xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision d0fed6c87ddc40a8bffa6f99e7433ddfc864dd83)
1 /*	$NetBSD: nfs_vnops.c,v 1.72 1997/03/05 02:08:06 mycroft Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
39  */
40 
41 
42 /*
43  * vnode op calls for Sun NFS version 2 and 3
44  */
45 
46 #include <sys/param.h>
47 #include <sys/proc.h>
48 #include <sys/kernel.h>
49 #include <sys/systm.h>
50 #include <sys/resourcevar.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/buf.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/conf.h>
57 #include <sys/namei.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 
63 #include <vm/vm.h>
64 
65 #include <miscfs/fifofs/fifo.h>
66 #include <miscfs/genfs/genfs.h>
67 #include <miscfs/specfs/specdev.h>
68 
69 #include <nfs/rpcv2.h>
70 #include <nfs/nfsproto.h>
71 #include <nfs/nfs.h>
72 #include <nfs/nfsnode.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nqnfs.h>
77 #include <nfs/nfs_var.h>
78 
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 
83 /* Defs */
84 #define	TRUE	1
85 #define	FALSE	0
86 
87 /*
88  * Global vfs data structures for nfs
89  */
90 int (**nfsv2_vnodeop_p) __P((void *));
91 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
92 	{ &vop_default_desc, vn_default_error },
93 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
94 	{ &vop_create_desc, nfs_create },		/* create */
95 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
96 	{ &vop_open_desc, nfs_open },			/* open */
97 	{ &vop_close_desc, nfs_close },			/* close */
98 	{ &vop_access_desc, nfs_access },		/* access */
99 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
100 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
101 	{ &vop_read_desc, nfs_read },			/* read */
102 	{ &vop_write_desc, nfs_write },			/* write */
103 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
104 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
105 	{ &vop_poll_desc, nfs_poll },			/* poll */
106 #ifdef Lite2_integrated
107 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
108 #endif
109 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
110 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
111 	{ &vop_seek_desc, nfs_seek },			/* seek */
112 	{ &vop_remove_desc, nfs_remove },		/* remove */
113 	{ &vop_link_desc, nfs_link },			/* link */
114 	{ &vop_rename_desc, nfs_rename },		/* rename */
115 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
116 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
117 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
118 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
119 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
120 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
121 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
122 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
123 	{ &vop_lock_desc, nfs_lock },			/* lock */
124 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
125 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
126 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
127 	{ &vop_print_desc, nfs_print },			/* print */
128 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
129 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
130 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
131 	{ &vop_blkatoff_desc, nfs_blkatoff },		/* blkatoff */
132 	{ &vop_valloc_desc, nfs_valloc },		/* valloc */
133 	{ &vop_reallocblks_desc, nfs_reallocblks },	/* reallocblks */
134 	{ &vop_vfree_desc, nfs_vfree },			/* vfree */
135 	{ &vop_truncate_desc, nfs_truncate },		/* truncate */
136 	{ &vop_update_desc, nfs_update },		/* update */
137 	{ &vop_bwrite_desc, nfs_bwrite },		/* bwrite */
138 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
139 };
140 struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
141 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
142 
143 /*
144  * Special device vnode ops
145  */
146 int (**spec_nfsv2nodeop_p) __P((void *));
147 struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
148 	{ &vop_default_desc, vn_default_error },
149 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
150 	{ &vop_create_desc, spec_create },		/* create */
151 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
152 	{ &vop_open_desc, spec_open },			/* open */
153 	{ &vop_close_desc, nfsspec_close },		/* close */
154 	{ &vop_access_desc, nfsspec_access },		/* access */
155 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
156 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
157 	{ &vop_read_desc, nfsspec_read },		/* read */
158 	{ &vop_write_desc, nfsspec_write },		/* write */
159 	{ &vop_lease_desc, spec_lease_check },		/* lease */
160 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
161 	{ &vop_poll_desc, spec_poll },			/* poll */
162 #ifdef Lite2_integrated
163 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
164 #endif
165 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
166 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
167 	{ &vop_seek_desc, spec_seek },			/* seek */
168 	{ &vop_remove_desc, spec_remove },		/* remove */
169 	{ &vop_link_desc, spec_link },			/* link */
170 	{ &vop_rename_desc, spec_rename },		/* rename */
171 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
172 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
173 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
174 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
175 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
176 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
177 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
178 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
179 	{ &vop_lock_desc, nfs_lock },			/* lock */
180 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
181 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
182 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
183 	{ &vop_print_desc, nfs_print },			/* print */
184 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
185 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
186 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
187 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
188 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
189 	{ &vop_reallocblks_desc, spec_reallocblks },	/* reallocblks */
190 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
191 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
192 	{ &vop_update_desc, nfs_update },		/* update */
193 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
194 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
195 };
196 struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
197 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
198 
199 #ifdef FIFO
200 int (**fifo_nfsv2nodeop_p) __P((void *));
201 struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
202 	{ &vop_default_desc, vn_default_error },
203 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
204 	{ &vop_create_desc, fifo_create },		/* create */
205 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
206 	{ &vop_open_desc, fifo_open },			/* open */
207 	{ &vop_close_desc, nfsfifo_close },		/* close */
208 	{ &vop_access_desc, nfsspec_access },		/* access */
209 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
210 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
211 	{ &vop_read_desc, nfsfifo_read },		/* read */
212 	{ &vop_write_desc, nfsfifo_write },		/* write */
213 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
214 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
215 	{ &vop_poll_desc, fifo_poll },			/* poll */
216 #ifdef Lite2_integrated
217 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
218 #endif
219 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
220 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
221 	{ &vop_seek_desc, fifo_seek },			/* seek */
222 	{ &vop_remove_desc, fifo_remove },		/* remove */
223 	{ &vop_link_desc, fifo_link },			/* link */
224 	{ &vop_rename_desc, fifo_rename },		/* rename */
225 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
226 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
227 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
228 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
229 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
230 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
231 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
232 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
233 	{ &vop_lock_desc, nfs_lock },			/* lock */
234 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
235 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
236 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
237 	{ &vop_print_desc, nfs_print },			/* print */
238 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
239 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
240 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
241 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
242 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
243 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
244 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
245 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
246 	{ &vop_update_desc, nfs_update },		/* update */
247 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
248 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
249 };
250 struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
251 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
252 #endif /* FIFO */
253 
254 /*
255  * Global variables
256  */
257 extern u_int32_t nfs_true, nfs_false;
258 extern u_int32_t nfs_xdrneg1;
259 extern struct nfsstats nfsstats;
260 extern nfstype nfsv3_type[9];
261 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
262 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
263 int nfs_numasync = 0;
264 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
265 
266 /*
267  * nfs null call from vfs.
268  */
269 int
270 nfs_null(vp, cred, procp)
271 	struct vnode *vp;
272 	struct ucred *cred;
273 	struct proc *procp;
274 {
275 	caddr_t bpos, dpos;
276 	int error = 0;
277 	struct mbuf *mreq, *mrep, *md, *mb;
278 
279 	nfsm_reqhead(vp, NFSPROC_NULL, 0);
280 	nfsm_request(vp, NFSPROC_NULL, procp, cred);
281 	nfsm_reqdone;
282 	return (error);
283 }
284 
285 /*
286  * nfs access vnode op.
287  * For nfs version 2, just return ok. File accesses may fail later.
288  * For nfs version 3, use the access rpc to check accessibility. If file modes
289  * are changed on the server, accesses might still fail later.
290  */
291 int
292 nfs_access(v)
293 	void *v;
294 {
295 	struct vop_access_args /* {
296 		struct vnode *a_vp;
297 		int  a_mode;
298 		struct ucred *a_cred;
299 		struct proc *a_p;
300 	} */ *ap = v;
301 	register struct vnode *vp = ap->a_vp;
302 	register u_int32_t *tl;
303 	register caddr_t cp;
304 	register int32_t t1, t2;
305 	caddr_t bpos, dpos, cp2;
306 	int error = 0, attrflag;
307 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
308 	u_int32_t mode, rmode;
309 	int v3 = NFS_ISV3(vp);
310 
311 	/*
312 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
313 	 * ufs_access() locally using the vattr. This may not be correct,
314 	 * since the server may apply other access criteria such as
315 	 * client uid-->server uid mapping that we do not know about, but
316 	 * this is better than just returning anything that is lying about
317 	 * in the cache.
318 	 */
319 	if (v3) {
320 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
321 		nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
322 		nfsm_fhtom(vp, v3);
323 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
324 		if (ap->a_mode & VREAD)
325 			mode = NFSV3ACCESS_READ;
326 		else
327 			mode = 0;
328 		if (vp->v_type == VDIR) {
329 			if (ap->a_mode & VWRITE)
330 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
331 					 NFSV3ACCESS_DELETE);
332 			if (ap->a_mode & VEXEC)
333 				mode |= NFSV3ACCESS_LOOKUP;
334 		} else {
335 			if (ap->a_mode & VWRITE)
336 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
337 			if (ap->a_mode & VEXEC)
338 				mode |= NFSV3ACCESS_EXECUTE;
339 		}
340 		*tl = txdr_unsigned(mode);
341 		nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
342 		nfsm_postop_attr(vp, attrflag);
343 		if (!error) {
344 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
345 			rmode = fxdr_unsigned(u_int32_t, *tl);
346 			/*
347 			 * The NFS V3 spec does not clarify whether or not
348 			 * the returned access bits can be a superset of
349 			 * the ones requested, so...
350 			 */
351 			if ((rmode & mode) != mode)
352 				error = EACCES;
353 		}
354 		nfsm_reqdone;
355 		if (error)
356 			return (error);
357 	} else
358 		return (nfsspec_access(ap));
359 	/*
360 	 * Disallow write attempts on filesystems mounted read-only;
361 	 * unless the file is a socket, fifo, or a block or character
362 	 * device resident on the filesystem.
363 	 */
364 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
365 		switch (vp->v_type) {
366 		case VREG:
367 		case VDIR:
368 		case VLNK:
369 			return (EROFS);
370 		default:
371 			break;
372 		}
373 	}
374 	return (0);
375 }
376 
377 /*
378  * nfs open vnode op
379  * Check to see if the type is ok
380  * and that deletion is not in progress.
381  * For paged in text files, you will need to flush the page cache
382  * if consistency is lost.
383  */
384 /* ARGSUSED */
385 int
386 nfs_open(v)
387 	void *v;
388 {
389 	struct vop_open_args /* {
390 		struct vnode *a_vp;
391 		int  a_mode;
392 		struct ucred *a_cred;
393 		struct proc *a_p;
394 	} */ *ap = v;
395 	register struct vnode *vp = ap->a_vp;
396 	struct nfsnode *np = VTONFS(vp);
397 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
398 	struct vattr vattr;
399 	int error;
400 
401 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
402 #ifdef DIAGNOSTIC
403 		printf("open eacces vtyp=%d\n",vp->v_type);
404 #endif
405 		return (EACCES);
406 	}
407 	/*
408 	 * Get a valid lease. If cached data is stale, flush it.
409 	 */
410 	if (nmp->nm_flag & NFSMNT_NQNFS) {
411 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
412 		    do {
413 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
414 			    ap->a_p);
415 		    } while (error == NQNFS_EXPIRED);
416 		    if (error)
417 			return (error);
418 		    if (np->n_lrev != np->n_brev ||
419 			(np->n_flag & NQNFSNONCACHE)) {
420 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
421 				ap->a_p, 1)) == EINTR)
422 				return (error);
423 			(void) vnode_pager_uncache(vp);
424 			np->n_brev = np->n_lrev;
425 		    }
426 		}
427 	} else {
428 		if (np->n_flag & NMODIFIED) {
429 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
430 				ap->a_p, 1)) == EINTR)
431 				return (error);
432 			(void) vnode_pager_uncache(vp);
433 			np->n_attrstamp = 0;
434 			if (vp->v_type == VDIR)
435 				np->n_direofoffset = 0;
436 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
437 			if (error)
438 				return (error);
439 			np->n_mtime = vattr.va_mtime.tv_sec;
440 		} else {
441 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
442 			if (error)
443 				return (error);
444 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
445 				if (vp->v_type == VDIR)
446 					np->n_direofoffset = 0;
447 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
448 					ap->a_cred, ap->a_p, 1)) == EINTR)
449 					return (error);
450 				(void) vnode_pager_uncache(vp);
451 				np->n_mtime = vattr.va_mtime.tv_sec;
452 			}
453 		}
454 	}
455 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
456 		np->n_attrstamp = 0; /* For Open/Close consistency */
457 	return (0);
458 }
459 
460 /*
461  * nfs close vnode op
462  * What an NFS client should do upon close after writing is a debatable issue.
463  * Most NFS clients push delayed writes to the server upon close, basically for
464  * two reasons:
465  * 1 - So that any write errors may be reported back to the client process
466  *     doing the close system call. By far the two most likely errors are
467  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
468  * 2 - To put a worst case upper bound on cache inconsistency between
469  *     multiple clients for the file.
470  * There is also a consistency problem for Version 2 of the protocol w.r.t.
471  * not being able to tell if other clients are writing a file concurrently,
472  * since there is no way of knowing if the changed modify time in the reply
473  * is only due to the write for this client.
474  * (NFS Version 3 provides weak cache consistency data in the reply that
475  *  should be sufficient to detect and handle this case.)
476  *
477  * The current code does the following:
478  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
479  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
480  *                     or commit them (this satisfies 1 and 2 except for the
481  *                     case where the server crashes after this close but
482  *                     before the commit RPC, which is felt to be "good
483  *                     enough". Changing the last argument to nfs_flush() to
484  *                     a 1 would force a commit operation, if it is felt a
485  *                     commit is necessary now.
486  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
487  *                     1 should be dealt with via an fsync() system call for
488  *                     cases where write errors are important.
489  */
490 /* ARGSUSED */
491 int
492 nfs_close(v)
493 	void *v;
494 {
495 	struct vop_close_args /* {
496 		struct vnodeop_desc *a_desc;
497 		struct vnode *a_vp;
498 		int  a_fflag;
499 		struct ucred *a_cred;
500 		struct proc *a_p;
501 	} */ *ap = v;
502 	register struct vnode *vp = ap->a_vp;
503 	register struct nfsnode *np = VTONFS(vp);
504 	int error = 0;
505 
506 	if (vp->v_type == VREG) {
507 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
508 		(np->n_flag & NMODIFIED)) {
509 		if (NFS_ISV3(vp)) {
510 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
511 		    np->n_flag &= ~NMODIFIED;
512 		} else
513 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
514 		np->n_attrstamp = 0;
515 	    }
516 	    if (np->n_flag & NWRITEERR) {
517 		np->n_flag &= ~NWRITEERR;
518 		error = np->n_error;
519 	    }
520 	}
521 	return (error);
522 }
523 
524 /*
525  * nfs getattr call from vfs.
526  */
527 int
528 nfs_getattr(v)
529 	void *v;
530 {
531 	struct vop_getattr_args /* {
532 		struct vnode *a_vp;
533 		struct vattr *a_vap;
534 		struct ucred *a_cred;
535 		struct proc *a_p;
536 	} */ *ap = v;
537 	register struct vnode *vp = ap->a_vp;
538 	register struct nfsnode *np = VTONFS(vp);
539 	register caddr_t cp;
540 	register u_int32_t *tl;
541 	register int32_t t1, t2;
542 	caddr_t bpos, dpos;
543 	int error = 0;
544 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
545 	int v3 = NFS_ISV3(vp);
546 
547 	/*
548 	 * Update local times for special files.
549 	 */
550 	if (np->n_flag & (NACC | NUPD))
551 		np->n_flag |= NCHG;
552 	/*
553 	 * First look in the cache.
554 	 */
555 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
556 		return (0);
557 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
558 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
559 	nfsm_fhtom(vp, v3);
560 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
561 	if (!error)
562 		nfsm_loadattr(vp, ap->a_vap);
563 	nfsm_reqdone;
564 	return (error);
565 }
566 
567 /*
568  * nfs setattr call.
569  */
570 int
571 nfs_setattr(v)
572 	void *v;
573 {
574 	struct vop_setattr_args /* {
575 		struct vnodeop_desc *a_desc;
576 		struct vnode *a_vp;
577 		struct vattr *a_vap;
578 		struct ucred *a_cred;
579 		struct proc *a_p;
580 	} */ *ap = v;
581 	register struct vnode *vp = ap->a_vp;
582 	register struct nfsnode *np = VTONFS(vp);
583 	register struct vattr *vap = ap->a_vap;
584 	int error = 0;
585 	u_quad_t tsize = 0;
586 
587 	/*
588 	 * Setting of flags is not supported.
589 	 */
590 	if (vap->va_flags != VNOVAL)
591 		return (EOPNOTSUPP);
592 
593 	/*
594 	 * Disallow write attempts if the filesystem is mounted read-only.
595 	 */
596   	if ((vap->va_uid != (uid_t)VNOVAL ||
597 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
598 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
599 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
600 		return (EROFS);
601 	if (vap->va_size != VNOVAL) {
602  		switch (vp->v_type) {
603  		case VDIR:
604  			return (EISDIR);
605  		case VCHR:
606  		case VBLK:
607  		case VSOCK:
608  		case VFIFO:
609 			if (vap->va_mtime.tv_sec == VNOVAL &&
610 			    vap->va_atime.tv_sec == VNOVAL &&
611 			    vap->va_mode == (u_short)VNOVAL &&
612 			    vap->va_uid == (uid_t)VNOVAL &&
613 			    vap->va_gid == (gid_t)VNOVAL)
614 				return (0);
615  			vap->va_size = VNOVAL;
616  			break;
617  		default:
618 			/*
619 			 * Disallow write attempts if the filesystem is
620 			 * mounted read-only.
621 			 */
622 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
623 				return (EROFS);
624  			vnode_pager_setsize(vp, (u_long)vap->va_size);
625  			if (vap->va_size == 0)
626  				error = nfs_vinvalbuf(vp, 0,
627  				     ap->a_cred, ap->a_p, 1);
628 			else
629 				error = nfs_vinvalbuf(vp, V_SAVE,
630 				     ap->a_cred, ap->a_p, 1);
631 			if (error) {
632 				vnode_pager_setsize(vp, (u_long)np->n_size);
633 				return (error);
634 			}
635  			tsize = np->n_size;
636  			np->n_size = np->n_vattr.va_size = vap->va_size;
637   		}
638   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
639 		vap->va_atime.tv_sec != VNOVAL) &&
640 		vp->v_type == VREG &&
641   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
642 		 ap->a_p, 1)) == EINTR)
643 		return (error);
644 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
645 	if (error && vap->va_size != VNOVAL) {
646 		np->n_size = np->n_vattr.va_size = tsize;
647 		vnode_pager_setsize(vp, (u_long)np->n_size);
648 	}
649 	return (error);
650 }
651 
652 /*
653  * Do an nfs setattr rpc.
654  */
655 int
656 nfs_setattrrpc(vp, vap, cred, procp)
657 	register struct vnode *vp;
658 	register struct vattr *vap;
659 	struct ucred *cred;
660 	struct proc *procp;
661 {
662 	register struct nfsv2_sattr *sp;
663 	register caddr_t cp;
664 	register int32_t t1, t2;
665 	caddr_t bpos, dpos, cp2;
666 	u_int32_t *tl;
667 	int error = 0, wccflag = NFSV3_WCCRATTR;
668 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
669 	int v3 = NFS_ISV3(vp);
670 
671 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
672 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
673 	nfsm_fhtom(vp, v3);
674 	if (v3) {
675 		if (vap->va_mode != (u_short)VNOVAL) {
676 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
677 			*tl++ = nfs_true;
678 			*tl = txdr_unsigned(vap->va_mode);
679 		} else {
680 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
681 			*tl = nfs_false;
682 		}
683 		if (vap->va_uid != (uid_t)VNOVAL) {
684 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
685 			*tl++ = nfs_true;
686 			*tl = txdr_unsigned(vap->va_uid);
687 		} else {
688 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
689 			*tl = nfs_false;
690 		}
691 		if (vap->va_gid != (gid_t)VNOVAL) {
692 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
693 			*tl++ = nfs_true;
694 			*tl = txdr_unsigned(vap->va_gid);
695 		} else {
696 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
697 			*tl = nfs_false;
698 		}
699 		if (vap->va_size != VNOVAL) {
700 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
701 			*tl++ = nfs_true;
702 			txdr_hyper(&vap->va_size, tl);
703 		} else {
704 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
705 			*tl = nfs_false;
706 		}
707 		if (vap->va_atime.tv_sec != VNOVAL) {
708 			if (vap->va_atime.tv_sec != time.tv_sec) {
709 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
710 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
711 				txdr_nfsv3time(&vap->va_atime, tl);
712 			} else {
713 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
714 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
715 			}
716 		} else {
717 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
718 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
719 		}
720 		if (vap->va_mtime.tv_sec != VNOVAL) {
721 			if (vap->va_mtime.tv_sec != time.tv_sec) {
722 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
723 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
724 				txdr_nfsv3time(&vap->va_mtime, tl);
725 			} else {
726 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
727 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
728 			}
729 		} else {
730 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
731 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
732 		}
733 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
734 		*tl = nfs_false;
735 	} else {
736 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
737 		if (vap->va_mode == (u_short)VNOVAL)
738 			sp->sa_mode = nfs_xdrneg1;
739 		else
740 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
741 		if (vap->va_uid == (uid_t)VNOVAL)
742 			sp->sa_uid = nfs_xdrneg1;
743 		else
744 			sp->sa_uid = txdr_unsigned(vap->va_uid);
745 		if (vap->va_gid == (gid_t)VNOVAL)
746 			sp->sa_gid = nfs_xdrneg1;
747 		else
748 			sp->sa_gid = txdr_unsigned(vap->va_gid);
749 		sp->sa_size = txdr_unsigned(vap->va_size);
750 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
751 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
752 	}
753 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
754 	if (v3) {
755 		nfsm_wcc_data(vp, wccflag);
756 	} else
757 		nfsm_loadattr(vp, (struct vattr *)0);
758 	nfsm_reqdone;
759 	return (error);
760 }
761 
762 /*
763  * nfs lookup call, one step at a time...
764  * First look in cache
765  * If not found, unlock the directory nfsnode and do the rpc
766  */
767 int
768 nfs_lookup(v)
769 	void *v;
770 {
771 	struct vop_lookup_args /* {
772 		struct vnodeop_desc *a_desc;
773 		struct vnode *a_dvp;
774 		struct vnode **a_vpp;
775 		struct componentname *a_cnp;
776 	} */ *ap = v;
777 	register struct componentname *cnp = ap->a_cnp;
778 	register struct vnode *dvp = ap->a_dvp;
779 	register struct vnode **vpp = ap->a_vpp;
780 	register int flags = cnp->cn_flags;
781 	register struct vnode *newvp;
782 	register u_int32_t *tl;
783 	register caddr_t cp;
784 	register int32_t t1, t2;
785 	struct nfsmount *nmp;
786 	caddr_t bpos, dpos, cp2;
787 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
788 	long len;
789 	nfsfh_t *fhp;
790 	struct nfsnode *np;
791 	int lockparent, wantparent, error = 0, attrflag, fhsize;
792 	int v3 = NFS_ISV3(dvp);
793 
794 	*vpp = NULLVP;
795 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
796 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
797 		return (EROFS);
798 	if (dvp->v_type != VDIR)
799 		return (ENOTDIR);
800 	lockparent = flags & LOCKPARENT;
801 	wantparent = flags & (LOCKPARENT|WANTPARENT);
802 	nmp = VFSTONFS(dvp->v_mount);
803 	np = VTONFS(dvp);
804 	if ((error = cache_lookup(dvp, vpp, cnp)) != 0 && error != ENOENT) {
805 		struct vattr vattr;
806 		int vpid;
807 
808 		newvp = *vpp;
809 		vpid = newvp->v_id;
810 		/*
811 		 * See the comment starting `Step through' in ufs/ufs_lookup.c
812 		 * for an explanation of the locking protocol
813 		 */
814 		if (dvp == newvp) {
815 			VREF(newvp);
816 			error = 0;
817 		} else
818 #ifdef Lite2_integrated
819 			error = vget(newvp, LK_EXCLUSIVE, p);
820 #else
821 			error = vget(newvp, 1);
822 #endif
823 		if (!error) {
824 			if (vpid == newvp->v_id) {
825 			   if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
826 			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
827 				nfsstats.lookupcache_hits++;
828 				if (cnp->cn_nameiop != LOOKUP &&
829 				    (flags & ISLASTCN))
830 					cnp->cn_flags |= SAVENAME;
831 				return (0);
832 			   }
833 			   cache_purge(newvp);
834 			}
835 			vrele(newvp);
836 		}
837 		*vpp = NULLVP;
838 	}
839 	error = 0;
840 	newvp = NULLVP;
841 	nfsstats.lookupcache_misses++;
842 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
843 	len = cnp->cn_namelen;
844 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
845 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
846 	nfsm_fhtom(dvp, v3);
847 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
848 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
849 	if (error) {
850 		nfsm_postop_attr(dvp, attrflag);
851 		m_freem(mrep);
852 		goto nfsmout;
853 	}
854 	nfsm_getfh(fhp, fhsize, v3);
855 
856 	/*
857 	 * Handle RENAME case...
858 	 */
859 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
860 		if (NFS_CMPFH(np, fhp, fhsize)) {
861 			m_freem(mrep);
862 			return (EISDIR);
863 		}
864 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
865 		if (error) {
866 			m_freem(mrep);
867 			return (error);
868 		}
869 		newvp = NFSTOV(np);
870 		if (v3) {
871 			nfsm_postop_attr(newvp, attrflag);
872 			nfsm_postop_attr(dvp, attrflag);
873 		} else
874 			nfsm_loadattr(newvp, (struct vattr *)0);
875 		*vpp = newvp;
876 		m_freem(mrep);
877 		cnp->cn_flags |= SAVENAME;
878 		return (0);
879 	}
880 
881 	if (NFS_CMPFH(np, fhp, fhsize)) {
882 		VREF(dvp);
883 		newvp = dvp;
884 	} else {
885 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
886 		if (error) {
887 			m_freem(mrep);
888 			return (error);
889 		}
890 		newvp = NFSTOV(np);
891 	}
892 	if (v3) {
893 		nfsm_postop_attr(newvp, attrflag);
894 		nfsm_postop_attr(dvp, attrflag);
895 	} else
896 		nfsm_loadattr(newvp, (struct vattr *)0);
897 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
898 		cnp->cn_flags |= SAVENAME;
899 	if ((cnp->cn_flags & MAKEENTRY) &&
900 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
901 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
902 		cache_enter(dvp, newvp, cnp);
903 	}
904 	*vpp = newvp;
905 	nfsm_reqdone;
906 	if (error) {
907 		if (newvp != NULLVP)
908 			vrele(newvp);
909 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
910 		    (flags & ISLASTCN) && error == ENOENT) {
911 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
912 				error = EROFS;
913 			else
914 				error = EJUSTRETURN;
915 		}
916 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
917 			cnp->cn_flags |= SAVENAME;
918 	}
919 	return (error);
920 }
921 
922 /*
923  * nfs read call.
924  * Just call nfs_bioread() to do the work.
925  */
926 int
927 nfs_read(v)
928 	void *v;
929 {
930 	struct vop_read_args /* {
931 		struct vnode *a_vp;
932 		struct uio *a_uio;
933 		int  a_ioflag;
934 		struct ucred *a_cred;
935 	} */ *ap = v;
936 	register struct vnode *vp = ap->a_vp;
937 
938 	if (vp->v_type != VREG)
939 		return (EPERM);
940 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
941 }
942 
943 /*
944  * nfs readlink call
945  */
946 int
947 nfs_readlink(v)
948 	void *v;
949 {
950 	struct vop_readlink_args /* {
951 		struct vnode *a_vp;
952 		struct uio *a_uio;
953 		struct ucred *a_cred;
954 	} */ *ap = v;
955 	register struct vnode *vp = ap->a_vp;
956 
957 	if (vp->v_type != VLNK)
958 		return (EPERM);
959 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
960 }
961 
962 /*
963  * Do a readlink rpc.
964  * Called by nfs_doio() from below the buffer cache.
965  */
966 int
967 nfs_readlinkrpc(vp, uiop, cred)
968 	register struct vnode *vp;
969 	struct uio *uiop;
970 	struct ucred *cred;
971 {
972 	register u_int32_t *tl;
973 	register caddr_t cp;
974 	register int32_t t1, t2;
975 	caddr_t bpos, dpos, cp2;
976 	int error = 0, len, attrflag;
977 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
978 	int v3 = NFS_ISV3(vp);
979 
980 	nfsstats.rpccnt[NFSPROC_READLINK]++;
981 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
982 	nfsm_fhtom(vp, v3);
983 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
984 	if (v3)
985 		nfsm_postop_attr(vp, attrflag);
986 	if (!error) {
987 		nfsm_strsiz(len, NFS_MAXPATHLEN);
988 		nfsm_mtouio(uiop, len);
989 	}
990 	nfsm_reqdone;
991 	return (error);
992 }
993 
994 /*
995  * nfs read rpc call
996  * Ditto above
997  */
998 int
999 nfs_readrpc(vp, uiop, cred)
1000 	register struct vnode *vp;
1001 	struct uio *uiop;
1002 	struct ucred *cred;
1003 {
1004 	register u_int32_t *tl;
1005 	register caddr_t cp;
1006 	register int32_t t1, t2;
1007 	caddr_t bpos, dpos, cp2;
1008 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1009 	struct nfsmount *nmp;
1010 	int error = 0, len, retlen, tsiz, eof, attrflag;
1011 	int v3 = NFS_ISV3(vp);
1012 
1013 #ifndef nolint
1014 	eof = 0;
1015 #endif
1016 	nmp = VFSTONFS(vp->v_mount);
1017 	tsiz = uiop->uio_resid;
1018 	if (uiop->uio_offset + tsiz > 0xffffffff && !v3)
1019 		return (EFBIG);
1020 	while (tsiz > 0) {
1021 		nfsstats.rpccnt[NFSPROC_READ]++;
1022 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1023 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1024 		nfsm_fhtom(vp, v3);
1025 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1026 		if (v3) {
1027 			txdr_hyper(&uiop->uio_offset, tl);
1028 			*(tl + 2) = txdr_unsigned(len);
1029 		} else {
1030 			*tl++ = txdr_unsigned(uiop->uio_offset);
1031 			*tl++ = txdr_unsigned(len);
1032 			*tl = 0;
1033 		}
1034 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp, cred);
1035 		if (v3) {
1036 			nfsm_postop_attr(vp, attrflag);
1037 			if (error) {
1038 				m_freem(mrep);
1039 				goto nfsmout;
1040 			}
1041 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1042 			eof = fxdr_unsigned(int, *(tl + 1));
1043 		} else
1044 			nfsm_loadattr(vp, (struct vattr *)0);
1045 		nfsm_strsiz(retlen, nmp->nm_rsize);
1046 		nfsm_mtouio(uiop, retlen);
1047 		m_freem(mrep);
1048 		tsiz -= retlen;
1049 		if (v3) {
1050 			if (eof || retlen == 0)
1051 				tsiz = 0;
1052 		} else if (retlen < len)
1053 			tsiz = 0;
1054 	}
1055 nfsmout:
1056 	return (error);
1057 }
1058 
1059 /*
1060  * nfs write call
1061  */
1062 int
1063 nfs_writerpc(vp, uiop, cred, iomode, must_commit)
1064 	register struct vnode *vp;
1065 	register struct uio *uiop;
1066 	struct ucred *cred;
1067 	int *iomode, *must_commit;
1068 {
1069 	register u_int32_t *tl;
1070 	register caddr_t cp;
1071 	register int32_t t1, t2, backup;
1072 	caddr_t bpos, dpos, cp2;
1073 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1074 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1075 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1076 	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1077 
1078 #ifndef DIAGNOSTIC
1079 	if (uiop->uio_iovcnt != 1)
1080 		panic("nfs: writerpc iovcnt > 1");
1081 #endif
1082 	*must_commit = 0;
1083 	tsiz = uiop->uio_resid;
1084 	if (uiop->uio_offset + tsiz > 0xffffffff && !v3)
1085 		return (EFBIG);
1086 	while (tsiz > 0) {
1087 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1088 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1089 		nfsm_reqhead(vp, NFSPROC_WRITE,
1090 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1091 		nfsm_fhtom(vp, v3);
1092 		if (v3) {
1093 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1094 			txdr_hyper(&uiop->uio_offset, tl);
1095 			tl += 2;
1096 			*tl++ = txdr_unsigned(len);
1097 			*tl++ = txdr_unsigned(*iomode);
1098 			*tl = txdr_unsigned(len);
1099 		} else {
1100 			register u_int32_t x;
1101 
1102 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1103 			/* Set both "begin" and "current" to non-garbage. */
1104 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1105 			*tl++ = x;      /* "begin offset" */
1106 			*tl++ = x;      /* "current offset" */
1107 			x = txdr_unsigned(len);
1108 			*tl++ = x;      /* total to this offset */
1109 			*tl = x;        /* size of this write */
1110 
1111 		}
1112 		nfsm_uiotom(uiop, len);
1113 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp, cred);
1114 		if (v3) {
1115 			wccflag = NFSV3_WCCCHK;
1116 			nfsm_wcc_data(vp, wccflag);
1117 			if (!error) {
1118 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1119 					+ NFSX_V3WRITEVERF);
1120 				rlen = fxdr_unsigned(int, *tl++);
1121 				if (rlen == 0) {
1122 					error = NFSERR_IO;
1123 					break;
1124 				} else if (rlen < len) {
1125 					backup = len - rlen;
1126 					uiop->uio_iov->iov_base -= backup;
1127 					uiop->uio_iov->iov_len += backup;
1128 					uiop->uio_offset -= backup;
1129 					uiop->uio_resid += backup;
1130 					len = rlen;
1131 				}
1132 				commit = fxdr_unsigned(int, *tl++);
1133 
1134 				/*
1135 				 * Return the lowest committment level
1136 				 * obtained by any of the RPCs.
1137 				 */
1138 				if (committed == NFSV3WRITE_FILESYNC)
1139 					committed = commit;
1140 				else if (committed == NFSV3WRITE_DATASYNC &&
1141 					commit == NFSV3WRITE_UNSTABLE)
1142 					committed = commit;
1143 				if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0) {
1144 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1145 					NFSX_V3WRITEVERF);
1146 				    nmp->nm_flag |= NFSMNT_HASWRITEVERF;
1147 				} else if (bcmp((caddr_t)tl,
1148 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1149 				    *must_commit = 1;
1150 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1151 					NFSX_V3WRITEVERF);
1152 				}
1153 			}
1154 		} else
1155 		    nfsm_loadattr(vp, (struct vattr *)0);
1156 		if (wccflag)
1157 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1158 		m_freem(mrep);
1159 		tsiz -= len;
1160 	}
1161 nfsmout:
1162 	*iomode = committed;
1163 	if (error)
1164 		uiop->uio_resid = tsiz;
1165 	return (error);
1166 }
1167 
1168 /*
1169  * nfs mknod rpc
1170  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1171  * mode set to specify the file type and the size field for rdev.
1172  */
1173 int
1174 nfs_mknodrpc(dvp, vpp, cnp, vap)
1175 	register struct vnode *dvp;
1176 	register struct vnode **vpp;
1177 	register struct componentname *cnp;
1178 	register struct vattr *vap;
1179 {
1180 	register struct nfsv2_sattr *sp;
1181 	register struct nfsv3_sattr *sp3;
1182 	register u_int32_t *tl;
1183 	register caddr_t cp;
1184 	register int32_t t1, t2;
1185 	struct vnode *newvp = (struct vnode *)0;
1186 	struct nfsnode *np;
1187 	char *cp2;
1188 	caddr_t bpos, dpos;
1189 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1190 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1191 	u_int32_t rdev;
1192 	int v3 = NFS_ISV3(dvp);
1193 
1194 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1195 		rdev = txdr_unsigned(vap->va_rdev);
1196 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1197 		rdev = nfs_xdrneg1;
1198 	else {
1199 		VOP_ABORTOP(dvp, cnp);
1200 		vput(dvp);
1201 		return (EOPNOTSUPP);
1202 	}
1203 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1204 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1205 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1206 	nfsm_fhtom(dvp, v3);
1207 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1208 	if (v3) {
1209 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3SRVSATTR);
1210 		*tl++ = vtonfsv3_type(vap->va_type);
1211 		sp3 = (struct nfsv3_sattr *)tl;
1212 		nfsm_v3sattr(sp3, vap);
1213 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1214 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1215 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1216 			*tl = txdr_unsigned(minor(vap->va_rdev));
1217 		}
1218 	} else {
1219 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1220 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1221 		sp->sa_uid = nfs_xdrneg1;
1222 		sp->sa_gid = nfs_xdrneg1;
1223 		sp->sa_size = rdev;
1224 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1225 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1226 	}
1227 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1228 	if (!error) {
1229 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1230 		if (!gotvp) {
1231 			if (newvp) {
1232 				vrele(newvp);
1233 				newvp = (struct vnode *)0;
1234 			}
1235 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1236 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1237 			if (!error)
1238 				newvp = NFSTOV(np);
1239 		}
1240 	}
1241 	if (v3)
1242 		nfsm_wcc_data(dvp, wccflag);
1243 	nfsm_reqdone;
1244 	if (error) {
1245 		if (newvp)
1246 			vrele(newvp);
1247 	} else {
1248 		if (cnp->cn_flags & MAKEENTRY)
1249 			cache_enter(dvp, newvp, cnp);
1250 		*vpp = newvp;
1251 	}
1252 	FREE(cnp->cn_pnbuf, M_NAMEI);
1253 	VTONFS(dvp)->n_flag |= NMODIFIED;
1254 	if (!wccflag)
1255 		VTONFS(dvp)->n_attrstamp = 0;
1256 	vrele(dvp);
1257 	return (error);
1258 }
1259 
1260 /*
1261  * nfs mknod vop
1262  * just call nfs_mknodrpc() to do the work.
1263  */
1264 /* ARGSUSED */
1265 int
1266 nfs_mknod(v)
1267 	void *v;
1268 {
1269 	struct vop_mknod_args /* {
1270 		struct vnode *a_dvp;
1271 		struct vnode **a_vpp;
1272 		struct componentname *a_cnp;
1273 		struct vattr *a_vap;
1274 	} */ *ap = v;
1275 	struct vnode *newvp;
1276 	int error;
1277 
1278 	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1279 	if (!error)
1280 		vrele(newvp);
1281 	return (error);
1282 }
1283 
1284 static u_long create_verf;
1285 /*
1286  * nfs file create call
1287  */
1288 int
1289 nfs_create(v)
1290 	void *v;
1291 {
1292 	struct vop_create_args /* {
1293 		struct vnode *a_dvp;
1294 		struct vnode **a_vpp;
1295 		struct componentname *a_cnp;
1296 		struct vattr *a_vap;
1297 	} */ *ap = v;
1298 	register struct vnode *dvp = ap->a_dvp;
1299 	register struct vattr *vap = ap->a_vap;
1300 	register struct componentname *cnp = ap->a_cnp;
1301 	register struct nfsv2_sattr *sp;
1302 	register struct nfsv3_sattr *sp3;
1303 	register u_int32_t *tl;
1304 	register caddr_t cp;
1305 	register int32_t t1, t2;
1306 	struct nfsnode *np = (struct nfsnode *)0;
1307 	struct vnode *newvp = (struct vnode *)0;
1308 	caddr_t bpos, dpos, cp2;
1309 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1310 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1311 	int v3 = NFS_ISV3(dvp);
1312 
1313 	/*
1314 	 * Oops, not for me..
1315 	 */
1316 	if (vap->va_type == VSOCK)
1317 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1318 
1319 #ifdef VA_EXCLUSIVE
1320 	if (vap->va_vaflags & VA_EXCLUSIVE)
1321 		fmode |= O_EXCL;
1322 #endif
1323 again:
1324 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1325 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1326 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1327 	nfsm_fhtom(dvp, v3);
1328 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1329 	if (v3) {
1330 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1331 		if (fmode & O_EXCL) {
1332 		    *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1333 		    nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1334 		    if (in_ifaddr.tqh_first)
1335 			*tl++ = in_ifaddr.tqh_first->ia_addr.sin_addr.s_addr;
1336 		    else
1337 			*tl++ = create_verf;
1338 		    *tl = ++create_verf;
1339 		} else {
1340 		    *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1341 		    nfsm_build(tl, u_int32_t *, NFSX_V3SRVSATTR);
1342 		    sp3 = (struct nfsv3_sattr *)tl;
1343 		    nfsm_v3sattr(sp3, vap);
1344 		}
1345 	} else {
1346 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1347 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1348 		sp->sa_uid = nfs_xdrneg1;
1349 		sp->sa_gid = nfs_xdrneg1;
1350 		sp->sa_size = 0;
1351 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1352 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1353 	}
1354 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1355 	if (!error) {
1356 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1357 		if (!gotvp) {
1358 			if (newvp) {
1359 				vrele(newvp);
1360 				newvp = (struct vnode *)0;
1361 			}
1362 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1363 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1364 			if (!error)
1365 				newvp = NFSTOV(np);
1366 		}
1367 	}
1368 	if (v3)
1369 		nfsm_wcc_data(dvp, wccflag);
1370 	nfsm_reqdone;
1371 	if (error) {
1372 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1373 			fmode &= ~O_EXCL;
1374 			goto again;
1375 		}
1376 		if (newvp)
1377 			vrele(newvp);
1378 	} else if (v3 && (fmode & O_EXCL))
1379 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1380 	if (!error) {
1381 		if (cnp->cn_flags & MAKEENTRY)
1382 			cache_enter(dvp, newvp, cnp);
1383 		*ap->a_vpp = newvp;
1384 	}
1385 	FREE(cnp->cn_pnbuf, M_NAMEI);
1386 	VTONFS(dvp)->n_flag |= NMODIFIED;
1387 	if (!wccflag)
1388 		VTONFS(dvp)->n_attrstamp = 0;
1389 	vrele(dvp);
1390 	return (error);
1391 }
1392 
1393 /*
1394  * nfs file remove call
1395  * To try and make nfs semantics closer to ufs semantics, a file that has
1396  * other processes using the vnode is renamed instead of removed and then
1397  * removed later on the last close.
1398  * - If v_usecount > 1
1399  *	  If a rename is not already in the works
1400  *	     call nfs_sillyrename() to set it up
1401  *     else
1402  *	  do the remove rpc
1403  */
1404 int
1405 nfs_remove(v)
1406 	void *v;
1407 {
1408 	struct vop_remove_args /* {
1409 		struct vnodeop_desc *a_desc;
1410 		struct vnode * a_dvp;
1411 		struct vnode * a_vp;
1412 		struct componentname * a_cnp;
1413 	} */ *ap = v;
1414 	register struct vnode *vp = ap->a_vp;
1415 	register struct vnode *dvp = ap->a_dvp;
1416 	register struct componentname *cnp = ap->a_cnp;
1417 	register struct nfsnode *np = VTONFS(vp);
1418 	int error = 0;
1419 	struct vattr vattr;
1420 
1421 #ifndef DIAGNOSTIC
1422 	if ((cnp->cn_flags & HASBUF) == 0)
1423 		panic("nfs_remove: no name");
1424 	if (vp->v_usecount < 1)
1425 		panic("nfs_remove: bad v_usecount");
1426 #endif
1427 	if (vp->v_usecount == 1 || (np->n_sillyrename &&
1428 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1429 	    vattr.va_nlink > 1)) {
1430 		/*
1431 		 * Purge the name cache so that the chance of a lookup for
1432 		 * the name succeeding while the remove is in progress is
1433 		 * minimized. Without node locking it can still happen, such
1434 		 * that an I/O op returns ESTALE, but since you get this if
1435 		 * another host removes the file..
1436 		 */
1437 		cache_purge(vp);
1438 		/*
1439 		 * throw away biocache buffers, mainly to avoid
1440 		 * unnecessary delayed writes later.
1441 		 */
1442 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1443 		/* Do the rpc */
1444 		if (error != EINTR)
1445 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1446 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1447 		/*
1448 		 * Kludge City: If the first reply to the remove rpc is lost..
1449 		 *   the reply to the retransmitted request will be ENOENT
1450 		 *   since the file was in fact removed
1451 		 *   Therefore, we cheat and return success.
1452 		 */
1453 		if (error == ENOENT)
1454 			error = 0;
1455 	} else if (!np->n_sillyrename)
1456 		error = nfs_sillyrename(dvp, vp, cnp);
1457 	FREE(cnp->cn_pnbuf, M_NAMEI);
1458 	np->n_attrstamp = 0;
1459 	vrele(dvp);
1460 	vrele(vp);
1461 	return (error);
1462 }
1463 
1464 /*
1465  * nfs file remove rpc called from nfs_inactive
1466  */
1467 int
1468 nfs_removeit(sp)
1469 	register struct sillyrename *sp;
1470 {
1471 
1472 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1473 		(struct proc *)0));
1474 }
1475 
1476 /*
1477  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1478  */
1479 int
1480 nfs_removerpc(dvp, name, namelen, cred, proc)
1481 	register struct vnode *dvp;
1482 	const char *name;
1483 	int namelen;
1484 	struct ucred *cred;
1485 	struct proc *proc;
1486 {
1487 	register u_int32_t *tl;
1488 	register caddr_t cp;
1489 	register int32_t t1, t2;
1490 	caddr_t bpos, dpos, cp2;
1491 	int error = 0, wccflag = NFSV3_WCCRATTR;
1492 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1493 	int v3 = NFS_ISV3(dvp);
1494 
1495 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1496 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1497 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1498 	nfsm_fhtom(dvp, v3);
1499 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1500 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1501 	if (v3)
1502 		nfsm_wcc_data(dvp, wccflag);
1503 	nfsm_reqdone;
1504 	VTONFS(dvp)->n_flag |= NMODIFIED;
1505 	if (!wccflag)
1506 		VTONFS(dvp)->n_attrstamp = 0;
1507 	return (error);
1508 }
1509 
1510 /*
1511  * nfs file rename call
1512  */
1513 int
1514 nfs_rename(v)
1515 	void *v;
1516 {
1517 	struct vop_rename_args  /* {
1518 		struct vnode *a_fdvp;
1519 		struct vnode *a_fvp;
1520 		struct componentname *a_fcnp;
1521 		struct vnode *a_tdvp;
1522 		struct vnode *a_tvp;
1523 		struct componentname *a_tcnp;
1524 	} */ *ap = v;
1525 	register struct vnode *fvp = ap->a_fvp;
1526 	register struct vnode *tvp = ap->a_tvp;
1527 	register struct vnode *fdvp = ap->a_fdvp;
1528 	register struct vnode *tdvp = ap->a_tdvp;
1529 	register struct componentname *tcnp = ap->a_tcnp;
1530 	register struct componentname *fcnp = ap->a_fcnp;
1531 	int error;
1532 
1533 #ifndef DIAGNOSTIC
1534 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1535 	    (fcnp->cn_flags & HASBUF) == 0)
1536 		panic("nfs_rename: no name");
1537 #endif
1538 	/* Check for cross-device rename */
1539 	if ((fvp->v_mount != tdvp->v_mount) ||
1540 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1541 		error = EXDEV;
1542 		goto out;
1543 	}
1544 
1545 	/*
1546 	 * If the tvp exists and is in use, sillyrename it before doing the
1547 	 * rename of the new file over it.
1548 	 */
1549 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1550 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1551 		vrele(tvp);
1552 		tvp = NULL;
1553 	}
1554 
1555 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1556 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1557 		tcnp->cn_proc);
1558 
1559 	if (fvp->v_type == VDIR) {
1560 		if (tvp != NULL && tvp->v_type == VDIR)
1561 			cache_purge(tdvp);
1562 		cache_purge(fdvp);
1563 	}
1564 out:
1565 	if (tdvp == tvp)
1566 		vrele(tdvp);
1567 	else
1568 		vput(tdvp);
1569 	if (tvp)
1570 		vput(tvp);
1571 	vrele(fdvp);
1572 	vrele(fvp);
1573 	/*
1574 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1575 	 */
1576 	if (error == ENOENT)
1577 		error = 0;
1578 	return (error);
1579 }
1580 
1581 /*
1582  * nfs file rename rpc called from nfs_remove() above
1583  */
1584 int
1585 nfs_renameit(sdvp, scnp, sp)
1586 	struct vnode *sdvp;
1587 	struct componentname *scnp;
1588 	register struct sillyrename *sp;
1589 {
1590 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1591 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1592 }
1593 
1594 /*
1595  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1596  */
1597 int
1598 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1599 	register struct vnode *fdvp;
1600 	const char *fnameptr;
1601 	int fnamelen;
1602 	register struct vnode *tdvp;
1603 	const char *tnameptr;
1604 	int tnamelen;
1605 	struct ucred *cred;
1606 	struct proc *proc;
1607 {
1608 	register u_int32_t *tl;
1609 	register caddr_t cp;
1610 	register int32_t t1, t2;
1611 	caddr_t bpos, dpos, cp2;
1612 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1613 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1614 	int v3 = NFS_ISV3(fdvp);
1615 
1616 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1617 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1618 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1619 		nfsm_rndup(tnamelen));
1620 	nfsm_fhtom(fdvp, v3);
1621 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1622 	nfsm_fhtom(tdvp, v3);
1623 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1624 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1625 	if (v3) {
1626 		nfsm_wcc_data(fdvp, fwccflag);
1627 		nfsm_wcc_data(tdvp, twccflag);
1628 	}
1629 	nfsm_reqdone;
1630 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1631 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1632 	if (!fwccflag)
1633 		VTONFS(fdvp)->n_attrstamp = 0;
1634 	if (!twccflag)
1635 		VTONFS(tdvp)->n_attrstamp = 0;
1636 	return (error);
1637 }
1638 
1639 /*
1640  * nfs hard link create call
1641  */
1642 int
1643 nfs_link(v)
1644 	void *v;
1645 {
1646 	struct vop_link_args /* {
1647 		struct vnode *a_dvp;
1648 		struct vnode *a_vp;
1649 		struct componentname *a_cnp;
1650 	} */ *ap = v;
1651 	register struct vnode *vp = ap->a_vp;
1652 	register struct vnode *dvp = ap->a_dvp;
1653 	register struct componentname *cnp = ap->a_cnp;
1654 	register u_int32_t *tl;
1655 	register caddr_t cp;
1656 	register int32_t t1, t2;
1657 	caddr_t bpos, dpos, cp2;
1658 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1659 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1660 	int v3;
1661 
1662 	if (dvp->v_mount != vp->v_mount) {
1663 		VOP_ABORTOP(vp, cnp);
1664 		vput(dvp);
1665 		return (EXDEV);
1666 	}
1667 
1668 	/*
1669 	 * Push all writes to the server, so that the attribute cache
1670 	 * doesn't get "out of sync" with the server.
1671 	 * XXX There should be a better way!
1672 	 */
1673 	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1674 
1675 	v3 = NFS_ISV3(vp);
1676 	nfsstats.rpccnt[NFSPROC_LINK]++;
1677 	nfsm_reqhead(vp, NFSPROC_LINK,
1678 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1679 	nfsm_fhtom(vp, v3);
1680 	nfsm_fhtom(dvp, v3);
1681 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1682 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1683 	if (v3) {
1684 		nfsm_postop_attr(vp, attrflag);
1685 		nfsm_wcc_data(dvp, wccflag);
1686 	}
1687 	nfsm_reqdone;
1688 	FREE(cnp->cn_pnbuf, M_NAMEI);
1689 	VTONFS(dvp)->n_flag |= NMODIFIED;
1690 	if (!attrflag)
1691 		VTONFS(vp)->n_attrstamp = 0;
1692 	if (!wccflag)
1693 		VTONFS(dvp)->n_attrstamp = 0;
1694 	vput(dvp);
1695 	/*
1696 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1697 	 */
1698 	if (error == EEXIST)
1699 		error = 0;
1700 	return (error);
1701 }
1702 
1703 /*
1704  * nfs symbolic link create call
1705  */
1706 int
1707 nfs_symlink(v)
1708 	void *v;
1709 {
1710 	struct vop_symlink_args /* {
1711 		struct vnode *a_dvp;
1712 		struct vnode **a_vpp;
1713 		struct componentname *a_cnp;
1714 		struct vattr *a_vap;
1715 		char *a_target;
1716 	} */ *ap = v;
1717 	register struct vnode *dvp = ap->a_dvp;
1718 	register struct vattr *vap = ap->a_vap;
1719 	register struct componentname *cnp = ap->a_cnp;
1720 	register struct nfsv2_sattr *sp;
1721 	register struct nfsv3_sattr *sp3;
1722 	register u_int32_t *tl;
1723 	register caddr_t cp;
1724 	register int32_t t1, t2;
1725 	caddr_t bpos, dpos, cp2;
1726 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1727 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1728 	struct vnode *newvp = (struct vnode *)0;
1729 	int v3 = NFS_ISV3(dvp);
1730 
1731 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1732 	slen = strlen(ap->a_target);
1733 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1734 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1735 	nfsm_fhtom(dvp, v3);
1736 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1737 	if (v3) {
1738 		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1739 		nfsm_v3sattr(sp3, vap);
1740 	}
1741 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1742 	if (!v3) {
1743 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1744 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1745 		sp->sa_uid = nfs_xdrneg1;
1746 		sp->sa_gid = nfs_xdrneg1;
1747 		sp->sa_size = nfs_xdrneg1;
1748 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1749 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1750 	}
1751 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1752 	if (v3) {
1753 		if (!error)
1754 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1755 		nfsm_wcc_data(dvp, wccflag);
1756 	}
1757 	nfsm_reqdone;
1758 	if (newvp)
1759 		vrele(newvp);
1760 	FREE(cnp->cn_pnbuf, M_NAMEI);
1761 	VTONFS(dvp)->n_flag |= NMODIFIED;
1762 	if (!wccflag)
1763 		VTONFS(dvp)->n_attrstamp = 0;
1764 	vrele(dvp);
1765 	/*
1766 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1767 	 */
1768 	if (error == EEXIST)
1769 		error = 0;
1770 	return (error);
1771 }
1772 
1773 /*
1774  * nfs make dir call
1775  */
1776 int
1777 nfs_mkdir(v)
1778 	void *v;
1779 {
1780 	struct vop_mkdir_args /* {
1781 		struct vnode *a_dvp;
1782 		struct vnode **a_vpp;
1783 		struct componentname *a_cnp;
1784 		struct vattr *a_vap;
1785 	} */ *ap = v;
1786 	register struct vnode *dvp = ap->a_dvp;
1787 	register struct vattr *vap = ap->a_vap;
1788 	register struct componentname *cnp = ap->a_cnp;
1789 	register struct nfsv2_sattr *sp;
1790 	register struct nfsv3_sattr *sp3;
1791 	register u_int32_t *tl;
1792 	register caddr_t cp;
1793 	register int32_t t1, t2;
1794 	register int len;
1795 	struct nfsnode *np = (struct nfsnode *)0;
1796 	struct vnode *newvp = (struct vnode *)0;
1797 	caddr_t bpos, dpos, cp2;
1798 	int error = 0, wccflag = NFSV3_WCCRATTR;
1799 	int gotvp = 0;
1800 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1801 	int v3 = NFS_ISV3(dvp);
1802 
1803 	len = cnp->cn_namelen;
1804 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1805 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1806 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1807 	nfsm_fhtom(dvp, v3);
1808 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1809 	if (v3) {
1810 		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1811 		nfsm_v3sattr(sp3, vap);
1812 	} else {
1813 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1814 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1815 		sp->sa_uid = nfs_xdrneg1;
1816 		sp->sa_gid = nfs_xdrneg1;
1817 		sp->sa_size = nfs_xdrneg1;
1818 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1819 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1820 	}
1821 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1822 	if (!error)
1823 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1824 	if (v3)
1825 		nfsm_wcc_data(dvp, wccflag);
1826 	nfsm_reqdone;
1827 	VTONFS(dvp)->n_flag |= NMODIFIED;
1828 	if (!wccflag)
1829 		VTONFS(dvp)->n_attrstamp = 0;
1830 	/*
1831 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1832 	 * if we can succeed in looking up the directory.
1833 	 */
1834 	if (error == EEXIST || (!error && !gotvp)) {
1835 		if (newvp) {
1836 			vrele(newvp);
1837 			newvp = (struct vnode *)0;
1838 		}
1839 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1840 			cnp->cn_proc, &np);
1841 		if (!error) {
1842 			newvp = NFSTOV(np);
1843 			if (newvp->v_type != VDIR)
1844 				error = EEXIST;
1845 		}
1846 	}
1847 	if (error) {
1848 		if (newvp)
1849 			vrele(newvp);
1850 	} else
1851 		*ap->a_vpp = newvp;
1852 	FREE(cnp->cn_pnbuf, M_NAMEI);
1853 	vrele(dvp);
1854 	return (error);
1855 }
1856 
1857 /*
1858  * nfs remove directory call
1859  */
1860 int
1861 nfs_rmdir(v)
1862 	void *v;
1863 {
1864 	struct vop_rmdir_args /* {
1865 		struct vnode *a_dvp;
1866 		struct vnode *a_vp;
1867 		struct componentname *a_cnp;
1868 	} */ *ap = v;
1869 	register struct vnode *vp = ap->a_vp;
1870 	register struct vnode *dvp = ap->a_dvp;
1871 	register struct componentname *cnp = ap->a_cnp;
1872 	register u_int32_t *tl;
1873 	register caddr_t cp;
1874 	register int32_t t1, t2;
1875 	caddr_t bpos, dpos, cp2;
1876 	int error = 0, wccflag = NFSV3_WCCRATTR;
1877 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1878 	int v3 = NFS_ISV3(dvp);
1879 
1880 	if (dvp == vp) {
1881 		vrele(dvp);
1882 		vrele(dvp);
1883 		FREE(cnp->cn_pnbuf, M_NAMEI);
1884 		return (EINVAL);
1885 	}
1886 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1887 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
1888 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1889 	nfsm_fhtom(dvp, v3);
1890 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1891 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
1892 	if (v3)
1893 		nfsm_wcc_data(dvp, wccflag);
1894 	nfsm_reqdone;
1895 	FREE(cnp->cn_pnbuf, M_NAMEI);
1896 	VTONFS(dvp)->n_flag |= NMODIFIED;
1897 	if (!wccflag)
1898 		VTONFS(dvp)->n_attrstamp = 0;
1899 	cache_purge(dvp);
1900 	cache_purge(vp);
1901 	vrele(vp);
1902 	vrele(dvp);
1903 	/*
1904 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1905 	 */
1906 	if (error == ENOENT)
1907 		error = 0;
1908 	return (error);
1909 }
1910 
1911 /*
1912  * nfs readdir call
1913  */
1914 int
1915 nfs_readdir(v)
1916 	void *v;
1917 {
1918 	struct vop_readdir_args /* {
1919 		struct vnode *a_vp;
1920 		struct uio *a_uio;
1921 		struct ucred *a_cred;
1922 		int *a_eofflag;
1923 		u_long *a_cookies;
1924 		int a_ncookies;
1925 	} */ *ap = v;
1926 	register struct vnode *vp = ap->a_vp;
1927 	register struct nfsnode *np = VTONFS(vp);
1928 	register struct uio *uio = ap->a_uio;
1929 	char *base = uio->uio_iov->iov_base;
1930 	off_t off = uio->uio_offset;
1931 	int tresid, error;
1932 	struct vattr vattr;
1933 
1934 	if (vp->v_type != VDIR)
1935 		return (EPERM);
1936 	/*
1937 	 * First, check for hit on the EOF offset cache
1938 	 */
1939 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1940 	    (np->n_flag & NMODIFIED) == 0) {
1941 		if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
1942 			if (NQNFS_CKCACHABLE(vp, ND_READ)) {
1943 				nfsstats.direofcache_hits++;
1944 				*ap->a_eofflag = 1;
1945 				return (0);
1946 			}
1947 		} else if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_procp) == 0 &&
1948 			np->n_mtime == vattr.va_mtime.tv_sec) {
1949 			nfsstats.direofcache_hits++;
1950 			*ap->a_eofflag = 1;
1951 			return (0);
1952 		}
1953 	}
1954 
1955 	/*
1956 	 * Call nfs_bioread() to do the real work.
1957 	 */
1958 	tresid = uio->uio_resid;
1959 	error = nfs_bioread(vp, uio, 0, ap->a_cred);
1960 
1961 	if (!error && uio->uio_resid == tresid) {
1962 		nfsstats.direofcache_misses++;
1963 		*ap->a_eofflag = 1;
1964 		return (0);
1965 	}
1966 
1967 	if (!error && ap->a_cookies) {
1968 		struct dirent *dp;
1969 		u_long *cookies = ap->a_cookies;
1970 		int ncookies = ap->a_ncookies;
1971 
1972 		/*
1973 		 * Only the NFS server and emulations use cookies, and they
1974 		 * load the directory block into system space, so we can
1975 		 * just look at it directly.
1976 		 */
1977 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1978 			panic("nfs_readdir: lost in space");
1979 		while (ncookies-- && base < uio->uio_iov->iov_base) {
1980 			dp = (struct dirent *) base;
1981 			if (dp->d_reclen == 0)
1982 				break;
1983 			off += dp->d_reclen;
1984 			*(cookies++) = off;
1985 			base += dp->d_reclen;
1986 		}
1987 		uio->uio_resid += (uio->uio_iov->iov_base - base);
1988 		uio->uio_iov->iov_len += (uio->uio_iov->iov_base - base);
1989 		uio->uio_iov->iov_base = base;
1990 	}
1991 
1992 	*ap->a_eofflag = 0;
1993 	return (error);
1994 }
1995 
1996 /*
1997  * Readdir rpc call.
1998  * Called from below the buffer cache by nfs_doio().
1999  */
2000 int
2001 nfs_readdirrpc(vp, uiop, cred)
2002 	struct vnode *vp;
2003 	register struct uio *uiop;
2004 	struct ucred *cred;
2005 {
2006 	register int len, left;
2007 	register struct dirent *dp = NULL;
2008 	register u_int32_t *tl;
2009 	register caddr_t cp;
2010 	register int32_t t1, t2;
2011 	register nfsuint64 *cookiep;
2012 	caddr_t bpos, dpos, cp2;
2013 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2014 	nfsuint64 cookie;
2015 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2016 	struct nfsnode *dnp = VTONFS(vp);
2017 	u_quad_t fileno;
2018 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2019 	int attrflag;
2020 	int v3 = NFS_ISV3(vp);
2021 
2022 #ifndef DIAGNOSTIC
2023 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (NFS_DIRBLKSIZ - 1)) ||
2024 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2025 		panic("nfs readdirrpc bad uio");
2026 #endif
2027 
2028 	/*
2029 	 * If there is no cookie, assume end of directory.
2030 	 */
2031 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2032 	if (cookiep)
2033 		cookie = *cookiep;
2034 	else
2035 		return (0);
2036 	/*
2037 	 * Loop around doing readdir rpc's of size nm_readdirsize
2038 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2039 	 * The stopping criteria is EOF or buffer full.
2040 	 */
2041 	while (more_dirs && bigenough) {
2042 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2043 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2044 			NFSX_READDIR(v3));
2045 		nfsm_fhtom(vp, v3);
2046 		if (v3) {
2047 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2048 			*tl++ = cookie.nfsuquad[0];
2049 			*tl++ = cookie.nfsuquad[1];
2050 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2051 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2052 		} else {
2053 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2054 			*tl++ = cookie.nfsuquad[0];
2055 		}
2056 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2057 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2058 		if (v3) {
2059 			nfsm_postop_attr(vp, attrflag);
2060 			if (!error) {
2061 				nfsm_dissect(tl, u_int32_t *,
2062 				    2 * NFSX_UNSIGNED);
2063 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2064 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2065 			} else {
2066 				m_freem(mrep);
2067 				goto nfsmout;
2068 			}
2069 		}
2070 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2071 		more_dirs = fxdr_unsigned(int, *tl);
2072 
2073 		/* loop thru the dir entries, doctoring them to 4bsd form */
2074 		while (more_dirs && bigenough) {
2075 			if (v3) {
2076 				nfsm_dissect(tl, u_int32_t *,
2077 				    3 * NFSX_UNSIGNED);
2078 				fxdr_hyper(tl, &fileno);
2079 				len = fxdr_unsigned(int, *(tl + 2));
2080 			} else {
2081 				nfsm_dissect(tl, u_int32_t *,
2082 				    2 * NFSX_UNSIGNED);
2083 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2084 				len = fxdr_unsigned(int, *tl);
2085 			}
2086 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2087 				error = EBADRPC;
2088 				m_freem(mrep);
2089 				goto nfsmout;
2090 			}
2091 			tlen = nfsm_rndup(len);
2092 			if (tlen == len)
2093 				tlen += 4;	/* To ensure null termination */
2094 			left = NFS_READDIRBLKSIZ - blksiz;
2095 			if ((tlen + DIRHDSIZ) > left) {
2096 				dp->d_reclen += left;
2097 				uiop->uio_iov->iov_base += left;
2098 				uiop->uio_iov->iov_len -= left;
2099 				uiop->uio_offset += left;
2100 				uiop->uio_resid -= left;
2101 				blksiz = 0;
2102 			}
2103 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2104 				bigenough = 0;
2105 			if (bigenough) {
2106 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2107 				dp->d_fileno = (int)fileno;
2108 				dp->d_namlen = len;
2109 				dp->d_reclen = tlen + DIRHDSIZ;
2110 				dp->d_type = DT_UNKNOWN;
2111 				blksiz += dp->d_reclen;
2112 				if (blksiz == NFS_READDIRBLKSIZ)
2113 					blksiz = 0;
2114 				uiop->uio_offset += DIRHDSIZ;
2115 				uiop->uio_resid -= DIRHDSIZ;
2116 				uiop->uio_iov->iov_base += DIRHDSIZ;
2117 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2118 				nfsm_mtouio(uiop, len);
2119 				cp = uiop->uio_iov->iov_base;
2120 				tlen -= len;
2121 				*cp = '\0';	/* null terminate */
2122 				uiop->uio_iov->iov_base += tlen;
2123 				uiop->uio_iov->iov_len -= tlen;
2124 				uiop->uio_offset += tlen;
2125 				uiop->uio_resid -= tlen;
2126 			} else
2127 				nfsm_adv(nfsm_rndup(len));
2128 			if (v3) {
2129 				nfsm_dissect(tl, u_int32_t *,
2130 				    3 * NFSX_UNSIGNED);
2131 			} else {
2132 				nfsm_dissect(tl, u_int32_t *,
2133 				    2 * NFSX_UNSIGNED);
2134 			}
2135 			if (bigenough) {
2136 				cookie.nfsuquad[0] = *tl++;
2137 				if (v3)
2138 					cookie.nfsuquad[1] = *tl++;
2139 			} else if (v3)
2140 				tl += 2;
2141 			else
2142 				tl++;
2143 			more_dirs = fxdr_unsigned(int, *tl);
2144 		}
2145 		/*
2146 		 * If at end of rpc data, get the eof boolean
2147 		 */
2148 		if (!more_dirs) {
2149 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2150 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2151 		}
2152 		m_freem(mrep);
2153 	}
2154 	/*
2155 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2156 	 * by increasing d_reclen for the last record.
2157 	 */
2158 	if (blksiz > 0) {
2159 		left = NFS_READDIRBLKSIZ - blksiz;
2160 		dp->d_reclen += left;
2161 		uiop->uio_iov->iov_base += left;
2162 		uiop->uio_iov->iov_len -= left;
2163 		uiop->uio_offset += left;
2164 		uiop->uio_resid -= left;
2165 	}
2166 
2167 	/*
2168 	 * We are now either at the end of the directory or have filled the
2169 	 * block.
2170 	 */
2171 	if (bigenough)
2172 		dnp->n_direofoffset = uiop->uio_offset;
2173 	else {
2174 		if (uiop->uio_resid > 0)
2175 			printf("EEK! readdirrpc resid > 0\n");
2176 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2177 		*cookiep = cookie;
2178 	}
2179 nfsmout:
2180 	return (error);
2181 }
2182 
2183 /*
2184  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2185  */
2186 int
2187 nfs_readdirplusrpc(vp, uiop, cred)
2188 	struct vnode *vp;
2189 	register struct uio *uiop;
2190 	struct ucred *cred;
2191 {
2192 	register int len, left;
2193 	register struct dirent *dp = NULL;
2194 	register u_int32_t *tl;
2195 	register caddr_t cp;
2196 	register int32_t t1, t2;
2197 	register struct vnode *newvp;
2198 	register nfsuint64 *cookiep;
2199 	caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2200 	struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2201 	struct nameidata nami, *ndp = &nami;
2202 	struct componentname *cnp = &ndp->ni_cnd;
2203 	nfsuint64 cookie;
2204 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2205 	struct nfsnode *dnp = VTONFS(vp), *np;
2206 	const unsigned char *hcp;
2207 	nfsfh_t *fhp;
2208 	u_quad_t fileno;
2209 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2210 	int attrflag, fhsize;
2211 
2212 #ifndef DIAGNOSTIC
2213 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (NFS_DIRBLKSIZ - 1)) ||
2214 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2215 		panic("nfs readdirplusrpc bad uio");
2216 #endif
2217 	ndp->ni_dvp = vp;
2218 	newvp = NULLVP;
2219 
2220 	/*
2221 	 * If there is no cookie, assume end of directory.
2222 	 */
2223 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2224 	if (cookiep)
2225 		cookie = *cookiep;
2226 	else
2227 		return (0);
2228 	/*
2229 	 * Loop around doing readdir rpc's of size nm_readdirsize
2230 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2231 	 * The stopping criteria is EOF or buffer full.
2232 	 */
2233 	while (more_dirs && bigenough) {
2234 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2235 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2236 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2237 		nfsm_fhtom(vp, 1);
2238  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2239 		*tl++ = cookie.nfsuquad[0];
2240 		*tl++ = cookie.nfsuquad[1];
2241 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2242 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2243 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2244 		*tl = txdr_unsigned(nmp->nm_rsize);
2245 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2246 		nfsm_postop_attr(vp, attrflag);
2247 		if (error) {
2248 			m_freem(mrep);
2249 			goto nfsmout;
2250 		}
2251 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2252 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2253 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2254 		more_dirs = fxdr_unsigned(int, *tl);
2255 
2256 		/* loop thru the dir entries, doctoring them to 4bsd form */
2257 		while (more_dirs && bigenough) {
2258 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2259 			fxdr_hyper(tl, &fileno);
2260 			len = fxdr_unsigned(int, *(tl + 2));
2261 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2262 				error = EBADRPC;
2263 				m_freem(mrep);
2264 				goto nfsmout;
2265 			}
2266 			tlen = nfsm_rndup(len);
2267 			if (tlen == len)
2268 				tlen += 4;	/* To ensure null termination*/
2269 			left = NFS_READDIRBLKSIZ - blksiz;
2270 			if ((tlen + DIRHDSIZ) > left) {
2271 				dp->d_reclen += left;
2272 				uiop->uio_iov->iov_base += left;
2273 				uiop->uio_iov->iov_len -= left;
2274 				uiop->uio_offset += left;
2275 				uiop->uio_resid -= left;
2276 				blksiz = 0;
2277 			}
2278 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2279 				bigenough = 0;
2280 			if (bigenough) {
2281 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2282 				dp->d_fileno = (int)fileno;
2283 				dp->d_namlen = len;
2284 				dp->d_reclen = tlen + DIRHDSIZ;
2285 				dp->d_type = DT_UNKNOWN;
2286 				blksiz += dp->d_reclen;
2287 				if (blksiz == NFS_READDIRBLKSIZ)
2288 					blksiz = 0;
2289 				uiop->uio_offset += DIRHDSIZ;
2290 				uiop->uio_resid -= DIRHDSIZ;
2291 				uiop->uio_iov->iov_base += DIRHDSIZ;
2292 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2293 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2294 				cnp->cn_namelen = len;
2295 				nfsm_mtouio(uiop, len);
2296 				cp = uiop->uio_iov->iov_base;
2297 				tlen -= len;
2298 				*cp = '\0';
2299 				uiop->uio_iov->iov_base += tlen;
2300 				uiop->uio_iov->iov_len -= tlen;
2301 				uiop->uio_offset += tlen;
2302 				uiop->uio_resid -= tlen;
2303 			} else
2304 				nfsm_adv(nfsm_rndup(len));
2305 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2306 			if (bigenough) {
2307 				cookie.nfsuquad[0] = *tl++;
2308 				cookie.nfsuquad[1] = *tl++;
2309 			} else
2310 				tl += 2;
2311 
2312 			/*
2313 			 * Since the attributes are before the file handle
2314 			 * (sigh), we must skip over the attributes and then
2315 			 * come back and get them.
2316 			 */
2317 			attrflag = fxdr_unsigned(int, *tl);
2318 			if (attrflag) {
2319 			    dpossav1 = dpos;
2320 			    mdsav1 = md;
2321 			    nfsm_adv(NFSX_V3FATTR);
2322 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2323 			    doit = fxdr_unsigned(int, *tl);
2324 			    if (doit) {
2325 				nfsm_getfh(fhp, fhsize, 1);
2326 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2327 				    VREF(vp);
2328 				    newvp = vp;
2329 				    np = dnp;
2330 				} else {
2331 				    error = nfs_nget(vp->v_mount, fhp,
2332 					fhsize, &np);
2333 				    if (error)
2334 					doit = 0;
2335 				    else
2336 					newvp = NFSTOV(np);
2337 				}
2338 			    }
2339 			    if (doit) {
2340 				dpossav2 = dpos;
2341 				dpos = dpossav1;
2342 				mdsav2 = md;
2343 				md = mdsav1;
2344 				nfsm_loadattr(newvp, (struct vattr *)0);
2345 				dpos = dpossav2;
2346 				md = mdsav2;
2347 				dp->d_type =
2348 				    IFTODT(VTTOIF(np->n_vattr.va_type));
2349 				ndp->ni_vp = newvp;
2350 				cnp->cn_hash = 0;
2351 				for (hcp = cnp->cn_nameptr, i = 1; i <= len;
2352 				    i++, hcp++)
2353 				    cnp->cn_hash += *hcp * i;
2354 				if (cnp->cn_namelen <= NCHNAMLEN)
2355 				    cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2356 			    }
2357 			} else {
2358 			    /* Just skip over the file handle */
2359 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2360 			    i = fxdr_unsigned(int, *tl);
2361 			    nfsm_adv(nfsm_rndup(i));
2362 			}
2363 			if (newvp != NULLVP) {
2364 			    vrele(newvp);
2365 			    newvp = NULLVP;
2366 			}
2367 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2368 			more_dirs = fxdr_unsigned(int, *tl);
2369 		}
2370 		/*
2371 		 * If at end of rpc data, get the eof boolean
2372 		 */
2373 		if (!more_dirs) {
2374 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2375 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2376 		}
2377 		m_freem(mrep);
2378 	}
2379 	/*
2380 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2381 	 * by increasing d_reclen for the last record.
2382 	 */
2383 	if (blksiz > 0) {
2384 		left = NFS_READDIRBLKSIZ - blksiz;
2385 		dp->d_reclen += left;
2386 		uiop->uio_iov->iov_base += left;
2387 		uiop->uio_iov->iov_len -= left;
2388 		uiop->uio_offset += left;
2389 		uiop->uio_resid -= left;
2390 	}
2391 
2392 	/*
2393 	 * We are now either at the end of the directory or have filled the
2394 	 * block.
2395 	 */
2396 	if (bigenough)
2397 		dnp->n_direofoffset = uiop->uio_offset;
2398 	else {
2399 		if (uiop->uio_resid > 0)
2400 			printf("EEK! readdirplusrpc resid > 0\n");
2401 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2402 		*cookiep = cookie;
2403 	}
2404 nfsmout:
2405 	if (newvp != NULLVP)
2406 		vrele(newvp);
2407 	return (error);
2408 }
2409 static char hextoasc[] = "0123456789abcdef";
2410 
2411 /*
2412  * Silly rename. To make the NFS filesystem that is stateless look a little
2413  * more like the "ufs" a remove of an active vnode is translated to a rename
2414  * to a funny looking filename that is removed by nfs_inactive on the
2415  * nfsnode. There is the potential for another process on a different client
2416  * to create the same funny name between the nfs_lookitup() fails and the
2417  * nfs_rename() completes, but...
2418  */
2419 int
2420 nfs_sillyrename(dvp, vp, cnp)
2421 	struct vnode *dvp, *vp;
2422 	struct componentname *cnp;
2423 {
2424 	register struct sillyrename *sp;
2425 	struct nfsnode *np;
2426 	int error;
2427 	short pid;
2428 
2429 	cache_purge(dvp);
2430 	np = VTONFS(vp);
2431 #ifndef DIAGNOSTIC
2432 	if (vp->v_type == VDIR)
2433 		panic("nfs: sillyrename dir");
2434 #endif
2435 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2436 		M_NFSREQ, M_WAITOK);
2437 	sp->s_cred = crdup(cnp->cn_cred);
2438 	sp->s_dvp = dvp;
2439 	VREF(dvp);
2440 
2441 	/* Fudge together a funny name */
2442 	pid = cnp->cn_proc->p_pid;
2443 	bcopy(".nfsAxxxx4.4", sp->s_name, 13);
2444 	sp->s_namlen = 12;
2445 	sp->s_name[8] = hextoasc[pid & 0xf];
2446 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2447 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2448 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2449 
2450 	/* Try lookitups until we get one that isn't there */
2451 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2452 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2453 		sp->s_name[4]++;
2454 		if (sp->s_name[4] > 'z') {
2455 			error = EINVAL;
2456 			goto bad;
2457 		}
2458 	}
2459 	error = nfs_renameit(dvp, cnp, sp);
2460 	if (error)
2461 		goto bad;
2462 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2463 		cnp->cn_proc, &np);
2464 	np->n_sillyrename = sp;
2465 	return (0);
2466 bad:
2467 	vrele(sp->s_dvp);
2468 	crfree(sp->s_cred);
2469 	free((caddr_t)sp, M_NFSREQ);
2470 	return (error);
2471 }
2472 
2473 /*
2474  * Look up a file name and optionally either update the file handle or
2475  * allocate an nfsnode, depending on the value of npp.
2476  * npp == NULL	--> just do the lookup
2477  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2478  *			handled too
2479  * *npp != NULL --> update the file handle in the vnode
2480  */
2481 int
2482 nfs_lookitup(dvp, name, len, cred, procp, npp)
2483 	register struct vnode *dvp;
2484 	const char *name;
2485 	int len;
2486 	struct ucred *cred;
2487 	struct proc *procp;
2488 	struct nfsnode **npp;
2489 {
2490 	register u_int32_t *tl;
2491 	register caddr_t cp;
2492 	register int32_t t1, t2;
2493 	struct vnode *newvp = (struct vnode *)0;
2494 	struct nfsnode *np, *dnp = VTONFS(dvp);
2495 	caddr_t bpos, dpos, cp2;
2496 	int error = 0, fhlen, attrflag;
2497 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2498 	nfsfh_t *nfhp;
2499 	int v3 = NFS_ISV3(dvp);
2500 
2501 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2502 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2503 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2504 	nfsm_fhtom(dvp, v3);
2505 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2506 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2507 	if (npp && !error) {
2508 		nfsm_getfh(nfhp, fhlen, v3);
2509 		if (*npp) {
2510 		    np = *npp;
2511 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2512 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2513 			np->n_fhp = &np->n_fh;
2514 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2515 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2516 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2517 		    np->n_fhsize = fhlen;
2518 		    newvp = NFSTOV(np);
2519 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2520 		    VREF(dvp);
2521 		    newvp = dvp;
2522 		} else {
2523 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2524 		    if (error) {
2525 			m_freem(mrep);
2526 			return (error);
2527 		    }
2528 		    newvp = NFSTOV(np);
2529 		}
2530 		if (v3) {
2531 			nfsm_postop_attr(newvp, attrflag);
2532 			if (!attrflag && *npp == NULL) {
2533 				m_freem(mrep);
2534 				vrele(newvp);
2535 				return (ENOENT);
2536 			}
2537 		} else
2538 			nfsm_loadattr(newvp, (struct vattr *)0);
2539 	}
2540 	nfsm_reqdone;
2541 	if (npp && *npp == NULL) {
2542 		if (error) {
2543 			if (newvp)
2544 				vrele(newvp);
2545 		} else
2546 			*npp = np;
2547 	}
2548 	return (error);
2549 }
2550 
2551 /*
2552  * Nfs Version 3 commit rpc
2553  */
2554 int
2555 nfs_commit(vp, offset, cnt, cred, procp)
2556 	register struct vnode *vp;
2557 	u_quad_t offset;
2558 	int cnt;
2559 	struct ucred *cred;
2560 	struct proc *procp;
2561 {
2562 	register caddr_t cp;
2563 	register u_int32_t *tl;
2564 	register int32_t t1, t2;
2565 	register struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2566 	caddr_t bpos, dpos, cp2;
2567 	int error = 0, wccflag = NFSV3_WCCRATTR;
2568 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2569 
2570 	if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0)
2571 		return (0);
2572 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2573 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2574 	nfsm_fhtom(vp, 1);
2575 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2576 	txdr_hyper(&offset, tl);
2577 	tl += 2;
2578 	*tl = txdr_unsigned(cnt);
2579 	nfsm_request(vp, NFSPROC_COMMIT, procp, cred);
2580 	nfsm_wcc_data(vp, wccflag);
2581 	if (!error) {
2582 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2583 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2584 			NFSX_V3WRITEVERF)) {
2585 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2586 				NFSX_V3WRITEVERF);
2587 			error = NFSERR_STALEWRITEVERF;
2588 		}
2589 	}
2590 	nfsm_reqdone;
2591 	return (error);
2592 }
2593 
2594 /*
2595  * Kludge City..
2596  * - make nfs_bmap() essentially a no-op that does no translation
2597  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2598  *   (Maybe I could use the process's page mapping, but I was concerned that
2599  *    Kernel Write might not be enabled and also figured copyout() would do
2600  *    a lot more work than bcopy() and also it currently happens in the
2601  *    context of the swapper process (2).
2602  */
2603 int
2604 nfs_bmap(v)
2605 	void *v;
2606 {
2607 	struct vop_bmap_args /* {
2608 		struct vnode *a_vp;
2609 		daddr_t  a_bn;
2610 		struct vnode **a_vpp;
2611 		daddr_t *a_bnp;
2612 		int *a_runp;
2613 	} */ *ap = v;
2614 	register struct vnode *vp = ap->a_vp;
2615 
2616 	if (ap->a_vpp != NULL)
2617 		*ap->a_vpp = vp;
2618 	if (ap->a_bnp != NULL)
2619 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2620 	return (0);
2621 }
2622 
2623 /*
2624  * Strategy routine.
2625  * For async requests when nfsiod(s) are running, queue the request by
2626  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2627  * request.
2628  */
2629 int
2630 nfs_strategy(v)
2631 	void *v;
2632 {
2633 	struct vop_strategy_args *ap = v;
2634 	register struct buf *bp = ap->a_bp;
2635 	struct ucred *cr;
2636 	struct proc *p;
2637 	int error = 0;
2638 
2639 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2640 		panic("nfs physio/async");
2641 	if (bp->b_flags & B_ASYNC)
2642 		p = (struct proc *)0;
2643 	else
2644 		p = curproc;	/* XXX */
2645 	if (bp->b_flags & B_READ)
2646 		cr = bp->b_rcred;
2647 	else
2648 		cr = bp->b_wcred;
2649 	/*
2650 	 * If the op is asynchronous and an i/o daemon is waiting
2651 	 * queue the request, wake it up and wait for completion
2652 	 * otherwise just do it ourselves.
2653 	 */
2654 	if ((bp->b_flags & B_ASYNC) == 0 ||
2655 		nfs_asyncio(bp, NOCRED))
2656 		error = nfs_doio(bp, cr, p);
2657 	return (error);
2658 }
2659 
2660 /*
2661  * Mmap a file
2662  *
2663  * NB Currently unsupported.
2664  */
2665 /* ARGSUSED */
2666 int
2667 nfs_mmap(v)
2668 	void *v;
2669 {
2670 #if 0
2671 	struct vop_mmap_args /* {
2672 		struct vnode *a_vp;
2673 		int a_fflags;
2674 		struct ucred *a_cred;
2675 		struct proc *a_p;
2676 	} */ *ap = v;
2677 #endif
2678 
2679 	return (EINVAL);
2680 }
2681 
2682 /*
2683  * fsync vnode op. Just call nfs_flush() with commit == 1.
2684  */
2685 /* ARGSUSED */
2686 int
2687 nfs_fsync(v)
2688 	void *v;
2689 {
2690 	struct vop_fsync_args /* {
2691 		struct vnodeop_desc *a_desc;
2692 		struct vnode * a_vp;
2693 		struct ucred * a_cred;
2694 		int  a_waitfor;
2695 		struct proc * a_p;
2696 	} */ *ap = v;
2697 
2698 	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2699 }
2700 
2701 /*
2702  * Flush all the blocks associated with a vnode.
2703  * 	Walk through the buffer pool and push any dirty pages
2704  *	associated with the vnode.
2705  */
2706 int
2707 nfs_flush(vp, cred, waitfor, p, commit)
2708 	register struct vnode *vp;
2709 	struct ucred *cred;
2710 	int waitfor;
2711 	struct proc *p;
2712 	int commit;
2713 {
2714 	register struct nfsnode *np = VTONFS(vp);
2715 	register struct buf *bp;
2716 	register int i;
2717 	struct buf *nbp;
2718 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2719 	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2720 	int passone = 1;
2721 	u_quad_t off = (u_quad_t)-1, endoff = 0, toff;
2722 	struct ucred* wcred = NULL;
2723 #ifndef NFS_COMMITBVECSIZ
2724 #define NFS_COMMITBVECSIZ	20
2725 #endif
2726 	struct buf *bvec[NFS_COMMITBVECSIZ];
2727 
2728 	if (nmp->nm_flag & NFSMNT_INT)
2729 		slpflag = PCATCH;
2730 	if (!commit)
2731 		passone = 0;
2732 	/*
2733 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2734 	 * server, but nas not been committed to stable storage on the server
2735 	 * yet. On the first pass, the byte range is worked out and the commit
2736 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2737 	 * job.
2738 	 */
2739 again:
2740 	bvecpos = 0;
2741 	if (NFS_ISV3(vp) && commit) {
2742 		s = splbio();
2743 		for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2744 			nbp = bp->b_vnbufs.le_next;
2745 			if (bvecpos >= NFS_COMMITBVECSIZ)
2746 				break;
2747 			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2748 				!= (B_DELWRI | B_NEEDCOMMIT))
2749 				continue;
2750 			bremfree(bp);
2751 			/*
2752 			 * Work out if all buffers are using the same cred
2753 			 * so we can deal with them all with one commit.
2754 			 */
2755 			if (wcred == NULL)
2756 				wcred = bp->b_wcred;
2757 			else if (wcred != bp->b_wcred)
2758 				wcred = NOCRED;
2759 			bp->b_flags |= (B_BUSY | B_WRITEINPROG);
2760 			/*
2761 			 * A list of these buffers is kept so that the
2762 			 * second loop knows which buffers have actually
2763 			 * been committed. This is necessary, since there
2764 			 * may be a race between the commit rpc and new
2765 			 * uncommitted writes on the file.
2766 			 */
2767 			bvec[bvecpos++] = bp;
2768 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2769 				bp->b_dirtyoff;
2770 			if (toff < off)
2771 				off = toff;
2772 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2773 			if (toff > endoff)
2774 				endoff = toff;
2775 		}
2776 		splx(s);
2777 	}
2778 	if (bvecpos > 0) {
2779 		/*
2780 		 * Commit data on the server, as required.
2781 		 * If all bufs are using the same wcred, then use that with
2782 		 * one call for all of them, otherwise commit each one
2783 		 * separately.
2784 		 */
2785 		if (wcred != NOCRED)
2786 			retv = nfs_commit(vp, off, (int)(endoff - off),
2787 					  wcred, p);
2788 		else {
2789 			retv = 0;
2790 			for (i = 0; i < bvecpos; i++) {
2791 				off_t off, size;
2792 				bp = bvec[i];
2793 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2794 					bp->b_dirtyoff;
2795 				size = (u_quad_t)(bp->b_dirtyend
2796 						  - bp->b_dirtyoff);
2797 				retv = nfs_commit(vp, off, (int)size,
2798 						  bp->b_wcred, p);
2799 				if (retv) break;
2800 			}
2801 		}
2802 
2803 		if (retv == NFSERR_STALEWRITEVERF)
2804 			nfs_clearcommit(vp->v_mount);
2805 		/*
2806 		 * Now, either mark the blocks I/O done or mark the
2807 		 * blocks dirty, depending on whether the commit
2808 		 * succeeded.
2809 		 */
2810 		for (i = 0; i < bvecpos; i++) {
2811 			bp = bvec[i];
2812 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG);
2813 			if (retv)
2814 			    brelse(bp);
2815 			else {
2816 			    vp->v_numoutput++;
2817 			    bp->b_flags |= B_ASYNC;
2818 			    bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
2819 			    bp->b_dirtyoff = bp->b_dirtyend = 0;
2820 			    reassignbuf(bp, vp);
2821 			    biodone(bp);
2822 			}
2823 		}
2824 	}
2825 
2826 	/*
2827 	 * Start/do any write(s) that are required.
2828 	 */
2829 loop:
2830 	s = splbio();
2831 	for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2832 		nbp = bp->b_vnbufs.le_next;
2833 		if (bp->b_flags & B_BUSY) {
2834 			if (waitfor != MNT_WAIT || passone)
2835 				continue;
2836 			bp->b_flags |= B_WANTED;
2837 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
2838 				"nfsfsync", slptimeo);
2839 			splx(s);
2840 			if (error) {
2841 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
2842 				return (EINTR);
2843 			    if (slpflag == PCATCH) {
2844 				slpflag = 0;
2845 				slptimeo = 2 * hz;
2846 			    }
2847 			}
2848 			goto loop;
2849 		}
2850 		if ((bp->b_flags & B_DELWRI) == 0)
2851 			panic("nfs_fsync: not dirty");
2852 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT))
2853 			continue;
2854 		bremfree(bp);
2855 		if (passone || !commit)
2856 		    bp->b_flags |= (B_BUSY|B_ASYNC);
2857 		else
2858 		    bp->b_flags |= (B_BUSY|B_ASYNC|B_WRITEINPROG|B_NEEDCOMMIT);
2859 		splx(s);
2860 		VOP_BWRITE(bp);
2861 		goto loop;
2862 	}
2863 	splx(s);
2864 	if (passone) {
2865 		passone = 0;
2866 		goto again;
2867 	}
2868 	if (waitfor == MNT_WAIT) {
2869 		while (vp->v_numoutput) {
2870 			vp->v_flag |= VBWAIT;
2871 			error = tsleep((caddr_t)&vp->v_numoutput,
2872 				slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
2873 			if (error) {
2874 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
2875 				return (EINTR);
2876 			    if (slpflag == PCATCH) {
2877 				slpflag = 0;
2878 				slptimeo = 2 * hz;
2879 			    }
2880 			}
2881 		}
2882 		if (vp->v_dirtyblkhd.lh_first && commit) {
2883 #if 0
2884 			vprint("nfs_fsync: dirty", vp);
2885 #endif
2886 			goto loop;
2887 		}
2888 	}
2889 	if (np->n_flag & NWRITEERR) {
2890 		error = np->n_error;
2891 		np->n_flag &= ~NWRITEERR;
2892 	}
2893 	return (error);
2894 }
2895 
2896 /*
2897  * Return POSIX pathconf information applicable to nfs.
2898  *
2899  * The NFS V2 protocol doesn't support this, so just return EINVAL
2900  * for V2.
2901  */
2902 /* ARGSUSED */
2903 int
2904 nfs_pathconf(v)
2905 	void *v;
2906 {
2907 #if 0
2908 	struct vop_pathconf_args /* {
2909 		struct vnode *a_vp;
2910 		int a_name;
2911 		register_t *a_retval;
2912 	} */ *ap = v;
2913 #endif
2914 
2915 	return (EINVAL);
2916 }
2917 
2918 /*
2919  * NFS advisory byte-level locks.
2920  */
2921 int
2922 nfs_advlock(v)
2923 	void *v;
2924 {
2925 	struct vop_advlock_args /* {
2926 		struct vnode *a_vp;
2927 		caddr_t  a_id;
2928 		int  a_op;
2929 		struct flock *a_fl;
2930 		int  a_flags;
2931 	} */ *ap = v;
2932 	register struct nfsnode *np = VTONFS(ap->a_vp);
2933 
2934 	return (lf_advlock(&np->n_lockf, np->n_size, ap->a_id, ap->a_op,
2935 	    ap->a_fl, ap->a_flags));
2936 }
2937 
2938 /*
2939  * Print out the contents of an nfsnode.
2940  */
2941 int
2942 nfs_print(v)
2943 	void *v;
2944 {
2945 	struct vop_print_args /* {
2946 		struct vnode *a_vp;
2947 	} */ *ap = v;
2948 	register struct vnode *vp = ap->a_vp;
2949 	register struct nfsnode *np = VTONFS(vp);
2950 
2951 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
2952 	    np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2953 #ifdef FIFO
2954 	if (vp->v_type == VFIFO)
2955 		fifo_printinfo(vp);
2956 #endif
2957 	printf("\n");
2958 	return (0);
2959 }
2960 
2961 /*
2962  * NFS file truncation.
2963  */
2964 int
2965 nfs_truncate(v)
2966 	void *v;
2967 {
2968 #if 0
2969 	struct vop_truncate_args /* {
2970 		struct vnode *a_vp;
2971 		off_t a_length;
2972 		int a_flags;
2973 		struct ucred *a_cred;
2974 		struct proc *a_p;
2975 	} */ *ap = v;
2976 #endif
2977 
2978 	/* Use nfs_setattr */
2979 	return (EOPNOTSUPP);
2980 }
2981 
2982 /*
2983  * NFS update.
2984  */
2985 int
2986 nfs_update(v)
2987 	void *v;
2988 #if 0
2989 	struct vop_update_args /* {
2990 		struct vnode *a_vp;
2991 		struct timespec *a_ta;
2992 		struct timespec *a_tm;
2993 		int a_waitfor;
2994 	} */ *ap = v;
2995 #endif
2996 {
2997 
2998 	/* Use nfs_setattr */
2999 	return (EOPNOTSUPP);
3000 }
3001 
3002 /*
3003  * Just call nfs_writebp() with the force argument set to 1.
3004  */
3005 int
3006 nfs_bwrite(v)
3007 	void *v;
3008 {
3009 	struct vop_bwrite_args /* {
3010 		struct vnode *a_bp;
3011 	} */ *ap = v;
3012 
3013 	return (nfs_writebp(ap->a_bp, 1));
3014 }
3015 
3016 /*
3017  * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3018  * the force flag is one and it also handles the B_NEEDCOMMIT flag.
3019  */
3020 int
3021 nfs_writebp(bp, force)
3022 	register struct buf *bp;
3023 	int force;
3024 {
3025 	register int oldflags = bp->b_flags, retv = 1;
3026 	register struct proc *p = curproc;	/* XXX */
3027 	off_t off;
3028 
3029 	if(!(bp->b_flags & B_BUSY))
3030 		panic("bwrite: buffer is not busy???");
3031 
3032 #ifdef fvdl_debug
3033 	printf("nfs_writebp(%x): vp %x voff %d vend %d doff %d dend %d\n",
3034 	    bp, bp->b_vp, bp->b_validoff, bp->b_validend, bp->b_dirtyoff,
3035 	    bp->b_dirtyend);
3036 #endif
3037 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
3038 
3039 	if (oldflags & B_ASYNC) {
3040 		if (oldflags & B_DELWRI) {
3041 			reassignbuf(bp, bp->b_vp);
3042 		} else if (p) {
3043 			++p->p_stats->p_ru.ru_oublock;
3044 		}
3045 	}
3046 	bp->b_vp->v_numoutput++;
3047 
3048 	/*
3049 	 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not
3050 	 * an actual write will have to be scheduled via. VOP_STRATEGY().
3051 	 * If B_WRITEINPROG is already set, then push it with a write anyhow.
3052 	 */
3053 	if ((oldflags & (B_NEEDCOMMIT | B_WRITEINPROG)) == B_NEEDCOMMIT) {
3054 		off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
3055 		bp->b_flags |= B_WRITEINPROG;
3056 		retv = nfs_commit(bp->b_vp, off, bp->b_dirtyend-bp->b_dirtyoff,
3057 			bp->b_wcred, bp->b_proc);
3058 		bp->b_flags &= ~B_WRITEINPROG;
3059 		if (!retv) {
3060 			bp->b_dirtyoff = bp->b_dirtyend = 0;
3061 			bp->b_flags &= ~B_NEEDCOMMIT;
3062 			biodone(bp);
3063 		} else if (retv == NFSERR_STALEWRITEVERF)
3064 			nfs_clearcommit(bp->b_vp->v_mount);
3065 	}
3066 	if (retv) {
3067 		if (force)
3068 			bp->b_flags |= B_WRITEINPROG;
3069 		VOP_STRATEGY(bp);
3070 	}
3071 
3072 	if( (oldflags & B_ASYNC) == 0) {
3073 		int rtval = biowait(bp);
3074 		if (oldflags & B_DELWRI) {
3075 			reassignbuf(bp, bp->b_vp);
3076 		} else if (p) {
3077 			++p->p_stats->p_ru.ru_oublock;
3078 		}
3079 		brelse(bp);
3080 		return (rtval);
3081 	}
3082 
3083 	return (0);
3084 }
3085 
3086 /*
3087  * nfs special file access vnode op.
3088  * Essentially just get vattr and then imitate iaccess() since the device is
3089  * local to the client.
3090  */
3091 int
3092 nfsspec_access(v)
3093 	void *v;
3094 {
3095 	struct vop_access_args /* {
3096 		struct vnode *a_vp;
3097 		int  a_mode;
3098 		struct ucred *a_cred;
3099 		struct proc *a_p;
3100 	} */ *ap = v;
3101 	struct vattr va;
3102 	struct vnode *vp = ap->a_vp;
3103 	int error;
3104 
3105 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3106 	if (error)
3107 		return (error);
3108 
3109         /*
3110 	 * Disallow write attempts on filesystems mounted read-only;
3111 	 * unless the file is a socket, fifo, or a block or character
3112 	 * device resident on the filesystem.
3113 	 */
3114 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3115 		switch (vp->v_type) {
3116 		case VREG:
3117 		case VDIR:
3118 		case VLNK:
3119 			return (EROFS);
3120 		default:
3121 			break;
3122 		}
3123 	}
3124 
3125 	return (vaccess(va.va_mode, va.va_uid, va.va_gid, ap->a_mode,
3126 	    ap->a_cred));
3127 }
3128 
3129 /*
3130  * Read wrapper for special devices.
3131  */
3132 int
3133 nfsspec_read(v)
3134 	void *v;
3135 {
3136 	struct vop_read_args /* {
3137 		struct vnode *a_vp;
3138 		struct uio *a_uio;
3139 		int  a_ioflag;
3140 		struct ucred *a_cred;
3141 	} */ *ap = v;
3142 	register struct nfsnode *np = VTONFS(ap->a_vp);
3143 
3144 	/*
3145 	 * Set access flag.
3146 	 */
3147 	np->n_flag |= NACC;
3148 	np->n_atim.tv_sec = time.tv_sec;
3149 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3150 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3151 }
3152 
3153 /*
3154  * Write wrapper for special devices.
3155  */
3156 int
3157 nfsspec_write(v)
3158 	void *v;
3159 {
3160 	struct vop_write_args /* {
3161 		struct vnode *a_vp;
3162 		struct uio *a_uio;
3163 		int  a_ioflag;
3164 		struct ucred *a_cred;
3165 	} */ *ap = v;
3166 	register struct nfsnode *np = VTONFS(ap->a_vp);
3167 
3168 	/*
3169 	 * Set update flag.
3170 	 */
3171 	np->n_flag |= NUPD;
3172 	np->n_mtim.tv_sec = time.tv_sec;
3173 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3174 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3175 }
3176 
3177 /*
3178  * Close wrapper for special devices.
3179  *
3180  * Update the times on the nfsnode then do device close.
3181  */
3182 int
3183 nfsspec_close(v)
3184 	void *v;
3185 {
3186 	struct vop_close_args /* {
3187 		struct vnode *a_vp;
3188 		int  a_fflag;
3189 		struct ucred *a_cred;
3190 		struct proc *a_p;
3191 	} */ *ap = v;
3192 	register struct vnode *vp = ap->a_vp;
3193 	register struct nfsnode *np = VTONFS(vp);
3194 	struct vattr vattr;
3195 
3196 	if (np->n_flag & (NACC | NUPD)) {
3197 		np->n_flag |= NCHG;
3198 		if (vp->v_usecount == 1 &&
3199 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3200 			VATTR_NULL(&vattr);
3201 			if (np->n_flag & NACC)
3202 				vattr.va_atime = np->n_atim;
3203 			if (np->n_flag & NUPD)
3204 				vattr.va_mtime = np->n_mtim;
3205 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3206 		}
3207 	}
3208 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3209 }
3210 
3211 #ifdef FIFO
3212 /*
3213  * Read wrapper for fifos.
3214  */
3215 int
3216 nfsfifo_read(v)
3217 	void *v;
3218 {
3219 	struct vop_read_args /* {
3220 		struct vnode *a_vp;
3221 		struct uio *a_uio;
3222 		int  a_ioflag;
3223 		struct ucred *a_cred;
3224 	} */ *ap = v;
3225 	extern int (**fifo_vnodeop_p) __P((void *));
3226 	register struct nfsnode *np = VTONFS(ap->a_vp);
3227 
3228 	/*
3229 	 * Set access flag.
3230 	 */
3231 	np->n_flag |= NACC;
3232 	np->n_atim.tv_sec = time.tv_sec;
3233 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3234 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3235 }
3236 
3237 /*
3238  * Write wrapper for fifos.
3239  */
3240 int
3241 nfsfifo_write(v)
3242 	void *v;
3243 {
3244 	struct vop_write_args /* {
3245 		struct vnode *a_vp;
3246 		struct uio *a_uio;
3247 		int  a_ioflag;
3248 		struct ucred *a_cred;
3249 	} */ *ap = v;
3250 	extern int (**fifo_vnodeop_p) __P((void *));
3251 	register struct nfsnode *np = VTONFS(ap->a_vp);
3252 
3253 	/*
3254 	 * Set update flag.
3255 	 */
3256 	np->n_flag |= NUPD;
3257 	np->n_mtim.tv_sec = time.tv_sec;
3258 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3259 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3260 }
3261 
3262 /*
3263  * Close wrapper for fifos.
3264  *
3265  * Update the times on the nfsnode then do fifo close.
3266  */
3267 int
3268 nfsfifo_close(v)
3269 	void *v;
3270 {
3271 	struct vop_close_args /* {
3272 		struct vnode *a_vp;
3273 		int  a_fflag;
3274 		struct ucred *a_cred;
3275 		struct proc *a_p;
3276 	} */ *ap = v;
3277 	register struct vnode *vp = ap->a_vp;
3278 	register struct nfsnode *np = VTONFS(vp);
3279 	struct vattr vattr;
3280 	extern int (**fifo_vnodeop_p) __P((void *));
3281 
3282 	if (np->n_flag & (NACC | NUPD)) {
3283 		if (np->n_flag & NACC) {
3284 			np->n_atim.tv_sec = time.tv_sec;
3285 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3286 		}
3287 		if (np->n_flag & NUPD) {
3288 			np->n_mtim.tv_sec = time.tv_sec;
3289 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3290 		}
3291 		np->n_flag |= NCHG;
3292 		if (vp->v_usecount == 1 &&
3293 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3294 			VATTR_NULL(&vattr);
3295 			if (np->n_flag & NACC)
3296 				vattr.va_atime = np->n_atim;
3297 			if (np->n_flag & NUPD)
3298 				vattr.va_mtime = np->n_mtim;
3299 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3300 		}
3301 	}
3302 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3303 }
3304 #endif /* ! FIFO */
3305