/*
1523 		 * Purge the name cache so that the chance of a lookup for
1524 		 * the name succeeding while the remove is in progress is
1525 		 * minimized. Without node locking it can still happen, such
1526 		 * that an I/O op returns ESTALE, but since you get this if
1527 		 * another host removes the file..
1528 		 */
1529 		cache_purge(vp);
1530 		/*
1531 		 * throw away biocache buffers, mainly to avoid
1532 		 * unnecessary delayed writes later.
1533 		 */
1534 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1535 		/* Do the rpc */
1536 		if (error != EINTR)
1537 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1538 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1539 		/*
1540 		 * Kludge City: If the first reply to the remove rpc is lost..
1541 		 *   the reply to the retransmitted request will be ENOENT
1542 		 *   since the file was in fact removed
1543 		 *   Therefore, we cheat and return success.
1544 		 */
1545 		if (error == ENOENT)
1546 			error = 0;
1547 	} else if (!np->n_sillyrename)
1548 		error = nfs_sillyrename(dvp, vp, cnp);
1549 	np->n_attrstamp = 0;
1550 	return (error);
1551 }
1552 
1553 /*
1554  * nfs file remove rpc called from nfs_inactive
1555  */
1556 int
1557 nfs_removeit(sp)
1558 	register struct sillyrename *sp;
1559 {
1560 
1561 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1562 		(struct proc *)0));
1563 }
1564 
1565 /*
1566  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1567  */
1568 static int
1569 nfs_removerpc(dvp, name, namelen, cred, proc)
1570 	register struct vnode *dvp;
1571 	const char *name;
1572 	int namelen;
1573 	struct ucred *cred;
1574 	struct proc *proc;
1575 {
1576 	register u_int32_t *tl;
1577 	register caddr_t cp;
1578 	register int32_t t1, t2;
1579 	caddr_t bpos, dpos, cp2;
1580 	int error = 0, wccflag = NFSV3_WCCRATTR;
1581 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1582 	int v3 = NFS_ISV3(dvp);
1583 
1584 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1585 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1586 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1587 	nfsm_fhtom(dvp, v3);
1588 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1589 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1590 	if (v3)
1591 		nfsm_wcc_data(dvp, wccflag);
1592 	nfsm_reqdone;
1593 	VTONFS(dvp)->n_flag |= NMODIFIED;
1594 	if (!wccflag)
1595 		VTONFS(dvp)->n_attrstamp = 0;
1596 	return (error);
1597 }
1598 
1599 /*
1600  * nfs file rename call
1601  */
1602 static int
1603 nfs_rename(ap)
1604 	struct vop_rename_args  /* {
1605 		struct vnode *a_fdvp;
1606 		struct vnode *a_fvp;
1607 		struct componentname *a_fcnp;
1608 		struct vnode *a_tdvp;
1609 		struct vnode *a_tvp;
1610 		struct componentname *a_tcnp;
1611 	} */ *ap;
1612 {
1613 	register struct vnode *fvp = ap->a_fvp;
1614 	register struct vnode *tvp = ap->a_tvp;
1615 	register struct vnode *fdvp = ap->a_fdvp;
1616 	register struct vnode *tdvp = ap->a_tdvp;
1617 	register struct componentname *tcnp = ap->a_tcnp;
1618 	register struct componentname *fcnp = ap->a_fcnp;
1619 	int error;
1620 
1621 #ifndef DIAGNOSTIC
1622 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1623 	    (fcnp->cn_flags & HASBUF) == 0)
1624 		panic("nfs_rename: no name");
1625 #endif
1626 	/* Check for cross-device rename */
1627 	if ((fvp->v_mount != tdvp->v_mount) ||
1628 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1629 		error = EXDEV;
1630 		goto out;
1631 	}
1632 
1633 	/*
1634 	 * We have to flush B_DELWRI data prior to renaming
1635 	 * the file.  If we don't, the delayed-write buffers
1636 	 * can be flushed out later after the file has gone stale
1637 	 * under NFSV3.  NFSV2 does not have this problem because
1638 	 * ( as far as I can tell ) it flushes dirty buffers more
1639 	 * often.
1640 	 */
1641 
1642 	VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_proc);
1643 	if (tvp)
1644 	    VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_proc);
1645 
1646 	/*
1647 	 * If the tvp exists and is in use, sillyrename it before doing the
1648 	 * rename of the new file over it.
1649 	 * XXX Can't sillyrename a directory.
1650 	 */
1651 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1652 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1653 		vput(tvp);
1654 		tvp = NULL;
1655 	}
1656 
1657 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1658 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1659 		tcnp->cn_proc);
1660 
1661 	if (fvp->v_type == VDIR) {
1662 		if (tvp != NULL && tvp->v_type == VDIR)
1663 			cache_purge(tdvp);
1664 		cache_purge(fdvp);
1665 	}
1666 
1667 out:
1668 	if (tdvp == tvp)
1669 		vrele(tdvp);
1670 	else
1671 		vput(tdvp);
1672 	if (tvp)
1673 		vput(tvp);
1674 	vrele(fdvp);
1675 	vrele(fvp);
1676 	/*
1677 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1678 	 */
1679 	if (error == ENOENT)
1680 		error = 0;
1681 	return (error);
1682 }
1683 
1684 /*
1685  * nfs file rename rpc called from nfs_remove() above
1686  */
1687 static int
1688 nfs_renameit(sdvp, scnp, sp)
1689 	struct vnode *sdvp;
1690 	struct componentname *scnp;
1691 	register struct sillyrename *sp;
1692 {
1693 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1694 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1695 }
1696 
1697 /*
1698  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1699  */
1700 static int
1701 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1702 	register struct vnode *fdvp;
1703 	const char *fnameptr;
1704 	int fnamelen;
1705 	register struct vnode *tdvp;
1706 	const char *tnameptr;
1707 	int tnamelen;
1708 	struct ucred *cred;
1709 	struct proc *proc;
1710 {
1711 	register u_int32_t *tl;
1712 	register caddr_t cp;
1713 	register int32_t t1, t2;
1714 	caddr_t bpos, dpos, cp2;
1715 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1716 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1717 	int v3 = NFS_ISV3(fdvp);
1718 
1719 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1720 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1721 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1722 		nfsm_rndup(tnamelen));
1723 	nfsm_fhtom(fdvp, v3);
1724 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1725 	nfsm_fhtom(tdvp, v3);
1726 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1727 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1728 	if (v3) {
1729 		nfsm_wcc_data(fdvp, fwccflag);
1730 		nfsm_wcc_data(tdvp, twccflag);
1731 	}
1732 	nfsm_reqdone;
1733 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1734 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1735 	if (!fwccflag)
1736 		VTONFS(fdvp)->n_attrstamp = 0;
1737 	if (!twccflag)
1738 		VTONFS(tdvp)->n_attrstamp = 0;
1739 	return (error);
1740 }
1741 
1742 /*
1743  * nfs hard link create call
1744  */
1745 static int
1746 nfs_link(ap)
1747 	struct vop_link_args /* {
1748 		struct vnode *a_tdvp;
1749 		struct vnode *a_vp;
1750 		struct componentname *a_cnp;
1751 	} */ *ap;
1752 {
1753 	register struct vnode *vp = ap->a_vp;
1754 	register struct vnode *tdvp = ap->a_tdvp;
1755 	register struct componentname *cnp = ap->a_cnp;
1756 	register u_int32_t *tl;
1757 	register caddr_t cp;
1758 	register int32_t t1, t2;
1759 	caddr_t bpos, dpos, cp2;
1760 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1761 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1762 	int v3;
1763 
1764 	if (vp->v_mount != tdvp->v_mount) {
1765 		return (EXDEV);
1766 	}
1767 
1768 	/*
1769 	 * Push all writes to the server, so that the attribute cache
1770 	 * doesn't get "out of sync" with the server.
1771 	 * XXX There should be a better way!
1772 	 */
1773 	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1774 
1775 	v3 = NFS_ISV3(vp);
1776 	nfsstats.rpccnt[NFSPROC_LINK]++;
1777 	nfsm_reqhead(vp, NFSPROC_LINK,
1778 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1779 	nfsm_fhtom(vp, v3);
1780 	nfsm_fhtom(tdvp, v3);
1781 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1782 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1783 	if (v3) {
1784 		nfsm_postop_attr(vp, attrflag);
1785 		nfsm_wcc_data(tdvp, wccflag);
1786 	}
1787 	nfsm_reqdone;
1788 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1789 	if (!attrflag)
1790 		VTONFS(vp)->n_attrstamp = 0;
1791 	if (!wccflag)
1792 		VTONFS(tdvp)->n_attrstamp = 0;
1793 	/*
1794 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1795 	 */
1796 	if (error == EEXIST)
1797 		error = 0;
1798 	return (error);
1799 }
1800 
1801 /*
1802  * nfs symbolic link create call
1803  */
1804 static int
1805 nfs_symlink(ap)
1806 	struct vop_symlink_args /* {
1807 		struct vnode *a_dvp;
1808 		struct vnode **a_vpp;
1809 		struct componentname *a_cnp;
1810 		struct vattr *a_vap;
1811 		char *a_target;
1812 	} */ *ap;
1813 {
1814 	register struct vnode *dvp = ap->a_dvp;
1815 	register struct vattr *vap = ap->a_vap;
1816 	register struct componentname *cnp = ap->a_cnp;
1817 	register struct nfsv2_sattr *sp;
1818 	register u_int32_t *tl;
1819 	register caddr_t cp;
1820 	register int32_t t1, t2;
1821 	caddr_t bpos, dpos, cp2;
1822 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1823 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1824 	struct vnode *newvp = (struct vnode *)0;
1825 	int v3 = NFS_ISV3(dvp);
1826 
1827 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1828 	slen = strlen(ap->a_target);
1829 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1830 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1831 	nfsm_fhtom(dvp, v3);
1832 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1833 	if (v3) {
1834 		nfsm_v3attrbuild(vap, FALSE);
1835 	}
1836 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1837 	if (!v3) {
1838 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1839 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1840 		sp->sa_uid = nfs_xdrneg1;
1841 		sp->sa_gid = nfs_xdrneg1;
1842 		sp->sa_size = nfs_xdrneg1;
1843 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1844 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1845 	}
1846 
1847 	/*
1848 	 * Issue the NFS request and get the rpc response.
1849 	 *
1850 	 * Only NFSv3 responses returning an error of 0 actually return
1851 	 * a file handle that can be converted into newvp without having
1852 	 * to do an extra lookup rpc.
1853 	 */
1854 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1855 	if (v3) {
1856 		if (error == 0)
1857 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1858 		nfsm_wcc_data(dvp, wccflag);
1859 	}
1860 
1861 	/*
1862 	 * out code jumps -> here, mrep is also freed.
1863 	 */
1864 
1865 	nfsm_reqdone;
1866 
1867 	/*
1868 	 * If we get an EEXIST error, silently convert it to no-error
1869 	 * in case of an NFS retry.
1870 	 */
1871 	if (error == EEXIST)
1872 		error = 0;
1873 
1874 	/*
1875 	 * If we do not have (or no longer have) an error, and we could
1876 	 * not extract the newvp from the response due to the request being
1877 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
1878 	 * to obtain a newvp to return.
1879 	 */
1880 	if (error == 0 && newvp == NULL) {
1881 		struct nfsnode *np = NULL;
1882 
1883 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1884 		    cnp->cn_cred, cnp->cn_proc, &np);
1885 		if (!error)
1886 			newvp = NFSTOV(np);
1887 	}
1888 	if (error) {
1889 		if (newvp)
1890 			vput(newvp);
1891 	} else {
1892 		*ap->a_vpp = newvp;
1893 	}
1894 	VTONFS(dvp)->n_flag |= NMODIFIED;
1895 	if (!wccflag)
1896 		VTONFS(dvp)->n_attrstamp = 0;
1897 	return (error);
1898 }
1899 
1900 /*
1901  * nfs make dir call
1902  */
1903 static int
1904 nfs_mkdir(ap)
1905 	struct vop_mkdir_args /* {
1906 		struct vnode *a_dvp;
1907 		struct vnode **a_vpp;
1908 		struct componentname *a_cnp;
1909 		struct vattr *a_vap;
1910 	} */ *ap;
1911 {
1912 	register struct vnode *dvp = ap->a_dvp;
1913 	register struct vattr *vap = ap->a_vap;
1914 	register struct componentname *cnp = ap->a_cnp;
1915 	register struct nfsv2_sattr *sp;
1916 	register u_int32_t *tl;
1917 	register caddr_t cp;
1918 	register int32_t t1, t2;
1919 	register int len;
1920 	struct nfsnode *np = (struct nfsnode *)0;
1921 	struct vnode *newvp = (struct vnode *)0;
1922 	caddr_t bpos, dpos, cp2;
1923 	int error = 0, wccflag = NFSV3_WCCRATTR;
1924 	int gotvp = 0;
1925 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1926 	struct vattr vattr;
1927 	int v3 = NFS_ISV3(dvp);
1928 
1929 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_proc)) != 0) {
1930 		return (error);
1931 	}
1932 	len = cnp->cn_namelen;
1933 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1934 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1935 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1936 	nfsm_fhtom(dvp, v3);
1937 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1938 	if (v3) {
1939 		nfsm_v3attrbuild(vap, FALSE);
1940 	} else {
1941 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1942 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1943 		sp->sa_uid = nfs_xdrneg1;
1944 		sp->sa_gid = nfs_xdrneg1;
1945 		sp->sa_size = nfs_xdrneg1;
1946 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1947 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1948 	}
1949 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1950 	if (!error)
1951 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1952 	if (v3)
1953 		nfsm_wcc_data(dvp, wccflag);
1954 	nfsm_reqdone;
1955 	VTONFS(dvp)->n_flag |= NMODIFIED;
1956 	if (!wccflag)
1957 		VTONFS(dvp)->n_attrstamp = 0;
1958 	/*
1959 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1960 	 * if we can succeed in looking up the directory.
1961 	 */
1962 	if (error == EEXIST || (!error && !gotvp)) {
1963 		if (newvp) {
1964 			vrele(newvp);
1965 			newvp = (struct vnode *)0;
1966 		}
1967 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1968 			cnp->cn_proc, &np);
1969 		if (!error) {
1970 			newvp = NFSTOV(np);
1971 			if (newvp->v_type != VDIR)
1972 				error = EEXIST;
1973 		}
1974 	}
1975 	if (error) {
1976 		if (newvp)
1977 			vrele(newvp);
1978 	} else
1979 		*ap->a_vpp = newvp;
1980 	return (error);
1981 }
1982 
1983 /*
1984  * nfs remove directory call
1985  */
1986 static int
1987 nfs_rmdir(ap)
1988 	struct vop_rmdir_args /* {
1989 		struct vnode *a_dvp;
1990 		struct vnode *a_vp;
1991 		struct componentname *a_cnp;
1992 	} */ *ap;
1993 {
1994 	register struct vnode *vp = ap->a_vp;
1995 	register struct vnode *dvp = ap->a_dvp;
1996 	register struct componentname *cnp = ap->a_cnp;
1997 	register u_int32_t *tl;
1998 	register caddr_t cp;
1999 	register int32_t t1, t2;
2000 	caddr_t bpos, dpos, cp2;
2001 	int error = 0, wccflag = NFSV3_WCCRATTR;
2002 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2003 	int v3 = NFS_ISV3(dvp);
2004 
2005 	if (dvp == vp)
2006 		return (EINVAL);
2007 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2008 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
2009 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2010 	nfsm_fhtom(dvp, v3);
2011 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2012 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2013 	if (v3)
2014 		nfsm_wcc_data(dvp, wccflag);
2015 	nfsm_reqdone;
2016 	VTONFS(dvp)->n_flag |= NMODIFIED;
2017 	if (!wccflag)
2018 		VTONFS(dvp)->n_attrstamp = 0;
2019 	cache_purge(dvp);
2020 	cache_purge(vp);
2021 	/*
2022 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2023 	 */
2024 	if (error == ENOENT)
2025 		error = 0;
2026 	return (error);
2027 }
2028 
2029 /*
2030  * nfs readdir call
2031  */
2032 static int
2033 nfs_readdir(ap)
2034 	struct vop_readdir_args /* {
2035 		struct vnode *a_vp;
2036 		struct uio *a_uio;
2037 		struct ucred *a_cred;
2038 	} */ *ap;
2039 {
2040 	register struct vnode *vp = ap->a_vp;
2041 	register struct nfsnode *np = VTONFS(vp);
2042 	register struct uio *uio = ap->a_uio;
2043 	int tresid, error;
2044 	struct vattr vattr;
2045 
2046 	if (vp->v_type != VDIR)
2047 		return (EPERM);
2048 	/*
2049 	 * First, check for hit on the EOF offset cache
2050 	 */
2051 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2052 	    (np->n_flag & NMODIFIED) == 0) {
2053 		if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
2054 			if (NQNFS_CKCACHABLE(vp, ND_READ)) {
2055 				nfsstats.direofcache_hits++;
2056 				return (0);
2057 			}
2058 		} else if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_procp) == 0 &&
2059 			np->n_mtime == vattr.va_mtime.tv_sec) {
2060 			nfsstats.direofcache_hits++;
2061 			return (0);
2062 		}
2063 	}
2064 
2065 	/*
2066 	 * Call nfs_bioread() to do the real work.
2067 	 */
2068 	tresid = uio->uio_resid;
2069 	error = nfs_bioread(vp, uio, 0, ap->a_cred);
2070 
2071 	if (!error && uio->uio_resid == tresid)
2072 		nfsstats.direofcache_misses++;
2073 	return (error);
2074 }
2075 
2076 /*
2077  * Readdir rpc call.
2078  * Called from below the buffer cache by nfs_doio().
2079  */
2080 int
2081 nfs_readdirrpc(vp, uiop, cred)
2082 	struct vnode *vp;
2083 	register struct uio *uiop;
2084 	struct ucred *cred;
2085 
2086 {
2087 	register int len, left;
2088 	register struct dirent *dp = NULL;
2089 	register u_int32_t *tl;
2090 	register caddr_t cp;
2091 	register int32_t t1, t2;
2092 	register nfsuint64 *cookiep;
2093 	caddr_t bpos, dpos, cp2;
2094 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2095 	nfsuint64 cookie;
2096 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2097 	struct nfsnode *dnp = VTONFS(vp);
2098 	u_quad_t fileno;
2099 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2100 	int attrflag;
2101 	int v3 = NFS_ISV3(vp);
2102 
2103 #ifndef DIAGNOSTIC
2104 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2105 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2106 		panic("nfs readdirrpc bad uio");
2107 #endif
2108 
2109 	/*
2110 	 * If there is no cookie, assume directory was stale.
2111 	 */
2112 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2113 	if (cookiep)
2114 		cookie = *cookiep;
2115 	else
2116 		return (NFSERR_BAD_COOKIE);
2117 	/*
2118 	 * Loop around doing readdir rpc's of size nm_readdirsize
2119 	 * truncated to a multiple of DIRBLKSIZ.
2120 	 * The stopping criteria is EOF or buffer full.
2121 	 */
2122 	while (more_dirs && bigenough) {
2123 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2124 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2125 			NFSX_READDIR(v3));
2126 		nfsm_fhtom(vp, v3);
2127 		if (v3) {
2128 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2129 			*tl++ = cookie.nfsuquad[0];
2130 			*tl++ = cookie.nfsuquad[1];
2131 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2132 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2133 		} else {
2134 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2135 			*tl++ = cookie.nfsuquad[0];
2136 		}
2137 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2138 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2139 		if (v3) {
2140 			nfsm_postop_attr(vp, attrflag);
2141 			if (!error) {
2142 				nfsm_dissect(tl, u_int32_t *,
2143 				    2 * NFSX_UNSIGNED);
2144 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2145 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2146 			} else {
2147 				m_freem(mrep);
2148 				goto nfsmout;
2149 			}
2150 		}
2151 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2152 		more_dirs = fxdr_unsigned(int, *tl);
2153 
2154 		/* loop thru the dir entries, doctoring them to 4bsd form */
2155 		while (more_dirs && bigenough) {
2156 			if (v3) {
2157 				nfsm_dissect(tl, u_int32_t *,
2158 				    3 * NFSX_UNSIGNED);
2159 				fileno = fxdr_hyper(tl);
2160 				len = fxdr_unsigned(int, *(tl + 2));
2161 			} else {
2162 				nfsm_dissect(tl, u_int32_t *,
2163 				    2 * NFSX_UNSIGNED);
2164 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2165 				len = fxdr_unsigned(int, *tl);
2166 			}
2167 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2168 				error = EBADRPC;
2169 				m_freem(mrep);
2170 				goto nfsmout;
2171 			}
2172 			tlen = nfsm_rndup(len);
2173 			if (tlen == len)
2174 				tlen += 4;	/* To ensure null termination */
2175 			left = DIRBLKSIZ - blksiz;
2176 			if ((tlen + DIRHDSIZ) > left) {
2177 				dp->d_reclen += left;
2178 				uiop->uio_iov->iov_base += left;
2179 				uiop->uio_iov->iov_len -= left;
2180 				uiop->uio_offset += left;
2181 				uiop->uio_resid -= left;
2182 				blksiz = 0;
2183 			}
2184 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2185 				bigenough = 0;
2186 			if (bigenough) {
2187 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2188 				dp->d_fileno = (int)fileno;
2189 				dp->d_namlen = len;
2190 				dp->d_reclen = tlen + DIRHDSIZ;
2191 				dp->d_type = DT_UNKNOWN;
2192 				blksiz += dp->d_reclen;
2193 				if (blksiz == DIRBLKSIZ)
2194 					blksiz = 0;
2195 				uiop->uio_offset += DIRHDSIZ;
2196 				uiop->uio_resid -= DIRHDSIZ;
2197 				uiop->uio_iov->iov_base += DIRHDSIZ;
2198 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2199 				nfsm_mtouio(uiop, len);
2200 				cp = uiop->uio_iov->iov_base;
2201 				tlen -= len;
2202 				*cp = '\0';	/* null terminate */
2203 				uiop->uio_iov->iov_base += tlen;
2204 				uiop->uio_iov->iov_len -= tlen;
2205 				uiop->uio_offset += tlen;
2206 				uiop->uio_resid -= tlen;
2207 			} else
2208 				nfsm_adv(nfsm_rndup(len));
2209 			if (v3) {
2210 				nfsm_dissect(tl, u_int32_t *,
2211 				    3 * NFSX_UNSIGNED);
2212 			} else {
2213 				nfsm_dissect(tl, u_int32_t *,
2214 				    2 * NFSX_UNSIGNED);
2215 			}
2216 			if (bigenough) {
2217 				cookie.nfsuquad[0] = *tl++;
2218 				if (v3)
2219 					cookie.nfsuquad[1] = *tl++;
2220 			} else if (v3)
2221 				tl += 2;
2222 			else
2223 				tl++;
2224 			more_dirs = fxdr_unsigned(int, *tl);
2225 		}
2226 		/*
2227 		 * If at end of rpc data, get the eof boolean
2228 		 */
2229 		if (!more_dirs) {
2230 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2231 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2232 		}
2233 		m_freem(mrep);
2234 	}
2235 	/*
2236 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2237 	 * by increasing d_reclen for the last record.
2238 	 */
2239 	if (blksiz > 0) {
2240 		left = DIRBLKSIZ - blksiz;
2241 		dp->d_reclen += left;
2242 		uiop->uio_iov->iov_base += left;
2243 		uiop->uio_iov->iov_len -= left;
2244 		uiop->uio_offset += left;
2245 		uiop->uio_resid -= left;
2246 	}
2247 
2248 	/*
2249 	 * We are now either at the end of the directory or have filled the
2250 	 * block.
2251 	 */
2252 	if (bigenough)
2253 		dnp->n_direofoffset = uiop->uio_offset;
2254 	else {
2255 		if (uiop->uio_resid > 0)
2256 			printf("EEK! readdirrpc resid > 0\n");
2257 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2258 		*cookiep = cookie;
2259 	}
2260 nfsmout:
2261 	return (error);
2262 }
2263 
2264 /*
2265  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2266  */
2267 int
2268 nfs_readdirplusrpc(vp, uiop, cred)
2269 	struct vnode *vp;
2270 	register struct uio *uiop;
2271 	struct ucred *cred;
2272 {
2273 	register int len, left;
2274 	register struct dirent *dp;
2275 	register u_int32_t *tl;
2276 	register caddr_t cp;
2277 	register int32_t t1, t2;
2278 	register struct vnode *newvp;
2279 	register nfsuint64 *cookiep;
2280 	caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2281 	struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2282 	struct nameidata nami, *ndp = &nami;
2283 	struct componentname *cnp = &ndp->ni_cnd;
2284 	nfsuint64 cookie;
2285 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2286 	struct nfsnode *dnp = VTONFS(vp), *np;
2287 	nfsfh_t *fhp;
2288 	u_quad_t fileno;
2289 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2290 	int attrflag, fhsize;
2291 
2292 #ifndef nolint
2293 	dp = (struct dirent *)0;
2294 #endif
2295 #ifndef DIAGNOSTIC
2296 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2297 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2298 		panic("nfs readdirplusrpc bad uio");
2299 #endif
2300 	ndp->ni_dvp = vp;
2301 	newvp = NULLVP;
2302 
2303 	/*
2304 	 * If there is no cookie, assume directory was stale.
2305 	 */
2306 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2307 	if (cookiep)
2308 		cookie = *cookiep;
2309 	else
2310 		return (NFSERR_BAD_COOKIE);
2311 	/*
2312 	 * Loop around doing readdir rpc's of size nm_readdirsize
2313 	 * truncated to a multiple of DIRBLKSIZ.
2314 	 * The stopping criteria is EOF or buffer full.
2315 	 */
2316 	while (more_dirs && bigenough) {
2317 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2318 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2319 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2320 		nfsm_fhtom(vp, 1);
2321  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2322 		*tl++ = cookie.nfsuquad[0];
2323 		*tl++ = cookie.nfsuquad[1];
2324 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2325 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2326 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2327 		*tl = txdr_unsigned(nmp->nm_rsize);
2328 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2329 		nfsm_postop_attr(vp, attrflag);
2330 		if (error) {
2331 			m_freem(mrep);
2332 			goto nfsmout;
2333 		}
2334 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2335 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2336 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2337 		more_dirs = fxdr_unsigned(int, *tl);
2338 
2339 		/* loop thru the dir entries, doctoring them to 4bsd form */
2340 		while (more_dirs && bigenough) {
2341 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2342 			fileno = fxdr_hyper(tl);
2343 			len = fxdr_unsigned(int, *(tl + 2));
2344 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2345 				error = EBADRPC;
2346 				m_freem(mrep);
2347 				goto nfsmout;
2348 			}
2349 			tlen = nfsm_rndup(len);
2350 			if (tlen == len)
2351 				tlen += 4;	/* To ensure null termination*/
2352 			left = DIRBLKSIZ - blksiz;
2353 			if ((tlen + DIRHDSIZ) > left) {
2354 				dp->d_reclen += left;
2355 				uiop->uio_iov->iov_base += left;
2356 				uiop->uio_iov->iov_len -= left;
2357 				uiop->uio_offset += left;
2358 				uiop->uio_resid -= left;
2359 				blksiz = 0;
2360 			}
2361 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2362 				bigenough = 0;
2363 			if (bigenough) {
2364 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2365 				dp->d_fileno = (int)fileno;
2366 				dp->d_namlen = len;
2367 				dp->d_reclen = tlen + DIRHDSIZ;
2368 				dp->d_type = DT_UNKNOWN;
2369 				blksiz += dp->d_reclen;
2370 				if (blksiz == DIRBLKSIZ)
2371 					blksiz = 0;
2372 				uiop->uio_offset += DIRHDSIZ;
2373 				uiop->uio_resid -= DIRHDSIZ;
2374 				uiop->uio_iov->iov_base += DIRHDSIZ;
2375 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2376 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2377 				cnp->cn_namelen = len;
2378 				nfsm_mtouio(uiop, len);
2379 				cp = uiop->uio_iov->iov_base;
2380 				tlen -= len;
2381 				*cp = '\0';
2382 				uiop->uio_iov->iov_base += tlen;
2383 				uiop->uio_iov->iov_len -= tlen;
2384 				uiop->uio_offset += tlen;
2385 				uiop->uio_resid -= tlen;
2386 			} else
2387 				nfsm_adv(nfsm_rndup(len));
2388 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2389 			if (bigenough) {
2390 				cookie.nfsuquad[0] = *tl++;
2391 				cookie.nfsuquad[1] = *tl++;
2392 			} else
2393 				tl += 2;
2394 
2395 			/*
2396 			 * Since the attributes are before the file handle
2397 			 * (sigh), we must skip over the attributes and then
2398 			 * come back and get them.
2399 			 */
2400 			attrflag = fxdr_unsigned(int, *tl);
2401 			if (attrflag) {
2402 			    dpossav1 = dpos;
2403 			    mdsav1 = md;
2404 			    nfsm_adv(NFSX_V3FATTR);
2405 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2406 			    doit = fxdr_unsigned(int, *tl);
2407 			    if (doit) {
2408 				nfsm_getfh(fhp, fhsize, 1);
2409 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2410 				    VREF(vp);
2411 				    newvp = vp;
2412 				    np = dnp;
2413 				} else {
2414 				    error = nfs_nget(vp->v_mount, fhp,
2415 					fhsize, &np);
2416 				    if (error)
2417 					doit = 0;
2418 				    else
2419 					newvp = NFSTOV(np);
2420 				}
2421 			    }
2422 			    if (doit && bigenough) {
2423 				dpossav2 = dpos;
2424 				dpos = dpossav1;
2425 				mdsav2 = md;
2426 				md = mdsav1;
2427 				nfsm_loadattr(newvp, (struct vattr *)0);
2428 				dpos = dpossav2;
2429 				md = mdsav2;
2430 				dp->d_type =
2431 				    IFTODT(VTTOIF(np->n_vattr.va_type));
2432 				ndp->ni_vp = newvp;
2433 			        cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2434 			    }
2435 			} else {
2436 			    /* Just skip over the file handle */
2437 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2438 			    i = fxdr_unsigned(int, *tl);
2439 			    nfsm_adv(nfsm_rndup(i));
2440 			}
2441 			if (newvp != NULLVP) {
2442 			    if (newvp == vp)
2443 				vrele(newvp);
2444 			    else
2445 				vput(newvp);
2446 			    newvp = NULLVP;
2447 			}
2448 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2449 			more_dirs = fxdr_unsigned(int, *tl);
2450 		}
2451 		/*
2452 		 * If at end of rpc data, get the eof boolean
2453 		 */
2454 		if (!more_dirs) {
2455 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2456 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2457 		}
2458 		m_freem(mrep);
2459 	}
2460 	/*
2461 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2462 	 * by increasing d_reclen for the last record.
2463 	 */
2464 	if (blksiz > 0) {
2465 		left = DIRBLKSIZ - blksiz;
2466 		dp->d_reclen += left;
2467 		uiop->uio_iov->iov_base += left;
2468 		uiop->uio_iov->iov_len -= left;
2469 		uiop->uio_offset += left;
2470 		uiop->uio_resid -= left;
2471 	}
2472 
2473 	/*
2474 	 * We are now either at the end of the directory or have filled the
2475 	 * block.
2476 	 */
2477 	if (bigenough)
2478 		dnp->n_direofoffset = uiop->uio_offset;
2479 	else {
2480 		if (uiop->uio_resid > 0)
2481 			printf("EEK! readdirplusrpc resid > 0\n");
2482 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2483 		*cookiep = cookie;
2484 	}
2485 nfsmout:
2486 	if (newvp != NULLVP) {
2487 	        if (newvp == vp)
2488 			vrele(newvp);
2489 		else
2490 			vput(newvp);
2491 		newvp = NULLVP;
2492 	}
2493 	return (error);
2494 }
2495 
2496 /*
2497  * Silly rename. To make the NFS filesystem that is stateless look a little
2498  * more like the "ufs" a remove of an active vnode is translated to a rename
2499  * to a funny looking filename that is removed by nfs_inactive on the
2500  * nfsnode. There is the potential for another process on a different client
2501  * to create the same funny name between the nfs_lookitup() fails and the
2502  * nfs_rename() completes, but...
2503  */
2504 static int
2505 nfs_sillyrename(dvp, vp, cnp)
2506 	struct vnode *dvp, *vp;
2507 	struct componentname *cnp;
2508 {
2509 	register struct sillyrename *sp;
2510 	struct nfsnode *np;
2511 	int error;
2512 	short pid;
2513 
2514 	cache_purge(dvp);
2515 	np = VTONFS(vp);
2516 #ifndef DIAGNOSTIC
2517 	if (vp->v_type == VDIR)
2518 		panic("nfs: sillyrename dir");
2519 #endif
2520 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2521 		M_NFSREQ, M_WAITOK);
2522 	sp->s_cred = crdup(cnp->cn_cred);
2523 	sp->s_dvp = dvp;
2524 	VREF(dvp);
2525 
2526 	/* Fudge together a funny name */
2527 	pid = cnp->cn_proc->p_pid;
2528 	sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
2529 
2530 	/* Try lookitups until we get one that isn't there */
2531 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2532 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2533 		sp->s_name[4]++;
2534 		if (sp->s_name[4] > 'z') {
2535 			error = EINVAL;
2536 			goto bad;
2537 		}
2538 	}
2539 	error = nfs_renameit(dvp, cnp, sp);
2540 	if (error)
2541 		goto bad;
2542 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2543 		cnp->cn_proc, &np);
2544 	np->n_sillyrename = sp;
2545 	return (0);
2546 bad:
2547 	vrele(sp->s_dvp);
2548 	crfree(sp->s_cred);
2549 	free((caddr_t)sp, M_NFSREQ);
2550 	return (error);
2551 }
2552 
2553 /*
2554  * Look up a file name and optionally either update the file handle or
2555  * allocate an nfsnode, depending on the value of npp.
2556  * npp == NULL	--> just do the lookup
2557  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2558  *			handled too
2559  * *npp != NULL --> update the file handle in the vnode
2560  */
2561 static int
2562 nfs_lookitup(dvp, name, len, cred, procp, npp)
2563 	register struct vnode *dvp;
2564 	const char *name;
2565 	int len;
2566 	struct ucred *cred;
2567 	struct proc *procp;
2568 	struct nfsnode **npp;
2569 {
2570 	register u_int32_t *tl;
2571 	register caddr_t cp;
2572 	register int32_t t1, t2;
2573 	struct vnode *newvp = (struct vnode *)0;
2574 	struct nfsnode *np, *dnp = VTONFS(dvp);
2575 	caddr_t bpos, dpos, cp2;
2576 	int error = 0, fhlen, attrflag;
2577 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2578 	nfsfh_t *nfhp;
2579 	int v3 = NFS_ISV3(dvp);
2580 
2581 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2582 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2583 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2584 	nfsm_fhtom(dvp, v3);
2585 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2586 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2587 	if (npp && !error) {
2588 		nfsm_getfh(nfhp, fhlen, v3);
2589 		if (*npp) {
2590 		    np = *npp;
2591 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2592 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2593 			np->n_fhp = &np->n_fh;
2594 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2595 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2596 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2597 		    np->n_fhsize = fhlen;
2598 		    newvp = NFSTOV(np);
2599 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2600 		    VREF(dvp);
2601 		    newvp = dvp;
2602 		} else {
2603 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2604 		    if (error) {
2605 			m_freem(mrep);
2606 			return (error);
2607 		    }
2608 		    newvp = NFSTOV(np);
2609 		}
2610 		if (v3) {
2611 			nfsm_postop_attr(newvp, attrflag);
2612 			if (!attrflag && *npp == NULL) {
2613 				m_freem(mrep);
2614 				if (newvp == dvp)
2615 					vrele(newvp);
2616 				else
2617 					vput(newvp);
2618 				return (ENOENT);
2619 			}
2620 		} else
2621 			nfsm_loadattr(newvp, (struct vattr *)0);
2622 	}
2623 	nfsm_reqdone;
2624 	if (npp && *npp == NULL) {
2625 		if (error) {
2626 			if (newvp) {
2627 				if (newvp == dvp)
2628 					vrele(newvp);
2629 				else
2630 					vput(newvp);
2631 			}
2632 		} else
2633 			*npp = np;
2634 	}
2635 	return (error);
2636 }
2637 
2638 /*
2639  * Nfs Version 3 commit rpc
2640  */
2641 int
2642 nfs_commit(vp, offset, cnt, cred, procp)
2643 	struct vnode *vp;
2644 	u_quad_t offset;
2645 	int cnt;
2646 	struct ucred *cred;
2647 	struct proc *procp;
2648 {
2649 	register caddr_t cp;
2650 	register u_int32_t *tl;
2651 	register int32_t t1, t2;
2652 	register struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2653 	caddr_t bpos, dpos, cp2;
2654 	int error = 0, wccflag = NFSV3_WCCRATTR;
2655 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2656 
2657 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2658 		return (0);
2659 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2660 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2661 	nfsm_fhtom(vp, 1);
2662 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2663 	txdr_hyper(offset, tl);
2664 	tl += 2;
2665 	*tl = txdr_unsigned(cnt);
2666 	nfsm_request(vp, NFSPROC_COMMIT, procp, cred);
2667 	nfsm_wcc_data(vp, wccflag);
2668 	if (!error) {
2669 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2670 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2671 			NFSX_V3WRITEVERF)) {
2672 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2673 				NFSX_V3WRITEVERF);
2674 			error = NFSERR_STALEWRITEVERF;
2675 		}
2676 	}
2677 	nfsm_reqdone;
2678 	return (error);
2679 }
2680 
2681 /*
2682  * Kludge City..
2683  * - make nfs_bmap() essentially a no-op that does no translation
2684  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2685  *   (Maybe I could use the process's page mapping, but I was concerned that
2686  *    Kernel Write might not be enabled and also figured copyout() would do
2687  *    a lot more work than bcopy() and also it currently happens in the
2688  *    context of the swapper process (2).
2689  */
2690 static int
2691 nfs_bmap(ap)
2692 	struct vop_bmap_args /* {
2693 		struct vnode *a_vp;
2694 		daddr_t  a_bn;
2695 		struct vnode **a_vpp;
2696 		daddr_t *a_bnp;
2697 		int *a_runp;
2698 		int *a_runb;
2699 	} */ *ap;
2700 {
2701 	register struct vnode *vp = ap->a_vp;
2702 
2703 	if (ap->a_vpp != NULL)
2704 		*ap->a_vpp = vp;
2705 	if (ap->a_bnp != NULL)
2706 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2707 	if (ap->a_runp != NULL)
2708 		*ap->a_runp = 0;
2709 	if (ap->a_runb != NULL)
2710 		*ap->a_runb = 0;
2711 	return (0);
2712 }
2713 
2714 /*
2715  * Strategy routine.
2716  * For async requests when nfsiod(s) are running, queue the request by
2717  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2718  * request.
2719  */
2720 static int
2721 nfs_strategy(ap)
2722 	struct vop_strategy_args *ap;
2723 {
2724 	register struct buf *bp = ap->a_bp;
2725 	struct ucred *cr;
2726 	struct proc *p;
2727 	int error = 0;
2728 
2729 	KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2730 	KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2731 
2732 	if (bp->b_flags & B_PHYS)
2733 		panic("nfs physio");
2734 
2735 	if (bp->b_flags & B_ASYNC)
2736 		p = (struct proc *)0;
2737 	else
2738 		p = curproc;	/* XXX */
2739 
2740 	if (bp->b_flags & B_READ)
2741 		cr = bp->b_rcred;
2742 	else
2743 		cr = bp->b_wcred;
2744 
2745 	/*
2746 	 * If the op is asynchronous and an i/o daemon is waiting
2747 	 * queue the request, wake it up and wait for completion
2748 	 * otherwise just do it ourselves.
2749 	 */
2750 	if ((bp->b_flags & B_ASYNC) == 0 ||
2751 		nfs_asyncio(bp, NOCRED, p))
2752 		error = nfs_doio(bp, cr, p);
2753 	return (error);
2754 }
2755 
2756 /*
2757  * Mmap a file
2758  *
2759  * NB Currently unsupported.
2760  */
2761 /* ARGSUSED */
2762 static int
2763 nfs_mmap(ap)
2764 	struct vop_mmap_args /* {
2765 		struct vnode *a_vp;
2766 		int  a_fflags;
2767 		struct ucred *a_cred;
2768 		struct proc *a_p;
2769 	} */ *ap;
2770 {
2771 
2772 	return (EINVAL);
2773 }
2774 
2775 /*
2776  * fsync vnode op. Just call nfs_flush() with commit == 1.
2777  */
2778 /* ARGSUSED */
2779 static int
2780 nfs_fsync(ap)
2781 	struct vop_fsync_args /* {
2782 		struct vnodeop_desc *a_desc;
2783 		struct vnode * a_vp;
2784 		struct ucred * a_cred;
2785 		int  a_waitfor;
2786 		struct proc * a_p;
2787 	} */ *ap;
2788 {
2789 
2790 	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2791 }
2792 
2793 /*
2794  * Flush all the blocks associated with a vnode.
2795  * 	Walk through the buffer pool and push any dirty pages
2796  *	associated with the vnode.
2797  */
2798 static int
2799 nfs_flush(vp, cred, waitfor, p, commit)
2800 	register struct vnode *vp;
2801 	struct ucred *cred;
2802 	int waitfor;
2803 	struct proc *p;
2804 	int commit;
2805 {
2806 	register struct nfsnode *np = VTONFS(vp);
2807 	register struct buf *bp;
2808 	register int i;
2809 	struct buf *nbp;
2810 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2811 	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2812 	int passone = 1;
2813 	u_quad_t off, endoff, toff;
2814 	struct ucred* wcred = NULL;
2815 	struct buf **bvec = NULL;
2816 #ifndef NFS_COMMITBVECSIZ
2817 #define NFS_COMMITBVECSIZ	20
2818 #endif
2819 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2820 	int bvecsize = 0, bveccount;
2821 
2822 	if (nmp->nm_flag & NFSMNT_INT)
2823 		slpflag = PCATCH;
2824 	if (!commit)
2825 		passone = 0;
2826 	/*
2827 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2828 	 * server, but nas not been committed to stable storage on the server
2829 	 * yet. On the first pass, the byte range is worked out and the commit
2830 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2831 	 * job.
2832 	 */
2833 again:
2834 	off = (u_quad_t)-1;
2835 	endoff = 0;
2836 	bvecpos = 0;
2837 	if (NFS_ISV3(vp) && commit) {
2838 		s = splbio();
2839 		/*
2840 		 * Count up how many buffers waiting for a commit.
2841 		 */
2842 		bveccount = 0;
2843 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2844 			nbp = TAILQ_NEXT(bp, b_vnbufs);
2845 			if (BUF_REFCNT(bp) == 0 &&
2846 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2847 				== (B_DELWRI | B_NEEDCOMMIT))
2848 				bveccount++;
2849 		}
2850 		/*
2851 		 * Allocate space to remember the list of bufs to commit.  It is
2852 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2853 		 * If we can't get memory (for whatever reason), we will end up
2854 		 * committing the buffers one-by-one in the loop below.
2855 		 */
2856 		if (bvec != NULL && bvec != bvec_on_stack)
2857 			free(bvec, M_TEMP);
2858 		if (bveccount > NFS_COMMITBVECSIZ) {
2859 			bvec = (struct buf **)
2860 				malloc(bveccount * sizeof(struct buf *),
2861 				       M_TEMP, M_NOWAIT);
2862 			if (bvec == NULL) {
2863 				bvec = bvec_on_stack;
2864 				bvecsize = NFS_COMMITBVECSIZ;
2865 			} else
2866 				bvecsize = bveccount;
2867 		} else {
2868 			bvec = bvec_on_stack;
2869 			bvecsize = NFS_COMMITBVECSIZ;
2870 		}
2871 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2872 			nbp = TAILQ_NEXT(bp, b_vnbufs);
2873 			if (bvecpos >= bvecsize)
2874 				break;
2875 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2876 			    (B_DELWRI | B_NEEDCOMMIT) ||
2877 			    BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
2878 				continue;
2879 			bremfree(bp);
2880 			/*
2881 			 * Work out if all buffers are using the same cred
2882 			 * so we can deal with them all with one commit.
2883 			 *
2884 			 * NOTE: we are not clearing B_DONE here, so we have
2885 			 * to do it later on in this routine if we intend to
2886 			 * initiate I/O on the bp.
2887 			 *
2888 			 * Note: to avoid loopback deadlocks, we do not
2889 			 * assign b_runningbufspace.
2890 			 */
2891 			if (wcred == NULL)
2892 				wcred = bp->b_wcred;
2893 			else if (wcred != bp->b_wcred)
2894 				wcred = NOCRED;
2895 			bp->b_flags |= B_WRITEINPROG;
2896 			vfs_busy_pages(bp, 1);
2897 
2898 			/*
2899 			 * bp is protected by being locked, but nbp is not
2900 			 * and vfs_busy_pages() may sleep.  We have to
2901 			 * recalculate nbp.
2902 			 */
2903 			nbp = TAILQ_NEXT(bp, b_vnbufs);
2904 
2905 			/*
2906 			 * A list of these buffers is kept so that the
2907 			 * second loop knows which buffers have actually
2908 			 * been committed. This is necessary, since there
2909 			 * may be a race between the commit rpc and new
2910 			 * uncommitted writes on the file.
2911 			 */
2912 			bvec[bvecpos++] = bp;
2913 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2914 				bp->b_dirtyoff;
2915 			if (toff < off)
2916 				off = toff;
2917 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2918 			if (toff > endoff)
2919 				endoff = toff;
2920 		}
2921 		splx(s);
2922 	}
2923 	if (bvecpos > 0) {
2924 		/*
2925 		 * Commit data on the server, as required.
2926 		 * If all bufs are using the same wcred, then use that with
2927 		 * one call for all of them, otherwise commit each one
2928 		 * separately.
2929 		 */
2930 		if (wcred != NOCRED)
2931 			retv = nfs_commit(vp, off, (int)(endoff - off),
2932 					  wcred, p);
2933 		else {
2934 			retv = 0;
2935 			for (i = 0; i < bvecpos; i++) {
2936 				off_t off, size;
2937 				bp = bvec[i];
2938 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2939 					bp->b_dirtyoff;
2940 				size = (u_quad_t)(bp->b_dirtyend
2941 						  - bp->b_dirtyoff);
2942 				retv = nfs_commit(vp, off, (int)size,
2943 						  bp->b_wcred, p);
2944 				if (retv) break;
2945 			}
2946 		}
2947 
2948 		if (retv == NFSERR_STALEWRITEVERF)
2949 			nfs_clearcommit(vp->v_mount);
2950 
2951 		/*
2952 		 * Now, either mark the blocks I/O done or mark the
2953 		 * blocks dirty, depending on whether the commit
2954 		 * succeeded.
2955 		 */
2956 		for (i = 0; i < bvecpos; i++) {
2957 			bp = bvec[i];
2958 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
2959 			if (retv) {
2960 				/*
2961 				 * Error, leave B_DELWRI intact
2962 				 */
2963 				vfs_unbusy_pages(bp);
2964 				brelse(bp);
2965 			} else {
2966 				/*
2967 				 * Success, remove B_DELWRI ( bundirty() ).
2968 				 *
2969 				 * b_dirtyoff/b_dirtyend seem to be NFS
2970 				 * specific.  We should probably move that
2971 				 * into bundirty(). XXX
2972 				 */
2973 				s = splbio();
2974 				vp->v_numoutput++;
2975 				bp->b_flags |= B_ASYNC;
2976 				bundirty(bp);
2977 				bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
2978 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2979 				splx(s);
2980 				biodone(bp);
2981 			}
2982 		}
2983 	}
2984 
2985 	/*
2986 	 * Start/do any write(s) that are required.
2987 	 */
2988 loop:
2989 	s = splbio();
2990 	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2991 		nbp = TAILQ_NEXT(bp, b_vnbufs);
2992 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
2993 			if (waitfor != MNT_WAIT || passone)
2994 				continue;
2995 			error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
2996 			    "nfsfsync", slpflag, slptimeo);
2997 			splx(s);
2998 			if (error == 0)
2999 				panic("nfs_fsync: inconsistent lock");
3000 			if (error == ENOLCK)
3001 				goto loop;
3002 			if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) {
3003 				error = EINTR;
3004 				goto done;
3005 			}
3006 			if (slpflag == PCATCH) {
3007 				slpflag = 0;
3008 				slptimeo = 2 * hz;
3009 			}
3010 			goto loop;
3011 		}
3012 		if ((bp->b_flags & B_DELWRI) == 0)
3013 			panic("nfs_fsync: not dirty");
3014 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3015 			BUF_UNLOCK(bp);
3016 			continue;
3017 		}
3018 		bremfree(bp);
3019 		if (passone || !commit)
3020 		    bp->b_flags |= B_ASYNC;
3021 		else
3022 		    bp->b_flags |= B_ASYNC | B_WRITEINPROG;
3023 		splx(s);
3024 		VOP_BWRITE(bp->b_vp, bp);
3025 		goto loop;
3026 	}
3027 	splx(s);
3028 	if (passone) {
3029 		passone = 0;
3030 		goto again;
3031 	}
3032 	if (waitfor == MNT_WAIT) {
3033 		while (vp->v_numoutput) {
3034 			vp->v_flag |= VBWAIT;
3035 			error = tsleep((caddr_t)&vp->v_numoutput,
3036 				slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
3037 			if (error) {
3038 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) {
3039 				error = EINTR;
3040 				goto done;
3041 			    }
3042 			    if (slpflag == PCATCH) {
3043 				slpflag = 0;
3044 				slptimeo = 2 * hz;
3045 			    }
3046 			}
3047 		}
3048 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
3049 			goto loop;
3050 		}
3051 	}
3052 	if (np->n_flag & NWRITEERR) {
3053 		error = np->n_error;
3054 		np->n_flag &= ~NWRITEERR;
3055 	}
3056 done:
3057 	if (bvec != NULL && bvec != bvec_on_stack)
3058 		free(bvec, M_TEMP);
3059 	return (error);
3060 }
3061 
3062 /*
3063  * NFS advisory byte-level locks.
3064  * Currently unsupported.
3065  */
3066 static int
3067 nfs_advlock(ap)
3068 	struct vop_advlock_args /* {
3069 		struct vnode *a_vp;
3070 		caddr_t  a_id;
3071 		int  a_op;
3072 		struct flock *a_fl;
3073 		int  a_flags;
3074 	} */ *ap;
3075 {
3076 	register struct nfsnode *np = VTONFS(ap->a_vp);
3077 
3078 	/*
3079 	 * The following kludge is to allow diskless support to work
3080 	 * until a real NFS lockd is implemented. Basically, just pretend
3081 	 * that this is a local lock.
3082 	 */
3083 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3084 }
3085 
3086 /*
3087  * Print out the contents of an nfsnode.
3088  */
3089 static int
3090 nfs_print(ap)
3091 	struct vop_print_args /* {
3092 		struct vnode *a_vp;
3093 	} */ *ap;
3094 {
3095 	register struct vnode *vp = ap->a_vp;
3096 	register struct nfsnode *np = VTONFS(vp);
3097 
3098 	printf("tag VT_NFS, fileid %ld fsid 0x%x",
3099 		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3100 	if (vp->v_type == VFIFO)
3101 		fifo_printinfo(vp);
3102 	printf("\n");
3103 	return (0);
3104 }
3105 
3106 /*
3107  * Just call nfs_writebp() with the force argument set to 1.
3108  *
3109  * NOTE: B_DONE may or may not be set in a_bp on call.
3110  */
3111 static int
3112 nfs_bwrite(ap)
3113 	struct vop_bwrite_args /* {
3114 		struct vnode *a_bp;
3115 	} */ *ap;
3116 {
3117 	return (nfs_writebp(ap->a_bp, 1, curproc));
3118 }
3119 
3120 /*
3121  * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3122  * the force flag is one and it also handles the B_NEEDCOMMIT flag.  We set
3123  * B_CACHE if this is a VMIO buffer.
3124  */
3125 int
3126 nfs_writebp(bp, force, procp)
3127 	register struct buf *bp;
3128 	int force;
3129 	struct proc *procp;
3130 {
3131 	int s;
3132 	int oldflags = bp->b_flags;
3133 #if 0
3134 	int retv = 1;
3135 	off_t off;
3136 #endif
3137 
3138 	if (BUF_REFCNT(bp) == 0)
3139 		panic("bwrite: buffer is not locked???");
3140 
3141 	if (bp->b_flags & B_INVAL) {
3142 		brelse(bp);
3143 		return(0);
3144 	}
3145 
3146 	bp->b_flags |= B_CACHE;
3147 
3148 	/*
3149 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3150 	 */
3151 
3152 	s = splbio();
3153 	bundirty(bp);
3154 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3155 
3156 	bp->b_vp->v_numoutput++;
3157 	curproc->p_stats->p_ru.ru_oublock++;
3158 	splx(s);
3159 
3160 	/*
3161 	 * Note: to avoid loopback deadlocks, we do not
3162 	 * assign b_runningbufspace.
3163 	 */
3164 	vfs_busy_pages(bp, 1);
3165 
3166 	if (force)
3167 		bp->b_flags |= B_WRITEINPROG;
3168 	BUF_KERNPROC(bp);
3169 	VOP_STRATEGY(bp->b_vp, bp);
3170 
3171 	if( (oldflags & B_ASYNC) == 0) {
3172 		int rtval = biowait(bp);
3173 
3174 		if (oldflags & B_DELWRI) {
3175 			s = splbio();
3176 			reassignbuf(bp, bp->b_vp);
3177 			splx(s);
3178 		}
3179 
3180 		brelse(bp);
3181 		return (rtval);
3182 	}
3183 
3184 	return (0);
3185 }
3186 
3187 /*
3188  * nfs special file access vnode op.
3189  * Essentially just get vattr and then imitate iaccess() since the device is
3190  * local to the client.
3191  */
3192 static int
3193 nfsspec_access(ap)
3194 	struct vop_access_args /* {
3195 		struct vnode *a_vp;
3196 		int  a_mode;
3197 		struct ucred *a_cred;
3198 		struct proc *a_p;
3199 	} */ *ap;
3200 {
3201 	register struct vattr *vap;
3202 	register gid_t *gp;
3203 	register struct ucred *cred = ap->a_cred;
3204 	struct vnode *vp = ap->a_vp;
3205 	mode_t mode = ap->a_mode;
3206 	struct vattr vattr;
3207 	register int i;
3208 	int error;
3209 
3210 	/*
3211 	 * Disallow write attempts on filesystems mounted read-only;
3212 	 * unless the file is a socket, fifo, or a block or character
3213 	 * device resident on the filesystem.
3214 	 */
3215 	if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3216 		switch (vp->v_type) {
3217 		case VREG:
3218 		case VDIR:
3219 		case VLNK:
3220 			return (EROFS);
3221 		default:
3222 			break;
3223 		}
3224 	}
3225 	/*
3226 	 * If you're the super-user,
3227 	 * you always get access.
3228 	 */
3229 	if (cred->cr_uid == 0)
3230 		return (0);
3231 	vap = &vattr;
3232 	error = VOP_GETATTR(vp, vap, cred, ap->a_p);
3233 	if (error)
3234 		return (error);
3235 	/*
3236 	 * Access check is based on only one of owner, group, public.
3237 	 * If not owner, then check group. If not a member of the
3238 	 * group, then check public access.
3239 	 */
3240 	if (cred->cr_uid != vap->va_uid) {
3241 		mode >>= 3;
3242 		gp = cred->cr_groups;
3243 		for (i = 0; i < cred->cr_ngroups; i++, gp++)
3244 			if (vap->va_gid == *gp)
3245 				goto found;
3246 		mode >>= 3;
3247 found:
3248 		;
3249 	}
3250 	error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3251 	return (error);
3252 }
3253 
3254 /*
3255  * Read wrapper for special devices.
3256  */
3257 static int
3258 nfsspec_read(ap)
3259 	struct vop_read_args /* {
3260 		struct vnode *a_vp;
3261 		struct uio *a_uio;
3262 		int  a_ioflag;
3263 		struct ucred *a_cred;
3264 	} */ *ap;
3265 {
3266 	register struct nfsnode *np = VTONFS(ap->a_vp);
3267 
3268 	/*
3269 	 * Set access flag.
3270 	 */
3271 	np->n_flag |= NACC;
3272 	getnanotime(&np->n_atim);
3273 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3274 }
3275 
3276 /*
3277  * Write wrapper for special devices.
3278  */
3279 static int
3280 nfsspec_write(ap)
3281 	struct vop_write_args /* {
3282 		struct vnode *a_vp;
3283 		struct uio *a_uio;
3284 		int  a_ioflag;
3285 		struct ucred *a_cred;
3286 	} */ *ap;
3287 {
3288 	register struct nfsnode *np = VTONFS(ap->a_vp);
3289 
3290 	/*
3291 	 * Set update flag.
3292 	 */
3293 	np->n_flag |= NUPD;
3294 	getnanotime(&np->n_mtim);
3295 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3296 }
3297 
3298 /*
3299  * Close wrapper for special devices.
3300  *
3301  * Update the times on the nfsnode then do device close.
3302  */
3303 static int
3304 nfsspec_close(ap)
3305 	struct vop_close_args /* {
3306 		struct vnode *a_vp;
3307 		int  a_fflag;
3308 		struct ucred *a_cred;
3309 		struct proc *a_p;
3310 	} */ *ap;
3311 {
3312 	register struct vnode *vp = ap->a_vp;
3313 	register struct nfsnode *np = VTONFS(vp);
3314 	struct vattr vattr;
3315 
3316 	if (np->n_flag & (NACC | NUPD)) {
3317 		np->n_flag |= NCHG;
3318 		if (vp->v_usecount == 1 &&
3319 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3320 			VATTR_NULL(&vattr);
3321 			if (np->n_flag & NACC)
3322 				vattr.va_atime = np->n_atim;
3323 			if (np->n_flag & NUPD)
3324 				vattr.va_mtime = np->n_mtim;
3325 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3326 		}
3327 	}
3328 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3329 }
3330 
3331 /*
3332  * Read wrapper for fifos.
3333  */
3334 static int
3335 nfsfifo_read(ap)
3336 	struct vop_read_args /* {
3337 		struct vnode *a_vp;
3338 		struct uio *a_uio;
3339 		int  a_ioflag;
3340 		struct ucred *a_cred;
3341 	} */ *ap;
3342 {
3343 	register struct nfsnode *np = VTONFS(ap->a_vp);
3344 
3345 	/*
3346 	 * Set access flag.
3347 	 */
3348 	np->n_flag |= NACC;
3349 	getnanotime(&np->n_atim);
3350 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3351 }
3352 
3353 /*
3354  * Write wrapper for fifos.
3355  */
3356 static int
3357 nfsfifo_write(ap)
3358 	struct vop_write_args /* {
3359 		struct vnode *a_vp;
3360 		struct uio *a_uio;
3361 		int  a_ioflag;
3362 		struct ucred *a_cred;
3363 	} */ *ap;
3364 {
3365 	register struct nfsnode *np = VTONFS(ap->a_vp);
3366 
3367 	/*
3368 	 * Set update flag.
3369 	 */
3370 	np->n_flag |= NUPD;
3371 	getnanotime(&np->n_mtim);
3372 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3373 }
3374 
3375 /*
3376  * Close wrapper for fifos.
3377  *
3378  * Update the times on the nfsnode then do fifo close.
3379  */
3380 static int
3381 nfsfifo_close(ap)
3382 	struct vop_close_args /* {
3383 		struct vnode *a_vp;
3384 		int  a_fflag;
3385 		struct ucred *a_cred;
3386 		struct proc *a_p;
3387 	} */ *ap;
3388 {
3389 	register struct vnode *vp = ap->a_vp;
3390 	register struct nfsnode *np = VTONFS(vp);
3391 	struct vattr vattr;
3392 	struct timespec ts;
3393 
3394 	if (np->n_flag & (NACC | NUPD)) {
3395 		getnanotime(&ts);
3396 		if (np->n_flag & NACC)
3397 			np->n_atim = ts;
3398 		if (np->n_flag & NUPD)
3399 			np->n_mtim = ts;
3400 		np->n_flag |= NCHG;
3401 		if (vp->v_usecount == 1 &&
3402 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3403 			VATTR_NULL(&vattr);
3404 			if (np->n_flag & NACC)
3405 				vattr.va_atime = np->n_atim;
3406 			if (np->n_flag & NUPD)
3407 				vattr.va_mtime = np->n_mtim;
3408 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3409 		}
3410 	}
3411 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3412 }
3413 
3414