xref: /openbsd/sys/nfs/nfs_vnops.c (revision 8932bfb7)
1 /*	$OpenBSD: nfs_vnops.c,v 1.138 2011/07/09 00:24:44 beck Exp $	*/
2 /*	$NetBSD: nfs_vnops.c,v 1.62.4.1 1996/07/08 20:26:52 jtc Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
36  */
37 
38 
39 /*
40  * vnode op calls for Sun NFS version 2 and 3
41  */
42 
43 #include <sys/param.h>
44 #include <sys/proc.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/poll.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/buf.h>
52 #include <sys/malloc.h>
53 #include <sys/pool.h>
54 #include <sys/mbuf.h>
55 #include <sys/conf.h>
56 #include <sys/namei.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/hash.h>
62 #include <sys/queue.h>
63 #include <sys/specdev.h>
64 
65 #include <uvm/uvm_extern.h>
66 
67 #include <miscfs/fifofs/fifo.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/nfs_var.h>
77 
78 #include <net/if.h>
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 
82 #include <dev/rndvar.h>
83 
84 void nfs_cache_enter(struct vnode *, struct vnode *, struct componentname *);
85 
86 /* Global vfs data structures for nfs. */
87 struct vops nfs_vops = {
88 	.vop_lookup	= nfs_lookup,
89 	.vop_create	= nfs_create,
90 	.vop_mknod	= nfs_mknod,
91 	.vop_open	= nfs_open,
92 	.vop_close	= nfs_close,
93 	.vop_access	= nfs_access,
94 	.vop_getattr	= nfs_getattr,
95 	.vop_setattr	= nfs_setattr,
96 	.vop_read	= nfs_read,
97 	.vop_write	= nfs_write,
98 	.vop_ioctl	= nfs_ioctl,
99 	.vop_poll	= nfs_poll,
100 	.vop_kqfilter	= nfs_kqfilter,
101 	.vop_revoke	= vop_generic_revoke,
102 	.vop_fsync	= nfs_fsync,
103 	.vop_remove	= nfs_remove,
104 	.vop_link	= nfs_link,
105 	.vop_rename	= nfs_rename,
106 	.vop_mkdir	= nfs_mkdir,
107 	.vop_rmdir	= nfs_rmdir,
108 	.vop_symlink	= nfs_symlink,
109 	.vop_readdir	= nfs_readdir,
110 	.vop_readlink	= nfs_readlink,
111 	.vop_abortop	= vop_generic_abortop,
112 	.vop_inactive	= nfs_inactive,
113 	.vop_reclaim	= nfs_reclaim,
114 	.vop_lock	= vop_generic_lock,	/* XXX: beck@ must fix this. */
115 	.vop_unlock	= vop_generic_unlock,
116 	.vop_bmap	= nfs_bmap,
117 	.vop_strategy	= nfs_strategy,
118 	.vop_print	= nfs_print,
119 	.vop_islocked	= vop_generic_islocked,
120 	.vop_pathconf	= nfs_pathconf,
121 	.vop_advlock	= nfs_advlock,
122 	.vop_bwrite	= nfs_bwrite
123 };
124 
125 /* Special device vnode ops. */
126 struct vops nfs_specvops = {
127 	.vop_close	= nfsspec_close,
128 	.vop_access	= nfsspec_access,
129 	.vop_getattr	= nfs_getattr,
130 	.vop_setattr	= nfs_setattr,
131 	.vop_read	= nfsspec_read,
132 	.vop_write	= nfsspec_write,
133 	.vop_fsync	= nfs_fsync,
134 	.vop_inactive	= nfs_inactive,
135 	.vop_reclaim	= nfs_reclaim,
136 	.vop_lock	= vop_generic_lock,
137 	.vop_unlock	= vop_generic_unlock,
138 	.vop_print	= nfs_print,
139 	.vop_islocked	= vop_generic_islocked,
140 
141 	/* XXX: Keep in sync with spec_vops. */
142 	.vop_lookup	= vop_generic_lookup,
143 	.vop_create	= spec_badop,
144 	.vop_mknod	= spec_badop,
145 	.vop_open	= spec_open,
146 	.vop_ioctl	= spec_ioctl,
147 	.vop_poll	= spec_poll,
148 	.vop_kqfilter	= spec_kqfilter,
149 	.vop_revoke	= vop_generic_revoke,
150 	.vop_remove	= spec_badop,
151 	.vop_link	= spec_badop,
152 	.vop_rename	= spec_badop,
153 	.vop_mkdir	= spec_badop,
154 	.vop_rmdir	= spec_badop,
155 	.vop_symlink	= spec_badop,
156 	.vop_readdir	= spec_badop,
157 	.vop_readlink	= spec_badop,
158 	.vop_abortop	= spec_badop,
159 	.vop_bmap	= vop_generic_bmap,
160 	.vop_strategy	= spec_strategy,
161 	.vop_pathconf	= spec_pathconf,
162 	.vop_advlock	= spec_advlock,
163 	.vop_bwrite	= vop_generic_bwrite,
164 };
165 
166 #ifdef FIFO
167 struct vops nfs_fifovops = {
168 	.vop_close	= nfsfifo_close,
169 	.vop_access	= nfsspec_access,
170 	.vop_getattr	= nfs_getattr,
171 	.vop_setattr	= nfs_setattr,
172 	.vop_read	= nfsfifo_read,
173 	.vop_write	= nfsfifo_write,
174 	.vop_fsync	= nfs_fsync,
175 	.vop_inactive	= nfs_inactive,
176 	.vop_reclaim	= nfsfifo_reclaim,
177 	.vop_lock	= vop_generic_lock,
178 	.vop_unlock	= vop_generic_unlock,
179 	.vop_print	= nfs_print,
180 	.vop_islocked	= vop_generic_islocked,
181 	.vop_bwrite	= vop_generic_bwrite,
182 
183 	/* XXX: Keep in sync with fifo_vops. */
184 	.vop_lookup	= vop_generic_lookup,
185 	.vop_create	= fifo_badop,
186 	.vop_mknod	= fifo_badop,
187 	.vop_open	= fifo_open,
188 	.vop_ioctl	= fifo_ioctl,
189 	.vop_poll	= fifo_poll,
190 	.vop_kqfilter	= fifo_kqfilter,
191 	.vop_revoke	= vop_generic_revoke,
192 	.vop_remove	= fifo_badop,
193 	.vop_link	= fifo_badop,
194 	.vop_rename	= fifo_badop,
195 	.vop_mkdir	= fifo_badop,
196 	.vop_rmdir	= fifo_badop,
197 	.vop_symlink	= fifo_badop,
198 	.vop_readdir	= fifo_badop,
199 	.vop_readlink	= fifo_badop,
200 	.vop_abortop	= fifo_badop,
201 	.vop_bmap	= vop_generic_bmap,
202 	.vop_strategy	= fifo_badop,
203 	.vop_pathconf	= fifo_pathconf,
204 	.vop_advlock	= fifo_advlock,
205 };
206 #endif /* FIFO */
207 
208 /*
209  * Global variables
210  */
211 extern u_int32_t nfs_true, nfs_false;
212 extern u_int32_t nfs_xdrneg1;
213 extern struct nfsstats nfsstats;
214 extern nfstype nfsv3_type[9];
215 int nfs_numasync = 0;
216 
217 void
218 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
219 {
220 	struct nfsnode *np;
221 
222 	if (vp != NULL) {
223 		np = VTONFS(vp);
224 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
225 	} else {
226 		np = VTONFS(dvp);
227 		if (!np->n_ctime)
228 			np->n_ctime = np->n_vattr.va_mtime.tv_sec;
229 	}
230 
231 	cache_enter(dvp, vp, cnp);
232 }
233 
234 /*
235  * nfs null call from vfs.
236  */
237 int
238 nfs_null(struct vnode *vp, struct ucred *cred, struct proc *procp)
239 {
240 	struct nfsm_info	 info;
241 	int			 error = 0;
242 
243 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(0);
244 	error = nfs_request(vp, NFSPROC_NULL, &info);
245 	m_freem(info.nmi_mrep);
246 	return (error);
247 }
248 
249 /*
250  * nfs access vnode op.
251  * For nfs version 2, just return ok. File accesses may fail later.
252  * For nfs version 3, use the access rpc to check accessibility. If file modes
253  * are changed on the server, accesses might still fail later.
254  */
255 int
256 nfs_access(void *v)
257 {
258 	struct vop_access_args *ap = v;
259 	struct vnode *vp = ap->a_vp;
260 	u_int32_t *tl;
261 	int32_t t1;
262 	caddr_t cp2;
263 	int error = 0, attrflag;
264 	u_int32_t mode, rmode;
265 	int v3 = NFS_ISV3(vp);
266 	int cachevalid;
267 	struct nfsm_info	info;
268 
269 	struct nfsnode *np = VTONFS(vp);
270 
271 	/*
272 	 * Disallow write attempts on filesystems mounted read-only;
273 	 * unless the file is a socket, fifo, or a block or character
274 	 * device resident on the filesystem.
275 	 */
276 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
277 		switch (vp->v_type) {
278 		case VREG:
279 		case VDIR:
280 		case VLNK:
281 			return (EROFS);
282 		default:
283 			break;
284 		}
285 	}
286 
287 	/*
288 	 * Check access cache first. If a request has been made for this uid
289 	 * shortly before, use the cached result.
290 	 */
291 	 cachevalid = (np->n_accstamp != -1 &&
292 	     (time_second - np->n_accstamp) < nfs_attrtimeo(np) &&
293 	     np->n_accuid == ap->a_cred->cr_uid);
294 
295 	if (cachevalid) {
296 		if (!np->n_accerror) {
297 			if ((np->n_accmode & ap->a_mode) == ap->a_mode)
298 				return (np->n_accerror);
299 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
300 			return (np->n_accerror);
301 	}
302 
303 	/*
304 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
305 	 * ufs_access() locally using the vattr. This may not be correct,
306 	 * since the server may apply other access criteria such as
307 	 * client uid-->server uid mapping that we do not know about, but
308 	 * this is better than just returning anything that is lying about
309 	 * in the cache.
310 	 */
311 	if (v3) {
312 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
313 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED);
314 		nfsm_fhtom(&info, vp, v3);
315 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
316 		if (ap->a_mode & VREAD)
317 			mode = NFSV3ACCESS_READ;
318 		else
319 			mode = 0;
320 		if (vp->v_type == VDIR) {
321 			if (ap->a_mode & VWRITE)
322 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
323 					 NFSV3ACCESS_DELETE);
324 			if (ap->a_mode & VEXEC)
325 				mode |= NFSV3ACCESS_LOOKUP;
326 		} else {
327 			if (ap->a_mode & VWRITE)
328 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
329 			if (ap->a_mode & VEXEC)
330 				mode |= NFSV3ACCESS_EXECUTE;
331 		}
332 		*tl = txdr_unsigned(mode);
333 
334 		info.nmi_procp = ap->a_p;
335 		info.nmi_cred = ap->a_cred;
336 		error = nfs_request(vp, NFSPROC_ACCESS, &info);
337 
338 		nfsm_postop_attr(vp, attrflag);
339 		if (error) {
340 			m_freem(info.nmi_mrep);
341 			goto nfsmout;
342 		}
343 
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 		m_freem(info.nmi_mrep);
355 	} else
356 		return (nfsspec_access(ap));
357 
358 
359 	/*
360 	 * If we got the same result as for a previous, different request, OR
361 	 * it in. Don't update the timestamp in that case.
362 	 */
363 	 if (!error || error == EACCES) {
364 		if (cachevalid && np->n_accstamp != -1 &&
365 		    error == np->n_accerror) {
366 			if (!error)
367 				np->n_accmode |= ap->a_mode;
368 			else {
369 				if ((np->n_accmode & ap->a_mode) == ap->a_mode)
370 					np->n_accmode = ap->a_mode;
371 			}
372 		} else {
373 			np->n_accstamp = time_second;
374 			np->n_accuid = ap->a_cred->cr_uid;
375 			np->n_accmode = ap->a_mode;
376 			np->n_accerror = error;
377 		}
378 	}
379 nfsmout:
380 	return (error);
381 }
382 
383 /*
384  * nfs open vnode op
385  * Check to see if the type is ok
386  * and that deletion is not in progress.
387  * For paged in text files, you will need to flush the page cache
388  * if consistency is lost.
389  */
390 int
391 nfs_open(void *v)
392 {
393 	struct vop_open_args *ap = v;
394 	struct vnode *vp = ap->a_vp;
395 	struct nfsnode *np = VTONFS(vp);
396 	struct vattr vattr;
397 	int error;
398 
399 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
400 #ifdef DIAGNOSTIC
401 		printf("open eacces vtyp=%d\n",vp->v_type);
402 #endif
403 		return (EACCES);
404 	}
405 
406 	/*
407 	 * Initialize read and write creds here, for swapfiles
408 	 * and other paths that don't set the creds themselves.
409 	 */
410 
411 	if (ap->a_mode & FREAD) {
412 		if (np->n_rcred) {
413 			crfree(np->n_rcred);
414 		}
415 		np->n_rcred = ap->a_cred;
416 		crhold(np->n_rcred);
417 	}
418 	if (ap->a_mode & FWRITE) {
419 		if (np->n_wcred) {
420 			crfree(np->n_wcred);
421 		}
422 		np->n_wcred = ap->a_cred;
423 		crhold(np->n_wcred);
424 	}
425 
426 	if (np->n_flag & NMODIFIED) {
427 		error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p);
428 		if (error == EINTR)
429 			return (error);
430 		uvm_vnp_uncache(vp);
431 		NFS_INVALIDATE_ATTRCACHE(np);
432 		if (vp->v_type == VDIR)
433 			np->n_direofoffset = 0;
434 		error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
435 		if (error)
436 			return (error);
437 		np->n_mtime = vattr.va_mtime;
438 	} else {
439 		error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
440 		if (error)
441 			return (error);
442 		if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) {
443 			if (vp->v_type == VDIR)
444 				np->n_direofoffset = 0;
445 			error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p);
446 			if (error == EINTR)
447 				return (error);
448 			uvm_vnp_uncache(vp);
449 			np->n_mtime = vattr.va_mtime;
450 		}
451 	}
452 	/* For open/close consistency. */
453 	NFS_INVALIDATE_ATTRCACHE(np);
454 	return (0);
455 }
456 
457 /*
458  * nfs close vnode op
459  * What an NFS client should do upon close after writing is a debatable issue.
460  * Most NFS clients push delayed writes to the server upon close, basically for
461  * two reasons:
462  * 1 - So that any write errors may be reported back to the client process
463  *     doing the close system call. By far the two most likely errors are
464  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
465  * 2 - To put a worst case upper bound on cache inconsistency between
466  *     multiple clients for the file.
467  * There is also a consistency problem for Version 2 of the protocol w.r.t.
468  * not being able to tell if other clients are writing a file concurrently,
469  * since there is no way of knowing if the changed modify time in the reply
470  * is only due to the write for this client.
471  * (NFS Version 3 provides weak cache consistency data in the reply that
472  *  should be sufficient to detect and handle this case.)
473  *
474  * The current code does the following:
475  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
476  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
477  *		       or commit them (this satisfies 1 and 2 except for the
478  *		       case where the server crashes after this close but
479  *		       before the commit RPC, which is felt to be "good
480  *		       enough". Changing the last argument to nfs_flush() to
481  *		       a 1 would force a commit operation, if it is felt a
482  *		       commit is necessary now.
483  */
484 int
485 nfs_close(void *v)
486 {
487 	struct vop_close_args *ap = v;
488 	struct vnode *vp = ap->a_vp;
489 	struct nfsnode *np = VTONFS(vp);
490 	int error = 0;
491 
492 	if (vp->v_type == VREG) {
493 	    if (np->n_flag & NMODIFIED) {
494 		if (NFS_ISV3(vp)) {
495 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
496 		    np->n_flag &= ~NMODIFIED;
497 		} else
498 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p);
499 		NFS_INVALIDATE_ATTRCACHE(np);
500 	    }
501 	    if (np->n_flag & NWRITEERR) {
502 		np->n_flag &= ~NWRITEERR;
503 		error = np->n_error;
504 	    }
505 	}
506 	return (error);
507 }
508 
509 /*
510  * nfs getattr call from vfs.
511  */
512 int
513 nfs_getattr(void *v)
514 {
515 	struct vop_getattr_args *ap = v;
516 	struct vnode *vp = ap->a_vp;
517 	struct nfsnode *np = VTONFS(vp);
518 	struct nfsm_info	info;
519 	int32_t t1;
520 	int error = 0;
521 
522 	info.nmi_v3 = NFS_ISV3(vp);
523 
524 	/*
525 	 * Update local times for special files.
526 	 */
527 	if (np->n_flag & (NACC | NUPD))
528 		np->n_flag |= NCHG;
529 	/*
530 	 * First look in the cache.
531 	 */
532 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
533 		return (0);
534 
535 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
536 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3));
537 	nfsm_fhtom(&info, vp, info.nmi_v3);
538 	info.nmi_procp = ap->a_p;
539 	info.nmi_cred = ap->a_cred;
540 	error = nfs_request(vp, NFSPROC_GETATTR, &info);
541 	if (!error)
542 		nfsm_loadattr(vp, ap->a_vap);
543 	m_freem(info.nmi_mrep);
544 nfsmout:
545 	return (error);
546 }
547 
548 /*
549  * nfs setattr call.
550  */
551 int
552 nfs_setattr(void *v)
553 {
554 	struct vop_setattr_args *ap = v;
555 	struct vnode *vp = ap->a_vp;
556 	struct nfsnode *np = VTONFS(vp);
557 	struct vattr *vap = ap->a_vap;
558 	int hint = NOTE_ATTRIB;
559 	int error = 0;
560 	u_quad_t tsize = 0;
561 
562 	/*
563 	 * Setting of flags is not supported.
564 	 */
565 	if (vap->va_flags != VNOVAL)
566 		return (EOPNOTSUPP);
567 
568 	/*
569 	 * Disallow write attempts if the filesystem is mounted read-only.
570 	 */
571 	if ((vap->va_uid != (uid_t)VNOVAL ||
572 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
573 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
574 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
575 		return (EROFS);
576 	if (vap->va_size != VNOVAL) {
577 		switch (vp->v_type) {
578 		case VDIR:
579 			return (EISDIR);
580 		case VCHR:
581 		case VBLK:
582 		case VSOCK:
583 		case VFIFO:
584 			if (vap->va_mtime.tv_sec == VNOVAL &&
585 			    vap->va_atime.tv_sec == VNOVAL &&
586 			    vap->va_mode == (mode_t)VNOVAL &&
587 			    vap->va_uid == (uid_t)VNOVAL &&
588 			    vap->va_gid == (gid_t)VNOVAL)
589 				return (0);
590 			vap->va_size = VNOVAL;
591 			break;
592 		default:
593 			/*
594 			 * Disallow write attempts if the filesystem is
595 			 * mounted read-only.
596 			 */
597 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
598 				return (EROFS);
599 			if (vap->va_size == 0)
600 				error = nfs_vinvalbuf(vp, 0,
601 				     ap->a_cred, ap->a_p);
602 			else
603 				error = nfs_vinvalbuf(vp, V_SAVE,
604 				     ap->a_cred, ap->a_p);
605 			if (error)
606 				return (error);
607 			tsize = np->n_size;
608 			np->n_size = np->n_vattr.va_size = vap->va_size;
609 			uvm_vnp_setsize(vp, np->n_size);
610 		};
611 	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
612 		vap->va_atime.tv_sec != VNOVAL) &&
613 		vp->v_type == VREG &&
614 		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
615 		    ap->a_p)) == EINTR)
616 		return (error);
617 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
618 	if (error && vap->va_size != VNOVAL) {
619 		np->n_size = np->n_vattr.va_size = tsize;
620 		uvm_vnp_setsize(vp, np->n_size);
621 	}
622 
623 	if (vap->va_size != VNOVAL && vap->va_size < tsize)
624 		hint |= NOTE_TRUNCATE;
625 
626 	VN_KNOTE(vp, hint); /* XXX setattrrpc? */
627 
628 	return (error);
629 }
630 
631 /*
632  * Do an nfs setattr rpc.
633  */
634 int
635 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
636     struct proc *procp)
637 {
638 	struct nfsv2_sattr *sp;
639 	struct nfsm_info	info;
640 	int32_t t1;
641 	caddr_t cp2;
642 	u_int32_t *tl;
643 	int error = 0, wccflag = NFSV3_WCCRATTR;
644 	int v3 = NFS_ISV3(vp);
645 
646 	info.nmi_v3 = NFS_ISV3(vp);
647 
648 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
649 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(v3) + NFSX_SATTR(v3));
650 	nfsm_fhtom(&info, vp, v3);
651 
652 	if (info.nmi_v3) {
653 		nfsm_v3attrbuild(&info.nmi_mb, vap, 1);
654 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
655 		*tl = nfs_false;
656 	} else {
657 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
658 		if (vap->va_mode == (mode_t)VNOVAL)
659 			sp->sa_mode = nfs_xdrneg1;
660 		else
661 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
662 		if (vap->va_uid == (uid_t)VNOVAL)
663 			sp->sa_uid = nfs_xdrneg1;
664 		else
665 			sp->sa_uid = txdr_unsigned(vap->va_uid);
666 		if (vap->va_gid == (gid_t)VNOVAL)
667 			sp->sa_gid = nfs_xdrneg1;
668 		else
669 			sp->sa_gid = txdr_unsigned(vap->va_gid);
670 		sp->sa_size = txdr_unsigned(vap->va_size);
671 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
672 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
673 	}
674 
675 	info.nmi_procp = procp;
676 	info.nmi_cred = cred;
677 	error = nfs_request(vp, NFSPROC_SETATTR, &info);
678 
679 	if (info.nmi_v3)
680 		nfsm_wcc_data(vp, wccflag);
681 	else if (error == 0)
682 		nfsm_loadattr(vp, NULL);
683 
684 	m_freem(info.nmi_mrep);
685 nfsmout:
686 	return (error);
687 }
688 
689 /*
690  * nfs lookup call, one step at a time...
691  * First look in cache
692  * If not found, unlock the directory nfsnode and do the rpc
693  */
694 int
695 nfs_lookup(void *v)
696 {
697 	struct vop_lookup_args *ap = v;
698 	struct componentname *cnp = ap->a_cnp;
699 	struct vnode *dvp = ap->a_dvp;
700 	struct vnode **vpp = ap->a_vpp;
701 	struct proc *p = cnp->cn_proc;
702 	struct nfsm_info	info;
703 	int flags;
704 	struct vnode *newvp;
705 	u_int32_t *tl;
706 	int32_t t1;
707 	struct nfsmount *nmp;
708 	caddr_t cp2;
709 	long len;
710 	nfsfh_t *fhp;
711 	struct nfsnode *np;
712 	int lockparent, wantparent, error = 0, attrflag, fhsize;
713 
714 	info.nmi_v3 = NFS_ISV3(dvp);
715 
716 	cnp->cn_flags &= ~PDIRUNLOCK;
717 	flags = cnp->cn_flags;
718 
719 	*vpp = NULLVP;
720 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
721 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
722 		return (EROFS);
723 	if (dvp->v_type != VDIR)
724 		return (ENOTDIR);
725 	lockparent = flags & LOCKPARENT;
726 	wantparent = flags & (LOCKPARENT|WANTPARENT);
727 	nmp = VFSTONFS(dvp->v_mount);
728 	np = VTONFS(dvp);
729 
730 	/*
731 	 * Before tediously performing a linear scan of the directory,
732 	 * check the name cache to see if the directory/name pair
733 	 * we are looking for is known already.
734 	 * If the directory/name pair is found in the name cache,
735 	 * we have to ensure the directory has not changed from
736 	 * the time the cache entry has been created. If it has,
737 	 * the cache entry has to be ignored.
738 	 */
739 	if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
740 		struct vattr vattr;
741 		int err2;
742 
743 		if (error && error != ENOENT) {
744 			*vpp = NULLVP;
745 			return (error);
746 		}
747 
748 		if (cnp->cn_flags & PDIRUNLOCK) {
749 			err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
750 			if (err2 != 0) {
751 				*vpp = NULLVP;
752 				return (err2);
753 			}
754 			cnp->cn_flags &= ~PDIRUNLOCK;
755 		}
756 
757 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
758 		if (err2 != 0) {
759 			if (error == 0) {
760 				if (*vpp != dvp)
761 					vput(*vpp);
762 				else
763 					vrele(*vpp);
764 			}
765 			*vpp = NULLVP;
766 			return (err2);
767 		}
768 
769 		if (error == ENOENT) {
770 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
771 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
772 			    VTONFS(dvp)->n_ctime)
773 				return (ENOENT);
774 			cache_purge(dvp);
775 			np->n_ctime = 0;
776 			goto dorpc;
777 		}
778 
779 		newvp = *vpp;
780 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
781 			&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
782 		{
783 			nfsstats.lookupcache_hits++;
784 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
785 				cnp->cn_flags |= SAVENAME;
786 			if ((!lockparent || !(flags & ISLASTCN)) &&
787 			     newvp != dvp)
788 				VOP_UNLOCK(dvp, 0, p);
789 			return (0);
790 		}
791 		cache_purge(newvp);
792 		if (newvp != dvp)
793 			vput(newvp);
794 		else
795 			vrele(newvp);
796 		*vpp = NULLVP;
797 	}
798 dorpc:
799 	error = 0;
800 	newvp = NULLVP;
801 	nfsstats.lookupcache_misses++;
802 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
803 	len = cnp->cn_namelen;
804 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
805 	    NFSX_UNSIGNED + nfsm_rndup(len));
806 	nfsm_fhtom(&info, dvp, info.nmi_v3);
807 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
808 
809 	info.nmi_procp = cnp->cn_proc;
810 	info.nmi_cred = cnp->cn_cred;
811 	error = nfs_request(dvp, NFSPROC_LOOKUP, &info);
812 
813 	if (error) {
814 		if (info.nmi_v3)
815 			nfsm_postop_attr(dvp, attrflag);
816 		m_freem(info.nmi_mrep);
817 		goto nfsmout;
818 	}
819 
820 	nfsm_getfh(fhp, fhsize, info.nmi_v3);
821 
822 	/*
823 	 * Handle RENAME case...
824 	 */
825 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
826 		if (NFS_CMPFH(np, fhp, fhsize)) {
827 			m_freem(info.nmi_mrep);
828 			return (EISDIR);
829 		}
830 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
831 		if (error) {
832 			m_freem(info.nmi_mrep);
833 			return (error);
834 		}
835 		newvp = NFSTOV(np);
836 		if (info.nmi_v3) {
837 			nfsm_postop_attr(newvp, attrflag);
838 			nfsm_postop_attr(dvp, attrflag);
839 		} else
840 			nfsm_loadattr(newvp, NULL);
841 		*vpp = newvp;
842 		m_freem(info.nmi_mrep);
843 		cnp->cn_flags |= SAVENAME;
844 		if (!lockparent) {
845 			VOP_UNLOCK(dvp, 0, p);
846 			cnp->cn_flags |= PDIRUNLOCK;
847 		}
848 		return (0);
849 	}
850 
851 	/*
852 	 * The postop attr handling is duplicated for each if case,
853 	 * because it should be done while dvp is locked (unlocking
854 	 * dvp is different for each case).
855 	 */
856 
857 	if (NFS_CMPFH(np, fhp, fhsize)) {
858 		vref(dvp);
859 		newvp = dvp;
860 		if (info.nmi_v3) {
861 			nfsm_postop_attr(newvp, attrflag);
862 			nfsm_postop_attr(dvp, attrflag);
863 		} else
864 			nfsm_loadattr(newvp, NULL);
865 	} else if (flags & ISDOTDOT) {
866 		VOP_UNLOCK(dvp, 0, p);
867 		cnp->cn_flags |= PDIRUNLOCK;
868 
869 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
870 		if (error) {
871 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
872 				cnp->cn_flags &= ~PDIRUNLOCK;
873 			m_freem(info.nmi_mrep);
874 			return (error);
875 		}
876 		newvp = NFSTOV(np);
877 
878 		if (info.nmi_v3) {
879 			nfsm_postop_attr(newvp, attrflag);
880 			nfsm_postop_attr(dvp, attrflag);
881 		} else
882 			nfsm_loadattr(newvp, NULL);
883 
884 		if (lockparent && (flags & ISLASTCN)) {
885 			if ((error = vn_lock(dvp, LK_EXCLUSIVE, p))) {
886 				m_freem(info.nmi_mrep);
887 				vput(newvp);
888 				return error;
889 			}
890 			cnp->cn_flags &= ~PDIRUNLOCK;
891 		}
892 
893 	} else {
894 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
895 		if (error) {
896 			m_freem(info.nmi_mrep);
897 			return error;
898 		}
899 		newvp = NFSTOV(np);
900 		if (info.nmi_v3) {
901 			nfsm_postop_attr(newvp, attrflag);
902 			nfsm_postop_attr(dvp, attrflag);
903 		} else
904 			nfsm_loadattr(newvp, NULL);
905 		if (!lockparent || !(flags & ISLASTCN)) {
906 			VOP_UNLOCK(dvp, 0, p);
907 			cnp->cn_flags |= PDIRUNLOCK;
908 		}
909 	}
910 
911 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
912 		cnp->cn_flags |= SAVENAME;
913 	if ((cnp->cn_flags & MAKEENTRY) &&
914 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
915 		nfs_cache_enter(dvp, newvp, cnp);
916 	}
917 
918 	*vpp = newvp;
919 	m_freem(info.nmi_mrep);
920 
921 nfsmout:
922 	if (error) {
923 		/*
924 		 * We get here only because of errors returned by
925 		 * the RPC. Otherwise we'll have returned above
926 		 * (the nfsm_* macros will jump to nfsmout
927 		 * on error).
928 		 */
929 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
930 		    cnp->cn_nameiop != CREATE) {
931 			nfs_cache_enter(dvp, NULL, cnp);
932 		}
933 		if (newvp != NULLVP) {
934 			vrele(newvp);
935 			if (newvp != dvp)
936 				VOP_UNLOCK(newvp, 0, p);
937 		}
938 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
939 		    (flags & ISLASTCN) && error == ENOENT) {
940 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
941 				error = EROFS;
942 			else
943 				error = EJUSTRETURN;
944 		}
945 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
946 			cnp->cn_flags |= SAVENAME;
947 		*vpp = NULL;
948 	}
949 	return (error);
950 }
951 
952 /*
953  * nfs read call.
954  * Just call nfs_bioread() to do the work.
955  */
956 int
957 nfs_read(void *v)
958 {
959 	struct vop_read_args *ap = v;
960 	struct vnode *vp = ap->a_vp;
961 
962 	if (vp->v_type != VREG)
963 		return (EPERM);
964 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
965 }
966 
967 /*
968  * nfs readlink call
969  */
970 int
971 nfs_readlink(void *v)
972 {
973 	struct vop_readlink_args *ap = v;
974 	struct vnode *vp = ap->a_vp;
975 
976 	if (vp->v_type != VLNK)
977 		return (EPERM);
978 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
979 }
980 
981 /*
982  * Do a readlink rpc.
983  * Called by nfs_doio() from below the buffer cache.
984  */
985 int
986 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
987 {
988 	struct nfsm_info	info;
989 	u_int32_t *tl;
990 	int32_t t1;
991 	caddr_t cp2;
992 	int error = 0, len, attrflag;
993 
994 	info.nmi_v3 = NFS_ISV3(vp);
995 
996 	nfsstats.rpccnt[NFSPROC_READLINK]++;
997 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3));
998 	nfsm_fhtom(&info, vp, info.nmi_v3);
999 
1000 	info.nmi_procp = curproc;
1001 	info.nmi_cred = cred;
1002 	error = nfs_request(vp, NFSPROC_READLINK, &info);
1003 
1004 	if (info.nmi_v3)
1005 		nfsm_postop_attr(vp, attrflag);
1006 	if (!error) {
1007 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1008 		nfsm_mtouio(uiop, len);
1009 	}
1010 
1011 	m_freem(info.nmi_mrep);
1012 
1013 nfsmout:
1014 	return (error);
1015 }
1016 
1017 /*
1018  * nfs read rpc call
1019  * Ditto above
1020  */
1021 int
1022 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1023 {
1024 	struct nfsm_info	info;
1025 	u_int32_t *tl;
1026 	int32_t t1;
1027 	caddr_t cp2;
1028 	struct nfsmount *nmp;
1029 	int error = 0, len, retlen, tsiz, eof, attrflag;
1030 
1031 	info.nmi_v3 = NFS_ISV3(vp);
1032 
1033 	eof = 0;
1034 
1035 	nmp = VFSTONFS(vp->v_mount);
1036 	tsiz = uiop->uio_resid;
1037 	if (uiop->uio_offset + tsiz > 0xffffffff && !info.nmi_v3)
1038 		return (EFBIG);
1039 	while (tsiz > 0) {
1040 		nfsstats.rpccnt[NFSPROC_READ]++;
1041 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1042 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1043 		    NFSX_UNSIGNED * 3);
1044 		nfsm_fhtom(&info, vp, info.nmi_v3);
1045 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED * 3);
1046 		if (info.nmi_v3) {
1047 			txdr_hyper(uiop->uio_offset, tl);
1048 			*(tl + 2) = txdr_unsigned(len);
1049 		} else {
1050 			*tl++ = txdr_unsigned(uiop->uio_offset);
1051 			*tl++ = txdr_unsigned(len);
1052 			*tl = 0;
1053 		}
1054 
1055 		info.nmi_procp = curproc;
1056 		info.nmi_cred = VTONFS(vp)->n_rcred;
1057 		error = nfs_request(vp, NFSPROC_READ, &info);
1058 		if (info.nmi_v3)
1059 			nfsm_postop_attr(vp, attrflag);
1060 		if (error) {
1061 			m_freem(info.nmi_mrep);
1062 			goto nfsmout;
1063 		}
1064 
1065 		if (info.nmi_v3) {
1066 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1067 			eof = fxdr_unsigned(int, *(tl + 1));
1068 		} else {
1069 			nfsm_loadattr(vp, NULL);
1070 		}
1071 
1072 		nfsm_strsiz(retlen, nmp->nm_rsize);
1073 		nfsm_mtouio(uiop, retlen);
1074 		m_freem(info.nmi_mrep);
1075 		tsiz -= retlen;
1076 		if (info.nmi_v3) {
1077 			if (eof || retlen == 0)
1078 				tsiz = 0;
1079 		} else if (retlen < len)
1080 			tsiz = 0;
1081 	}
1082 
1083 nfsmout:
1084 	return (error);
1085 }
1086 
1087 /*
1088  * nfs write call
1089  */
1090 int
1091 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1092 {
1093 	struct nfsm_info	info;
1094 	u_int32_t *tl;
1095 	int32_t t1, backup;
1096 	caddr_t cp2;
1097 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1098 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1099 	int committed = NFSV3WRITE_FILESYNC;
1100 
1101 	info.nmi_v3 = NFS_ISV3(vp);
1102 
1103 #ifdef DIAGNOSTIC
1104 	if (uiop->uio_iovcnt != 1)
1105 		panic("nfs: writerpc iovcnt > 1");
1106 #endif
1107 	*must_commit = 0;
1108 	tsiz = uiop->uio_resid;
1109 	if (uiop->uio_offset + tsiz > 0xffffffff && !info.nmi_v3)
1110 		return (EFBIG);
1111 	while (tsiz > 0) {
1112 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1113 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1114 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3)
1115 		    + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1116 		nfsm_fhtom(&info, vp, info.nmi_v3);
1117 		if (info.nmi_v3) {
1118 			tl = nfsm_build(&info.nmi_mb, 5 * NFSX_UNSIGNED);
1119 			txdr_hyper(uiop->uio_offset, tl);
1120 			tl += 2;
1121 			*tl++ = txdr_unsigned(len);
1122 			*tl++ = txdr_unsigned(*iomode);
1123 			*tl = txdr_unsigned(len);
1124 		} else {
1125 			u_int32_t x;
1126 
1127 			tl = nfsm_build(&info.nmi_mb, 4 * NFSX_UNSIGNED);
1128 			/* Set both "begin" and "current" to non-garbage. */
1129 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1130 			*tl++ = x;	/* "begin offset" */
1131 			*tl++ = x;	/* "current offset" */
1132 			x = txdr_unsigned(len);
1133 			*tl++ = x;	/* total to this offset */
1134 			*tl = x;	/* size of this write */
1135 
1136 		}
1137 		nfsm_uiotombuf(&info.nmi_mb, uiop, len);
1138 
1139 		info.nmi_procp = curproc;
1140 		info.nmi_cred = VTONFS(vp)->n_wcred;
1141 		error = nfs_request(vp, NFSPROC_WRITE, &info);
1142 		if (info.nmi_v3) {
1143 			wccflag = NFSV3_WCCCHK;
1144 			nfsm_wcc_data(vp, wccflag);
1145 		}
1146 
1147 		if (error) {
1148 			m_freem(info.nmi_mrep);
1149 			goto nfsmout;
1150 		}
1151 
1152 		if (info.nmi_v3) {
1153 			wccflag = NFSV3_WCCCHK;
1154 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1155 				+ NFSX_V3WRITEVERF);
1156 			rlen = fxdr_unsigned(int, *tl++);
1157 			if (rlen == 0) {
1158 				error = NFSERR_IO;
1159 				break;
1160 			} else if (rlen < len) {
1161 				backup = len - rlen;
1162 				uiop->uio_iov->iov_base =
1163 				    (char *)uiop->uio_iov->iov_base -
1164 				    backup;
1165 				uiop->uio_iov->iov_len += backup;
1166 				uiop->uio_offset -= backup;
1167 				uiop->uio_resid += backup;
1168 				len = rlen;
1169 			}
1170 			commit = fxdr_unsigned(int, *tl++);
1171 
1172 			/*
1173 			 * Return the lowest committment level
1174 			 * obtained by any of the RPCs.
1175 			 */
1176 			if (committed == NFSV3WRITE_FILESYNC)
1177 				committed = commit;
1178 			else if (committed == NFSV3WRITE_DATASYNC &&
1179 				commit == NFSV3WRITE_UNSTABLE)
1180 				committed = commit;
1181 			if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0) {
1182 				bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1183 				    NFSX_V3WRITEVERF);
1184 				nmp->nm_flag |= NFSMNT_HASWRITEVERF;
1185 			} else if (bcmp((caddr_t)tl,
1186 			    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1187 				*must_commit = 1;
1188 				bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1189 				    NFSX_V3WRITEVERF);
1190 			}
1191 		} else {
1192 			nfsm_loadattr(vp, NULL);
1193 		}
1194 		if (wccflag)
1195 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1196 		m_freem(info.nmi_mrep);
1197 		tsiz -= len;
1198 	}
1199 nfsmout:
1200 	*iomode = committed;
1201 	if (error)
1202 		uiop->uio_resid = tsiz;
1203 	return (error);
1204 }
1205 
1206 /*
1207  * nfs mknod rpc
1208  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1209  * mode set to specify the file type and the size field for rdev.
1210  */
1211 int
1212 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1213     struct vattr *vap)
1214 {
1215 	struct nfsv2_sattr *sp;
1216 	struct nfsm_info	info;
1217 	u_int32_t *tl;
1218 	int32_t t1;
1219 	struct vnode *newvp = NULL;
1220 	struct nfsnode *np = NULL;
1221 	char *cp2;
1222 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1223 	u_int32_t rdev;
1224 
1225 	info.nmi_v3 = NFS_ISV3(dvp);
1226 
1227 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1228 		rdev = txdr_unsigned(vap->va_rdev);
1229 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1230 		rdev = nfs_xdrneg1;
1231 	else {
1232 		VOP_ABORTOP(dvp, cnp);
1233 		vput(dvp);
1234 		return (EOPNOTSUPP);
1235 	}
1236 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1237 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1238 	    4 * NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen) +
1239 	    NFSX_SATTR(info.nmi_v3));
1240 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1241 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1242 
1243 	if (info.nmi_v3) {
1244 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
1245 		*tl++ = vtonfsv3_type(vap->va_type);
1246 		nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1247 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1248 			tl = nfsm_build(&info.nmi_mb, 2 * NFSX_UNSIGNED);
1249 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1250 			*tl = txdr_unsigned(minor(vap->va_rdev));
1251 		}
1252 	} else {
1253 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1254 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1255 		sp->sa_uid = nfs_xdrneg1;
1256 		sp->sa_gid = nfs_xdrneg1;
1257 		sp->sa_size = rdev;
1258 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1259 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1260 	}
1261 
1262 	KASSERT(cnp->cn_proc == curproc);
1263 	info.nmi_procp = cnp->cn_proc;
1264 	info.nmi_cred = cnp->cn_cred;
1265 	error = nfs_request(dvp, NFSPROC_MKNOD, &info);
1266 	if (!error) {
1267 		nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1268 		if (!gotvp) {
1269 			if (newvp) {
1270 				vrele(newvp);
1271 				newvp = NULL;
1272 			}
1273 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1274 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1275 			if (!error)
1276 				newvp = NFSTOV(np);
1277 		}
1278 	}
1279 	if (info.nmi_v3)
1280 		nfsm_wcc_data(dvp, wccflag);
1281 	m_freem(info.nmi_mrep);
1282 
1283 nfsmout:
1284 	if (error) {
1285 		if (newvp)
1286 			vrele(newvp);
1287 	} else {
1288 		if (cnp->cn_flags & MAKEENTRY)
1289 			nfs_cache_enter(dvp, newvp, cnp);
1290 		*vpp = newvp;
1291 	}
1292 	pool_put(&namei_pool, cnp->cn_pnbuf);
1293 	VTONFS(dvp)->n_flag |= NMODIFIED;
1294 	if (!wccflag)
1295 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1296 	vrele(dvp);
1297 	return (error);
1298 }
1299 
1300 /*
1301  * nfs mknod vop
1302  * just call nfs_mknodrpc() to do the work.
1303  */
1304 int
1305 nfs_mknod(void *v)
1306 {
1307 	struct vop_mknod_args *ap = v;
1308 	struct vnode *newvp;
1309 	int error;
1310 
1311 	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1312 	if (!error)
1313 		vrele(newvp);
1314 
1315 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1316 
1317 	return (error);
1318 }
1319 
1320 int
1321 nfs_create(void *v)
1322 {
1323 	struct vop_create_args *ap = v;
1324 	struct vnode *dvp = ap->a_dvp;
1325 	struct vattr *vap = ap->a_vap;
1326 	struct componentname *cnp = ap->a_cnp;
1327 	struct nfsv2_sattr *sp;
1328 	struct nfsm_info	info;
1329 	u_int32_t *tl;
1330 	int32_t t1;
1331 	struct nfsnode *np = NULL;
1332 	struct vnode *newvp = NULL;
1333 	caddr_t cp2;
1334 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1335 
1336 	info.nmi_v3 = NFS_ISV3(dvp);
1337 
1338 	/*
1339 	 * Oops, not for me..
1340 	 */
1341 	if (vap->va_type == VSOCK)
1342 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1343 
1344 	if (vap->va_vaflags & VA_EXCLUSIVE)
1345 		fmode |= O_EXCL;
1346 
1347 again:
1348 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1349 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1350 	    2 * NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen) +
1351 	    NFSX_SATTR(info.nmi_v3));
1352 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1353 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1354 	if (info.nmi_v3) {
1355 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
1356 		if (fmode & O_EXCL) {
1357 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1358 			tl = nfsm_build(&info.nmi_mb, NFSX_V3CREATEVERF);
1359 			*tl++ = arc4random();
1360 			*tl = arc4random();
1361 		} else {
1362 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1363 			nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1364 		}
1365 	} else {
1366 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1367 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1368 		sp->sa_uid = nfs_xdrneg1;
1369 		sp->sa_gid = nfs_xdrneg1;
1370 		sp->sa_size = 0;
1371 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1372 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1373 	}
1374 
1375 	KASSERT(cnp->cn_proc == curproc);
1376 	info.nmi_procp = cnp->cn_proc;
1377 	info.nmi_cred = cnp->cn_cred;
1378 	error = nfs_request(dvp, NFSPROC_CREATE, &info);
1379 	if (!error) {
1380 		nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1381 		if (!gotvp) {
1382 			if (newvp) {
1383 				vrele(newvp);
1384 				newvp = NULL;
1385 			}
1386 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1387 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1388 			if (!error)
1389 				newvp = NFSTOV(np);
1390 		}
1391 	}
1392 	if (info.nmi_v3)
1393 		nfsm_wcc_data(dvp, wccflag);
1394 	m_freem(info.nmi_mrep);
1395 
1396 nfsmout:
1397 	if (error) {
1398 		if (info.nmi_v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1399 			fmode &= ~O_EXCL;
1400 			goto again;
1401 		}
1402 		if (newvp)
1403 			vrele(newvp);
1404 	} else if (info.nmi_v3 && (fmode & O_EXCL))
1405 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1406 	if (!error) {
1407 		if (cnp->cn_flags & MAKEENTRY)
1408 			nfs_cache_enter(dvp, newvp, cnp);
1409 		*ap->a_vpp = newvp;
1410 	}
1411 	pool_put(&namei_pool, cnp->cn_pnbuf);
1412 	VTONFS(dvp)->n_flag |= NMODIFIED;
1413 	if (!wccflag)
1414 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1415 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1416 	vrele(dvp);
1417 	return (error);
1418 }
1419 
1420 /*
1421  * nfs file remove call
1422  * To try and make nfs semantics closer to ufs semantics, a file that has
1423  * other processes using the vnode is renamed instead of removed and then
1424  * removed later on the last close.
1425  * - If v_usecount > 1
1426  *	  If a rename is not already in the works
1427  *	     call nfs_sillyrename() to set it up
1428  *     else
1429  *	  do the remove rpc
1430  */
1431 int
1432 nfs_remove(void *v)
1433 {
1434 	struct vop_remove_args *ap = v;
1435 	struct vnode *vp = ap->a_vp;
1436 	struct vnode *dvp = ap->a_dvp;
1437 	struct componentname *cnp = ap->a_cnp;
1438 	struct nfsnode *np = VTONFS(vp);
1439 	int error = 0;
1440 	struct vattr vattr;
1441 
1442 #ifdef DIAGNOSTIC
1443 	if ((cnp->cn_flags & HASBUF) == 0)
1444 		panic("nfs_remove: no name");
1445 	if (vp->v_usecount < 1)
1446 		panic("nfs_remove: bad v_usecount");
1447 #endif
1448 	if (vp->v_type == VDIR)
1449 		error = EPERM;
1450 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1451 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1452 	    vattr.va_nlink > 1)) {
1453 		/*
1454 		 * Purge the name cache so that the chance of a lookup for
1455 		 * the name succeeding while the remove is in progress is
1456 		 * minimized. Without node locking it can still happen, such
1457 		 * that an I/O op returns ESTALE, but since you get this if
1458 		 * another host removes the file..
1459 		 */
1460 		cache_purge(vp);
1461 		/*
1462 		 * throw away biocache buffers, mainly to avoid
1463 		 * unnecessary delayed writes later.
1464 		 */
1465 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc);
1466 		/* Do the rpc */
1467 		if (error != EINTR)
1468 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1469 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1470 		/*
1471 		 * Kludge City: If the first reply to the remove rpc is lost..
1472 		 *   the reply to the retransmitted request will be ENOENT
1473 		 *   since the file was in fact removed
1474 		 *   Therefore, we cheat and return success.
1475 		 */
1476 		if (error == ENOENT)
1477 			error = 0;
1478 	} else if (!np->n_sillyrename)
1479 		error = nfs_sillyrename(dvp, vp, cnp);
1480 	pool_put(&namei_pool, cnp->cn_pnbuf);
1481 	NFS_INVALIDATE_ATTRCACHE(np);
1482 	vrele(dvp);
1483 	vrele(vp);
1484 
1485 	VN_KNOTE(vp, NOTE_DELETE);
1486 	VN_KNOTE(dvp, NOTE_WRITE);
1487 
1488 	return (error);
1489 }
1490 
1491 /*
1492  * nfs file remove rpc called from nfs_inactive
1493  */
1494 int
1495 nfs_removeit(struct sillyrename *sp)
1496 {
1497 	/*
1498 	 * Make sure that the directory vnode is still valid.
1499 	 * XXX we should lock sp->s_dvp here.
1500 	 *
1501 	 * NFS can potentially try to nuke a silly *after* the directory
1502 	 * has already been pushed out on a forced unmount. Since the silly
1503 	 * is going to go away anyway, this is fine.
1504 	 */
1505 	if (sp->s_dvp->v_type == VBAD)
1506 		return (0);
1507 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1508 		NULL));
1509 }
1510 
1511 /*
1512  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1513  */
1514 int
1515 nfs_removerpc(struct vnode *dvp, char *name, int namelen, struct ucred *cred,
1516     struct proc *proc)
1517 {
1518 	struct nfsm_info	info;
1519 	u_int32_t *tl;
1520 	int32_t t1;
1521 	caddr_t cp2;
1522 	int error = 0, wccflag = NFSV3_WCCRATTR;
1523 
1524 	info.nmi_v3 = NFS_ISV3(dvp);
1525 
1526 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1527 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1528 	     NFSX_UNSIGNED + nfsm_rndup(namelen));
1529 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1530 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1531 
1532 	info.nmi_procp = proc;
1533 	info.nmi_cred = cred;
1534 	error = nfs_request(dvp, NFSPROC_REMOVE, &info);
1535 	if (info.nmi_v3)
1536 		nfsm_wcc_data(dvp, wccflag);
1537 	m_freem(info.nmi_mrep);
1538 
1539 nfsmout:
1540 	VTONFS(dvp)->n_flag |= NMODIFIED;
1541 	if (!wccflag)
1542 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1543 	return (error);
1544 }
1545 
1546 /*
1547  * nfs file rename call
1548  */
1549 int
1550 nfs_rename(void *v)
1551 {
1552 	struct vop_rename_args	*ap = v;
1553 	struct vnode *fvp = ap->a_fvp;
1554 	struct vnode *tvp = ap->a_tvp;
1555 	struct vnode *fdvp = ap->a_fdvp;
1556 	struct vnode *tdvp = ap->a_tdvp;
1557 	struct componentname *tcnp = ap->a_tcnp;
1558 	struct componentname *fcnp = ap->a_fcnp;
1559 	int error;
1560 
1561 #ifdef DIAGNOSTIC
1562 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1563 	    (fcnp->cn_flags & HASBUF) == 0)
1564 		panic("nfs_rename: no name");
1565 #endif
1566 	/* Check for cross-device rename */
1567 	if ((fvp->v_mount != tdvp->v_mount) ||
1568 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1569 		error = EXDEV;
1570 		goto out;
1571 	}
1572 
1573 	/*
1574 	 * If the tvp exists and is in use, sillyrename it before doing the
1575 	 * rename of the new file over it.
1576 	 */
1577 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1578 	    tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1579 		VN_KNOTE(tvp, NOTE_DELETE);
1580 		vrele(tvp);
1581 		tvp = NULL;
1582 	}
1583 
1584 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1585 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1586 		tcnp->cn_proc);
1587 
1588 	VN_KNOTE(fdvp, NOTE_WRITE);
1589 	VN_KNOTE(tdvp, NOTE_WRITE);
1590 
1591 	if (fvp->v_type == VDIR) {
1592 		if (tvp != NULL && tvp->v_type == VDIR)
1593 			cache_purge(tdvp);
1594 		cache_purge(fdvp);
1595 	}
1596 out:
1597 	if (tdvp == tvp)
1598 		vrele(tdvp);
1599 	else
1600 		vput(tdvp);
1601 	if (tvp)
1602 		vput(tvp);
1603 	vrele(fdvp);
1604 	vrele(fvp);
1605 	/*
1606 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1607 	 */
1608 	if (error == ENOENT)
1609 		error = 0;
1610 	return (error);
1611 }
1612 
1613 /*
1614  * nfs file rename rpc called from nfs_remove() above
1615  */
1616 int
1617 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1618     struct sillyrename *sp)
1619 {
1620 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1621 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curproc));
1622 }
1623 
1624 /*
1625  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1626  */
1627 int
1628 nfs_renamerpc(struct vnode *fdvp, char *fnameptr, int fnamelen,
1629     struct vnode *tdvp, char *tnameptr, int tnamelen, struct ucred *cred,
1630     struct proc *proc)
1631 {
1632 	struct nfsm_info	info;
1633 	u_int32_t *tl;
1634 	int32_t t1;
1635 	caddr_t cp2;
1636 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1637 
1638 	info.nmi_v3 = NFS_ISV3(fdvp);
1639 
1640 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1641 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead((NFSX_FH(info.nmi_v3) +
1642 	    NFSX_UNSIGNED) * 2 + nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1643 	nfsm_fhtom(&info, fdvp, info.nmi_v3);
1644 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1645 	nfsm_fhtom(&info, tdvp, info.nmi_v3);
1646 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1647 
1648 	info.nmi_procp = proc;
1649 	info.nmi_cred = cred;
1650 	error = nfs_request(fdvp, NFSPROC_RENAME, &info);
1651 	if (info.nmi_v3) {
1652 		nfsm_wcc_data(fdvp, fwccflag);
1653 		nfsm_wcc_data(tdvp, twccflag);
1654 	}
1655 	m_freem(info.nmi_mrep);
1656 
1657 nfsmout:
1658 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1659 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1660 	if (!fwccflag)
1661 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
1662 	if (!twccflag)
1663 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
1664 	return (error);
1665 }
1666 
1667 /*
1668  * nfs hard link create call
1669  */
1670 int
1671 nfs_link(void *v)
1672 {
1673 	struct vop_link_args *ap = v;
1674 	struct vnode *vp = ap->a_vp;
1675 	struct vnode *dvp = ap->a_dvp;
1676 	struct componentname *cnp = ap->a_cnp;
1677 	struct nfsm_info	info;
1678 	u_int32_t *tl;
1679 	int32_t t1;
1680 	caddr_t cp2;
1681 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1682 
1683 	info.nmi_v3 = NFS_ISV3(vp);
1684 
1685 	if (dvp->v_mount != vp->v_mount) {
1686 		pool_put(&namei_pool, cnp->cn_pnbuf);
1687 		if (vp == dvp)
1688 			vrele(dvp);
1689 		else
1690 			vput(dvp);
1691 		return (EXDEV);
1692 	}
1693 
1694 	/*
1695 	 * Push all writes to the server, so that the attribute cache
1696 	 * doesn't get "out of sync" with the server.
1697 	 * XXX There should be a better way!
1698 	 */
1699 	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1700 
1701 	nfsstats.rpccnt[NFSPROC_LINK]++;
1702 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(2 * NFSX_FH(info.nmi_v3) +
1703 	    NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1704 	nfsm_fhtom(&info, vp, info.nmi_v3);
1705 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1706 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1707 
1708 	info.nmi_procp = cnp->cn_proc;
1709 	info.nmi_cred = cnp->cn_cred;
1710 	error = nfs_request(vp, NFSPROC_LINK, &info);
1711 	if (info.nmi_v3) {
1712 		nfsm_postop_attr(vp, attrflag);
1713 		nfsm_wcc_data(dvp, wccflag);
1714 	}
1715 	m_freem(info.nmi_mrep);
1716 nfsmout:
1717 	pool_put(&namei_pool, cnp->cn_pnbuf);
1718 	VTONFS(dvp)->n_flag |= NMODIFIED;
1719 	if (!attrflag)
1720 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
1721 	if (!wccflag)
1722 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1723 
1724 	VN_KNOTE(vp, NOTE_LINK);
1725 	VN_KNOTE(dvp, NOTE_WRITE);
1726 	vput(dvp);
1727 	return (error);
1728 }
1729 
1730 /*
1731  * nfs symbolic link create call
1732  */
1733 int
1734 nfs_symlink(void *v)
1735 {
1736 	struct vop_symlink_args *ap = v;
1737 	struct vnode *dvp = ap->a_dvp;
1738 	struct vattr *vap = ap->a_vap;
1739 	struct componentname *cnp = ap->a_cnp;
1740 	struct nfsv2_sattr *sp;
1741 	struct nfsm_info	info;
1742 	u_int32_t *tl;
1743 	int32_t t1;
1744 	caddr_t cp2;
1745 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1746 	struct vnode *newvp = NULL;
1747 
1748 	info.nmi_v3 = NFS_ISV3(dvp);
1749 
1750 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1751 	slen = strlen(ap->a_target);
1752 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1753 	    2 * NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) +
1754 	    NFSX_SATTR(info.nmi_v3));
1755 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1756 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1757 	if (info.nmi_v3)
1758 		nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1759 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1760 	if (!info.nmi_v3) {
1761 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1762 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1763 		sp->sa_uid = nfs_xdrneg1;
1764 		sp->sa_gid = nfs_xdrneg1;
1765 		sp->sa_size = nfs_xdrneg1;
1766 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1767 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1768 	}
1769 
1770 	info.nmi_procp = cnp->cn_proc;
1771 	info.nmi_cred = cnp->cn_cred;
1772 	error = nfs_request(dvp, NFSPROC_SYMLINK, &info);
1773 	if (info.nmi_v3) {
1774 		if (!error)
1775 			nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1776 		nfsm_wcc_data(dvp, wccflag);
1777 	}
1778 	m_freem(info.nmi_mrep);
1779 
1780 nfsmout:
1781 	if (newvp)
1782 		vrele(newvp);
1783 	pool_put(&namei_pool, cnp->cn_pnbuf);
1784 	VTONFS(dvp)->n_flag |= NMODIFIED;
1785 	if (!wccflag)
1786 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1787 	VN_KNOTE(dvp, NOTE_WRITE);
1788 	vrele(dvp);
1789 	return (error);
1790 }
1791 
1792 /*
1793  * nfs make dir call
1794  */
1795 int
1796 nfs_mkdir(void *v)
1797 {
1798 	struct vop_mkdir_args *ap = v;
1799 	struct vnode *dvp = ap->a_dvp;
1800 	struct vattr *vap = ap->a_vap;
1801 	struct componentname *cnp = ap->a_cnp;
1802 	struct nfsv2_sattr *sp;
1803 	struct nfsm_info	info;
1804 	u_int32_t *tl;
1805 	int32_t t1;
1806 	int len;
1807 	struct nfsnode *np = NULL;
1808 	struct vnode *newvp = NULL;
1809 	caddr_t cp2;
1810 	int error = 0, wccflag = NFSV3_WCCRATTR;
1811 	int gotvp = 0;
1812 
1813 	info.nmi_v3 = NFS_ISV3(dvp);
1814 
1815 	len = cnp->cn_namelen;
1816 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1817 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1818 	    NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(info.nmi_v3));
1819 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1820 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1821 
1822 	if (info.nmi_v3) {
1823 		nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1824 	} else {
1825 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1826 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1827 		sp->sa_uid = nfs_xdrneg1;
1828 		sp->sa_gid = nfs_xdrneg1;
1829 		sp->sa_size = nfs_xdrneg1;
1830 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1831 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1832 	}
1833 
1834 	info.nmi_procp = cnp->cn_proc;
1835 	info.nmi_cred = cnp->cn_cred;
1836 	error = nfs_request(dvp, NFSPROC_MKDIR, &info);
1837 	if (!error)
1838 		nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1839 	if (info.nmi_v3)
1840 		nfsm_wcc_data(dvp, wccflag);
1841 	m_freem(info.nmi_mrep);
1842 
1843 nfsmout:
1844 	VTONFS(dvp)->n_flag |= NMODIFIED;
1845 	if (!wccflag)
1846 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1847 
1848 	if (error == 0 && newvp == NULL) {
1849 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1850 			cnp->cn_proc, &np);
1851 		if (!error) {
1852 			newvp = NFSTOV(np);
1853 			if (newvp->v_type != VDIR)
1854 				error = EEXIST;
1855 		}
1856 	}
1857 	if (error) {
1858 		if (newvp)
1859 			vrele(newvp);
1860 	} else {
1861 		VN_KNOTE(dvp, NOTE_WRITE|NOTE_LINK);
1862 		if (cnp->cn_flags & MAKEENTRY)
1863 			nfs_cache_enter(dvp, newvp, cnp);
1864 		*ap->a_vpp = newvp;
1865 	}
1866 	pool_put(&namei_pool, cnp->cn_pnbuf);
1867 	vrele(dvp);
1868 	return (error);
1869 }
1870 
1871 /*
1872  * nfs remove directory call
1873  */
1874 int
1875 nfs_rmdir(void *v)
1876 {
1877 	struct vop_rmdir_args *ap = v;
1878 	struct vnode *vp = ap->a_vp;
1879 	struct vnode *dvp = ap->a_dvp;
1880 	struct componentname *cnp = ap->a_cnp;
1881 	struct nfsm_info	info;
1882 	u_int32_t *tl;
1883 	int32_t t1;
1884 	caddr_t cp2;
1885 	int error = 0, wccflag = NFSV3_WCCRATTR;
1886 
1887 	info.nmi_v3 = NFS_ISV3(dvp);
1888 
1889 	if (dvp == vp) {
1890 		vrele(dvp);
1891 		vrele(dvp);
1892 		pool_put(&namei_pool, cnp->cn_pnbuf);
1893 		return (EINVAL);
1894 	}
1895 
1896 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1897 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1898 	    NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1899 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1900 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1901 
1902 	info.nmi_procp = cnp->cn_proc;
1903 	info.nmi_cred = cnp->cn_cred;
1904 	error = nfs_request(dvp,  NFSPROC_RMDIR, &info);
1905 	if (info.nmi_v3)
1906 		nfsm_wcc_data(dvp, wccflag);
1907 	m_freem(info.nmi_mrep);
1908 
1909 nfsmout:
1910 	pool_put(&namei_pool, cnp->cn_pnbuf);
1911 	VTONFS(dvp)->n_flag |= NMODIFIED;
1912 	if (!wccflag)
1913 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1914 
1915 	VN_KNOTE(dvp, NOTE_WRITE|NOTE_LINK);
1916 	VN_KNOTE(vp, NOTE_DELETE);
1917 
1918 	cache_purge(vp);
1919 	vrele(vp);
1920 	vrele(dvp);
1921 	/*
1922 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1923 	 */
1924 	if (error == ENOENT)
1925 		error = 0;
1926 	return (error);
1927 }
1928 
1929 
1930 /*
1931  * The readdir logic below has a big design bug. It stores the NFS cookie in
1932  * the returned uio->uio_offset but does not store the verifier (it cannot).
1933  * Instead, the code stores the verifier in the nfsnode and applies that
1934  * verifies to all cookies, no matter what verifier was originally with
1935  * the cookie.
1936  *
1937  * From a practical standpoint, this is not a problem since almost all
1938  * NFS servers do not change the validity of cookies across deletes
1939  * and inserts.
1940  */
1941 
1942 struct nfs_dirent {
1943 	u_int32_t cookie[2];
1944 	struct dirent dirent;
1945 };
1946 
1947 #define	NFS_DIRHDSIZ	(sizeof (struct nfs_dirent) - (MAXNAMLEN + 1))
1948 #define NFS_DIRENT_OVERHEAD  offsetof(struct nfs_dirent, dirent)
1949 
1950 /*
1951  * nfs readdir call
1952  */
1953 int
1954 nfs_readdir(void *v)
1955 {
1956 	struct vop_readdir_args *ap = v;
1957 	struct vnode *vp = ap->a_vp;
1958 	struct nfsnode *np = VTONFS(vp);
1959 	struct uio *uio = ap->a_uio;
1960 	int tresid, error = 0;
1961 	struct vattr vattr;
1962 	u_long *cookies = NULL;
1963 	int ncookies = 0, cnt;
1964 	u_int64_t  newoff = uio->uio_offset;
1965 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1966 	struct uio readdir_uio;
1967 	struct iovec readdir_iovec;
1968 	struct proc * p = uio->uio_procp;
1969 	int done = 0, eof = 0;
1970 	struct ucred *cred = ap->a_cred;
1971 	void *data;
1972 
1973 	if (vp->v_type != VDIR)
1974 		return (EPERM);
1975 	/*
1976 	 * First, check for hit on the EOF offset cache
1977 	 */
1978 	if (np->n_direofoffset != 0 &&
1979 	    uio->uio_offset == np->n_direofoffset) {
1980 		if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_procp) == 0 &&
1981 		    timespeccmp(&np->n_mtime, &vattr.va_mtime, ==)) {
1982 			nfsstats.direofcache_hits++;
1983 			*ap->a_eofflag = 1;
1984 			return (0);
1985 		}
1986 	}
1987 
1988 	if (uio->uio_resid < NFS_FABLKSIZE)
1989 		return (EINVAL);
1990 
1991 	tresid = uio->uio_resid;
1992 
1993 	if (uio->uio_rw != UIO_READ)
1994 		return (EINVAL);
1995 
1996 	if (ap->a_cookies) {
1997 		ncookies = uio->uio_resid / 20;
1998 
1999 		cookies = malloc(sizeof(*cookies) * ncookies, M_TEMP,
2000 		    M_WAITOK);
2001 		*ap->a_ncookies = ncookies;
2002 		*ap->a_cookies = cookies;
2003 	}
2004 
2005 	if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3)
2006 		(void)nfs_fsinfo(nmp, vp, cred, p);
2007 
2008 	cnt = 5;
2009 
2010 	data = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
2011 	do {
2012 		struct nfs_dirent *ndp = data;
2013 
2014 		readdir_iovec.iov_len = NFS_DIRBLKSIZ;
2015 		readdir_iovec.iov_base = data;
2016 		readdir_uio.uio_offset = newoff;
2017 		readdir_uio.uio_iov = &readdir_iovec;
2018 		readdir_uio.uio_iovcnt = 1;
2019 		readdir_uio.uio_segflg = UIO_SYSSPACE;
2020 		readdir_uio.uio_rw = UIO_READ;
2021 		readdir_uio.uio_resid = NFS_DIRBLKSIZ;
2022 		readdir_uio.uio_procp = curproc;
2023 
2024 		if (nmp->nm_flag & NFSMNT_RDIRPLUS) {
2025 			error = nfs_readdirplusrpc(vp, &readdir_uio, cred,
2026 			    &eof);
2027 			if (error == NFSERR_NOTSUPP)
2028 				nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
2029 		}
2030 		if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0)
2031 			error = nfs_readdirrpc(vp, &readdir_uio, cred, &eof);
2032 
2033 		if (error == NFSERR_BAD_COOKIE)
2034 			error = EINVAL;
2035 
2036 		while (error == 0 &&
2037 		    (ap->a_cookies == NULL || ncookies != 0) &&
2038 		    ndp < (struct nfs_dirent *)readdir_iovec.iov_base) {
2039 			struct dirent *dp = &ndp->dirent;
2040 			int reclen = dp->d_reclen;
2041 
2042 			dp->d_reclen -= NFS_DIRENT_OVERHEAD;
2043 
2044 			if (uio->uio_resid < dp->d_reclen) {
2045 				eof = 0;
2046 				done = 1;
2047 				break;
2048 			}
2049 
2050 			error = uiomove((caddr_t)dp, dp->d_reclen, uio);
2051 			if (error)
2052 				break;
2053 
2054 			newoff = fxdr_hyper(&ndp->cookie[0]);
2055 
2056 			if (ap->a_cookies != NULL) {
2057 				*cookies = newoff;
2058 				cookies++;
2059 				ncookies--;
2060 			}
2061 
2062 			ndp = (struct nfs_dirent *)((u_int8_t *)ndp + reclen);
2063 		}
2064 	} while (!error && !done && !eof && cnt--);
2065 
2066 	free(data, M_TEMP);
2067 	data = NULL;
2068 
2069 	if (ap->a_cookies) {
2070 		if (error) {
2071 			free(*ap->a_cookies, M_TEMP);
2072 			*ap->a_cookies = NULL;
2073 			*ap->a_ncookies = 0;
2074 		} else {
2075 			*ap->a_ncookies -= ncookies;
2076 		}
2077 	}
2078 
2079 	if (!error)
2080 		uio->uio_offset = newoff;
2081 
2082 	if (!error && (eof || uio->uio_resid == tresid)) {
2083 		nfsstats.direofcache_misses++;
2084 		*ap->a_eofflag = 1;
2085 		return (0);
2086 	}
2087 
2088 	*ap->a_eofflag = 0;
2089 	return (error);
2090 }
2091 
2092 
2093 /*
2094  * The function below stuff the cookies in after the name
2095  */
2096 
2097 /*
2098  * Readdir rpc call.
2099  */
2100 int
2101 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2102     int *end_of_directory)
2103 {
2104 	int len, left;
2105 	struct nfs_dirent *ndp = NULL;
2106 	struct dirent *dp = NULL;
2107 	struct nfsm_info	info;
2108 	u_int32_t *tl;
2109 	caddr_t cp;
2110 	int32_t t1;
2111 	caddr_t cp2;
2112 	nfsuint64 cookie;
2113 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2114 	struct nfsnode *dnp = VTONFS(vp);
2115 	u_quad_t fileno;
2116 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2117 	int attrflag;
2118 
2119 	info.nmi_v3 = NFS_ISV3(vp);
2120 
2121 #ifdef DIAGNOSTIC
2122 	if (uiop->uio_iovcnt != 1 ||
2123 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2124 		panic("nfs readdirrpc bad uio");
2125 #endif
2126 
2127 	txdr_hyper(uiop->uio_offset, &cookie.nfsuquad[0]);
2128 
2129 	/*
2130 	 * Loop around doing readdir rpc's of size nm_readdirsize
2131 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2132 	 * The stopping criteria is EOF or buffer full.
2133 	 */
2134 	while (more_dirs && bigenough) {
2135 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2136 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3)
2137 		    + NFSX_READDIR(info.nmi_v3));
2138 		nfsm_fhtom(&info, vp, info.nmi_v3);
2139 		if (info.nmi_v3) {
2140 			tl = nfsm_build(&info.nmi_mb, 5 * NFSX_UNSIGNED);
2141 			*tl++ = cookie.nfsuquad[0];
2142 			*tl++ = cookie.nfsuquad[1];
2143 			if (cookie.nfsuquad[0] == 0 &&
2144 			    cookie.nfsuquad[1] == 0) {
2145 				*tl++ = 0;
2146 				*tl++ = 0;
2147 			} else {
2148 				*tl++ = dnp->n_cookieverf.nfsuquad[0];
2149 				*tl++ = dnp->n_cookieverf.nfsuquad[1];
2150 			}
2151 		} else {
2152 			tl = nfsm_build(&info.nmi_mb, 2 * NFSX_UNSIGNED);
2153 			*tl++ = cookie.nfsuquad[1];
2154 		}
2155 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2156 
2157 		info.nmi_procp = uiop->uio_procp;
2158 		info.nmi_cred = cred;
2159 		error = nfs_request(vp, NFSPROC_READDIR, &info);
2160 		if (info.nmi_v3)
2161 			nfsm_postop_attr(vp, attrflag);
2162 
2163 		if (error) {
2164 			m_freem(info.nmi_mrep);
2165 			goto nfsmout;
2166 		}
2167 
2168 		if (info.nmi_v3) {
2169 			nfsm_dissect(tl, u_int32_t *,
2170 			    2 * NFSX_UNSIGNED);
2171 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2172 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2173 		}
2174 
2175 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2176 		more_dirs = fxdr_unsigned(int, *tl);
2177 
2178 		/* loop thru the dir entries, doctoring them to 4bsd form */
2179 		while (more_dirs && bigenough) {
2180 			if (info.nmi_v3) {
2181 				nfsm_dissect(tl, u_int32_t *,
2182 				    3 * NFSX_UNSIGNED);
2183 				fileno = fxdr_hyper(tl);
2184 				len = fxdr_unsigned(int, *(tl + 2));
2185 			} else {
2186 				nfsm_dissect(tl, u_int32_t *,
2187 				    2 * NFSX_UNSIGNED);
2188 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2189 				len = fxdr_unsigned(int, *tl);
2190 			}
2191 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2192 				error = EBADRPC;
2193 				m_freem(info.nmi_mrep);
2194 				goto nfsmout;
2195 			}
2196 			tlen = nfsm_rndup(len + 1);
2197 			left = NFS_READDIRBLKSIZ - blksiz;
2198 			if ((tlen + NFS_DIRHDSIZ) > left) {
2199 				dp->d_reclen += left;
2200 				uiop->uio_iov->iov_base += left;
2201 				uiop->uio_iov->iov_len -= left;
2202 				uiop->uio_resid -= left;
2203 				blksiz = 0;
2204 			}
2205 			if ((tlen + NFS_DIRHDSIZ) > uiop->uio_resid)
2206 				bigenough = 0;
2207 			if (bigenough) {
2208 				ndp = (struct nfs_dirent *)
2209 				    uiop->uio_iov->iov_base;
2210 				dp = &ndp->dirent;
2211 				dp->d_fileno = (int)fileno;
2212 				dp->d_namlen = len;
2213 				dp->d_reclen = tlen + NFS_DIRHDSIZ;
2214 				dp->d_type = DT_UNKNOWN;
2215 				blksiz += dp->d_reclen;
2216 				if (blksiz == NFS_READDIRBLKSIZ)
2217 					blksiz = 0;
2218 				uiop->uio_resid -= NFS_DIRHDSIZ;
2219 				uiop->uio_iov->iov_base =
2220 				    (char *)uiop->uio_iov->iov_base +
2221 				    NFS_DIRHDSIZ;
2222 				uiop->uio_iov->iov_len -= NFS_DIRHDSIZ;
2223 				nfsm_mtouio(uiop, len);
2224 				cp = uiop->uio_iov->iov_base;
2225 				tlen -= len;
2226 				*cp = '\0';	/* null terminate */
2227 				uiop->uio_iov->iov_base += tlen;
2228 				uiop->uio_iov->iov_len -= tlen;
2229 				uiop->uio_resid -= tlen;
2230 			} else
2231 				nfsm_adv(nfsm_rndup(len));
2232 			if (info.nmi_v3) {
2233 				nfsm_dissect(tl, u_int32_t *,
2234 				    3 * NFSX_UNSIGNED);
2235 			} else {
2236 				nfsm_dissect(tl, u_int32_t *,
2237 				    2 * NFSX_UNSIGNED);
2238 			}
2239 			if (bigenough) {
2240 				if (info.nmi_v3) {
2241 					ndp->cookie[0] = cookie.nfsuquad[0] =
2242 					    *tl++;
2243 				} else
2244 					ndp->cookie[0] = 0;
2245 
2246 				ndp->cookie[1] = cookie.nfsuquad[1] = *tl++;
2247 			} else if (info.nmi_v3)
2248 				tl += 2;
2249 			else
2250 				tl++;
2251 			more_dirs = fxdr_unsigned(int, *tl);
2252 		}
2253 		/*
2254 		 * If at end of rpc data, get the eof boolean
2255 		 */
2256 		if (!more_dirs) {
2257 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2258 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2259 		}
2260 		m_freem(info.nmi_mrep);
2261 	}
2262 	/*
2263 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2264 	 * by increasing d_reclen for the last record.
2265 	 */
2266 	if (blksiz > 0) {
2267 		left = NFS_READDIRBLKSIZ - blksiz;
2268 		dp->d_reclen += left;
2269 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
2270 		    left;
2271 		uiop->uio_iov->iov_len -= left;
2272 		uiop->uio_resid -= left;
2273 	}
2274 
2275 	/*
2276 	 * We are now either at the end of the directory or have filled the
2277 	 * block.
2278 	 */
2279 	if (bigenough) {
2280 		dnp->n_direofoffset = fxdr_hyper(&cookie.nfsuquad[0]);
2281 		if (end_of_directory) *end_of_directory = 1;
2282 	} else {
2283 		if (uiop->uio_resid > 0)
2284 			printf("EEK! readdirrpc resid > 0\n");
2285 	}
2286 
2287 nfsmout:
2288 	return (error);
2289 }
2290 
2291 /*
2292  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2293  */
2294 int
2295 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2296     int *end_of_directory)
2297 {
2298 	int len, left;
2299 	struct nfs_dirent *ndirp = NULL;
2300 	struct dirent *dp = NULL;
2301 	struct nfsm_info	info;
2302 	u_int32_t *tl;
2303 	caddr_t cp;
2304 	int32_t t1;
2305 	struct vnode *newvp;
2306 	caddr_t cp2, dpossav1, dpossav2;
2307 	struct mbuf *mdsav1, *mdsav2;
2308 	struct nameidata nami, *ndp = &nami;
2309 	struct componentname *cnp = &ndp->ni_cnd;
2310 	nfsuint64 cookie;
2311 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2312 	struct nfsnode *dnp = VTONFS(vp), *np;
2313 	nfsfh_t *fhp;
2314 	u_quad_t fileno;
2315 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2316 	int attrflag, fhsize;
2317 
2318 #ifdef DIAGNOSTIC
2319 	if (uiop->uio_iovcnt != 1 ||
2320 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2321 		panic("nfs readdirplusrpc bad uio");
2322 #endif
2323 	ndp->ni_dvp = vp;
2324 	newvp = NULLVP;
2325 
2326 	txdr_hyper(uiop->uio_offset, &cookie.nfsuquad[0]);
2327 
2328 	/*
2329 	 * Loop around doing readdir rpc's of size nm_readdirsize
2330 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2331 	 * The stopping criteria is EOF or buffer full.
2332 	 */
2333 	while (more_dirs && bigenough) {
2334 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2335 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2336 		nfsm_fhtom(&info, vp, 1);
2337 		tl = nfsm_build(&info.nmi_mb, 6 * NFSX_UNSIGNED);
2338 		*tl++ = cookie.nfsuquad[0];
2339 		*tl++ = cookie.nfsuquad[1];
2340 		if (cookie.nfsuquad[0] == 0 &&
2341 		    cookie.nfsuquad[1] == 0) {
2342 			*tl++ = 0;
2343 			*tl++ = 0;
2344 		} else {
2345 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2346 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2347 		}
2348 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2349 		*tl = txdr_unsigned(nmp->nm_rsize);
2350 
2351 		info.nmi_procp = uiop->uio_procp;
2352 		info.nmi_cred = cred;
2353 		error = nfs_request(vp, NFSPROC_READDIRPLUS, &info);
2354 		nfsm_postop_attr(vp, attrflag);
2355 		if (error) {
2356 			m_freem(info.nmi_mrep);
2357 			goto nfsmout;
2358 		}
2359 
2360 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2361 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2362 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2363 		more_dirs = fxdr_unsigned(int, *tl);
2364 
2365 		/* loop thru the dir entries, doctoring them to 4bsd form */
2366 		while (more_dirs && bigenough) {
2367 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2368 			fileno = fxdr_hyper(tl);
2369 			len = fxdr_unsigned(int, *(tl + 2));
2370 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2371 				error = EBADRPC;
2372 				m_freem(info.nmi_mrep);
2373 				goto nfsmout;
2374 			}
2375 			tlen = nfsm_rndup(len + 1);
2376 			left = NFS_READDIRBLKSIZ - blksiz;
2377 			if ((tlen + NFS_DIRHDSIZ) > left) {
2378 				dp->d_reclen += left;
2379 				uiop->uio_iov->iov_base =
2380 				    (char *)uiop->uio_iov->iov_base + left;
2381 				uiop->uio_iov->iov_len -= left;
2382 				uiop->uio_resid -= left;
2383 				blksiz = 0;
2384 			}
2385 			if ((tlen + NFS_DIRHDSIZ) > uiop->uio_resid)
2386 				bigenough = 0;
2387 			if (bigenough) {
2388 				ndirp = (struct nfs_dirent *)
2389 				    uiop->uio_iov->iov_base;
2390 				dp = &ndirp->dirent;
2391 				dp->d_fileno = (int)fileno;
2392 				dp->d_namlen = len;
2393 				dp->d_reclen = tlen + NFS_DIRHDSIZ;
2394 				dp->d_type = DT_UNKNOWN;
2395 				blksiz += dp->d_reclen;
2396 				if (blksiz == NFS_READDIRBLKSIZ)
2397 					blksiz = 0;
2398 				uiop->uio_resid -= NFS_DIRHDSIZ;
2399 				uiop->uio_iov->iov_base =
2400 				    (char *)uiop->uio_iov->iov_base +
2401 				    NFS_DIRHDSIZ;
2402 				uiop->uio_iov->iov_len -= NFS_DIRHDSIZ;
2403 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2404 				cnp->cn_namelen = len;
2405 				nfsm_mtouio(uiop, len);
2406 				cp = uiop->uio_iov->iov_base;
2407 				tlen -= len;
2408 				*cp = '\0';
2409 				uiop->uio_iov->iov_base += tlen;
2410 				uiop->uio_iov->iov_len -= tlen;
2411 				uiop->uio_resid -= tlen;
2412 			} else
2413 				nfsm_adv(nfsm_rndup(len));
2414 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2415 			if (bigenough) {
2416 				ndirp->cookie[0] = cookie.nfsuquad[0] = *tl++;
2417 				ndirp->cookie[1] = cookie.nfsuquad[1] = *tl++;
2418 			} else
2419 				tl += 2;
2420 
2421 			/*
2422 			 * Since the attributes are before the file handle
2423 			 * (sigh), we must skip over the attributes and then
2424 			 * come back and get them.
2425 			 */
2426 			attrflag = fxdr_unsigned(int, *tl);
2427 			if (attrflag) {
2428 				dpossav1 = info.nmi_dpos;
2429 				mdsav1 = info.nmi_md;
2430 				nfsm_adv(NFSX_V3FATTR);
2431 				nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2432 				doit = fxdr_unsigned(int, *tl);
2433 				if (doit) {
2434 					nfsm_getfh(fhp, fhsize, 1);
2435 					if (NFS_CMPFH(dnp, fhp, fhsize)) {
2436 						vref(vp);
2437 						newvp = vp;
2438 						np = dnp;
2439 					} else {
2440 						error = nfs_nget(vp->v_mount,
2441 						    fhp, fhsize, &np);
2442 						if (error)
2443 							doit = 0;
2444 						else
2445 							newvp = NFSTOV(np);
2446 					}
2447 				}
2448 				if (doit && bigenough) {
2449 					dpossav2 = info.nmi_dpos;
2450 					info.nmi_dpos = dpossav1;
2451 					mdsav2 = info.nmi_md;
2452 					info.nmi_md = mdsav1;
2453 					nfsm_loadattr(newvp, NULL);
2454 					info.nmi_dpos = dpossav2;
2455 					info.nmi_md = mdsav2;
2456 					dp->d_type = IFTODT(
2457 						VTTOIF(np->n_vattr.va_type));
2458 					if (cnp->cn_namelen <= NCHNAMLEN) {
2459 						ndp->ni_vp = newvp;
2460 						cache_purge(ndp->ni_dvp);
2461 						nfs_cache_enter(ndp->ni_dvp,
2462 						    ndp->ni_vp, cnp);
2463 					}
2464 				}
2465 			} else {
2466 				/* Just skip over the file handle */
2467 				nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2468 				i = fxdr_unsigned(int, *tl);
2469 				nfsm_adv(nfsm_rndup(i));
2470 			}
2471 			if (newvp != NULLVP) {
2472 				vrele(newvp);
2473 				newvp = NULLVP;
2474 			}
2475 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2476 			more_dirs = fxdr_unsigned(int, *tl);
2477 		}
2478 		/*
2479 		 * If at end of rpc data, get the eof boolean
2480 		 */
2481 		if (!more_dirs) {
2482 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2483 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2484 		}
2485 		m_freem(info.nmi_mrep);
2486 	}
2487 	/*
2488 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2489 	 * by increasing d_reclen for the last record.
2490 	 */
2491 	if (blksiz > 0) {
2492 		left = NFS_READDIRBLKSIZ - blksiz;
2493 		dp->d_reclen += left;
2494 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
2495 		    left;
2496 		uiop->uio_iov->iov_len -= left;
2497 		uiop->uio_resid -= left;
2498 	}
2499 
2500 	/*
2501 	 * We are now either at the end of the directory or have filled the
2502 	 * block.
2503 	 */
2504 	if (bigenough) {
2505 		dnp->n_direofoffset = fxdr_hyper(&cookie.nfsuquad[0]);
2506 		if (end_of_directory) *end_of_directory = 1;
2507 	} else {
2508 		if (uiop->uio_resid > 0)
2509 			printf("EEK! readdirplusrpc resid > 0\n");
2510 	}
2511 
2512 nfsmout:
2513 	if (newvp != NULLVP)
2514 		vrele(newvp);
2515 	return (error);
2516 }
2517 
2518 /*
2519  * Silly rename. To make the NFS filesystem that is stateless look a little
2520  * more like the "ufs" a remove of an active vnode is translated to a rename
2521  * to a funny looking filename that is removed by nfs_inactive on the
2522  * nfsnode. There is the potential for another process on a different client
2523  * to create the same funny name between the nfs_lookitup() fails and the
2524  * nfs_rename() completes, but...
2525  */
2526 int
2527 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2528 {
2529 	struct sillyrename *sp;
2530 	struct nfsnode *np;
2531 	int error;
2532 
2533 	cache_purge(dvp);
2534 	np = VTONFS(vp);
2535 	sp = malloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
2536 	sp->s_cred = crdup(cnp->cn_cred);
2537 	sp->s_dvp = dvp;
2538 	vref(dvp);
2539 
2540 	if (vp->v_type == VDIR) {
2541 #ifdef DIAGNOSTIC
2542 		printf("nfs: sillyrename dir\n");
2543 #endif
2544 		error = EINVAL;
2545 		goto bad;
2546 	}
2547 
2548 	/* Try lookitups until we get one that isn't there */
2549 	while (1) {
2550 		/* Fudge together a funny name */
2551 		sp->s_namlen = snprintf(sp->s_name, sizeof sp->s_name,
2552 		    ".nfs%08X%08X", arc4random(), arc4random());
2553 		if (sp->s_namlen > sizeof sp->s_name)
2554 			sp->s_namlen = strlen(sp->s_name);
2555 
2556 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2557 		    cnp->cn_proc, NULL))
2558 			break;
2559 	}
2560 
2561 	error = nfs_renameit(dvp, cnp, sp);
2562 	if (error)
2563 		goto bad;
2564 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2565 		cnp->cn_proc, &np);
2566 	np->n_sillyrename = sp;
2567 	return (0);
2568 bad:
2569 	vrele(sp->s_dvp);
2570 	crfree(sp->s_cred);
2571 	free(sp, M_NFSREQ);
2572 	return (error);
2573 }
2574 
2575 /*
2576  * Look up a file name and optionally either update the file handle or
2577  * allocate an nfsnode, depending on the value of npp.
2578  * npp == NULL	--> just do the lookup
2579  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2580  *			handled too
2581  * *npp != NULL --> update the file handle in the vnode
2582  */
2583 int
2584 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2585     struct proc *procp, struct nfsnode **npp)
2586 {
2587 	struct nfsm_info	info;
2588 	u_int32_t *tl;
2589 	int32_t t1;
2590 	struct vnode *newvp = NULL;
2591 	struct nfsnode *np, *dnp = VTONFS(dvp);
2592 	caddr_t cp2;
2593 	int error = 0, fhlen, attrflag;
2594 	nfsfh_t *nfhp;
2595 
2596 	info.nmi_v3 = NFS_ISV3(dvp);
2597 
2598 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2599 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) + NFSX_UNSIGNED +
2600 	    nfsm_rndup(len));
2601 	nfsm_fhtom(&info, dvp, info.nmi_v3);
2602 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2603 
2604 	info.nmi_procp = procp;
2605 	info.nmi_cred = cred;
2606 	error = nfs_request(dvp, NFSPROC_LOOKUP, &info);
2607 	if (error && !info.nmi_v3) {
2608 		m_freem(info.nmi_mrep);
2609 		goto nfsmout;
2610 	}
2611 
2612 	if (npp && !error) {
2613 		nfsm_getfh(nfhp, fhlen, info.nmi_v3);
2614 		if (*npp) {
2615 			np = *npp;
2616 			np->n_fhp = &np->n_fh;
2617 			bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2618 			np->n_fhsize = fhlen;
2619 			newvp = NFSTOV(np);
2620 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2621 			vref(dvp);
2622 			newvp = dvp;
2623 			np = dnp;
2624 		} else {
2625 			error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2626 			if (error) {
2627 				m_freem(info.nmi_mrep);
2628 				return (error);
2629 			}
2630 			newvp = NFSTOV(np);
2631 		}
2632 		if (info.nmi_v3) {
2633 			nfsm_postop_attr(newvp, attrflag);
2634 			if (!attrflag && *npp == NULL) {
2635 				m_freem(info.nmi_mrep);
2636 				vrele(newvp);
2637 				return (ENOENT);
2638 			}
2639 		} else
2640 			nfsm_loadattr(newvp, NULL);
2641 	}
2642 	m_freem(info.nmi_mrep);
2643 nfsmout:
2644 	if (npp && *npp == NULL) {
2645 		if (error) {
2646 			if (newvp)
2647 				vrele(newvp);
2648 		} else
2649 			*npp = np;
2650 	}
2651 	return (error);
2652 }
2653 
2654 /*
2655  * Nfs Version 3 commit rpc
2656  */
2657 int
2658 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct proc *procp)
2659 {
2660 	struct nfsm_info	info;
2661 	u_int32_t *tl;
2662 	int32_t t1;
2663 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2664 	caddr_t cp2;
2665 	int error = 0, wccflag = NFSV3_WCCRATTR;
2666 
2667 	if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0)
2668 		return (0);
2669 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2670 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(1));
2671 	nfsm_fhtom(&info, vp, 1);
2672 
2673 	tl = nfsm_build(&info.nmi_mb, 3 * NFSX_UNSIGNED);
2674 	txdr_hyper(offset, tl);
2675 	tl += 2;
2676 	*tl = txdr_unsigned(cnt);
2677 
2678 	info.nmi_procp = procp;
2679 	info.nmi_cred = VTONFS(vp)->n_wcred;
2680 	error = nfs_request(vp, NFSPROC_COMMIT, &info);
2681 	nfsm_wcc_data(vp, wccflag);
2682 
2683 	if (!error) {
2684 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2685 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2686 			NFSX_V3WRITEVERF)) {
2687 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2688 				NFSX_V3WRITEVERF);
2689 			error = NFSERR_STALEWRITEVERF;
2690 		}
2691 	}
2692 	m_freem(info.nmi_mrep);
2693 
2694 nfsmout:
2695 	return (error);
2696 }
2697 
2698 /*
2699  * Kludge City..
2700  * - make nfs_bmap() essentially a no-op that does no translation
2701  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2702  *   (Maybe I could use the process's page mapping, but I was concerned that
2703  *    Kernel Write might not be enabled and also figured copyout() would do
2704  *    a lot more work than bcopy() and also it currently happens in the
2705  *    context of the swapper process (2).
2706  */
2707 int
2708 nfs_bmap(void *v)
2709 {
2710 	struct vop_bmap_args *ap = v;
2711 	struct vnode *vp = ap->a_vp;
2712 
2713 	if (ap->a_vpp != NULL)
2714 		*ap->a_vpp = vp;
2715 	if (ap->a_bnp != NULL)
2716 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2717 	return (0);
2718 }
2719 
2720 /*
2721  * Strategy routine.
2722  * For async requests when nfsiod(s) are running, queue the request by
2723  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2724  * request.
2725  */
2726 int
2727 nfs_strategy(void *v)
2728 {
2729 	struct vop_strategy_args *ap = v;
2730 	struct buf *bp = ap->a_bp;
2731 	struct proc *p;
2732 	int error = 0;
2733 
2734 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2735 		panic("nfs physio/async");
2736 	if (bp->b_flags & B_ASYNC)
2737 		p = NULL;
2738 	else
2739 		p = curproc;	/* XXX */
2740 	/*
2741 	 * If the op is asynchronous and an i/o daemon is waiting
2742 	 * queue the request, wake it up and wait for completion
2743 	 * otherwise just do it ourselves.
2744 	 */
2745 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp, 0))
2746 		error = nfs_doio(bp, p);
2747 	return (error);
2748 }
2749 
2750 /*
2751  * fsync vnode op. Just call nfs_flush() with commit == 1.
2752  */
2753 int
2754 nfs_fsync(void *v)
2755 {
2756 	struct vop_fsync_args *ap = v;
2757 
2758 	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2759 }
2760 
2761 /*
2762  * Flush all the blocks associated with a vnode.
2763  *	Walk through the buffer pool and push any dirty pages
2764  *	associated with the vnode.
2765  */
2766 int
2767 nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct proc *p,
2768     int commit)
2769 {
2770 	struct nfsnode *np = VTONFS(vp);
2771 	struct buf *bp;
2772 	int i;
2773 	struct buf *nbp;
2774 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2775 	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2776 	int passone = 1;
2777 	u_quad_t off = (u_quad_t)-1, endoff = 0, toff;
2778 #ifndef NFS_COMMITBVECSIZ
2779 #define NFS_COMMITBVECSIZ	20
2780 #endif
2781 	struct buf *bvec[NFS_COMMITBVECSIZ];
2782 
2783 	if (nmp->nm_flag & NFSMNT_INT)
2784 		slpflag = PCATCH;
2785 	if (!commit)
2786 		passone = 0;
2787 	/*
2788 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2789 	 * server, but nas not been committed to stable storage on the server
2790 	 * yet. On the first pass, the byte range is worked out and the commit
2791 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2792 	 * job.
2793 	 */
2794 again:
2795 	bvecpos = 0;
2796 	if (NFS_ISV3(vp) && commit) {
2797 		s = splbio();
2798 		for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp != NULL; bp = nbp) {
2799 			nbp = LIST_NEXT(bp, b_vnbufs);
2800 			if (bvecpos >= NFS_COMMITBVECSIZ)
2801 				break;
2802 			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2803 				!= (B_DELWRI | B_NEEDCOMMIT))
2804 				continue;
2805 			bremfree(bp);
2806 			bp->b_flags |= B_WRITEINPROG;
2807 			buf_acquire(bp);
2808 			/*
2809 			 * A list of these buffers is kept so that the
2810 			 * second loop knows which buffers have actually
2811 			 * been committed. This is necessary, since there
2812 			 * may be a race between the commit rpc and new
2813 			 * uncommitted writes on the file.
2814 			 */
2815 			bvec[bvecpos++] = bp;
2816 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2817 				bp->b_dirtyoff;
2818 			if (toff < off)
2819 				off = toff;
2820 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2821 			if (toff > endoff)
2822 				endoff = toff;
2823 		}
2824 		splx(s);
2825 	}
2826 	if (bvecpos > 0) {
2827 		/*
2828 		 * Commit data on the server, as required.
2829 		 */
2830 		bcstats.pendingwrites++;
2831 		bcstats.numwrites++;
2832 		retv = nfs_commit(vp, off, (int)(endoff - off), p);
2833 		if (retv == NFSERR_STALEWRITEVERF)
2834 			nfs_clearcommit(vp->v_mount);
2835 		/*
2836 		 * Now, either mark the blocks I/O done or mark the
2837 		 * blocks dirty, depending on whether the commit
2838 		 * succeeded.
2839 		 */
2840 		for (i = 0; i < bvecpos; i++) {
2841 			bp = bvec[i];
2842 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG);
2843 			if (retv) {
2844 				if (i == 0)
2845 					bcstats.pendingwrites--;
2846 				brelse(bp);
2847 			} else {
2848 				if (i > 0)
2849 					bcstats.pendingwrites++;
2850 				s = splbio();
2851 				buf_undirty(bp);
2852 				vp->v_numoutput++;
2853 				bp->b_flags |= B_ASYNC;
2854 				bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
2855 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2856 				biodone(bp);
2857 				splx(s);
2858 			}
2859 		}
2860 	}
2861 
2862 	/*
2863 	 * Start/do any write(s) that are required.
2864 	 */
2865 loop:
2866 	s = splbio();
2867 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp != NULL; bp = nbp) {
2868 		nbp = LIST_NEXT(bp, b_vnbufs);
2869 		if (bp->b_flags & B_BUSY) {
2870 			if (waitfor != MNT_WAIT || passone)
2871 				continue;
2872 			bp->b_flags |= B_WANTED;
2873 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
2874 				"nfsfsync", slptimeo);
2875 			splx(s);
2876 			if (error) {
2877 				if (nfs_sigintr(nmp, NULL, p))
2878 					return (EINTR);
2879 				if (slpflag == PCATCH) {
2880 					slpflag = 0;
2881 					slptimeo = 2 * hz;
2882 				}
2883 			}
2884 			goto loop;
2885 		}
2886 		if ((bp->b_flags & B_DELWRI) == 0)
2887 			panic("nfs_fsync: not dirty");
2888 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT))
2889 			continue;
2890 		bremfree(bp);
2891 		if (passone || !commit) {
2892 			bp->b_flags |= B_ASYNC;
2893 		} else {
2894 			bp->b_flags |= (B_ASYNC|B_WRITEINPROG|B_NEEDCOMMIT);
2895 		}
2896 		buf_acquire(bp);
2897 		splx(s);
2898 		VOP_BWRITE(bp);
2899 		goto loop;
2900 	}
2901 	splx(s);
2902 	if (passone) {
2903 		passone = 0;
2904 		goto again;
2905 	}
2906 	if (waitfor == MNT_WAIT) {
2907  loop2:
2908 		s = splbio();
2909 		error = vwaitforio(vp, slpflag, "nfs_fsync", slptimeo);
2910 		splx(s);
2911 		if (error) {
2912 			if (nfs_sigintr(nmp, NULL, p))
2913 				return (EINTR);
2914 			if (slpflag == PCATCH) {
2915 				slpflag = 0;
2916 				slptimeo = 2 * hz;
2917 			}
2918 			goto loop2;
2919 		}
2920 
2921 		if (LIST_FIRST(&vp->v_dirtyblkhd) && commit) {
2922 #if 0
2923 			vprint("nfs_fsync: dirty", vp);
2924 #endif
2925 			goto loop;
2926 		}
2927 	}
2928 	if (np->n_flag & NWRITEERR) {
2929 		error = np->n_error;
2930 		np->n_flag &= ~NWRITEERR;
2931 	}
2932 	return (error);
2933 }
2934 
2935 /*
2936  * Return POSIX pathconf information applicable to nfs.
2937  *
2938  * The NFS V2 protocol doesn't support this, so just return EINVAL
2939  * for V2.
2940  */
2941 /* ARGSUSED */
2942 int
2943 nfs_pathconf(void *v)
2944 {
2945 #if 0
2946 	struct vop_pathconf_args *ap = v;
2947 #endif
2948 
2949 	return (EINVAL);
2950 }
2951 
2952 /*
2953  * NFS advisory byte-level locks.
2954  */
2955 int
2956 nfs_advlock(void *v)
2957 {
2958 	struct vop_advlock_args *ap = v;
2959 	struct nfsnode *np = VTONFS(ap->a_vp);
2960 
2961 	return (lf_advlock(&np->n_lockf, np->n_size, ap->a_id, ap->a_op,
2962 	    ap->a_fl, ap->a_flags));
2963 }
2964 
2965 /*
2966  * Print out the contents of an nfsnode.
2967  */
2968 int
2969 nfs_print(void *v)
2970 {
2971 	struct vop_print_args *ap = v;
2972 	struct vnode *vp = ap->a_vp;
2973 	struct nfsnode *np = VTONFS(vp);
2974 
2975 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
2976 		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2977 #ifdef FIFO
2978 	if (vp->v_type == VFIFO)
2979 		fifo_printinfo(vp);
2980 #endif
2981 	printf("\n");
2982 	return (0);
2983 }
2984 
2985 /*
2986  * Just call nfs_writebp() with the force argument set to 1.
2987  */
2988 int
2989 nfs_bwrite(void *v)
2990 {
2991 	struct vop_bwrite_args *ap = v;
2992 
2993 	return (nfs_writebp(ap->a_bp, 1));
2994 }
2995 
2996 /*
2997  * This is a clone of vop_generic_bwrite(), except that B_WRITEINPROG isn't set unless
2998  * the force flag is one and it also handles the B_NEEDCOMMIT flag.
2999  */
3000 int
3001 nfs_writebp(struct buf *bp, int force)
3002 {
3003 	int oldflags = bp->b_flags, retv = 1;
3004 	struct proc *p = curproc;	/* XXX */
3005 	off_t off;
3006 	size_t cnt;
3007 	int   s;
3008 	struct vnode *vp;
3009 	struct nfsnode *np;
3010 
3011 	if(!(bp->b_flags & B_BUSY))
3012 		panic("bwrite: buffer is not busy???");
3013 
3014 	vp = bp->b_vp;
3015 	np = VTONFS(vp);
3016 
3017 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3018 
3019 	s = splbio();
3020 	buf_undirty(bp);
3021 
3022 	if ((oldflags & B_ASYNC) && !(oldflags & B_DELWRI) && p)
3023 		++p->p_stats->p_ru.ru_oublock;
3024 
3025 	bp->b_vp->v_numoutput++;
3026 	splx(s);
3027 
3028 	/*
3029 	 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not
3030 	 * an actual write will have to be scheduled via. VOP_STRATEGY().
3031 	 * If B_WRITEINPROG is already set, then push it with a write anyhow.
3032 	 */
3033 	if ((oldflags & (B_NEEDCOMMIT | B_WRITEINPROG)) == B_NEEDCOMMIT) {
3034 		off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
3035 		cnt = bp->b_dirtyend - bp->b_dirtyoff;
3036 
3037 		rw_enter_write(&np->n_commitlock);
3038 		if (!(bp->b_flags & B_NEEDCOMMIT)) {
3039 			rw_exit_write(&np->n_commitlock);
3040 			return (0);
3041 		}
3042 
3043 		/*
3044 		 * If it's already been commited by somebody else,
3045 		 * bail.
3046 		 */
3047 		if (!nfs_in_committed_range(vp, bp)) {
3048 			int pushedrange = 0;
3049 			/*
3050 			 * Since we're going to do this, push as much
3051 			 * as we can.
3052 			 */
3053 
3054 			if (nfs_in_tobecommitted_range(vp, bp)) {
3055 				pushedrange = 1;
3056 				off = np->n_pushlo;
3057 				cnt = np->n_pushhi - np->n_pushlo;
3058 			}
3059 
3060 			bp->b_flags |= B_WRITEINPROG;
3061 			bcstats.pendingwrites++;
3062 			bcstats.numwrites++;
3063 			retv = nfs_commit(bp->b_vp, off, cnt, curproc);
3064 			bp->b_flags &= ~B_WRITEINPROG;
3065 
3066 			if (retv == 0) {
3067 				if (pushedrange)
3068 					nfs_merge_commit_ranges(vp);
3069 				else
3070 					nfs_add_committed_range(vp, bp);
3071 			} else
3072 				bcstats.pendingwrites--;
3073 		} else
3074 			retv = 0; /* It has already been commited. */
3075 
3076 		rw_exit_write(&np->n_commitlock);
3077 		if (!retv) {
3078 			bp->b_dirtyoff = bp->b_dirtyend = 0;
3079 			bp->b_flags &= ~B_NEEDCOMMIT;
3080 			s = splbio();
3081 			biodone(bp);
3082 			splx(s);
3083 		} else if (retv == NFSERR_STALEWRITEVERF)
3084 			nfs_clearcommit(bp->b_vp->v_mount);
3085 	}
3086 	if (retv) {
3087 		if (force)
3088 			bp->b_flags |= B_WRITEINPROG;
3089 		VOP_STRATEGY(bp);
3090 	}
3091 
3092 	if( (oldflags & B_ASYNC) == 0) {
3093 		int rtval;
3094 
3095 		bp->b_flags |= B_RAW;
3096 		rtval = biowait(bp);
3097 		if (!(oldflags & B_DELWRI) && p) {
3098 			++p->p_stats->p_ru.ru_oublock;
3099 		}
3100 		brelse(bp);
3101 		return (rtval);
3102 	}
3103 
3104 	return (0);
3105 }
3106 
3107 /*
3108  * nfs special file access vnode op.
3109  * Essentially just get vattr and then imitate iaccess() since the device is
3110  * local to the client.
3111  */
3112 int
3113 nfsspec_access(void *v)
3114 {
3115 	struct vop_access_args *ap = v;
3116 	struct vattr va;
3117 	struct vnode *vp = ap->a_vp;
3118 	int error;
3119 
3120 	/*
3121 	 * Disallow write attempts on filesystems mounted read-only;
3122 	 * unless the file is a socket, fifo, or a block or character
3123 	 * device resident on the filesystem.
3124 	 */
3125 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3126 		switch (vp->v_type) {
3127 		case VREG:
3128 		case VDIR:
3129 		case VLNK:
3130 			return (EROFS);
3131 		default:
3132 			break;
3133 		}
3134 	}
3135 
3136 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3137 	if (error)
3138 		return (error);
3139 
3140 	return (vaccess(vp->v_type, va.va_mode, va.va_uid, va.va_gid,
3141 	    ap->a_mode, ap->a_cred));
3142 }
3143 
3144 int
3145 nfs_poll(void *v)
3146 {
3147 	struct vop_poll_args *ap = v;
3148 
3149 	/*
3150 	 * We should really check to see if I/O is possible.
3151 	 */
3152 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
3153 }
3154 
3155 /*
3156  * Read wrapper for special devices.
3157  */
3158 int
3159 nfsspec_read(void *v)
3160 {
3161 	struct vop_read_args *ap = v;
3162 	struct nfsnode *np = VTONFS(ap->a_vp);
3163 
3164 	/*
3165 	 * Set access flag.
3166 	 */
3167 	np->n_flag |= NACC;
3168 	getnanotime(&np->n_atim);
3169 	return (spec_read(ap));
3170 }
3171 
3172 /*
3173  * Write wrapper for special devices.
3174  */
3175 int
3176 nfsspec_write(void *v)
3177 {
3178 	struct vop_write_args *ap = v;
3179 	struct nfsnode *np = VTONFS(ap->a_vp);
3180 
3181 	/*
3182 	 * Set update flag.
3183 	 */
3184 	np->n_flag |= NUPD;
3185 	getnanotime(&np->n_mtim);
3186 	return (spec_write(ap));
3187 }
3188 
3189 /*
3190  * Close wrapper for special devices.
3191  *
3192  * Update the times on the nfsnode then do device close.
3193  */
3194 int
3195 nfsspec_close(void *v)
3196 {
3197 	struct vop_close_args *ap = v;
3198 	struct vnode *vp = ap->a_vp;
3199 	struct nfsnode *np = VTONFS(vp);
3200 	struct vattr vattr;
3201 
3202 	if (np->n_flag & (NACC | NUPD)) {
3203 		np->n_flag |= NCHG;
3204 		if (vp->v_usecount == 1 &&
3205 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3206 			VATTR_NULL(&vattr);
3207 			if (np->n_flag & NACC)
3208 				vattr.va_atime = np->n_atim;
3209 			if (np->n_flag & NUPD)
3210 				vattr.va_mtime = np->n_mtim;
3211 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3212 		}
3213 	}
3214 	return (spec_close(ap));
3215 }
3216 
3217 #ifdef FIFO
3218 /*
3219  * Read wrapper for fifos.
3220  */
3221 int
3222 nfsfifo_read(void *v)
3223 {
3224 	struct vop_read_args *ap = v;
3225 	struct nfsnode *np = VTONFS(ap->a_vp);
3226 
3227 	/*
3228 	 * Set access flag.
3229 	 */
3230 	np->n_flag |= NACC;
3231 	getnanotime(&np->n_atim);
3232 	return (fifo_read(ap));
3233 }
3234 
3235 /*
3236  * Write wrapper for fifos.
3237  */
3238 int
3239 nfsfifo_write(void *v)
3240 {
3241 	struct vop_write_args *ap = v;
3242 	struct nfsnode *np = VTONFS(ap->a_vp);
3243 
3244 	/*
3245 	 * Set update flag.
3246 	 */
3247 	np->n_flag |= NUPD;
3248 	getnanotime(&np->n_mtim);
3249 	return (fifo_write(ap));
3250 }
3251 
3252 /*
3253  * Close wrapper for fifos.
3254  *
3255  * Update the times on the nfsnode then do fifo close.
3256  */
3257 int
3258 nfsfifo_close(void *v)
3259 {
3260 	struct vop_close_args *ap = v;
3261 	struct vnode *vp = ap->a_vp;
3262 	struct nfsnode *np = VTONFS(vp);
3263 	struct vattr vattr;
3264 
3265 	if (np->n_flag & (NACC | NUPD)) {
3266 		if (np->n_flag & NACC) {
3267 			getnanotime(&np->n_atim);
3268 		}
3269 		if (np->n_flag & NUPD) {
3270 			getnanotime(&np->n_mtim);
3271 		}
3272 		np->n_flag |= NCHG;
3273 		if (vp->v_usecount == 1 &&
3274 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3275 			VATTR_NULL(&vattr);
3276 			if (np->n_flag & NACC)
3277 				vattr.va_atime = np->n_atim;
3278 			if (np->n_flag & NUPD)
3279 				vattr.va_mtime = np->n_mtim;
3280 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3281 		}
3282 	}
3283 	return (fifo_close(ap));
3284 }
3285 
3286 int
3287 nfsfifo_reclaim(void *v)
3288 {
3289 	fifo_reclaim(v);
3290 	return (nfs_reclaim(v));
3291 }
3292 #endif /* ! FIFO */
3293