xref: /dflybsd-src/sys/vfs/nfs/nfs_vnops.c (revision 2d8a3be70850f8655e3b289a16da332c46c17b92)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
37  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38  * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.14 2003/10/10 22:01:13 dillon Exp $
39  */
40 
41 
42 /*
43  * vnode op calls for Sun NFS version 2 and 3
44  */
45 
46 #include "opt_inet.h"
47 
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/conf.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_zone.h>
70 
71 #include <sys/buf2.h>
72 
73 #include <vfs/fifofs/fifo.h>
74 
75 #include "rpcv2.h"
76 #include "nfsproto.h"
77 #include "nfs.h"
78 #include "nfsmount.h"
79 #include "nfsnode.h"
80 #include "xdr_subs.h"
81 #include "nfsm_subs.h"
82 #include "nqnfs.h"
83 
84 #include <net/if.h>
85 #include <netinet/in.h>
86 #include <netinet/in_var.h>
87 
88 /* Defs */
89 #define	TRUE	1
90 #define	FALSE	0
91 
92 /*
93  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
94  * calls are not in getblk() and brelse() so that they would not be necessary
95  * here.
96  */
97 #ifndef B_VMIO
98 #define vfs_busy_pages(bp, f)
99 #endif
100 
101 static int	nfsspec_read (struct vop_read_args *);
102 static int	nfsspec_write (struct vop_write_args *);
103 static int	nfsfifo_read (struct vop_read_args *);
104 static int	nfsfifo_write (struct vop_write_args *);
105 static int	nfsspec_close (struct vop_close_args *);
106 static int	nfsfifo_close (struct vop_close_args *);
107 #define nfs_poll vop_nopoll
108 static int	nfs_flush (struct vnode *,int,struct thread *,int);
109 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
110 static	int	nfs_lookup (struct vop_lookup_args *);
111 static	int	nfs_create (struct vop_create_args *);
112 static	int	nfs_mknod (struct vop_mknod_args *);
113 static	int	nfs_open (struct vop_open_args *);
114 static	int	nfs_close (struct vop_close_args *);
115 static	int	nfs_access (struct vop_access_args *);
116 static	int	nfs_getattr (struct vop_getattr_args *);
117 static	int	nfs_setattr (struct vop_setattr_args *);
118 static	int	nfs_read (struct vop_read_args *);
119 static	int	nfs_mmap (struct vop_mmap_args *);
120 static	int	nfs_fsync (struct vop_fsync_args *);
121 static	int	nfs_remove (struct vop_remove_args *);
122 static	int	nfs_link (struct vop_link_args *);
123 static	int	nfs_rename (struct vop_rename_args *);
124 static	int	nfs_mkdir (struct vop_mkdir_args *);
125 static	int	nfs_rmdir (struct vop_rmdir_args *);
126 static	int	nfs_symlink (struct vop_symlink_args *);
127 static	int	nfs_readdir (struct vop_readdir_args *);
128 static	int	nfs_bmap (struct vop_bmap_args *);
129 static	int	nfs_strategy (struct vop_strategy_args *);
130 static	int	nfs_lookitup (struct vnode *, const char *, int,
131 			struct ucred *, struct thread *, struct nfsnode **);
132 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
133 static int	nfsspec_access (struct vop_access_args *);
134 static int	nfs_readlink (struct vop_readlink_args *);
135 static int	nfs_print (struct vop_print_args *);
136 static int	nfs_advlock (struct vop_advlock_args *);
137 static int	nfs_bwrite (struct vop_bwrite_args *);
138 /*
139  * Global vfs data structures for nfs
140  */
141 vop_t **nfsv2_vnodeop_p;
142 static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
143 	{ &vop_default_desc,		(vop_t *) vop_defaultop },
144 	{ &vop_access_desc,		(vop_t *) nfs_access },
145 	{ &vop_advlock_desc,		(vop_t *) nfs_advlock },
146 	{ &vop_bmap_desc,		(vop_t *) nfs_bmap },
147 	{ &vop_bwrite_desc,		(vop_t *) nfs_bwrite },
148 	{ &vop_close_desc,		(vop_t *) nfs_close },
149 	{ &vop_create_desc,		(vop_t *) nfs_create },
150 	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
151 	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
152 	{ &vop_getpages_desc,		(vop_t *) nfs_getpages },
153 	{ &vop_putpages_desc,		(vop_t *) nfs_putpages },
154 	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
155 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
156 	{ &vop_lease_desc,		(vop_t *) vop_null },
157 	{ &vop_link_desc,		(vop_t *) nfs_link },
158 	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
159 	{ &vop_lookup_desc,		(vop_t *) nfs_lookup },
160 	{ &vop_mkdir_desc,		(vop_t *) nfs_mkdir },
161 	{ &vop_mknod_desc,		(vop_t *) nfs_mknod },
162 	{ &vop_mmap_desc,		(vop_t *) nfs_mmap },
163 	{ &vop_open_desc,		(vop_t *) nfs_open },
164 	{ &vop_poll_desc,		(vop_t *) nfs_poll },
165 	{ &vop_print_desc,		(vop_t *) nfs_print },
166 	{ &vop_read_desc,		(vop_t *) nfs_read },
167 	{ &vop_readdir_desc,		(vop_t *) nfs_readdir },
168 	{ &vop_readlink_desc,		(vop_t *) nfs_readlink },
169 	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
170 	{ &vop_remove_desc,		(vop_t *) nfs_remove },
171 	{ &vop_rename_desc,		(vop_t *) nfs_rename },
172 	{ &vop_rmdir_desc,		(vop_t *) nfs_rmdir },
173 	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
174 	{ &vop_strategy_desc,		(vop_t *) nfs_strategy },
175 	{ &vop_symlink_desc,		(vop_t *) nfs_symlink },
176 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
177 	{ &vop_write_desc,		(vop_t *) nfs_write },
178 	{ NULL, NULL }
179 };
180 static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
181 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
182 VNODEOP_SET(nfsv2_vnodeop_opv_desc);
183 
184 /*
185  * Special device vnode ops
186  */
187 vop_t **spec_nfsv2nodeop_p;
188 static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
189 	{ &vop_default_desc,		(vop_t *) spec_vnoperate },
190 	{ &vop_access_desc,		(vop_t *) nfsspec_access },
191 	{ &vop_close_desc,		(vop_t *) nfsspec_close },
192 	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
193 	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
194 	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
195 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
196 	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
197 	{ &vop_print_desc,		(vop_t *) nfs_print },
198 	{ &vop_read_desc,		(vop_t *) nfsspec_read },
199 	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
200 	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
201 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
202 	{ &vop_write_desc,		(vop_t *) nfsspec_write },
203 	{ NULL, NULL }
204 };
205 static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
206 	{ &spec_nfsv2nodeop_p, nfsv2_specop_entries };
207 VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
208 
209 vop_t **fifo_nfsv2nodeop_p;
210 static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
211 	{ &vop_default_desc,		(vop_t *) fifo_vnoperate },
212 	{ &vop_access_desc,		(vop_t *) nfsspec_access },
213 	{ &vop_close_desc,		(vop_t *) nfsfifo_close },
214 	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
215 	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
216 	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
217 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
218 	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
219 	{ &vop_print_desc,		(vop_t *) nfs_print },
220 	{ &vop_read_desc,		(vop_t *) nfsfifo_read },
221 	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
222 	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
223 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
224 	{ &vop_write_desc,		(vop_t *) nfsfifo_write },
225 	{ NULL, NULL }
226 };
227 static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
228 	{ &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
229 VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
230 
231 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
232 				  struct componentname *cnp,
233 				  struct vattr *vap);
234 static int	nfs_removerpc (struct vnode *dvp, const char *name,
235 				   int namelen,
236 				   struct ucred *cred, struct thread *td);
237 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
238 				   int fnamelen, struct vnode *tdvp,
239 				   const char *tnameptr, int tnamelen,
240 				   struct ucred *cred, struct thread *td);
241 static int	nfs_renameit (struct vnode *sdvp,
242 				  struct componentname *scnp,
243 				  struct sillyrename *sp);
244 
245 /*
246  * Global variables
247  */
248 extern u_int32_t nfs_true, nfs_false;
249 extern u_int32_t nfs_xdrneg1;
250 extern struct nfsstats nfsstats;
251 extern nfstype nfsv3_type[9];
252 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
253 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
254 int nfs_numasync = 0;
255 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
256 
257 SYSCTL_DECL(_vfs_nfs);
258 
259 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
260 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
261 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
262 
263 static int	nfsv3_commit_on_close = 0;
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
265 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
266 #if 0
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
268 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
269 
270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
271 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
272 #endif
273 
274 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
275 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
276 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
277 static int
278 nfs3_access_otw(struct vnode *vp, int wmode,
279 	struct thread *td, struct ucred *cred)
280 {
281 	const int v3 = 1;
282 	u_int32_t *tl;
283 	int error = 0, attrflag;
284 
285 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
286 	caddr_t bpos, dpos, cp2;
287 	int32_t t1, t2;
288 	caddr_t cp;
289 	u_int32_t rmode;
290 	struct nfsnode *np = VTONFS(vp);
291 
292 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
293 	nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
294 	nfsm_fhtom(vp, v3);
295 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
296 	*tl = txdr_unsigned(wmode);
297 	nfsm_request(vp, NFSPROC_ACCESS, td, cred);
298 	nfsm_postop_attr(vp, attrflag);
299 	if (!error) {
300 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
301 		rmode = fxdr_unsigned(u_int32_t, *tl);
302 		np->n_mode = rmode;
303 		np->n_modeuid = cred->cr_uid;
304 		np->n_modestamp = time_second;
305 	}
306 	nfsm_reqdone;
307 	return error;
308 }
309 
310 /*
311  * nfs access vnode op.
312  * For nfs version 2, just return ok. File accesses may fail later.
313  * For nfs version 3, use the access rpc to check accessibility. If file modes
314  * are changed on the server, accesses might still fail later.
315  */
316 static int
317 nfs_access(ap)
318 	struct vop_access_args /* {
319 		struct vnode *a_vp;
320 		int  a_mode;
321 		struct ucred *a_cred;
322 		struct thread *a_td;
323 	} */ *ap;
324 {
325 	struct vnode *vp = ap->a_vp;
326 	int error = 0;
327 	u_int32_t mode, wmode;
328 	int v3 = NFS_ISV3(vp);
329 	struct nfsnode *np = VTONFS(vp);
330 
331 	/*
332 	 * Disallow write attempts on filesystems mounted read-only;
333 	 * unless the file is a socket, fifo, or a block or character
334 	 * device resident on the filesystem.
335 	 */
336 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
337 		switch (vp->v_type) {
338 		case VREG:
339 		case VDIR:
340 		case VLNK:
341 			return (EROFS);
342 		default:
343 			break;
344 		}
345 	}
346 	/*
347 	 * For nfs v3, check to see if we have done this recently, and if
348 	 * so return our cached result instead of making an ACCESS call.
349 	 * If not, do an access rpc, otherwise you are stuck emulating
350 	 * ufs_access() locally using the vattr. This may not be correct,
351 	 * since the server may apply other access criteria such as
352 	 * client uid-->server uid mapping that we do not know about.
353 	 */
354 	if (v3) {
355 		if (ap->a_mode & VREAD)
356 			mode = NFSV3ACCESS_READ;
357 		else
358 			mode = 0;
359 		if (vp->v_type != VDIR) {
360 			if (ap->a_mode & VWRITE)
361 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
362 			if (ap->a_mode & VEXEC)
363 				mode |= NFSV3ACCESS_EXECUTE;
364 		} else {
365 			if (ap->a_mode & VWRITE)
366 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
367 					 NFSV3ACCESS_DELETE);
368 			if (ap->a_mode & VEXEC)
369 				mode |= NFSV3ACCESS_LOOKUP;
370 		}
371 		/* XXX safety belt, only make blanket request if caching */
372 		if (nfsaccess_cache_timeout > 0) {
373 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
374 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
375 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
376 		} else {
377 			wmode = mode;
378 		}
379 
380 		/*
381 		 * Does our cached result allow us to give a definite yes to
382 		 * this request?
383 		 */
384 		if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
385 		    (ap->a_cred->cr_uid == np->n_modeuid) &&
386 		    ((np->n_mode & mode) == mode)) {
387 			nfsstats.accesscache_hits++;
388 		} else {
389 			/*
390 			 * Either a no, or a don't know.  Go to the wire.
391 			 */
392 			nfsstats.accesscache_misses++;
393 		        error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
394 			if (!error) {
395 				if ((np->n_mode & mode) != mode) {
396 					error = EACCES;
397 				}
398 			}
399 		}
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_td = ap->a_td;
426 
427 			if (vp->v_type == VREG) {
428 				error = nfs_readrpc(vp, &auio);
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);
435 				free(bp, M_TEMP);
436 			} else if (vp->v_type == VLNK) {
437 				error = nfs_readlinkrpc(vp, &auio);
438 			} else {
439 				error = EACCES;
440 			}
441 		}
442 	}
443 	/*
444 	 * [re]record creds for reading and/or writing if access
445 	 * was granted.
446 	 */
447 	if (error == 0) {
448 		if ((ap->a_mode & VREAD) && ap->a_cred != np->n_rucred) {
449 			crhold(ap->a_cred);
450 			if (np->n_rucred)
451 				crfree(np->n_rucred);
452 			np->n_rucred = ap->a_cred;
453 		}
454 		if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
455 			crhold(ap->a_cred);
456 			if (np->n_wucred)
457 				crfree(np->n_wucred);
458 			np->n_wucred = ap->a_cred;
459 		}
460 	}
461 	return(error);
462 }
463 
464 /*
465  * nfs open vnode op
466  * Check to see if the type is ok
467  * and that deletion is not in progress.
468  * For paged in text files, you will need to flush the page cache
469  * if consistency is lost.
470  */
471 /* ARGSUSED */
472 static int
473 nfs_open(ap)
474 	struct vop_open_args /* {
475 		struct vnode *a_vp;
476 		int  a_mode;
477 		struct ucred *a_cred;
478 		struct thread *a_td;
479 	} */ *ap;
480 {
481 	struct vnode *vp = ap->a_vp;
482 	struct nfsnode *np = VTONFS(vp);
483 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
484 	struct vattr vattr;
485 	int error;
486 
487 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
488 #ifdef DIAGNOSTIC
489 		printf("open eacces vtyp=%d\n",vp->v_type);
490 #endif
491 		return (EACCES);
492 	}
493 	/*
494 	 * Get a valid lease. If cached data is stale, flush it.
495 	 */
496 	if (nmp->nm_flag & NFSMNT_NQNFS) {
497 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
498 		    do {
499 			error = nqnfs_getlease(vp, ND_READ, ap->a_td);
500 		    } while (error == NQNFS_EXPIRED);
501 		    if (error)
502 			return (error);
503 		    if (np->n_lrev != np->n_brev ||
504 			(np->n_flag & NQNFSNONCACHE)) {
505 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1))
506 			    == EINTR) {
507 				return (error);
508 			}
509 			np->n_brev = np->n_lrev;
510 		    }
511 		}
512 	} else {
513 		if (np->n_flag & NMODIFIED) {
514 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1))
515 			    == EINTR) {
516 				return (error);
517 			}
518 			np->n_attrstamp = 0;
519 			if (vp->v_type == VDIR)
520 				np->n_direofoffset = 0;
521 			error = VOP_GETATTR(vp, &vattr, ap->a_td);
522 			if (error)
523 				return (error);
524 			np->n_mtime = vattr.va_mtime.tv_sec;
525 		} else {
526 			error = VOP_GETATTR(vp, &vattr, ap->a_td);
527 			if (error)
528 				return (error);
529 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
530 				if (vp->v_type == VDIR)
531 					np->n_direofoffset = 0;
532 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
533 				    ap->a_td, 1)) == EINTR) {
534 					return (error);
535 				}
536 				np->n_mtime = vattr.va_mtime.tv_sec;
537 			}
538 		}
539 	}
540 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
541 		np->n_attrstamp = 0; /* For Open/Close consistency */
542 	return (0);
543 }
544 
545 /*
546  * nfs close vnode op
547  * What an NFS client should do upon close after writing is a debatable issue.
548  * Most NFS clients push delayed writes to the server upon close, basically for
549  * two reasons:
550  * 1 - So that any write errors may be reported back to the client process
551  *     doing the close system call. By far the two most likely errors are
552  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
553  * 2 - To put a worst case upper bound on cache inconsistency between
554  *     multiple clients for the file.
555  * There is also a consistency problem for Version 2 of the protocol w.r.t.
556  * not being able to tell if other clients are writing a file concurrently,
557  * since there is no way of knowing if the changed modify time in the reply
558  * is only due to the write for this client.
559  * (NFS Version 3 provides weak cache consistency data in the reply that
560  *  should be sufficient to detect and handle this case.)
561  *
562  * The current code does the following:
563  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
564  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
565  *                     or commit them (this satisfies 1 and 2 except for the
566  *                     case where the server crashes after this close but
567  *                     before the commit RPC, which is felt to be "good
568  *                     enough". Changing the last argument to nfs_flush() to
569  *                     a 1 would force a commit operation, if it is felt a
570  *                     commit is necessary now.
571  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
572  *                     1 should be dealt with via an fsync() system call for
573  *                     cases where write errors are important.
574  */
575 /* ARGSUSED */
576 static int
577 nfs_close(ap)
578 	struct vop_close_args /* {
579 		struct vnodeop_desc *a_desc;
580 		struct vnode *a_vp;
581 		int  a_fflag;
582 		struct ucred *a_cred;
583 		struct thread *a_td;
584 	} */ *ap;
585 {
586 	struct vnode *vp = ap->a_vp;
587 	struct nfsnode *np = VTONFS(vp);
588 	int error = 0;
589 
590 	if (vp->v_type == VREG) {
591 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
592 		(np->n_flag & NMODIFIED)) {
593 		if (NFS_ISV3(vp)) {
594 		    /*
595 		     * Under NFSv3 we have dirty buffers to dispose of.  We
596 		     * must flush them to the NFS server.  We have the option
597 		     * of waiting all the way through the commit rpc or just
598 		     * waiting for the initial write.  The default is to only
599 		     * wait through the initial write so the data is in the
600 		     * server's cache, which is roughly similar to the state
601 		     * a standard disk subsystem leaves the file in on close().
602 		     *
603 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
604 		     * potential races with other processes, and certainly
605 		     * cannot clear it if we don't commit.
606 		     */
607 		    int cm = nfsv3_commit_on_close ? 1 : 0;
608 		    error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
609 		    /* np->n_flag &= ~NMODIFIED; */
610 		} else {
611 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
612 		}
613 		np->n_attrstamp = 0;
614 	    }
615 	    if (np->n_flag & NWRITEERR) {
616 		np->n_flag &= ~NWRITEERR;
617 		error = np->n_error;
618 	    }
619 	}
620 	return (error);
621 }
622 
623 /*
624  * nfs getattr call from vfs.
625  */
626 static int
627 nfs_getattr(ap)
628 	struct vop_getattr_args /* {
629 		struct vnode *a_vp;
630 		struct vattr *a_vap;
631 		struct ucred *a_cred;
632 		struct thread *a_td;
633 	} */ *ap;
634 {
635 	struct vnode *vp = ap->a_vp;
636 	struct nfsnode *np = VTONFS(vp);
637 	caddr_t cp;
638 	u_int32_t *tl;
639 	int32_t t1, t2;
640 	caddr_t bpos, dpos;
641 	int error = 0;
642 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
643 	int v3 = NFS_ISV3(vp);
644 
645 	/*
646 	 * Update local times for special files.
647 	 */
648 	if (np->n_flag & (NACC | NUPD))
649 		np->n_flag |= NCHG;
650 	/*
651 	 * First look in the cache.
652 	 */
653 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
654 		return (0);
655 
656 	if (v3 && nfsaccess_cache_timeout > 0) {
657 		nfsstats.accesscache_misses++;
658 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
659 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
660 			return (0);
661 	}
662 
663 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
664 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
665 	nfsm_fhtom(vp, v3);
666 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
667 	if (!error) {
668 		nfsm_loadattr(vp, ap->a_vap);
669 	}
670 	nfsm_reqdone;
671 	return (error);
672 }
673 
674 /*
675  * nfs setattr call.
676  */
677 static int
678 nfs_setattr(ap)
679 	struct vop_setattr_args /* {
680 		struct vnodeop_desc *a_desc;
681 		struct vnode *a_vp;
682 		struct vattr *a_vap;
683 		struct ucred *a_cred;
684 		struct thread *a_td;
685 	} */ *ap;
686 {
687 	struct vnode *vp = ap->a_vp;
688 	struct nfsnode *np = VTONFS(vp);
689 	struct vattr *vap = ap->a_vap;
690 	int error = 0;
691 	u_quad_t tsize;
692 
693 #ifndef nolint
694 	tsize = (u_quad_t)0;
695 #endif
696 
697 	/*
698 	 * Setting of flags is not supported.
699 	 */
700 	if (vap->va_flags != VNOVAL)
701 		return (EOPNOTSUPP);
702 
703 	/*
704 	 * Disallow write attempts if the filesystem is mounted read-only.
705 	 */
706   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
707 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
708 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
709 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
710 		return (EROFS);
711 	if (vap->va_size != VNOVAL) {
712  		switch (vp->v_type) {
713  		case VDIR:
714  			return (EISDIR);
715  		case VCHR:
716  		case VBLK:
717  		case VSOCK:
718  		case VFIFO:
719 			if (vap->va_mtime.tv_sec == VNOVAL &&
720 			    vap->va_atime.tv_sec == VNOVAL &&
721 			    vap->va_mode == (mode_t)VNOVAL &&
722 			    vap->va_uid == (uid_t)VNOVAL &&
723 			    vap->va_gid == (gid_t)VNOVAL)
724 				return (0);
725  			vap->va_size = VNOVAL;
726  			break;
727  		default:
728 			/*
729 			 * Disallow write attempts if the filesystem is
730 			 * mounted read-only.
731 			 */
732 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
733 				return (EROFS);
734 
735 			/*
736 			 * We run vnode_pager_setsize() early (why?),
737 			 * we must set np->n_size now to avoid vinvalbuf
738 			 * V_SAVE races that might setsize a lower
739 			 * value.
740 			 */
741 
742 			tsize = np->n_size;
743 			error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
744 
745  			if (np->n_flag & NMODIFIED) {
746  			    if (vap->va_size == 0)
747  				error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
748  			    else
749  				error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
750  			    if (error) {
751 				np->n_size = tsize;
752 				vnode_pager_setsize(vp, np->n_size);
753  				return (error);
754 			    }
755  			}
756 			/* np->n_size has already been set to vap->va_size
757 			 * in nfs_meta_setsize(). We must set it again since
758 			 * nfs_loadattrcache() could be called through
759 			 * nfs_meta_setsize() and could modify np->n_size.
760 			 */
761 			np->n_vattr.va_size = np->n_size = vap->va_size;
762 		};
763   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
764 		vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
765 		vp->v_type == VREG &&
766   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR)
767 		return (error);
768 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
769 	if (error && vap->va_size != VNOVAL) {
770 		np->n_size = np->n_vattr.va_size = tsize;
771 		vnode_pager_setsize(vp, np->n_size);
772 	}
773 	return (error);
774 }
775 
776 /*
777  * Do an nfs setattr rpc.
778  */
779 static int
780 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
781 	struct ucred *cred, struct thread *td)
782 {
783 	struct nfsv2_sattr *sp;
784 	caddr_t cp;
785 	int32_t t1, t2;
786 	caddr_t bpos, dpos, cp2;
787 	u_int32_t *tl;
788 	int error = 0, wccflag = NFSV3_WCCRATTR;
789 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
790 	int v3 = NFS_ISV3(vp);
791 
792 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
793 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
794 	nfsm_fhtom(vp, v3);
795 	if (v3) {
796 		nfsm_v3attrbuild(vap, TRUE);
797 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
798 		*tl = nfs_false;
799 	} else {
800 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
801 		if (vap->va_mode == (mode_t)VNOVAL)
802 			sp->sa_mode = nfs_xdrneg1;
803 		else
804 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
805 		if (vap->va_uid == (uid_t)VNOVAL)
806 			sp->sa_uid = nfs_xdrneg1;
807 		else
808 			sp->sa_uid = txdr_unsigned(vap->va_uid);
809 		if (vap->va_gid == (gid_t)VNOVAL)
810 			sp->sa_gid = nfs_xdrneg1;
811 		else
812 			sp->sa_gid = txdr_unsigned(vap->va_gid);
813 		sp->sa_size = txdr_unsigned(vap->va_size);
814 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
815 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
816 	}
817 	nfsm_request(vp, NFSPROC_SETATTR, td, cred);
818 	if (v3) {
819 		nfsm_wcc_data(vp, wccflag);
820 	} else
821 		nfsm_loadattr(vp, (struct vattr *)0);
822 	nfsm_reqdone;
823 	return (error);
824 }
825 
826 /*
827  * nfs lookup call, one step at a time...
828  * First look in cache
829  * If not found, unlock the directory nfsnode and do the rpc
830  */
831 static int
832 nfs_lookup(ap)
833 	struct vop_lookup_args /* {
834 		struct vnodeop_desc *a_desc;
835 		struct vnode *a_dvp;
836 		struct vnode **a_vpp;
837 		struct componentname *a_cnp;
838 	} */ *ap;
839 {
840 	struct componentname *cnp = ap->a_cnp;
841 	struct vnode *dvp = ap->a_dvp;
842 	struct vnode **vpp = ap->a_vpp;
843 	int flags = cnp->cn_flags;
844 	struct vnode *newvp;
845 	u_int32_t *tl;
846 	caddr_t cp;
847 	int32_t t1, t2;
848 	struct nfsmount *nmp;
849 	caddr_t bpos, dpos, cp2;
850 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
851 	long len;
852 	nfsfh_t *fhp;
853 	struct nfsnode *np;
854 	int lockparent, wantparent, error = 0, attrflag, fhsize;
855 	int v3 = NFS_ISV3(dvp);
856 	struct thread *td = cnp->cn_td;
857 
858 	*vpp = NULLVP;
859 	if ((flags & CNP_ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
860 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
861 		return (EROFS);
862 	if (dvp->v_type != VDIR)
863 		return (ENOTDIR);
864 	lockparent = flags & CNP_LOCKPARENT;
865 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
866 	nmp = VFSTONFS(dvp->v_mount);
867 	np = VTONFS(dvp);
868 	if ((error = cache_lookup(dvp, NCPNULL, vpp, NCPPNULL, cnp)) && error != ENOENT) {
869 		struct vattr vattr;
870 		int vpid;
871 
872 		if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
873 			*vpp = NULLVP;
874 			return (error);
875 		}
876 
877 		newvp = *vpp;
878 		vpid = newvp->v_id;
879 		/*
880 		 * See the comment starting `Step through' in ufs/ufs_lookup.c
881 		 * for an explanation of the locking protocol
882 		 */
883 		if (dvp == newvp) {
884 			VREF(newvp);
885 			error = 0;
886 		} else if (flags & CNP_ISDOTDOT) {
887 			VOP_UNLOCK(dvp, 0, td);
888 			error = vget(newvp, LK_EXCLUSIVE, td);
889 			if (!error && lockparent && (flags & CNP_ISLASTCN))
890 				error = vn_lock(dvp, LK_EXCLUSIVE, td);
891 		} else {
892 			error = vget(newvp, LK_EXCLUSIVE, td);
893 			if (!lockparent || error || !(flags & CNP_ISLASTCN))
894 				VOP_UNLOCK(dvp, 0, td);
895 		}
896 		if (!error) {
897 			if (vpid == newvp->v_id) {
898 			   if (!VOP_GETATTR(newvp, &vattr, td)
899 			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
900 				nfsstats.lookupcache_hits++;
901 				if (cnp->cn_nameiop != NAMEI_LOOKUP &&
902 				    (flags & CNP_ISLASTCN))
903 					cnp->cn_flags |= CNP_SAVENAME;
904 				return (0);
905 			   }
906 			   cache_purge(newvp);
907 			}
908 			vput(newvp);
909 			if (lockparent && dvp != newvp && (flags & CNP_ISLASTCN))
910 				VOP_UNLOCK(dvp, 0, td);
911 		}
912 		error = vn_lock(dvp, LK_EXCLUSIVE, td);
913 		*vpp = NULLVP;
914 		if (error)
915 			return (error);
916 	}
917 	error = 0;
918 	newvp = NULLVP;
919 	nfsstats.lookupcache_misses++;
920 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
921 	len = cnp->cn_namelen;
922 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
923 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
924 	nfsm_fhtom(dvp, v3);
925 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
926 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
927 	if (error) {
928 		nfsm_postop_attr(dvp, attrflag);
929 		m_freem(mrep);
930 		goto nfsmout;
931 	}
932 	nfsm_getfh(fhp, fhsize, v3);
933 
934 	/*
935 	 * Handle RENAME case...
936 	 */
937 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent && (flags & CNP_ISLASTCN)) {
938 		if (NFS_CMPFH(np, fhp, fhsize)) {
939 			m_freem(mrep);
940 			return (EISDIR);
941 		}
942 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
943 		if (error) {
944 			m_freem(mrep);
945 			return (error);
946 		}
947 		newvp = NFSTOV(np);
948 		if (v3) {
949 			nfsm_postop_attr(newvp, attrflag);
950 			nfsm_postop_attr(dvp, attrflag);
951 		} else
952 			nfsm_loadattr(newvp, (struct vattr *)0);
953 		*vpp = newvp;
954 		m_freem(mrep);
955 		cnp->cn_flags |= CNP_SAVENAME;
956 		if (!lockparent)
957 			VOP_UNLOCK(dvp, 0, td);
958 		return (0);
959 	}
960 
961 	if (flags & CNP_ISDOTDOT) {
962 		VOP_UNLOCK(dvp, 0, td);
963 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
964 		if (error) {
965 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
966 			return (error);
967 		}
968 		newvp = NFSTOV(np);
969 		if (lockparent && (flags & CNP_ISLASTCN) &&
970 		    (error = vn_lock(dvp, LK_EXCLUSIVE, td))) {
971 		    	vput(newvp);
972 			return (error);
973 		}
974 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
975 		VREF(dvp);
976 		newvp = dvp;
977 	} else {
978 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
979 		if (error) {
980 			m_freem(mrep);
981 			return (error);
982 		}
983 		if (!lockparent || !(flags & CNP_ISLASTCN))
984 			VOP_UNLOCK(dvp, 0, td);
985 		newvp = NFSTOV(np);
986 	}
987 	if (v3) {
988 		nfsm_postop_attr(newvp, attrflag);
989 		nfsm_postop_attr(dvp, attrflag);
990 	} else
991 		nfsm_loadattr(newvp, (struct vattr *)0);
992 	if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN))
993 		cnp->cn_flags |= CNP_SAVENAME;
994 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
995 	    (cnp->cn_nameiop != NAMEI_DELETE || !(flags & CNP_ISLASTCN))) {
996 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
997 		cache_enter(dvp, NCPNULL, newvp, cnp);
998 	}
999 	*vpp = newvp;
1000 	nfsm_reqdone;
1001 	if (error) {
1002 		if (newvp != NULLVP) {
1003 			vrele(newvp);
1004 			*vpp = NULLVP;
1005 		}
1006 		if ((cnp->cn_nameiop == NAMEI_CREATE || cnp->cn_nameiop == NAMEI_RENAME) &&
1007 		    (flags & CNP_ISLASTCN) && error == ENOENT) {
1008 			if (!lockparent)
1009 				VOP_UNLOCK(dvp, 0, td);
1010 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1011 				error = EROFS;
1012 			else
1013 				error = EJUSTRETURN;
1014 		}
1015 		if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN))
1016 			cnp->cn_flags |= CNP_SAVENAME;
1017 	}
1018 	return (error);
1019 }
1020 
1021 /*
1022  * nfs read call.
1023  * Just call nfs_bioread() to do the work.
1024  */
1025 static int
1026 nfs_read(ap)
1027 	struct vop_read_args /* {
1028 		struct vnode *a_vp;
1029 		struct uio *a_uio;
1030 		int  a_ioflag;
1031 		struct ucred *a_cred;
1032 	} */ *ap;
1033 {
1034 	struct vnode *vp = ap->a_vp;
1035 
1036 	if (vp->v_type != VREG)
1037 		return (EPERM);
1038 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1039 }
1040 
1041 /*
1042  * nfs readlink call
1043  */
1044 static int
1045 nfs_readlink(ap)
1046 	struct vop_readlink_args /* {
1047 		struct vnode *a_vp;
1048 		struct uio *a_uio;
1049 		struct ucred *a_cred;
1050 	} */ *ap;
1051 {
1052 	struct vnode *vp = ap->a_vp;
1053 
1054 	if (vp->v_type != VLNK)
1055 		return (EINVAL);
1056 	return (nfs_bioread(vp, ap->a_uio, 0));
1057 }
1058 
1059 /*
1060  * Do a readlink rpc.
1061  * Called by nfs_doio() from below the buffer cache.
1062  */
1063 int
1064 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1065 {
1066 	u_int32_t *tl;
1067 	caddr_t cp;
1068 	int32_t t1, t2;
1069 	caddr_t bpos, dpos, cp2;
1070 	int error = 0, len, attrflag;
1071 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1072 	int v3 = NFS_ISV3(vp);
1073 
1074 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1075 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1076 	nfsm_fhtom(vp, v3);
1077 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1078 	if (v3)
1079 		nfsm_postop_attr(vp, attrflag);
1080 	if (!error) {
1081 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1082 		if (len == NFS_MAXPATHLEN) {
1083 			struct nfsnode *np = VTONFS(vp);
1084 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1085 				len = np->n_size;
1086 		}
1087 		nfsm_mtouio(uiop, len);
1088 	}
1089 	nfsm_reqdone;
1090 	return (error);
1091 }
1092 
1093 /*
1094  * nfs read rpc call
1095  * Ditto above
1096  */
1097 int
1098 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1099 {
1100 	u_int32_t *tl;
1101 	caddr_t cp;
1102 	int32_t t1, t2;
1103 	caddr_t bpos, dpos, cp2;
1104 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1105 	struct nfsmount *nmp;
1106 	int error = 0, len, retlen, tsiz, eof, attrflag;
1107 	int v3 = NFS_ISV3(vp);
1108 
1109 #ifndef nolint
1110 	eof = 0;
1111 #endif
1112 	nmp = VFSTONFS(vp->v_mount);
1113 	tsiz = uiop->uio_resid;
1114 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1115 		return (EFBIG);
1116 	while (tsiz > 0) {
1117 		nfsstats.rpccnt[NFSPROC_READ]++;
1118 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1119 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1120 		nfsm_fhtom(vp, v3);
1121 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1122 		if (v3) {
1123 			txdr_hyper(uiop->uio_offset, tl);
1124 			*(tl + 2) = txdr_unsigned(len);
1125 		} else {
1126 			*tl++ = txdr_unsigned(uiop->uio_offset);
1127 			*tl++ = txdr_unsigned(len);
1128 			*tl = 0;
1129 		}
1130 		nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1131 		if (v3) {
1132 			nfsm_postop_attr(vp, attrflag);
1133 			if (error) {
1134 				m_freem(mrep);
1135 				goto nfsmout;
1136 			}
1137 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1138 			eof = fxdr_unsigned(int, *(tl + 1));
1139 		} else
1140 			nfsm_loadattr(vp, (struct vattr *)0);
1141 		nfsm_strsiz(retlen, nmp->nm_rsize);
1142 		nfsm_mtouio(uiop, retlen);
1143 		m_freem(mrep);
1144 		tsiz -= retlen;
1145 		if (v3) {
1146 			if (eof || retlen == 0) {
1147 				tsiz = 0;
1148 			}
1149 		} else if (retlen < len) {
1150 			tsiz = 0;
1151 		}
1152 	}
1153 nfsmout:
1154 	return (error);
1155 }
1156 
1157 /*
1158  * nfs write call
1159  */
1160 int
1161 nfs_writerpc(vp, uiop, iomode, must_commit)
1162 	struct vnode *vp;
1163 	struct uio *uiop;
1164 	int *iomode, *must_commit;
1165 {
1166 	u_int32_t *tl;
1167 	caddr_t cp;
1168 	int32_t t1, t2, backup;
1169 	caddr_t bpos, dpos, cp2;
1170 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1171 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1172 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1173 	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1174 
1175 #ifndef DIAGNOSTIC
1176 	if (uiop->uio_iovcnt != 1)
1177 		panic("nfs: writerpc iovcnt > 1");
1178 #endif
1179 	*must_commit = 0;
1180 	tsiz = uiop->uio_resid;
1181 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1182 		return (EFBIG);
1183 	while (tsiz > 0) {
1184 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1185 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1186 		nfsm_reqhead(vp, NFSPROC_WRITE,
1187 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1188 		nfsm_fhtom(vp, v3);
1189 		if (v3) {
1190 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1191 			txdr_hyper(uiop->uio_offset, tl);
1192 			tl += 2;
1193 			*tl++ = txdr_unsigned(len);
1194 			*tl++ = txdr_unsigned(*iomode);
1195 			*tl = txdr_unsigned(len);
1196 		} else {
1197 			u_int32_t x;
1198 
1199 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1200 			/* Set both "begin" and "current" to non-garbage. */
1201 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1202 			*tl++ = x;	/* "begin offset" */
1203 			*tl++ = x;	/* "current offset" */
1204 			x = txdr_unsigned(len);
1205 			*tl++ = x;	/* total to this offset */
1206 			*tl = x;	/* size of this write */
1207 		}
1208 		nfsm_uiotom(uiop, len);
1209 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1210 		if (v3) {
1211 			wccflag = NFSV3_WCCCHK;
1212 			nfsm_wcc_data(vp, wccflag);
1213 			if (!error) {
1214 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1215 					+ NFSX_V3WRITEVERF);
1216 				rlen = fxdr_unsigned(int, *tl++);
1217 				if (rlen == 0) {
1218 					error = NFSERR_IO;
1219 					m_freem(mrep);
1220 					break;
1221 				} else if (rlen < len) {
1222 					backup = len - rlen;
1223 					uiop->uio_iov->iov_base -= backup;
1224 					uiop->uio_iov->iov_len += backup;
1225 					uiop->uio_offset -= backup;
1226 					uiop->uio_resid += backup;
1227 					len = rlen;
1228 				}
1229 				commit = fxdr_unsigned(int, *tl++);
1230 
1231 				/*
1232 				 * Return the lowest committment level
1233 				 * obtained by any of the RPCs.
1234 				 */
1235 				if (committed == NFSV3WRITE_FILESYNC)
1236 					committed = commit;
1237 				else if (committed == NFSV3WRITE_DATASYNC &&
1238 					commit == NFSV3WRITE_UNSTABLE)
1239 					committed = commit;
1240 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1241 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1242 					NFSX_V3WRITEVERF);
1243 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1244 				} else if (bcmp((caddr_t)tl,
1245 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1246 				    *must_commit = 1;
1247 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1248 					NFSX_V3WRITEVERF);
1249 				}
1250 			}
1251 		} else
1252 		    nfsm_loadattr(vp, (struct vattr *)0);
1253 		if (wccflag)
1254 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1255 		m_freem(mrep);
1256 		if (error)
1257 			break;
1258 		tsiz -= len;
1259 	}
1260 nfsmout:
1261 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1262 		committed = NFSV3WRITE_FILESYNC;
1263 	*iomode = committed;
1264 	if (error)
1265 		uiop->uio_resid = tsiz;
1266 	return (error);
1267 }
1268 
1269 /*
1270  * nfs mknod rpc
1271  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1272  * mode set to specify the file type and the size field for rdev.
1273  */
1274 static int
1275 nfs_mknodrpc(dvp, vpp, cnp, vap)
1276 	struct vnode *dvp;
1277 	struct vnode **vpp;
1278 	struct componentname *cnp;
1279 	struct vattr *vap;
1280 {
1281 	struct nfsv2_sattr *sp;
1282 	u_int32_t *tl;
1283 	caddr_t cp;
1284 	int32_t t1, t2;
1285 	struct vnode *newvp = (struct vnode *)0;
1286 	struct nfsnode *np = (struct nfsnode *)0;
1287 	struct vattr vattr;
1288 	char *cp2;
1289 	caddr_t bpos, dpos;
1290 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1291 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1292 	u_int32_t rdev;
1293 	int v3 = NFS_ISV3(dvp);
1294 
1295 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1296 		rdev = txdr_unsigned(vap->va_rdev);
1297 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1298 		rdev = nfs_xdrneg1;
1299 	else {
1300 		return (EOPNOTSUPP);
1301 	}
1302 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1303 		return (error);
1304 	}
1305 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1306 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1307 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1308 	nfsm_fhtom(dvp, v3);
1309 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1310 	if (v3) {
1311 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1312 		*tl++ = vtonfsv3_type(vap->va_type);
1313 		nfsm_v3attrbuild(vap, FALSE);
1314 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1315 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1316 			*tl++ = txdr_unsigned(umajor(vap->va_rdev));
1317 			*tl = txdr_unsigned(uminor(vap->va_rdev));
1318 		}
1319 	} else {
1320 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1321 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1322 		sp->sa_uid = nfs_xdrneg1;
1323 		sp->sa_gid = nfs_xdrneg1;
1324 		sp->sa_size = rdev;
1325 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1326 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1327 	}
1328 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1329 	if (!error) {
1330 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1331 		if (!gotvp) {
1332 			if (newvp) {
1333 				vput(newvp);
1334 				newvp = (struct vnode *)0;
1335 			}
1336 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1337 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1338 			if (!error)
1339 				newvp = NFSTOV(np);
1340 		}
1341 	}
1342 	if (v3)
1343 		nfsm_wcc_data(dvp, wccflag);
1344 	nfsm_reqdone;
1345 	if (error) {
1346 		if (newvp)
1347 			vput(newvp);
1348 	} else {
1349 		if (cnp->cn_flags & CNP_MAKEENTRY)
1350 			cache_enter(dvp, NCPNULL, newvp, cnp);
1351 		*vpp = newvp;
1352 	}
1353 	VTONFS(dvp)->n_flag |= NMODIFIED;
1354 	if (!wccflag)
1355 		VTONFS(dvp)->n_attrstamp = 0;
1356 	return (error);
1357 }
1358 
1359 /*
1360  * nfs mknod vop
1361  * just call nfs_mknodrpc() to do the work.
1362  */
1363 /* ARGSUSED */
1364 static int
1365 nfs_mknod(ap)
1366 	struct vop_mknod_args /* {
1367 		struct vnode *a_dvp;
1368 		struct vnode **a_vpp;
1369 		struct componentname *a_cnp;
1370 		struct vattr *a_vap;
1371 	} */ *ap;
1372 {
1373 	return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1374 }
1375 
1376 static u_long create_verf;
1377 /*
1378  * nfs file create call
1379  */
1380 static int
1381 nfs_create(ap)
1382 	struct vop_create_args /* {
1383 		struct vnode *a_dvp;
1384 		struct vnode **a_vpp;
1385 		struct componentname *a_cnp;
1386 		struct vattr *a_vap;
1387 	} */ *ap;
1388 {
1389 	struct vnode *dvp = ap->a_dvp;
1390 	struct vattr *vap = ap->a_vap;
1391 	struct componentname *cnp = ap->a_cnp;
1392 	struct nfsv2_sattr *sp;
1393 	u_int32_t *tl;
1394 	caddr_t cp;
1395 	int32_t t1, t2;
1396 	struct nfsnode *np = (struct nfsnode *)0;
1397 	struct vnode *newvp = (struct vnode *)0;
1398 	caddr_t bpos, dpos, cp2;
1399 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1400 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1401 	struct vattr vattr;
1402 	int v3 = NFS_ISV3(dvp);
1403 
1404 	/*
1405 	 * Oops, not for me..
1406 	 */
1407 	if (vap->va_type == VSOCK)
1408 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1409 
1410 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1411 		return (error);
1412 	}
1413 	if (vap->va_vaflags & VA_EXCLUSIVE)
1414 		fmode |= O_EXCL;
1415 again:
1416 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1417 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1418 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1419 	nfsm_fhtom(dvp, v3);
1420 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1421 	if (v3) {
1422 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1423 		if (fmode & O_EXCL) {
1424 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1425 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1426 #ifdef INET
1427 			if (!TAILQ_EMPTY(&in_ifaddrhead))
1428 				*tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
1429 			else
1430 #endif
1431 				*tl++ = create_verf;
1432 			*tl = ++create_verf;
1433 		} else {
1434 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1435 			nfsm_v3attrbuild(vap, FALSE);
1436 		}
1437 	} else {
1438 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1439 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1440 		sp->sa_uid = nfs_xdrneg1;
1441 		sp->sa_gid = nfs_xdrneg1;
1442 		sp->sa_size = 0;
1443 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1444 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1445 	}
1446 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1447 	if (!error) {
1448 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1449 		if (!gotvp) {
1450 			if (newvp) {
1451 				vput(newvp);
1452 				newvp = (struct vnode *)0;
1453 			}
1454 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1455 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1456 			if (!error)
1457 				newvp = NFSTOV(np);
1458 		}
1459 	}
1460 	if (v3)
1461 		nfsm_wcc_data(dvp, wccflag);
1462 	nfsm_reqdone;
1463 	if (error) {
1464 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1465 			fmode &= ~O_EXCL;
1466 			goto again;
1467 		}
1468 		if (newvp)
1469 			vput(newvp);
1470 	} else if (v3 && (fmode & O_EXCL)) {
1471 		/*
1472 		 * We are normally called with only a partially initialized
1473 		 * VAP.  Since the NFSv3 spec says that server may use the
1474 		 * file attributes to store the verifier, the spec requires
1475 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1476 		 * in atime, but we can't really assume that all servers will
1477 		 * so we ensure that our SETATTR sets both atime and mtime.
1478 		 */
1479 		if (vap->va_mtime.tv_sec == VNOVAL)
1480 			vfs_timestamp(&vap->va_mtime);
1481 		if (vap->va_atime.tv_sec == VNOVAL)
1482 			vap->va_atime = vap->va_mtime;
1483 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1484 	}
1485 	if (!error) {
1486 		if (cnp->cn_flags & CNP_MAKEENTRY)
1487 			cache_enter(dvp, NCPNULL, newvp, cnp);
1488 		/*
1489 		 * The new np may have enough info for access
1490 		 * checks, make sure rucred and wucred are
1491 		 * initialized for read and write rpc's.
1492 		 */
1493 		np = VTONFS(newvp);
1494 		if (np->n_rucred == NULL)
1495 			np->n_rucred = crhold(cnp->cn_cred);
1496 		if (np->n_wucred == NULL)
1497 			np->n_wucred = crhold(cnp->cn_cred);
1498 		*ap->a_vpp = newvp;
1499 	}
1500 	VTONFS(dvp)->n_flag |= NMODIFIED;
1501 	if (!wccflag)
1502 		VTONFS(dvp)->n_attrstamp = 0;
1503 	return (error);
1504 }
1505 
1506 /*
1507  * nfs file remove call
1508  * To try and make nfs semantics closer to ufs semantics, a file that has
1509  * other processes using the vnode is renamed instead of removed and then
1510  * removed later on the last close.
1511  * - If v_usecount > 1
1512  *	  If a rename is not already in the works
1513  *	     call nfs_sillyrename() to set it up
1514  *     else
1515  *	  do the remove rpc
1516  */
1517 static int
1518 nfs_remove(ap)
1519 	struct vop_remove_args /* {
1520 		struct vnodeop_desc *a_desc;
1521 		struct vnode * a_dvp;
1522 		struct vnode * a_vp;
1523 		struct componentname * a_cnp;
1524 	} */ *ap;
1525 {
1526 	struct vnode *vp = ap->a_vp;
1527 	struct vnode *dvp = ap->a_dvp;
1528 	struct componentname *cnp = ap->a_cnp;
1529 	struct nfsnode *np = VTONFS(vp);
1530 	int error = 0;
1531 	struct vattr vattr;
1532 
1533 #ifndef DIAGNOSTIC
1534 	if ((cnp->cn_flags & CNP_HASBUF) == 0)
1535 		panic("nfs_remove: no name");
1536 	if (vp->v_usecount < 1)
1537 		panic("nfs_remove: bad v_usecount");
1538 #endif
1539 	if (vp->v_type == VDIR)
1540 		error = EPERM;
1541 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1542 	    VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
1543 	    vattr.va_nlink > 1)) {
1544 		/*
1545 		 * Purge the name cache so that the chance of a lookup for
1546 		 * the name succeeding while the remove is in progress is
1547 		 * minimized. Without node locking it can still happen, such
1548 		 * that an I/O op returns ESTALE, but since you get this if
1549 		 * another host removes the file..
1550 		 */
1551 		cache_purge(vp);
1552 		/*
1553 		 * throw away biocache buffers, mainly to avoid
1554 		 * unnecessary delayed writes later.
1555 		 */
1556 		error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
1557 		/* Do the rpc */
1558 		if (error != EINTR)
1559 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1560 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1561 		/*
1562 		 * Kludge City: If the first reply to the remove rpc is lost..
1563 		 *   the reply to the retransmitted request will be ENOENT
1564 		 *   since the file was in fact removed
1565 		 *   Therefore, we cheat and return success.
1566 		 */
1567 		if (error == ENOENT)
1568 			error = 0;
1569 	} else if (!np->n_sillyrename)
1570 		error = nfs_sillyrename(dvp, vp, cnp);
1571 	np->n_attrstamp = 0;
1572 	return (error);
1573 }
1574 
1575 /*
1576  * nfs file remove rpc called from nfs_inactive
1577  */
1578 int
1579 nfs_removeit(struct sillyrename *sp)
1580 {
1581 
1582 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1583 		sp->s_cred, NULL));
1584 }
1585 
1586 /*
1587  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1588  */
1589 static int
1590 nfs_removerpc(dvp, name, namelen, cred, td)
1591 	struct vnode *dvp;
1592 	const char *name;
1593 	int namelen;
1594 	struct ucred *cred;
1595 	struct thread *td;
1596 {
1597 	u_int32_t *tl;
1598 	caddr_t cp;
1599 	int32_t t1, t2;
1600 	caddr_t bpos, dpos, cp2;
1601 	int error = 0, wccflag = NFSV3_WCCRATTR;
1602 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1603 	int v3 = NFS_ISV3(dvp);
1604 
1605 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1606 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1607 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1608 	nfsm_fhtom(dvp, v3);
1609 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1610 	nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1611 	if (v3)
1612 		nfsm_wcc_data(dvp, wccflag);
1613 	nfsm_reqdone;
1614 	VTONFS(dvp)->n_flag |= NMODIFIED;
1615 	if (!wccflag)
1616 		VTONFS(dvp)->n_attrstamp = 0;
1617 	return (error);
1618 }
1619 
1620 /*
1621  * nfs file rename call
1622  */
1623 static int
1624 nfs_rename(ap)
1625 	struct vop_rename_args  /* {
1626 		struct vnode *a_fdvp;
1627 		struct vnode *a_fvp;
1628 		struct componentname *a_fcnp;
1629 		struct vnode *a_tdvp;
1630 		struct vnode *a_tvp;
1631 		struct componentname *a_tcnp;
1632 	} */ *ap;
1633 {
1634 	struct vnode *fvp = ap->a_fvp;
1635 	struct vnode *tvp = ap->a_tvp;
1636 	struct vnode *fdvp = ap->a_fdvp;
1637 	struct vnode *tdvp = ap->a_tdvp;
1638 	struct componentname *tcnp = ap->a_tcnp;
1639 	struct componentname *fcnp = ap->a_fcnp;
1640 	int error;
1641 
1642 #ifndef DIAGNOSTIC
1643 	if ((tcnp->cn_flags & CNP_HASBUF) == 0 ||
1644 	    (fcnp->cn_flags & CNP_HASBUF) == 0)
1645 		panic("nfs_rename: no name");
1646 #endif
1647 	/* Check for cross-device rename */
1648 	if ((fvp->v_mount != tdvp->v_mount) ||
1649 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1650 		error = EXDEV;
1651 		goto out;
1652 	}
1653 
1654 	/*
1655 	 * We have to flush B_DELWRI data prior to renaming
1656 	 * the file.  If we don't, the delayed-write buffers
1657 	 * can be flushed out later after the file has gone stale
1658 	 * under NFSV3.  NFSV2 does not have this problem because
1659 	 * ( as far as I can tell ) it flushes dirty buffers more
1660 	 * often.
1661 	 */
1662 
1663 	VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
1664 	if (tvp)
1665 	    VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
1666 
1667 	/*
1668 	 * If the tvp exists and is in use, sillyrename it before doing the
1669 	 * rename of the new file over it.
1670 	 * XXX Can't sillyrename a directory.
1671 	 */
1672 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1673 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1674 		vput(tvp);
1675 		tvp = NULL;
1676 	}
1677 
1678 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1679 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1680 		tcnp->cn_td);
1681 
1682 	if (fvp->v_type == VDIR) {
1683 		if (tvp != NULL && tvp->v_type == VDIR)
1684 			cache_purge(tdvp);
1685 		cache_purge(fdvp);
1686 	}
1687 
1688 out:
1689 	if (tdvp == tvp)
1690 		vrele(tdvp);
1691 	else
1692 		vput(tdvp);
1693 	if (tvp)
1694 		vput(tvp);
1695 	vrele(fdvp);
1696 	vrele(fvp);
1697 	/*
1698 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1699 	 */
1700 	if (error == ENOENT)
1701 		error = 0;
1702 	return (error);
1703 }
1704 
1705 /*
1706  * nfs file rename rpc called from nfs_remove() above
1707  */
1708 static int
1709 nfs_renameit(sdvp, scnp, sp)
1710 	struct vnode *sdvp;
1711 	struct componentname *scnp;
1712 	struct sillyrename *sp;
1713 {
1714 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1715 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1716 }
1717 
1718 /*
1719  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1720  */
1721 static int
1722 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, td)
1723 	struct vnode *fdvp;
1724 	const char *fnameptr;
1725 	int fnamelen;
1726 	struct vnode *tdvp;
1727 	const char *tnameptr;
1728 	int tnamelen;
1729 	struct ucred *cred;
1730 	struct thread *td;
1731 {
1732 	u_int32_t *tl;
1733 	caddr_t cp;
1734 	int32_t t1, t2;
1735 	caddr_t bpos, dpos, cp2;
1736 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1737 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1738 	int v3 = NFS_ISV3(fdvp);
1739 
1740 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1741 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1742 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1743 		nfsm_rndup(tnamelen));
1744 	nfsm_fhtom(fdvp, v3);
1745 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1746 	nfsm_fhtom(tdvp, v3);
1747 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1748 	nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1749 	if (v3) {
1750 		nfsm_wcc_data(fdvp, fwccflag);
1751 		nfsm_wcc_data(tdvp, twccflag);
1752 	}
1753 	nfsm_reqdone;
1754 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1755 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1756 	if (!fwccflag)
1757 		VTONFS(fdvp)->n_attrstamp = 0;
1758 	if (!twccflag)
1759 		VTONFS(tdvp)->n_attrstamp = 0;
1760 	return (error);
1761 }
1762 
1763 /*
1764  * nfs hard link create call
1765  */
1766 static int
1767 nfs_link(ap)
1768 	struct vop_link_args /* {
1769 		struct vnode *a_tdvp;
1770 		struct vnode *a_vp;
1771 		struct componentname *a_cnp;
1772 	} */ *ap;
1773 {
1774 	struct vnode *vp = ap->a_vp;
1775 	struct vnode *tdvp = ap->a_tdvp;
1776 	struct componentname *cnp = ap->a_cnp;
1777 	u_int32_t *tl;
1778 	caddr_t cp;
1779 	int32_t t1, t2;
1780 	caddr_t bpos, dpos, cp2;
1781 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1782 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1783 	int v3;
1784 
1785 	if (vp->v_mount != tdvp->v_mount) {
1786 		return (EXDEV);
1787 	}
1788 
1789 	/*
1790 	 * Push all writes to the server, so that the attribute cache
1791 	 * doesn't get "out of sync" with the server.
1792 	 * XXX There should be a better way!
1793 	 */
1794 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
1795 
1796 	v3 = NFS_ISV3(vp);
1797 	nfsstats.rpccnt[NFSPROC_LINK]++;
1798 	nfsm_reqhead(vp, NFSPROC_LINK,
1799 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1800 	nfsm_fhtom(vp, v3);
1801 	nfsm_fhtom(tdvp, v3);
1802 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1803 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1804 	if (v3) {
1805 		nfsm_postop_attr(vp, attrflag);
1806 		nfsm_wcc_data(tdvp, wccflag);
1807 	}
1808 	nfsm_reqdone;
1809 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1810 	if (!attrflag)
1811 		VTONFS(vp)->n_attrstamp = 0;
1812 	if (!wccflag)
1813 		VTONFS(tdvp)->n_attrstamp = 0;
1814 	/*
1815 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1816 	 */
1817 	if (error == EEXIST)
1818 		error = 0;
1819 	return (error);
1820 }
1821 
1822 /*
1823  * nfs symbolic link create call
1824  */
1825 static int
1826 nfs_symlink(ap)
1827 	struct vop_symlink_args /* {
1828 		struct vnode *a_dvp;
1829 		struct vnode **a_vpp;
1830 		struct componentname *a_cnp;
1831 		struct vattr *a_vap;
1832 		char *a_target;
1833 	} */ *ap;
1834 {
1835 	struct vnode *dvp = ap->a_dvp;
1836 	struct vattr *vap = ap->a_vap;
1837 	struct componentname *cnp = ap->a_cnp;
1838 	struct nfsv2_sattr *sp;
1839 	u_int32_t *tl;
1840 	caddr_t cp;
1841 	int32_t t1, t2;
1842 	caddr_t bpos, dpos, cp2;
1843 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1844 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1845 	struct vnode *newvp = (struct vnode *)0;
1846 	int v3 = NFS_ISV3(dvp);
1847 
1848 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1849 	slen = strlen(ap->a_target);
1850 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1851 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1852 	nfsm_fhtom(dvp, v3);
1853 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1854 	if (v3) {
1855 		nfsm_v3attrbuild(vap, FALSE);
1856 	}
1857 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1858 	if (!v3) {
1859 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1860 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1861 		sp->sa_uid = nfs_xdrneg1;
1862 		sp->sa_gid = nfs_xdrneg1;
1863 		sp->sa_size = nfs_xdrneg1;
1864 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1865 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1866 	}
1867 
1868 	/*
1869 	 * Issue the NFS request and get the rpc response.
1870 	 *
1871 	 * Only NFSv3 responses returning an error of 0 actually return
1872 	 * a file handle that can be converted into newvp without having
1873 	 * to do an extra lookup rpc.
1874 	 */
1875 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1876 	if (v3) {
1877 		if (error == 0)
1878 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1879 		nfsm_wcc_data(dvp, wccflag);
1880 	}
1881 
1882 	/*
1883 	 * out code jumps -> here, mrep is also freed.
1884 	 */
1885 
1886 	nfsm_reqdone;
1887 
1888 	/*
1889 	 * If we get an EEXIST error, silently convert it to no-error
1890 	 * in case of an NFS retry.
1891 	 */
1892 	if (error == EEXIST)
1893 		error = 0;
1894 
1895 	/*
1896 	 * If we do not have (or no longer have) an error, and we could
1897 	 * not extract the newvp from the response due to the request being
1898 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
1899 	 * to obtain a newvp to return.
1900 	 */
1901 	if (error == 0 && newvp == NULL) {
1902 		struct nfsnode *np = NULL;
1903 
1904 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1905 		    cnp->cn_cred, cnp->cn_td, &np);
1906 		if (!error)
1907 			newvp = NFSTOV(np);
1908 	}
1909 	if (error) {
1910 		if (newvp)
1911 			vput(newvp);
1912 	} else {
1913 		*ap->a_vpp = newvp;
1914 	}
1915 	VTONFS(dvp)->n_flag |= NMODIFIED;
1916 	if (!wccflag)
1917 		VTONFS(dvp)->n_attrstamp = 0;
1918 	return (error);
1919 }
1920 
1921 /*
1922  * nfs make dir call
1923  */
1924 static int
1925 nfs_mkdir(ap)
1926 	struct vop_mkdir_args /* {
1927 		struct vnode *a_dvp;
1928 		struct vnode **a_vpp;
1929 		struct componentname *a_cnp;
1930 		struct vattr *a_vap;
1931 	} */ *ap;
1932 {
1933 	struct vnode *dvp = ap->a_dvp;
1934 	struct vattr *vap = ap->a_vap;
1935 	struct componentname *cnp = ap->a_cnp;
1936 	struct nfsv2_sattr *sp;
1937 	u_int32_t *tl;
1938 	caddr_t cp;
1939 	int32_t t1, t2;
1940 	int len;
1941 	struct nfsnode *np = (struct nfsnode *)0;
1942 	struct vnode *newvp = (struct vnode *)0;
1943 	caddr_t bpos, dpos, cp2;
1944 	int error = 0, wccflag = NFSV3_WCCRATTR;
1945 	int gotvp = 0;
1946 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1947 	struct vattr vattr;
1948 	int v3 = NFS_ISV3(dvp);
1949 
1950 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1951 		return (error);
1952 	}
1953 	len = cnp->cn_namelen;
1954 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1955 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1956 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1957 	nfsm_fhtom(dvp, v3);
1958 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1959 	if (v3) {
1960 		nfsm_v3attrbuild(vap, FALSE);
1961 	} else {
1962 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1963 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1964 		sp->sa_uid = nfs_xdrneg1;
1965 		sp->sa_gid = nfs_xdrneg1;
1966 		sp->sa_size = nfs_xdrneg1;
1967 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1968 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1969 	}
1970 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
1971 	if (!error)
1972 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1973 	if (v3)
1974 		nfsm_wcc_data(dvp, wccflag);
1975 	nfsm_reqdone;
1976 	VTONFS(dvp)->n_flag |= NMODIFIED;
1977 	if (!wccflag)
1978 		VTONFS(dvp)->n_attrstamp = 0;
1979 	/*
1980 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1981 	 * if we can succeed in looking up the directory.
1982 	 */
1983 	if (error == EEXIST || (!error && !gotvp)) {
1984 		if (newvp) {
1985 			vrele(newvp);
1986 			newvp = (struct vnode *)0;
1987 		}
1988 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1989 			cnp->cn_td, &np);
1990 		if (!error) {
1991 			newvp = NFSTOV(np);
1992 			if (newvp->v_type != VDIR)
1993 				error = EEXIST;
1994 		}
1995 	}
1996 	if (error) {
1997 		if (newvp)
1998 			vrele(newvp);
1999 	} else
2000 		*ap->a_vpp = newvp;
2001 	return (error);
2002 }
2003 
2004 /*
2005  * nfs remove directory call
2006  */
2007 static int
2008 nfs_rmdir(ap)
2009 	struct vop_rmdir_args /* {
2010 		struct vnode *a_dvp;
2011 		struct vnode *a_vp;
2012 		struct componentname *a_cnp;
2013 	} */ *ap;
2014 {
2015 	struct vnode *vp = ap->a_vp;
2016 	struct vnode *dvp = ap->a_dvp;
2017 	struct componentname *cnp = ap->a_cnp;
2018 	u_int32_t *tl;
2019 	caddr_t cp;
2020 	int32_t t1, t2;
2021 	caddr_t bpos, dpos, cp2;
2022 	int error = 0, wccflag = NFSV3_WCCRATTR;
2023 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2024 	int v3 = NFS_ISV3(dvp);
2025 
2026 	if (dvp == vp)
2027 		return (EINVAL);
2028 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2029 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
2030 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2031 	nfsm_fhtom(dvp, v3);
2032 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2033 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2034 	if (v3)
2035 		nfsm_wcc_data(dvp, wccflag);
2036 	nfsm_reqdone;
2037 	VTONFS(dvp)->n_flag |= NMODIFIED;
2038 	if (!wccflag)
2039 		VTONFS(dvp)->n_attrstamp = 0;
2040 	cache_purge(dvp);
2041 	cache_purge(vp);
2042 	/*
2043 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2044 	 */
2045 	if (error == ENOENT)
2046 		error = 0;
2047 	return (error);
2048 }
2049 
2050 /*
2051  * nfs readdir call
2052  */
2053 static int
2054 nfs_readdir(ap)
2055 	struct vop_readdir_args /* {
2056 		struct vnode *a_vp;
2057 		struct uio *a_uio;
2058 		struct ucred *a_cred;
2059 	} */ *ap;
2060 {
2061 	struct vnode *vp = ap->a_vp;
2062 	struct nfsnode *np = VTONFS(vp);
2063 	struct uio *uio = ap->a_uio;
2064 	int tresid, error;
2065 	struct vattr vattr;
2066 
2067 	if (vp->v_type != VDIR)
2068 		return (EPERM);
2069 	/*
2070 	 * First, check for hit on the EOF offset cache
2071 	 */
2072 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2073 	    (np->n_flag & NMODIFIED) == 0) {
2074 		if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
2075 			if (NQNFS_CKCACHABLE(vp, ND_READ)) {
2076 				nfsstats.direofcache_hits++;
2077 				return (0);
2078 			}
2079 		} else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
2080 			np->n_mtime == vattr.va_mtime.tv_sec) {
2081 			nfsstats.direofcache_hits++;
2082 			return (0);
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * Call nfs_bioread() to do the real work.
2088 	 */
2089 	tresid = uio->uio_resid;
2090 	error = nfs_bioread(vp, uio, 0);
2091 
2092 	if (!error && uio->uio_resid == tresid)
2093 		nfsstats.direofcache_misses++;
2094 	return (error);
2095 }
2096 
2097 /*
2098  * Readdir rpc call.
2099  * Called from below the buffer cache by nfs_doio().
2100  */
2101 int
2102 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2103 {
2104 	int len, left;
2105 	struct dirent *dp = NULL;
2106 	u_int32_t *tl;
2107 	caddr_t cp;
2108 	int32_t t1, t2;
2109 	nfsuint64 *cookiep;
2110 	caddr_t bpos, dpos, cp2;
2111 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2112 	nfsuint64 cookie;
2113 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2114 	struct nfsnode *dnp = VTONFS(vp);
2115 	u_quad_t fileno;
2116 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2117 	int attrflag;
2118 	int v3 = NFS_ISV3(vp);
2119 
2120 #ifndef DIAGNOSTIC
2121 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2122 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2123 		panic("nfs readdirrpc bad uio");
2124 #endif
2125 
2126 	/*
2127 	 * If there is no cookie, assume directory was stale.
2128 	 */
2129 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2130 	if (cookiep)
2131 		cookie = *cookiep;
2132 	else
2133 		return (NFSERR_BAD_COOKIE);
2134 	/*
2135 	 * Loop around doing readdir rpc's of size nm_readdirsize
2136 	 * truncated to a multiple of DIRBLKSIZ.
2137 	 * The stopping criteria is EOF or buffer full.
2138 	 */
2139 	while (more_dirs && bigenough) {
2140 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2141 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2142 			NFSX_READDIR(v3));
2143 		nfsm_fhtom(vp, v3);
2144 		if (v3) {
2145 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2146 			*tl++ = cookie.nfsuquad[0];
2147 			*tl++ = cookie.nfsuquad[1];
2148 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2149 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2150 		} else {
2151 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2152 			*tl++ = cookie.nfsuquad[0];
2153 		}
2154 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2155 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2156 		if (v3) {
2157 			nfsm_postop_attr(vp, attrflag);
2158 			if (!error) {
2159 				nfsm_dissect(tl, u_int32_t *,
2160 				    2 * NFSX_UNSIGNED);
2161 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2162 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2163 			} else {
2164 				m_freem(mrep);
2165 				goto nfsmout;
2166 			}
2167 		}
2168 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2169 		more_dirs = fxdr_unsigned(int, *tl);
2170 
2171 		/* loop thru the dir entries, doctoring them to 4bsd form */
2172 		while (more_dirs && bigenough) {
2173 			if (v3) {
2174 				nfsm_dissect(tl, u_int32_t *,
2175 				    3 * NFSX_UNSIGNED);
2176 				fileno = fxdr_hyper(tl);
2177 				len = fxdr_unsigned(int, *(tl + 2));
2178 			} else {
2179 				nfsm_dissect(tl, u_int32_t *,
2180 				    2 * NFSX_UNSIGNED);
2181 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2182 				len = fxdr_unsigned(int, *tl);
2183 			}
2184 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2185 				error = EBADRPC;
2186 				m_freem(mrep);
2187 				goto nfsmout;
2188 			}
2189 			tlen = nfsm_rndup(len);
2190 			if (tlen == len)
2191 				tlen += 4;	/* To ensure null termination */
2192 			left = DIRBLKSIZ - blksiz;
2193 			if ((tlen + DIRHDSIZ) > left) {
2194 				dp->d_reclen += left;
2195 				uiop->uio_iov->iov_base += left;
2196 				uiop->uio_iov->iov_len -= left;
2197 				uiop->uio_offset += left;
2198 				uiop->uio_resid -= left;
2199 				blksiz = 0;
2200 			}
2201 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2202 				bigenough = 0;
2203 			if (bigenough) {
2204 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2205 				dp->d_fileno = (int)fileno;
2206 				dp->d_namlen = len;
2207 				dp->d_reclen = tlen + DIRHDSIZ;
2208 				dp->d_type = DT_UNKNOWN;
2209 				blksiz += dp->d_reclen;
2210 				if (blksiz == DIRBLKSIZ)
2211 					blksiz = 0;
2212 				uiop->uio_offset += DIRHDSIZ;
2213 				uiop->uio_resid -= DIRHDSIZ;
2214 				uiop->uio_iov->iov_base += DIRHDSIZ;
2215 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2216 				nfsm_mtouio(uiop, len);
2217 				cp = uiop->uio_iov->iov_base;
2218 				tlen -= len;
2219 				*cp = '\0';	/* null terminate */
2220 				uiop->uio_iov->iov_base += tlen;
2221 				uiop->uio_iov->iov_len -= tlen;
2222 				uiop->uio_offset += tlen;
2223 				uiop->uio_resid -= tlen;
2224 			} else
2225 				nfsm_adv(nfsm_rndup(len));
2226 			if (v3) {
2227 				nfsm_dissect(tl, u_int32_t *,
2228 				    3 * NFSX_UNSIGNED);
2229 			} else {
2230 				nfsm_dissect(tl, u_int32_t *,
2231 				    2 * NFSX_UNSIGNED);
2232 			}
2233 			if (bigenough) {
2234 				cookie.nfsuquad[0] = *tl++;
2235 				if (v3)
2236 					cookie.nfsuquad[1] = *tl++;
2237 			} else if (v3)
2238 				tl += 2;
2239 			else
2240 				tl++;
2241 			more_dirs = fxdr_unsigned(int, *tl);
2242 		}
2243 		/*
2244 		 * If at end of rpc data, get the eof boolean
2245 		 */
2246 		if (!more_dirs) {
2247 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2248 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2249 		}
2250 		m_freem(mrep);
2251 	}
2252 	/*
2253 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2254 	 * by increasing d_reclen for the last record.
2255 	 */
2256 	if (blksiz > 0) {
2257 		left = DIRBLKSIZ - blksiz;
2258 		dp->d_reclen += left;
2259 		uiop->uio_iov->iov_base += left;
2260 		uiop->uio_iov->iov_len -= left;
2261 		uiop->uio_offset += left;
2262 		uiop->uio_resid -= left;
2263 	}
2264 
2265 	/*
2266 	 * We are now either at the end of the directory or have filled the
2267 	 * block.
2268 	 */
2269 	if (bigenough)
2270 		dnp->n_direofoffset = uiop->uio_offset;
2271 	else {
2272 		if (uiop->uio_resid > 0)
2273 			printf("EEK! readdirrpc resid > 0\n");
2274 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2275 		*cookiep = cookie;
2276 	}
2277 nfsmout:
2278 	return (error);
2279 }
2280 
2281 /*
2282  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2283  */
2284 int
2285 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2286 {
2287 	int len, left;
2288 	struct dirent *dp;
2289 	u_int32_t *tl;
2290 	caddr_t cp;
2291 	int32_t t1, t2;
2292 	struct vnode *newvp;
2293 	nfsuint64 *cookiep;
2294 	caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2295 	struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2296 	struct nameidata nami, *ndp = &nami;
2297 	struct componentname *cnp = &ndp->ni_cnd;
2298 	nfsuint64 cookie;
2299 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2300 	struct nfsnode *dnp = VTONFS(vp), *np;
2301 	nfsfh_t *fhp;
2302 	u_quad_t fileno;
2303 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2304 	int attrflag, fhsize;
2305 
2306 #ifndef nolint
2307 	dp = (struct dirent *)0;
2308 #endif
2309 #ifndef DIAGNOSTIC
2310 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2311 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2312 		panic("nfs readdirplusrpc bad uio");
2313 #endif
2314 	ndp->ni_dvp = vp;
2315 	newvp = NULLVP;
2316 
2317 	/*
2318 	 * If there is no cookie, assume directory was stale.
2319 	 */
2320 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2321 	if (cookiep)
2322 		cookie = *cookiep;
2323 	else
2324 		return (NFSERR_BAD_COOKIE);
2325 	/*
2326 	 * Loop around doing readdir rpc's of size nm_readdirsize
2327 	 * truncated to a multiple of DIRBLKSIZ.
2328 	 * The stopping criteria is EOF or buffer full.
2329 	 */
2330 	while (more_dirs && bigenough) {
2331 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2332 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2333 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2334 		nfsm_fhtom(vp, 1);
2335  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2336 		*tl++ = cookie.nfsuquad[0];
2337 		*tl++ = cookie.nfsuquad[1];
2338 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2339 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2340 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2341 		*tl = txdr_unsigned(nmp->nm_rsize);
2342 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2343 		nfsm_postop_attr(vp, attrflag);
2344 		if (error) {
2345 			m_freem(mrep);
2346 			goto nfsmout;
2347 		}
2348 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2349 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2350 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2351 		more_dirs = fxdr_unsigned(int, *tl);
2352 
2353 		/* loop thru the dir entries, doctoring them to 4bsd form */
2354 		while (more_dirs && bigenough) {
2355 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2356 			fileno = fxdr_hyper(tl);
2357 			len = fxdr_unsigned(int, *(tl + 2));
2358 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2359 				error = EBADRPC;
2360 				m_freem(mrep);
2361 				goto nfsmout;
2362 			}
2363 			tlen = nfsm_rndup(len);
2364 			if (tlen == len)
2365 				tlen += 4;	/* To ensure null termination*/
2366 			left = DIRBLKSIZ - blksiz;
2367 			if ((tlen + DIRHDSIZ) > left) {
2368 				dp->d_reclen += left;
2369 				uiop->uio_iov->iov_base += left;
2370 				uiop->uio_iov->iov_len -= left;
2371 				uiop->uio_offset += left;
2372 				uiop->uio_resid -= left;
2373 				blksiz = 0;
2374 			}
2375 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2376 				bigenough = 0;
2377 			if (bigenough) {
2378 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2379 				dp->d_fileno = (int)fileno;
2380 				dp->d_namlen = len;
2381 				dp->d_reclen = tlen + DIRHDSIZ;
2382 				dp->d_type = DT_UNKNOWN;
2383 				blksiz += dp->d_reclen;
2384 				if (blksiz == DIRBLKSIZ)
2385 					blksiz = 0;
2386 				uiop->uio_offset += DIRHDSIZ;
2387 				uiop->uio_resid -= DIRHDSIZ;
2388 				uiop->uio_iov->iov_base += DIRHDSIZ;
2389 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2390 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2391 				cnp->cn_namelen = len;
2392 				nfsm_mtouio(uiop, len);
2393 				cp = uiop->uio_iov->iov_base;
2394 				tlen -= len;
2395 				*cp = '\0';
2396 				uiop->uio_iov->iov_base += tlen;
2397 				uiop->uio_iov->iov_len -= tlen;
2398 				uiop->uio_offset += tlen;
2399 				uiop->uio_resid -= tlen;
2400 			} else
2401 				nfsm_adv(nfsm_rndup(len));
2402 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2403 			if (bigenough) {
2404 				cookie.nfsuquad[0] = *tl++;
2405 				cookie.nfsuquad[1] = *tl++;
2406 			} else
2407 				tl += 2;
2408 
2409 			/*
2410 			 * Since the attributes are before the file handle
2411 			 * (sigh), we must skip over the attributes and then
2412 			 * come back and get them.
2413 			 */
2414 			attrflag = fxdr_unsigned(int, *tl);
2415 			if (attrflag) {
2416 			    dpossav1 = dpos;
2417 			    mdsav1 = md;
2418 			    nfsm_adv(NFSX_V3FATTR);
2419 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2420 			    doit = fxdr_unsigned(int, *tl);
2421 			    if (doit) {
2422 				nfsm_getfh(fhp, fhsize, 1);
2423 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2424 				    VREF(vp);
2425 				    newvp = vp;
2426 				    np = dnp;
2427 				} else {
2428 				    error = nfs_nget(vp->v_mount, fhp,
2429 					fhsize, &np);
2430 				    if (error)
2431 					doit = 0;
2432 				    else
2433 					newvp = NFSTOV(np);
2434 				}
2435 			    }
2436 			    if (doit && bigenough) {
2437 				dpossav2 = dpos;
2438 				dpos = dpossav1;
2439 				mdsav2 = md;
2440 				md = mdsav1;
2441 				nfsm_loadattr(newvp, (struct vattr *)0);
2442 				dpos = dpossav2;
2443 				md = mdsav2;
2444 				dp->d_type =
2445 				    IFTODT(VTTOIF(np->n_vattr.va_type));
2446 				ndp->ni_vp = newvp;
2447 			        cache_enter(ndp->ni_dvp, NCPNULL, ndp->ni_vp, cnp);
2448 			    }
2449 			} else {
2450 			    /* Just skip over the file handle */
2451 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2452 			    i = fxdr_unsigned(int, *tl);
2453 			    nfsm_adv(nfsm_rndup(i));
2454 			}
2455 			if (newvp != NULLVP) {
2456 			    if (newvp == vp)
2457 				vrele(newvp);
2458 			    else
2459 				vput(newvp);
2460 			    newvp = NULLVP;
2461 			}
2462 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2463 			more_dirs = fxdr_unsigned(int, *tl);
2464 		}
2465 		/*
2466 		 * If at end of rpc data, get the eof boolean
2467 		 */
2468 		if (!more_dirs) {
2469 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2470 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2471 		}
2472 		m_freem(mrep);
2473 	}
2474 	/*
2475 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2476 	 * by increasing d_reclen for the last record.
2477 	 */
2478 	if (blksiz > 0) {
2479 		left = DIRBLKSIZ - blksiz;
2480 		dp->d_reclen += left;
2481 		uiop->uio_iov->iov_base += left;
2482 		uiop->uio_iov->iov_len -= left;
2483 		uiop->uio_offset += left;
2484 		uiop->uio_resid -= left;
2485 	}
2486 
2487 	/*
2488 	 * We are now either at the end of the directory or have filled the
2489 	 * block.
2490 	 */
2491 	if (bigenough)
2492 		dnp->n_direofoffset = uiop->uio_offset;
2493 	else {
2494 		if (uiop->uio_resid > 0)
2495 			printf("EEK! readdirplusrpc resid > 0\n");
2496 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2497 		*cookiep = cookie;
2498 	}
2499 nfsmout:
2500 	if (newvp != NULLVP) {
2501 	        if (newvp == vp)
2502 			vrele(newvp);
2503 		else
2504 			vput(newvp);
2505 		newvp = NULLVP;
2506 	}
2507 	return (error);
2508 }
2509 
2510 /*
2511  * Silly rename. To make the NFS filesystem that is stateless look a little
2512  * more like the "ufs" a remove of an active vnode is translated to a rename
2513  * to a funny looking filename that is removed by nfs_inactive on the
2514  * nfsnode. There is the potential for another process on a different client
2515  * to create the same funny name between the nfs_lookitup() fails and the
2516  * nfs_rename() completes, but...
2517  */
2518 static int
2519 nfs_sillyrename(dvp, vp, cnp)
2520 	struct vnode *dvp, *vp;
2521 	struct componentname *cnp;
2522 {
2523 	struct sillyrename *sp;
2524 	struct nfsnode *np;
2525 	int error;
2526 
2527 	cache_purge(dvp);
2528 	np = VTONFS(vp);
2529 #ifndef DIAGNOSTIC
2530 	if (vp->v_type == VDIR)
2531 		panic("nfs: sillyrename dir");
2532 #endif
2533 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2534 		M_NFSREQ, M_WAITOK);
2535 	sp->s_cred = crdup(cnp->cn_cred);
2536 	sp->s_dvp = dvp;
2537 	VREF(dvp);
2538 
2539 	/* Fudge together a funny name */
2540 	sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2541 
2542 	/* Try lookitups until we get one that isn't there */
2543 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2544 		cnp->cn_td, (struct nfsnode **)0) == 0) {
2545 		sp->s_name[4]++;
2546 		if (sp->s_name[4] > 'z') {
2547 			error = EINVAL;
2548 			goto bad;
2549 		}
2550 	}
2551 	error = nfs_renameit(dvp, cnp, sp);
2552 	if (error)
2553 		goto bad;
2554 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2555 		cnp->cn_td, &np);
2556 	np->n_sillyrename = sp;
2557 	return (0);
2558 bad:
2559 	vrele(sp->s_dvp);
2560 	crfree(sp->s_cred);
2561 	free((caddr_t)sp, M_NFSREQ);
2562 	return (error);
2563 }
2564 
2565 /*
2566  * Look up a file name and optionally either update the file handle or
2567  * allocate an nfsnode, depending on the value of npp.
2568  * npp == NULL	--> just do the lookup
2569  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2570  *			handled too
2571  * *npp != NULL --> update the file handle in the vnode
2572  */
2573 static int
2574 nfs_lookitup(dvp, name, len, cred, td, npp)
2575 	struct vnode *dvp;
2576 	const char *name;
2577 	int len;
2578 	struct ucred *cred;
2579 	struct thread *td;
2580 	struct nfsnode **npp;
2581 {
2582 	u_int32_t *tl;
2583 	caddr_t cp;
2584 	int32_t t1, t2;
2585 	struct vnode *newvp = (struct vnode *)0;
2586 	struct nfsnode *np, *dnp = VTONFS(dvp);
2587 	caddr_t bpos, dpos, cp2;
2588 	int error = 0, fhlen, attrflag;
2589 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2590 	nfsfh_t *nfhp;
2591 	int v3 = NFS_ISV3(dvp);
2592 
2593 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2594 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2595 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2596 	nfsm_fhtom(dvp, v3);
2597 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2598 	nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2599 	if (npp && !error) {
2600 		nfsm_getfh(nfhp, fhlen, v3);
2601 		if (*npp) {
2602 		    np = *npp;
2603 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2604 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2605 			np->n_fhp = &np->n_fh;
2606 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2607 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2608 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2609 		    np->n_fhsize = fhlen;
2610 		    newvp = NFSTOV(np);
2611 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2612 		    VREF(dvp);
2613 		    newvp = dvp;
2614 		} else {
2615 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2616 		    if (error) {
2617 			m_freem(mrep);
2618 			return (error);
2619 		    }
2620 		    newvp = NFSTOV(np);
2621 		}
2622 		if (v3) {
2623 			nfsm_postop_attr(newvp, attrflag);
2624 			if (!attrflag && *npp == NULL) {
2625 				m_freem(mrep);
2626 				if (newvp == dvp)
2627 					vrele(newvp);
2628 				else
2629 					vput(newvp);
2630 				return (ENOENT);
2631 			}
2632 		} else
2633 			nfsm_loadattr(newvp, (struct vattr *)0);
2634 	}
2635 	nfsm_reqdone;
2636 	if (npp && *npp == NULL) {
2637 		if (error) {
2638 			if (newvp) {
2639 				if (newvp == dvp)
2640 					vrele(newvp);
2641 				else
2642 					vput(newvp);
2643 			}
2644 		} else
2645 			*npp = np;
2646 	}
2647 	return (error);
2648 }
2649 
2650 /*
2651  * Nfs Version 3 commit rpc
2652  */
2653 int
2654 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2655 {
2656 	caddr_t cp;
2657 	u_int32_t *tl;
2658 	int32_t t1, t2;
2659 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2660 	caddr_t bpos, dpos, cp2;
2661 	int error = 0, wccflag = NFSV3_WCCRATTR;
2662 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2663 
2664 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2665 		return (0);
2666 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2667 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2668 	nfsm_fhtom(vp, 1);
2669 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2670 	txdr_hyper(offset, tl);
2671 	tl += 2;
2672 	*tl = txdr_unsigned(cnt);
2673 	nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2674 	nfsm_wcc_data(vp, wccflag);
2675 	if (!error) {
2676 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2677 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2678 			NFSX_V3WRITEVERF)) {
2679 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2680 				NFSX_V3WRITEVERF);
2681 			error = NFSERR_STALEWRITEVERF;
2682 		}
2683 	}
2684 	nfsm_reqdone;
2685 	return (error);
2686 }
2687 
2688 /*
2689  * Kludge City..
2690  * - make nfs_bmap() essentially a no-op that does no translation
2691  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2692  *   (Maybe I could use the process's page mapping, but I was concerned that
2693  *    Kernel Write might not be enabled and also figured copyout() would do
2694  *    a lot more work than bcopy() and also it currently happens in the
2695  *    context of the swapper process (2).
2696  */
2697 static int
2698 nfs_bmap(ap)
2699 	struct vop_bmap_args /* {
2700 		struct vnode *a_vp;
2701 		daddr_t  a_bn;
2702 		struct vnode **a_vpp;
2703 		daddr_t *a_bnp;
2704 		int *a_runp;
2705 		int *a_runb;
2706 	} */ *ap;
2707 {
2708 	struct vnode *vp = ap->a_vp;
2709 
2710 	if (ap->a_vpp != NULL)
2711 		*ap->a_vpp = vp;
2712 	if (ap->a_bnp != NULL)
2713 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2714 	if (ap->a_runp != NULL)
2715 		*ap->a_runp = 0;
2716 	if (ap->a_runb != NULL)
2717 		*ap->a_runb = 0;
2718 	return (0);
2719 }
2720 
2721 /*
2722  * Strategy routine.
2723  * For async requests when nfsiod(s) are running, queue the request by
2724  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2725  * request.
2726  */
2727 static int
2728 nfs_strategy(ap)
2729 	struct vop_strategy_args *ap;
2730 {
2731 	struct buf *bp = ap->a_bp;
2732 	struct thread *td;
2733 	int error = 0;
2734 
2735 	KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2736 	KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2737 
2738 	if (bp->b_flags & B_PHYS)
2739 		panic("nfs physio");
2740 
2741 	if (bp->b_flags & B_ASYNC)
2742 		td = NULL;
2743 	else
2744 		td = curthread;	/* XXX */
2745 
2746 	/*
2747 	 * If the op is asynchronous and an i/o daemon is waiting
2748 	 * queue the request, wake it up and wait for completion
2749 	 * otherwise just do it ourselves.
2750 	 */
2751 	if ((bp->b_flags & B_ASYNC) == 0 ||
2752 		nfs_asyncio(bp, td))
2753 		error = nfs_doio(bp, td);
2754 	return (error);
2755 }
2756 
2757 /*
2758  * Mmap a file
2759  *
2760  * NB Currently unsupported.
2761  */
2762 /* ARGSUSED */
2763 static int
2764 nfs_mmap(ap)
2765 	struct vop_mmap_args /* {
2766 		struct vnode *a_vp;
2767 		int  a_fflags;
2768 		struct ucred *a_cred;
2769 		struct thread *a_td;
2770 	} */ *ap;
2771 {
2772 
2773 	return (EINVAL);
2774 }
2775 
2776 /*
2777  * fsync vnode op. Just call nfs_flush() with commit == 1.
2778  */
2779 /* ARGSUSED */
2780 static int
2781 nfs_fsync(ap)
2782 	struct vop_fsync_args /* {
2783 		struct vnodeop_desc *a_desc;
2784 		struct vnode * a_vp;
2785 		struct ucred * a_cred;
2786 		int  a_waitfor;
2787 		struct thread * a_td;
2788 	} */ *ap;
2789 {
2790 
2791 	return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2792 }
2793 
2794 /*
2795  * Flush all the blocks associated with a vnode.
2796  * 	Walk through the buffer pool and push any dirty pages
2797  *	associated with the vnode.
2798  */
2799 static int
2800 nfs_flush(vp, waitfor, td, commit)
2801 	struct vnode *vp;
2802 	int waitfor;
2803 	struct thread *td;
2804 	int commit;
2805 {
2806 	struct nfsnode *np = VTONFS(vp);
2807 	struct buf *bp;
2808 	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 buf **bvec = NULL;
2815 #ifndef NFS_COMMITBVECSIZ
2816 #define NFS_COMMITBVECSIZ	20
2817 #endif
2818 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2819 	int bvecsize = 0, bveccount;
2820 
2821 	if (nmp->nm_flag & NFSMNT_INT)
2822 		slpflag = PCATCH;
2823 	if (!commit)
2824 		passone = 0;
2825 	/*
2826 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2827 	 * server, but nas not been committed to stable storage on the server
2828 	 * yet. On the first pass, the byte range is worked out and the commit
2829 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2830 	 * job.
2831 	 */
2832 again:
2833 	off = (u_quad_t)-1;
2834 	endoff = 0;
2835 	bvecpos = 0;
2836 	if (NFS_ISV3(vp) && commit) {
2837 		s = splbio();
2838 		/*
2839 		 * Count up how many buffers waiting for a commit.
2840 		 */
2841 		bveccount = 0;
2842 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2843 			nbp = TAILQ_NEXT(bp, b_vnbufs);
2844 			if (BUF_REFCNT(bp) == 0 &&
2845 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2846 				== (B_DELWRI | B_NEEDCOMMIT))
2847 				bveccount++;
2848 		}
2849 		/*
2850 		 * Allocate space to remember the list of bufs to commit.  It is
2851 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2852 		 * If we can't get memory (for whatever reason), we will end up
2853 		 * committing the buffers one-by-one in the loop below.
2854 		 */
2855 		if (bvec != NULL && bvec != bvec_on_stack)
2856 			free(bvec, M_TEMP);
2857 		if (bveccount > NFS_COMMITBVECSIZ) {
2858 			bvec = (struct buf **)
2859 				malloc(bveccount * sizeof(struct buf *),
2860 				       M_TEMP, M_NOWAIT);
2861 			if (bvec == NULL) {
2862 				bvec = bvec_on_stack;
2863 				bvecsize = NFS_COMMITBVECSIZ;
2864 			} else
2865 				bvecsize = bveccount;
2866 		} else {
2867 			bvec = bvec_on_stack;
2868 			bvecsize = NFS_COMMITBVECSIZ;
2869 		}
2870 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2871 			nbp = TAILQ_NEXT(bp, b_vnbufs);
2872 			if (bvecpos >= bvecsize)
2873 				break;
2874 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2875 			    (B_DELWRI | B_NEEDCOMMIT) ||
2876 			    BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
2877 				continue;
2878 			bremfree(bp);
2879 			/*
2880 			 * NOTE: we are not clearing B_DONE here, so we have
2881 			 * to do it later on in this routine if we intend to
2882 			 * initiate I/O on the bp.
2883 			 *
2884 			 * Note: to avoid loopback deadlocks, we do not
2885 			 * assign b_runningbufspace.
2886 			 */
2887 			bp->b_flags |= B_WRITEINPROG;
2888 			vfs_busy_pages(bp, 1);
2889 
2890 			/*
2891 			 * bp is protected by being locked, but nbp is not
2892 			 * and vfs_busy_pages() may sleep.  We have to
2893 			 * recalculate nbp.
2894 			 */
2895 			nbp = TAILQ_NEXT(bp, b_vnbufs);
2896 
2897 			/*
2898 			 * A list of these buffers is kept so that the
2899 			 * second loop knows which buffers have actually
2900 			 * been committed. This is necessary, since there
2901 			 * may be a race between the commit rpc and new
2902 			 * uncommitted writes on the file.
2903 			 */
2904 			bvec[bvecpos++] = bp;
2905 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2906 				bp->b_dirtyoff;
2907 			if (toff < off)
2908 				off = toff;
2909 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2910 			if (toff > endoff)
2911 				endoff = toff;
2912 		}
2913 		splx(s);
2914 	}
2915 	if (bvecpos > 0) {
2916 		/*
2917 		 * Commit data on the server, as required.  Note that
2918 		 * nfs_commit will use the vnode's cred for the commit.
2919 		 */
2920 		retv = nfs_commit(vp, off, (int)(endoff - off), td);
2921 
2922 		if (retv == NFSERR_STALEWRITEVERF)
2923 			nfs_clearcommit(vp->v_mount);
2924 
2925 		/*
2926 		 * Now, either mark the blocks I/O done or mark the
2927 		 * blocks dirty, depending on whether the commit
2928 		 * succeeded.
2929 		 */
2930 		for (i = 0; i < bvecpos; i++) {
2931 			bp = bvec[i];
2932 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
2933 			if (retv) {
2934 				/*
2935 				 * Error, leave B_DELWRI intact
2936 				 */
2937 				vfs_unbusy_pages(bp);
2938 				brelse(bp);
2939 			} else {
2940 				/*
2941 				 * Success, remove B_DELWRI ( bundirty() ).
2942 				 *
2943 				 * b_dirtyoff/b_dirtyend seem to be NFS
2944 				 * specific.  We should probably move that
2945 				 * into bundirty(). XXX
2946 				 */
2947 				s = splbio();
2948 				vp->v_numoutput++;
2949 				bp->b_flags |= B_ASYNC;
2950 				bundirty(bp);
2951 				bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
2952 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2953 				splx(s);
2954 				biodone(bp);
2955 			}
2956 		}
2957 	}
2958 
2959 	/*
2960 	 * Start/do any write(s) that are required.
2961 	 */
2962 loop:
2963 	s = splbio();
2964 	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2965 		nbp = TAILQ_NEXT(bp, b_vnbufs);
2966 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
2967 			if (waitfor != MNT_WAIT || passone)
2968 				continue;
2969 			error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
2970 			    "nfsfsync", slpflag, slptimeo);
2971 			splx(s);
2972 			if (error == 0)
2973 				panic("nfs_fsync: inconsistent lock");
2974 			if (error == ENOLCK)
2975 				goto loop;
2976 			if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2977 				error = EINTR;
2978 				goto done;
2979 			}
2980 			if (slpflag == PCATCH) {
2981 				slpflag = 0;
2982 				slptimeo = 2 * hz;
2983 			}
2984 			goto loop;
2985 		}
2986 		if ((bp->b_flags & B_DELWRI) == 0)
2987 			panic("nfs_fsync: not dirty");
2988 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2989 			BUF_UNLOCK(bp);
2990 			continue;
2991 		}
2992 		bremfree(bp);
2993 		if (passone || !commit)
2994 		    bp->b_flags |= B_ASYNC;
2995 		else
2996 		    bp->b_flags |= B_ASYNC | B_WRITEINPROG;
2997 		splx(s);
2998 		VOP_BWRITE(bp->b_vp, bp);
2999 		goto loop;
3000 	}
3001 	splx(s);
3002 	if (passone) {
3003 		passone = 0;
3004 		goto again;
3005 	}
3006 	if (waitfor == MNT_WAIT) {
3007 		while (vp->v_numoutput) {
3008 			vp->v_flag |= VBWAIT;
3009 			error = tsleep((caddr_t)&vp->v_numoutput,
3010 				slpflag, "nfsfsync", slptimeo);
3011 			if (error) {
3012 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3013 				error = EINTR;
3014 				goto done;
3015 			    }
3016 			    if (slpflag == PCATCH) {
3017 				slpflag = 0;
3018 				slptimeo = 2 * hz;
3019 			    }
3020 			}
3021 		}
3022 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
3023 			goto loop;
3024 		}
3025 	}
3026 	if (np->n_flag & NWRITEERR) {
3027 		error = np->n_error;
3028 		np->n_flag &= ~NWRITEERR;
3029 	}
3030 done:
3031 	if (bvec != NULL && bvec != bvec_on_stack)
3032 		free(bvec, M_TEMP);
3033 	return (error);
3034 }
3035 
3036 /*
3037  * NFS advisory byte-level locks.
3038  * Currently unsupported.
3039  */
3040 static int
3041 nfs_advlock(ap)
3042 	struct vop_advlock_args /* {
3043 		struct vnode *a_vp;
3044 		caddr_t  a_id;
3045 		int  a_op;
3046 		struct flock *a_fl;
3047 		int  a_flags;
3048 	} */ *ap;
3049 {
3050 	struct nfsnode *np = VTONFS(ap->a_vp);
3051 
3052 	/*
3053 	 * The following kludge is to allow diskless support to work
3054 	 * until a real NFS lockd is implemented. Basically, just pretend
3055 	 * that this is a local lock.
3056 	 */
3057 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3058 }
3059 
3060 /*
3061  * Print out the contents of an nfsnode.
3062  */
3063 static int
3064 nfs_print(ap)
3065 	struct vop_print_args /* {
3066 		struct vnode *a_vp;
3067 	} */ *ap;
3068 {
3069 	struct vnode *vp = ap->a_vp;
3070 	struct nfsnode *np = VTONFS(vp);
3071 
3072 	printf("tag VT_NFS, fileid %ld fsid 0x%x",
3073 		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3074 	if (vp->v_type == VFIFO)
3075 		fifo_printinfo(vp);
3076 	printf("\n");
3077 	return (0);
3078 }
3079 
3080 /*
3081  * Just call nfs_writebp() with the force argument set to 1.
3082  *
3083  * NOTE: B_DONE may or may not be set in a_bp on call.
3084  */
3085 static int
3086 nfs_bwrite(ap)
3087 	struct vop_bwrite_args /* {
3088 		struct vnode *a_bp;
3089 	} */ *ap;
3090 {
3091 	return (nfs_writebp(ap->a_bp, 1, curthread));
3092 }
3093 
3094 /*
3095  * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3096  * the force flag is one and it also handles the B_NEEDCOMMIT flag.  We set
3097  * B_CACHE if this is a VMIO buffer.
3098  */
3099 int
3100 nfs_writebp(bp, force, td)
3101 	struct buf *bp;
3102 	int force;
3103 	struct thread *td;
3104 {
3105 	int s;
3106 	int oldflags = bp->b_flags;
3107 #if 0
3108 	int retv = 1;
3109 	off_t off;
3110 #endif
3111 
3112 	if (BUF_REFCNT(bp) == 0)
3113 		panic("bwrite: buffer is not locked???");
3114 
3115 	if (bp->b_flags & B_INVAL) {
3116 		brelse(bp);
3117 		return(0);
3118 	}
3119 
3120 	bp->b_flags |= B_CACHE;
3121 
3122 	/*
3123 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3124 	 */
3125 
3126 	s = splbio();
3127 	bundirty(bp);
3128 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3129 
3130 	bp->b_vp->v_numoutput++;
3131 	splx(s);
3132 
3133 	/*
3134 	 * Note: to avoid loopback deadlocks, we do not
3135 	 * assign b_runningbufspace.
3136 	 */
3137 	vfs_busy_pages(bp, 1);
3138 
3139 	if (force)
3140 		bp->b_flags |= B_WRITEINPROG;
3141 	BUF_KERNPROC(bp);
3142 	VOP_STRATEGY(bp->b_vp, bp);
3143 
3144 	if( (oldflags & B_ASYNC) == 0) {
3145 		int rtval = biowait(bp);
3146 
3147 		if (oldflags & B_DELWRI) {
3148 			s = splbio();
3149 			reassignbuf(bp, bp->b_vp);
3150 			splx(s);
3151 		}
3152 
3153 		brelse(bp);
3154 		return (rtval);
3155 	}
3156 
3157 	return (0);
3158 }
3159 
3160 /*
3161  * nfs special file access vnode op.
3162  * Essentially just get vattr and then imitate iaccess() since the device is
3163  * local to the client.
3164  */
3165 static int
3166 nfsspec_access(ap)
3167 	struct vop_access_args /* {
3168 		struct vnode *a_vp;
3169 		int  a_mode;
3170 		struct ucred *a_cred;
3171 		struct thread *a_td;
3172 	} */ *ap;
3173 {
3174 	struct vattr *vap;
3175 	gid_t *gp;
3176 	struct ucred *cred = ap->a_cred;
3177 	struct vnode *vp = ap->a_vp;
3178 	mode_t mode = ap->a_mode;
3179 	struct vattr vattr;
3180 	int i;
3181 	int error;
3182 
3183 	/*
3184 	 * Disallow write attempts on filesystems mounted read-only;
3185 	 * unless the file is a socket, fifo, or a block or character
3186 	 * device resident on the filesystem.
3187 	 */
3188 	if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3189 		switch (vp->v_type) {
3190 		case VREG:
3191 		case VDIR:
3192 		case VLNK:
3193 			return (EROFS);
3194 		default:
3195 			break;
3196 		}
3197 	}
3198 	/*
3199 	 * If you're the super-user,
3200 	 * you always get access.
3201 	 */
3202 	if (cred->cr_uid == 0)
3203 		return (0);
3204 	vap = &vattr;
3205 	error = VOP_GETATTR(vp, vap, ap->a_td);
3206 	if (error)
3207 		return (error);
3208 	/*
3209 	 * Access check is based on only one of owner, group, public.
3210 	 * If not owner, then check group. If not a member of the
3211 	 * group, then check public access.
3212 	 */
3213 	if (cred->cr_uid != vap->va_uid) {
3214 		mode >>= 3;
3215 		gp = cred->cr_groups;
3216 		for (i = 0; i < cred->cr_ngroups; i++, gp++)
3217 			if (vap->va_gid == *gp)
3218 				goto found;
3219 		mode >>= 3;
3220 found:
3221 		;
3222 	}
3223 	error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3224 	return (error);
3225 }
3226 
3227 /*
3228  * Read wrapper for special devices.
3229  */
3230 static int
3231 nfsspec_read(ap)
3232 	struct vop_read_args /* {
3233 		struct vnode *a_vp;
3234 		struct uio *a_uio;
3235 		int  a_ioflag;
3236 		struct ucred *a_cred;
3237 	} */ *ap;
3238 {
3239 	struct nfsnode *np = VTONFS(ap->a_vp);
3240 
3241 	/*
3242 	 * Set access flag.
3243 	 */
3244 	np->n_flag |= NACC;
3245 	getnanotime(&np->n_atim);
3246 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3247 }
3248 
3249 /*
3250  * Write wrapper for special devices.
3251  */
3252 static int
3253 nfsspec_write(ap)
3254 	struct vop_write_args /* {
3255 		struct vnode *a_vp;
3256 		struct uio *a_uio;
3257 		int  a_ioflag;
3258 		struct ucred *a_cred;
3259 	} */ *ap;
3260 {
3261 	struct nfsnode *np = VTONFS(ap->a_vp);
3262 
3263 	/*
3264 	 * Set update flag.
3265 	 */
3266 	np->n_flag |= NUPD;
3267 	getnanotime(&np->n_mtim);
3268 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3269 }
3270 
3271 /*
3272  * Close wrapper for special devices.
3273  *
3274  * Update the times on the nfsnode then do device close.
3275  */
3276 static int
3277 nfsspec_close(ap)
3278 	struct vop_close_args /* {
3279 		struct vnode *a_vp;
3280 		int  a_fflag;
3281 		struct ucred *a_cred;
3282 		struct thread *a_td;
3283 	} */ *ap;
3284 {
3285 	struct vnode *vp = ap->a_vp;
3286 	struct nfsnode *np = VTONFS(vp);
3287 	struct vattr vattr;
3288 
3289 	if (np->n_flag & (NACC | NUPD)) {
3290 		np->n_flag |= NCHG;
3291 		if (vp->v_usecount == 1 &&
3292 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3293 			VATTR_NULL(&vattr);
3294 			if (np->n_flag & NACC)
3295 				vattr.va_atime = np->n_atim;
3296 			if (np->n_flag & NUPD)
3297 				vattr.va_mtime = np->n_mtim;
3298 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3299 		}
3300 	}
3301 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3302 }
3303 
3304 /*
3305  * Read wrapper for fifos.
3306  */
3307 static int
3308 nfsfifo_read(ap)
3309 	struct vop_read_args /* {
3310 		struct vnode *a_vp;
3311 		struct uio *a_uio;
3312 		int  a_ioflag;
3313 		struct ucred *a_cred;
3314 	} */ *ap;
3315 {
3316 	struct nfsnode *np = VTONFS(ap->a_vp);
3317 
3318 	/*
3319 	 * Set access flag.
3320 	 */
3321 	np->n_flag |= NACC;
3322 	getnanotime(&np->n_atim);
3323 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3324 }
3325 
3326 /*
3327  * Write wrapper for fifos.
3328  */
3329 static int
3330 nfsfifo_write(ap)
3331 	struct vop_write_args /* {
3332 		struct vnode *a_vp;
3333 		struct uio *a_uio;
3334 		int  a_ioflag;
3335 		struct ucred *a_cred;
3336 	} */ *ap;
3337 {
3338 	struct nfsnode *np = VTONFS(ap->a_vp);
3339 
3340 	/*
3341 	 * Set update flag.
3342 	 */
3343 	np->n_flag |= NUPD;
3344 	getnanotime(&np->n_mtim);
3345 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3346 }
3347 
3348 /*
3349  * Close wrapper for fifos.
3350  *
3351  * Update the times on the nfsnode then do fifo close.
3352  */
3353 static int
3354 nfsfifo_close(ap)
3355 	struct vop_close_args /* {
3356 		struct vnode *a_vp;
3357 		int  a_fflag;
3358 		struct thread *a_td;
3359 	} */ *ap;
3360 {
3361 	struct vnode *vp = ap->a_vp;
3362 	struct nfsnode *np = VTONFS(vp);
3363 	struct vattr vattr;
3364 	struct timespec ts;
3365 
3366 	if (np->n_flag & (NACC | NUPD)) {
3367 		getnanotime(&ts);
3368 		if (np->n_flag & NACC)
3369 			np->n_atim = ts;
3370 		if (np->n_flag & NUPD)
3371 			np->n_mtim = ts;
3372 		np->n_flag |= NCHG;
3373 		if (vp->v_usecount == 1 &&
3374 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3375 			VATTR_NULL(&vattr);
3376 			if (np->n_flag & NACC)
3377 				vattr.va_atime = np->n_atim;
3378 			if (np->n_flag & NUPD)
3379 				vattr.va_mtime = np->n_mtim;
3380 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3381 		}
3382 	}
3383 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3384 }
3385 
3386