xref: /openbsd-src/sys/nfs/nfs_vnops.c (revision 25c4e8bd056e974b28f4a0ffd39d76c190a56013)
1 /*	$OpenBSD: nfs_vnops.c,v 1.189 2022/06/26 05:20:42 visa 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/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/resourcevar.h>
47 #include <sys/proc.h>
48 #include <sys/mount.h>
49 #include <sys/buf.h>
50 #include <sys/malloc.h>
51 #include <sys/pool.h>
52 #include <sys/mbuf.h>
53 #include <sys/conf.h>
54 #include <sys/namei.h>
55 #include <sys/vnode.h>
56 #include <sys/lock.h>
57 #include <sys/dirent.h>
58 #include <sys/fcntl.h>
59 #include <sys/lockf.h>
60 #include <sys/queue.h>
61 #include <sys/specdev.h>
62 #include <sys/unistd.h>
63 
64 #include <miscfs/fifofs/fifo.h>
65 
66 #include <nfs/rpcv2.h>
67 #include <nfs/nfsproto.h>
68 #include <nfs/nfs.h>
69 #include <nfs/nfsnode.h>
70 #include <nfs/nfsmount.h>
71 #include <nfs/xdr_subs.h>
72 #include <nfs/nfsm_subs.h>
73 #include <nfs/nfs_var.h>
74 
75 #include <uvm/uvm_extern.h>
76 
77 #include <netinet/in.h>
78 
79 int nfs_access(void *);
80 int nfs_advlock(void *);
81 int nfs_bmap(void *);
82 int nfs_bwrite(void *);
83 int nfs_close(void *);
84 int nfs_commit(struct vnode *, u_quad_t, int, struct proc *);
85 int nfs_create(void *);
86 int nfs_flush(struct vnode *, struct ucred *, int, struct proc *, int);
87 int nfs_fsync(void *);
88 int nfs_getattr(void *);
89 int nfs_getreq(struct nfsrv_descript *, struct nfsd *, int);
90 int nfs_islocked(void *);
91 int nfs_link(void *);
92 int nfs_lock(void *);
93 int nfs_lookitup(struct vnode *, char *, int, struct ucred *, struct proc *,
94 	struct nfsnode **);
95 int nfs_lookup(void *);
96 int nfs_mkdir(void *);
97 int nfs_mknod(void *);
98 int nfs_mknodrpc(struct vnode *, struct vnode **, struct componentname *,
99 	struct vattr *);
100 int nfs_null(struct vnode *, struct ucred *, struct proc *);
101 int nfs_open(void *);
102 int nfs_pathconf(void *);
103 int nfs_print(void *);
104 int nfs_read(void *);
105 int nfs_readdir(void *);
106 int nfs_readdirplusrpc(struct vnode *, struct uio *, struct ucred *, int *,
107 	struct proc *);
108 int nfs_readdirrpc(struct vnode *, struct uio *, struct ucred *, int *);
109 int nfs_remove(void *);
110 int nfs_removerpc(struct vnode *, char *, int, struct ucred *, struct proc *);
111 int nfs_rename(void *);
112 int nfs_renameit(struct vnode *, struct componentname *, struct sillyrename *);
113 int nfs_renamerpc(struct vnode *, char *, int, struct vnode *, char *, int,
114 	struct ucred *, struct proc *);
115 int nfs_rmdir(void *);
116 int nfs_setattr(void *);
117 int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
118 	struct proc *);
119 int nfs_sillyrename(struct vnode *, struct vnode *,
120 			 struct componentname *);
121 int nfs_strategy(void *);
122 int nfs_symlink(void *);
123 int nfs_unlock(void *);
124 
125 void nfs_cache_enter(struct vnode *, struct vnode *, struct componentname *);
126 
127 int nfsfifo_close(void *);
128 int nfsfifo_read(void *);
129 int nfsfifo_reclaim(void *);
130 int nfsfifo_write(void *);
131 
132 int nfsspec_access(void *);
133 int nfsspec_close(void *);
134 int nfsspec_read(void *);
135 int nfsspec_write(void *);
136 
137 /* Global vfs data structures for nfs. */
138 const struct vops nfs_vops = {
139 	.vop_lookup	= nfs_lookup,
140 	.vop_create	= nfs_create,
141 	.vop_mknod	= nfs_mknod,
142 	.vop_open	= nfs_open,
143 	.vop_close	= nfs_close,
144 	.vop_access	= nfs_access,
145 	.vop_getattr	= nfs_getattr,
146 	.vop_setattr	= nfs_setattr,
147 	.vop_read	= nfs_read,
148 	.vop_write	= nfs_write,
149 	.vop_ioctl	= nfs_ioctl,
150 	.vop_kqfilter	= nfs_kqfilter,
151 	.vop_revoke	= vop_generic_revoke,
152 	.vop_fsync	= nfs_fsync,
153 	.vop_remove	= nfs_remove,
154 	.vop_link	= nfs_link,
155 	.vop_rename	= nfs_rename,
156 	.vop_mkdir	= nfs_mkdir,
157 	.vop_rmdir	= nfs_rmdir,
158 	.vop_symlink	= nfs_symlink,
159 	.vop_readdir	= nfs_readdir,
160 	.vop_readlink	= nfs_readlink,
161 	.vop_abortop	= vop_generic_abortop,
162 	.vop_inactive	= nfs_inactive,
163 	.vop_reclaim	= nfs_reclaim,
164 	.vop_lock	= nfs_lock,
165 	.vop_unlock	= nfs_unlock,
166 	.vop_bmap	= nfs_bmap,
167 	.vop_strategy	= nfs_strategy,
168 	.vop_print	= nfs_print,
169 	.vop_islocked	= nfs_islocked,
170 	.vop_pathconf	= nfs_pathconf,
171 	.vop_advlock	= nfs_advlock,
172 	.vop_bwrite	= nfs_bwrite
173 };
174 
175 /* Special device vnode ops. */
176 const struct vops nfs_specvops = {
177 	.vop_close	= nfsspec_close,
178 	.vop_access	= nfsspec_access,
179 	.vop_getattr	= nfs_getattr,
180 	.vop_setattr	= nfs_setattr,
181 	.vop_read	= nfsspec_read,
182 	.vop_write	= nfsspec_write,
183 	.vop_fsync	= nfs_fsync,
184 	.vop_inactive	= nfs_inactive,
185 	.vop_reclaim	= nfs_reclaim,
186 	.vop_lock	= nfs_lock,
187 	.vop_unlock	= nfs_unlock,
188 	.vop_print	= nfs_print,
189 	.vop_islocked	= nfs_islocked,
190 
191 	/* XXX: Keep in sync with spec_vops. */
192 	.vop_lookup	= vop_generic_lookup,
193 	.vop_create	= vop_generic_badop,
194 	.vop_mknod	= vop_generic_badop,
195 	.vop_open	= spec_open,
196 	.vop_ioctl	= spec_ioctl,
197 	.vop_kqfilter	= spec_kqfilter,
198 	.vop_revoke	= vop_generic_revoke,
199 	.vop_remove	= vop_generic_badop,
200 	.vop_link	= vop_generic_badop,
201 	.vop_rename	= vop_generic_badop,
202 	.vop_mkdir	= vop_generic_badop,
203 	.vop_rmdir	= vop_generic_badop,
204 	.vop_symlink	= vop_generic_badop,
205 	.vop_readdir	= vop_generic_badop,
206 	.vop_readlink	= vop_generic_badop,
207 	.vop_abortop	= vop_generic_badop,
208 	.vop_bmap	= vop_generic_bmap,
209 	.vop_strategy	= spec_strategy,
210 	.vop_pathconf	= spec_pathconf,
211 	.vop_advlock	= spec_advlock,
212 	.vop_bwrite	= vop_generic_bwrite,
213 };
214 
215 #ifdef FIFO
216 const struct vops nfs_fifovops = {
217 	.vop_close	= nfsfifo_close,
218 	.vop_access	= nfsspec_access,
219 	.vop_getattr	= nfs_getattr,
220 	.vop_setattr	= nfs_setattr,
221 	.vop_read	= nfsfifo_read,
222 	.vop_write	= nfsfifo_write,
223 	.vop_fsync	= nfs_fsync,
224 	.vop_inactive	= nfs_inactive,
225 	.vop_reclaim	= nfsfifo_reclaim,
226 	.vop_lock	= nfs_lock,
227 	.vop_unlock	= nfs_unlock,
228 	.vop_print	= nfs_print,
229 	.vop_islocked	= nfs_islocked,
230 	.vop_bwrite	= vop_generic_bwrite,
231 
232 	/* XXX: Keep in sync with fifo_vops. */
233 	.vop_lookup	= vop_generic_lookup,
234 	.vop_create	= vop_generic_badop,
235 	.vop_mknod	= vop_generic_badop,
236 	.vop_open	= fifo_open,
237 	.vop_ioctl	= fifo_ioctl,
238 	.vop_kqfilter	= fifo_kqfilter,
239 	.vop_revoke	= vop_generic_revoke,
240 	.vop_remove	= vop_generic_badop,
241 	.vop_link	= vop_generic_badop,
242 	.vop_rename	= vop_generic_badop,
243 	.vop_mkdir	= vop_generic_badop,
244 	.vop_rmdir	= vop_generic_badop,
245 	.vop_symlink	= vop_generic_badop,
246 	.vop_readdir	= vop_generic_badop,
247 	.vop_readlink	= vop_generic_badop,
248 	.vop_abortop	= vop_generic_badop,
249 	.vop_bmap	= vop_generic_bmap,
250 	.vop_strategy	= vop_generic_badop,
251 	.vop_pathconf	= fifo_pathconf,
252 	.vop_advlock	= fifo_advlock,
253 };
254 #endif /* FIFO */
255 
256 /*
257  * Global variables
258  */
259 extern u_int32_t nfs_true, nfs_false;
260 extern u_int32_t nfs_xdrneg1;
261 extern struct nfsstats nfsstats;
262 extern nfstype nfsv3_type[9];
263 int nfs_numasync = 0;
264 
265 void
266 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
267 {
268 	struct nfsnode *np;
269 
270 	if (vp != NULL) {
271 		np = VTONFS(vp);
272 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
273 	} else {
274 		np = VTONFS(dvp);
275 		if (!np->n_ctime)
276 			np->n_ctime = np->n_vattr.va_mtime.tv_sec;
277 	}
278 
279 	cache_enter(dvp, vp, cnp);
280 }
281 
282 /*
283  * nfs null call from vfs.
284  */
285 int
286 nfs_null(struct vnode *vp, struct ucred *cred, struct proc *procp)
287 {
288 	struct nfsm_info	 info;
289 	int			 error = 0;
290 
291 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(0);
292 	error = nfs_request(vp, NFSPROC_NULL, &info);
293 	m_freem(info.nmi_mrep);
294 	return (error);
295 }
296 
297 /*
298  * nfs access vnode op.
299  * For nfs version 2, just return ok. File accesses may fail later.
300  * For nfs version 3, use the access rpc to check accessibility. If file modes
301  * are changed on the server, accesses might still fail later.
302  */
303 int
304 nfs_access(void *v)
305 {
306 	struct vop_access_args *ap = v;
307 	struct vnode *vp = ap->a_vp;
308 	u_int32_t *tl;
309 	int32_t t1;
310 	caddr_t cp2;
311 	int error = 0, attrflag;
312 	u_int32_t mode, rmode;
313 	int v3 = NFS_ISV3(vp);
314 	int cachevalid;
315 	struct nfsm_info	info;
316 
317 	struct nfsnode *np = VTONFS(vp);
318 
319 	/*
320 	 * Disallow write attempts on filesystems mounted read-only;
321 	 * unless the file is a socket, fifo, or a block or character
322 	 * device resident on the filesystem.
323 	 */
324 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
325 		switch (vp->v_type) {
326 		case VREG:
327 		case VDIR:
328 		case VLNK:
329 			return (EROFS);
330 		default:
331 			break;
332 		}
333 	}
334 
335 	/*
336 	 * Check access cache first. If a request has been made for this uid
337 	 * shortly before, use the cached result.
338 	 */
339 	 cachevalid = (np->n_accstamp != -1 &&
340 	     (gettime() - np->n_accstamp) < nfs_attrtimeo(np) &&
341 	     np->n_accuid == ap->a_cred->cr_uid);
342 
343 	if (cachevalid) {
344 		if (!np->n_accerror) {
345 			if ((np->n_accmode & ap->a_mode) == ap->a_mode)
346 				return (np->n_accerror);
347 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
348 			return (np->n_accerror);
349 	}
350 
351 	/*
352 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
353 	 * ufs_access() locally using the vattr. This may not be correct,
354 	 * since the server may apply other access criteria such as
355 	 * client uid-->server uid mapping that we do not know about, but
356 	 * this is better than just returning anything that is lying about
357 	 * in the cache.
358 	 */
359 	if (v3) {
360 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
361 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED);
362 		nfsm_fhtom(&info, vp, v3);
363 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
364 		if (ap->a_mode & VREAD)
365 			mode = NFSV3ACCESS_READ;
366 		else
367 			mode = 0;
368 		if (vp->v_type == VDIR) {
369 			if (ap->a_mode & VWRITE)
370 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
371 					 NFSV3ACCESS_DELETE);
372 			if (ap->a_mode & VEXEC)
373 				mode |= NFSV3ACCESS_LOOKUP;
374 		} else {
375 			if (ap->a_mode & VWRITE)
376 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
377 			if (ap->a_mode & VEXEC)
378 				mode |= NFSV3ACCESS_EXECUTE;
379 		}
380 		*tl = txdr_unsigned(mode);
381 
382 		info.nmi_procp = ap->a_p;
383 		info.nmi_cred = ap->a_cred;
384 		error = nfs_request(vp, NFSPROC_ACCESS, &info);
385 
386 		nfsm_postop_attr(vp, attrflag);
387 		if (error) {
388 			m_freem(info.nmi_mrep);
389 			goto nfsmout;
390 		}
391 
392 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
393 		rmode = fxdr_unsigned(u_int32_t, *tl);
394 		/*
395 		 * The NFS V3 spec does not clarify whether or not
396 		 * the returned access bits can be a superset of
397 		 * the ones requested, so...
398 		 */
399 		if ((rmode & mode) != mode)
400 			error = EACCES;
401 
402 		m_freem(info.nmi_mrep);
403 	} else
404 		return (nfsspec_access(ap));
405 
406 
407 	/*
408 	 * If we got the same result as for a previous, different request, OR
409 	 * it in. Don't update the timestamp in that case.
410 	 */
411 	 if (!error || error == EACCES) {
412 		if (cachevalid && np->n_accstamp != -1 &&
413 		    error == np->n_accerror) {
414 			if (!error)
415 				np->n_accmode |= ap->a_mode;
416 			else {
417 				if ((np->n_accmode & ap->a_mode) == ap->a_mode)
418 					np->n_accmode = ap->a_mode;
419 			}
420 		} else {
421 			np->n_accstamp = gettime();
422 			np->n_accuid = ap->a_cred->cr_uid;
423 			np->n_accmode = ap->a_mode;
424 			np->n_accerror = error;
425 		}
426 	}
427 nfsmout:
428 	return (error);
429 }
430 
431 /*
432  * nfs open vnode op
433  * Check to see if the type is ok
434  * and that deletion is not in progress.
435  * For paged in text files, you will need to flush the page cache
436  * if consistency is lost.
437  */
438 int
439 nfs_open(void *v)
440 {
441 	struct vop_open_args *ap = v;
442 	struct vnode *vp = ap->a_vp;
443 	struct nfsnode *np = VTONFS(vp);
444 	struct vattr vattr;
445 	int error;
446 
447 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
448 #ifdef DIAGNOSTIC
449 		printf("open eacces vtyp=%d\n",vp->v_type);
450 #endif
451 		return (EACCES);
452 	}
453 
454 	/*
455 	 * Initialize read and write creds here, for swapfiles
456 	 * and other paths that don't set the creds themselves.
457 	 */
458 
459 	if (ap->a_mode & FREAD) {
460 		if (np->n_rcred) {
461 			crfree(np->n_rcred);
462 		}
463 		np->n_rcred = ap->a_cred;
464 		crhold(np->n_rcred);
465 	}
466 	if (ap->a_mode & FWRITE) {
467 		if (np->n_wcred) {
468 			crfree(np->n_wcred);
469 		}
470 		np->n_wcred = ap->a_cred;
471 		crhold(np->n_wcred);
472 	}
473 
474 	if (np->n_flag & NMODIFIED) {
475 		error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p);
476 		if (error == EINTR)
477 			return (error);
478 		uvm_vnp_uncache(vp);
479 		NFS_INVALIDATE_ATTRCACHE(np);
480 		if (vp->v_type == VDIR)
481 			np->n_direofoffset = 0;
482 		error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
483 		if (error)
484 			return (error);
485 		np->n_mtime = vattr.va_mtime;
486 	} else {
487 		error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
488 		if (error)
489 			return (error);
490 		if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) {
491 			if (vp->v_type == VDIR)
492 				np->n_direofoffset = 0;
493 			error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p);
494 			if (error == EINTR)
495 				return (error);
496 			uvm_vnp_uncache(vp);
497 			np->n_mtime = vattr.va_mtime;
498 		}
499 	}
500 	/* For open/close consistency. */
501 	NFS_INVALIDATE_ATTRCACHE(np);
502 	return (0);
503 }
504 
505 /*
506  * nfs close vnode op
507  * What an NFS client should do upon close after writing is a debatable issue.
508  * Most NFS clients push delayed writes to the server upon close, basically for
509  * two reasons:
510  * 1 - So that any write errors may be reported back to the client process
511  *     doing the close system call. By far the two most likely errors are
512  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
513  * 2 - To put a worst case upper bound on cache inconsistency between
514  *     multiple clients for the file.
515  * There is also a consistency problem for Version 2 of the protocol w.r.t.
516  * not being able to tell if other clients are writing a file concurrently,
517  * since there is no way of knowing if the changed modify time in the reply
518  * is only due to the write for this client.
519  * (NFS Version 3 provides weak cache consistency data in the reply that
520  *  should be sufficient to detect and handle this case.)
521  *
522  * The current code does the following:
523  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
524  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
525  *		       or commit them (this satisfies 1 and 2 except for the
526  *		       case where the server crashes after this close but
527  *		       before the commit RPC, which is felt to be "good
528  *		       enough". Changing the last argument to nfs_flush() to
529  *		       a 1 would force a commit operation, if it is felt a
530  *		       commit is necessary now.
531  */
532 int
533 nfs_close(void *v)
534 {
535 	struct vop_close_args *ap = v;
536 	struct vnode *vp = ap->a_vp;
537 	struct nfsnode *np = VTONFS(vp);
538 	int error = 0;
539 
540 	if (vp->v_type == VREG) {
541 	    if (np->n_flag & NMODIFIED) {
542 		if (NFS_ISV3(vp)) {
543 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
544 		    np->n_flag &= ~NMODIFIED;
545 		} else
546 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p);
547 		NFS_INVALIDATE_ATTRCACHE(np);
548 	    }
549 	    if (np->n_flag & NWRITEERR) {
550 		np->n_flag &= ~NWRITEERR;
551 		error = np->n_error;
552 	    }
553 	}
554 	return (error);
555 }
556 
557 /*
558  * nfs getattr call from vfs.
559  */
560 int
561 nfs_getattr(void *v)
562 {
563 	struct vop_getattr_args *ap = v;
564 	struct vnode *vp = ap->a_vp;
565 	struct nfsnode *np = VTONFS(vp);
566 	struct nfsm_info	info;
567 	int32_t t1;
568 	int error = 0;
569 
570 	info.nmi_v3 = NFS_ISV3(vp);
571 
572 	/*
573 	 * Update local times for special files.
574 	 */
575 	if (np->n_flag & (NACC | NUPD))
576 		np->n_flag |= NCHG;
577 	/*
578 	 * First look in the cache.
579 	 */
580 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
581 		return (0);
582 
583 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
584 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3));
585 	nfsm_fhtom(&info, vp, info.nmi_v3);
586 	info.nmi_procp = ap->a_p;
587 	info.nmi_cred = ap->a_cred;
588 	error = nfs_request(vp, NFSPROC_GETATTR, &info);
589 	if (!error)
590 		nfsm_loadattr(vp, ap->a_vap);
591 	m_freem(info.nmi_mrep);
592 nfsmout:
593 	return (error);
594 }
595 
596 /*
597  * nfs setattr call.
598  */
599 int
600 nfs_setattr(void *v)
601 {
602 	struct vop_setattr_args *ap = v;
603 	struct vnode *vp = ap->a_vp;
604 	struct nfsnode *np = VTONFS(vp);
605 	struct vattr *vap = ap->a_vap;
606 	int hint = NOTE_ATTRIB;
607 	int error = 0;
608 	u_quad_t tsize = 0;
609 
610 	/*
611 	 * Setting of flags is not supported.
612 	 */
613 	if (vap->va_flags != VNOVAL)
614 		return (EOPNOTSUPP);
615 
616 	/*
617 	 * Disallow write attempts if the filesystem is mounted read-only.
618 	 */
619 	if ((vap->va_uid != (uid_t)VNOVAL ||
620 	    vap->va_gid != (gid_t)VNOVAL ||
621 	    vap->va_atime.tv_nsec != VNOVAL ||
622 	    vap->va_mtime.tv_nsec != VNOVAL ||
623 	    vap->va_mode != (mode_t)VNOVAL) &&
624 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
625 		return (EROFS);
626 	if (vap->va_size != VNOVAL) {
627 		switch (vp->v_type) {
628 		case VDIR:
629 			return (EISDIR);
630 		case VCHR:
631 		case VBLK:
632 		case VSOCK:
633 		case VFIFO:
634 			if (vap->va_mtime.tv_nsec == VNOVAL &&
635 			    vap->va_atime.tv_nsec == VNOVAL &&
636 			    vap->va_mode == (mode_t)VNOVAL &&
637 			    vap->va_uid == (uid_t)VNOVAL &&
638 			    vap->va_gid == (gid_t)VNOVAL)
639 				return (0);
640 			vap->va_size = VNOVAL;
641 			break;
642 		default:
643 			/*
644 			 * Disallow write attempts if the filesystem is
645 			 * mounted read-only.
646 			 */
647 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
648 				return (EROFS);
649 			if (vap->va_size == 0)
650 				error = nfs_vinvalbuf(vp, 0,
651 				     ap->a_cred, ap->a_p);
652 			else
653 				error = nfs_vinvalbuf(vp, V_SAVE,
654 				     ap->a_cred, ap->a_p);
655 			if (error)
656 				return (error);
657 			tsize = np->n_size;
658 			np->n_size = np->n_vattr.va_size = vap->va_size;
659 			uvm_vnp_setsize(vp, np->n_size);
660 		};
661 	} else if ((vap->va_mtime.tv_nsec != VNOVAL ||
662 		vap->va_atime.tv_nsec != VNOVAL) &&
663 		vp->v_type == VREG &&
664 		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
665 		    ap->a_p)) == EINTR)
666 		return (error);
667 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
668 	if (error && vap->va_size != VNOVAL) {
669 		np->n_size = np->n_vattr.va_size = tsize;
670 		uvm_vnp_setsize(vp, np->n_size);
671 	}
672 
673 	if (vap->va_size != VNOVAL && vap->va_size < tsize)
674 		hint |= NOTE_TRUNCATE;
675 
676 	VN_KNOTE(vp, hint); /* XXX setattrrpc? */
677 
678 	return (error);
679 }
680 
681 /*
682  * Do an nfs setattr rpc.
683  */
684 int
685 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
686     struct proc *procp)
687 {
688 	struct nfsv2_sattr *sp;
689 	struct nfsm_info	info;
690 	int32_t t1;
691 	caddr_t cp2;
692 	u_int32_t *tl;
693 	int error = 0, wccflag = NFSV3_WCCRATTR;
694 	int v3 = NFS_ISV3(vp);
695 
696 	info.nmi_v3 = NFS_ISV3(vp);
697 
698 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
699 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(v3) + NFSX_SATTR(v3));
700 	nfsm_fhtom(&info, vp, v3);
701 
702 	if (info.nmi_v3) {
703 		nfsm_v3attrbuild(&info.nmi_mb, vap, 1);
704 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
705 		*tl = nfs_false;
706 	} else {
707 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
708 		if (vap->va_mode == (mode_t)VNOVAL)
709 			sp->sa_mode = nfs_xdrneg1;
710 		else
711 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
712 		if (vap->va_uid == (uid_t)VNOVAL)
713 			sp->sa_uid = nfs_xdrneg1;
714 		else
715 			sp->sa_uid = txdr_unsigned(vap->va_uid);
716 		if (vap->va_gid == (gid_t)VNOVAL)
717 			sp->sa_gid = nfs_xdrneg1;
718 		else
719 			sp->sa_gid = txdr_unsigned(vap->va_gid);
720 		sp->sa_size = txdr_unsigned(vap->va_size);
721 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
722 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
723 	}
724 
725 	info.nmi_procp = procp;
726 	info.nmi_cred = cred;
727 	error = nfs_request(vp, NFSPROC_SETATTR, &info);
728 
729 	if (info.nmi_v3)
730 		nfsm_wcc_data(vp, wccflag);
731 	else if (error == 0)
732 		nfsm_loadattr(vp, NULL);
733 
734 	m_freem(info.nmi_mrep);
735 nfsmout:
736 	return (error);
737 }
738 
739 /*
740  * nfs lookup call, one step at a time...
741  * First look in cache
742  * If not found, unlock the directory nfsnode and do the rpc
743  */
744 int
745 nfs_lookup(void *v)
746 {
747 	struct vop_lookup_args *ap = v;
748 	struct componentname *cnp = ap->a_cnp;
749 	struct vnode *dvp = ap->a_dvp;
750 	struct vnode **vpp = ap->a_vpp;
751 	struct nfsm_info	info;
752 	int flags;
753 	struct vnode *newvp;
754 	u_int32_t *tl;
755 	int32_t t1;
756 	struct nfsmount *nmp;
757 	caddr_t cp2;
758 	long len;
759 	nfsfh_t *fhp;
760 	struct nfsnode *np;
761 	int lockparent, wantparent, error = 0, attrflag, fhsize;
762 
763 	info.nmi_v3 = NFS_ISV3(dvp);
764 
765 	cnp->cn_flags &= ~PDIRUNLOCK;
766 	flags = cnp->cn_flags;
767 
768 	*vpp = NULLVP;
769 	newvp = NULLVP;
770 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
771 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
772 		return (EROFS);
773 	if (dvp->v_type != VDIR)
774 		return (ENOTDIR);
775 	lockparent = flags & LOCKPARENT;
776 	wantparent = flags & (LOCKPARENT|WANTPARENT);
777 	nmp = VFSTONFS(dvp->v_mount);
778 	np = VTONFS(dvp);
779 
780 	/*
781 	 * Before tediously performing a linear scan of the directory,
782 	 * check the name cache to see if the directory/name pair
783 	 * we are looking for is known already.
784 	 * If the directory/name pair is found in the name cache,
785 	 * we have to ensure the directory has not changed from
786 	 * the time the cache entry has been created. If it has,
787 	 * the cache entry has to be ignored.
788 	 */
789 	if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
790 		struct vattr vattr;
791 		int err2;
792 
793 		if (error && error != ENOENT) {
794 			*vpp = NULLVP;
795 			return (error);
796 		}
797 
798 		if (cnp->cn_flags & PDIRUNLOCK) {
799 			err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
800 			if (err2 != 0) {
801 				*vpp = NULLVP;
802 				return (err2);
803 			}
804 			cnp->cn_flags &= ~PDIRUNLOCK;
805 		}
806 
807 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
808 		if (err2 != 0) {
809 			if (error == 0) {
810 				if (*vpp != dvp)
811 					vput(*vpp);
812 				else
813 					vrele(*vpp);
814 			}
815 			*vpp = NULLVP;
816 			return (err2);
817 		}
818 
819 		if (error == ENOENT) {
820 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
821 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
822 			    VTONFS(dvp)->n_ctime)
823 				return (ENOENT);
824 			cache_purge(dvp);
825 			np->n_ctime = 0;
826 			goto dorpc;
827 		}
828 
829 		newvp = *vpp;
830 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
831 			&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
832 		{
833 			nfsstats.lookupcache_hits++;
834 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
835 				cnp->cn_flags |= SAVENAME;
836 			if ((!lockparent || !(flags & ISLASTCN)) &&
837 			     newvp != dvp) {
838 				VOP_UNLOCK(dvp);
839 				cnp->cn_flags |= PDIRUNLOCK;
840 			}
841 			return (0);
842 		}
843 		cache_purge(newvp);
844 		if (newvp != dvp)
845 			vput(newvp);
846 		else
847 			vrele(newvp);
848 		*vpp = NULLVP;
849 	}
850 dorpc:
851 	error = 0;
852 	newvp = NULLVP;
853 	nfsstats.lookupcache_misses++;
854 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
855 	len = cnp->cn_namelen;
856 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
857 	    NFSX_UNSIGNED + nfsm_rndup(len));
858 	nfsm_fhtom(&info, dvp, info.nmi_v3);
859 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
860 
861 	info.nmi_procp = cnp->cn_proc;
862 	info.nmi_cred = cnp->cn_cred;
863 	error = nfs_request(dvp, NFSPROC_LOOKUP, &info);
864 
865 	if (error) {
866 		if (info.nmi_v3)
867 			nfsm_postop_attr(dvp, attrflag);
868 		m_freem(info.nmi_mrep);
869 		goto nfsmout;
870 	}
871 
872 	nfsm_getfh(fhp, fhsize, info.nmi_v3);
873 
874 	/*
875 	 * Handle RENAME case...
876 	 */
877 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
878 		if (NFS_CMPFH(np, fhp, fhsize)) {
879 			m_freem(info.nmi_mrep);
880 			return (EISDIR);
881 		}
882 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
883 		if (error) {
884 			m_freem(info.nmi_mrep);
885 			return (error);
886 		}
887 		newvp = NFSTOV(np);
888 		if (info.nmi_v3) {
889 			nfsm_postop_attr(newvp, attrflag);
890 			nfsm_postop_attr(dvp, attrflag);
891 		} else
892 			nfsm_loadattr(newvp, NULL);
893 		*vpp = newvp;
894 		m_freem(info.nmi_mrep);
895 		cnp->cn_flags |= SAVENAME;
896 		if (!lockparent) {
897 			VOP_UNLOCK(dvp);
898 			cnp->cn_flags |= PDIRUNLOCK;
899 		}
900 		return (0);
901 	}
902 
903 	/*
904 	 * The postop attr handling is duplicated for each if case,
905 	 * because it should be done while dvp is locked (unlocking
906 	 * dvp is different for each case).
907 	 */
908 
909 	if (NFS_CMPFH(np, fhp, fhsize)) {
910 		vref(dvp);
911 		newvp = dvp;
912 		if (info.nmi_v3) {
913 			nfsm_postop_attr(newvp, attrflag);
914 			nfsm_postop_attr(dvp, attrflag);
915 		} else
916 			nfsm_loadattr(newvp, NULL);
917 	} else if (flags & ISDOTDOT) {
918 		VOP_UNLOCK(dvp);
919 		cnp->cn_flags |= PDIRUNLOCK;
920 
921 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
922 		if (error) {
923 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
924 				cnp->cn_flags &= ~PDIRUNLOCK;
925 			m_freem(info.nmi_mrep);
926 			return (error);
927 		}
928 		newvp = NFSTOV(np);
929 
930 		if (info.nmi_v3) {
931 			nfsm_postop_attr(newvp, attrflag);
932 			nfsm_postop_attr(dvp, attrflag);
933 		} else
934 			nfsm_loadattr(newvp, NULL);
935 
936 		if (lockparent && (flags & ISLASTCN)) {
937 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
938 				m_freem(info.nmi_mrep);
939 				vput(newvp);
940 				return error;
941 			}
942 			cnp->cn_flags &= ~PDIRUNLOCK;
943 		}
944 
945 	} else {
946 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
947 		if (error) {
948 			m_freem(info.nmi_mrep);
949 			return error;
950 		}
951 		newvp = NFSTOV(np);
952 		if (info.nmi_v3) {
953 			nfsm_postop_attr(newvp, attrflag);
954 			nfsm_postop_attr(dvp, attrflag);
955 		} else
956 			nfsm_loadattr(newvp, NULL);
957 		if (!lockparent || !(flags & ISLASTCN)) {
958 			VOP_UNLOCK(dvp);
959 			cnp->cn_flags |= PDIRUNLOCK;
960 		}
961 	}
962 
963 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
964 		cnp->cn_flags |= SAVENAME;
965 	if ((cnp->cn_flags & MAKEENTRY) &&
966 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
967 		nfs_cache_enter(dvp, newvp, cnp);
968 	}
969 
970 	*vpp = newvp;
971 	m_freem(info.nmi_mrep);
972 
973 nfsmout:
974 	if (error) {
975 		/*
976 		 * We get here only because of errors returned by
977 		 * the RPC. Otherwise we'll have returned above
978 		 * (the nfsm_* macros will jump to nfsmout
979 		 * on error).
980 		 */
981 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
982 		    cnp->cn_nameiop != CREATE) {
983 			nfs_cache_enter(dvp, NULL, cnp);
984 		}
985 		if (newvp != NULLVP) {
986 			if (newvp != dvp)
987 				vput(newvp);
988 			else
989 				vrele(newvp);
990 		}
991 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
992 		    (flags & ISLASTCN) && error == ENOENT) {
993 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
994 				error = EROFS;
995 			else
996 				error = EJUSTRETURN;
997 		}
998 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
999 			cnp->cn_flags |= SAVENAME;
1000 		*vpp = NULL;
1001 	}
1002 	return (error);
1003 }
1004 
1005 /*
1006  * nfs read call.
1007  * Just call nfs_bioread() to do the work.
1008  */
1009 int
1010 nfs_read(void *v)
1011 {
1012 	struct vop_read_args *ap = v;
1013 	struct vnode *vp = ap->a_vp;
1014 
1015 	if (vp->v_type != VREG)
1016 		return (EPERM);
1017 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1018 }
1019 
1020 /*
1021  * nfs readlink call
1022  */
1023 int
1024 nfs_readlink(void *v)
1025 {
1026 	struct vop_readlink_args *ap = v;
1027 	struct vnode *vp = ap->a_vp;
1028 
1029 	if (vp->v_type != VLNK)
1030 		return (EPERM);
1031 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1032 }
1033 
1034 /*
1035  * Lock an inode.
1036  */
1037 int
1038 nfs_lock(void *v)
1039 {
1040 	struct vop_lock_args *ap = v;
1041 	struct vnode *vp = ap->a_vp;
1042 
1043 	return rrw_enter(&VTONFS(vp)->n_lock, ap->a_flags & LK_RWFLAGS);
1044 }
1045 
1046 /*
1047  * Unlock an inode.
1048  */
1049 int
1050 nfs_unlock(void *v)
1051 {
1052 	struct vop_unlock_args *ap = v;
1053 	struct vnode *vp = ap->a_vp;
1054 
1055 	rrw_exit(&VTONFS(vp)->n_lock);
1056 	return 0;
1057 }
1058 
1059 /*
1060  * Check for a locked inode.
1061  */
1062 int
1063 nfs_islocked(void *v)
1064 {
1065 	struct vop_islocked_args *ap = v;
1066 
1067 	return rrw_status(&VTONFS(ap->a_vp)->n_lock);
1068 }
1069 
1070 /*
1071  * Do a readlink rpc.
1072  * Called by nfs_doio() from below the buffer cache.
1073  */
1074 int
1075 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1076 {
1077 	struct nfsm_info	info;
1078 	u_int32_t *tl;
1079 	int32_t t1;
1080 	caddr_t cp2;
1081 	int error = 0, len, attrflag;
1082 
1083 	info.nmi_v3 = NFS_ISV3(vp);
1084 
1085 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1086 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3));
1087 	nfsm_fhtom(&info, vp, info.nmi_v3);
1088 
1089 	info.nmi_procp = curproc;
1090 	info.nmi_cred = cred;
1091 	error = nfs_request(vp, NFSPROC_READLINK, &info);
1092 
1093 	if (info.nmi_v3)
1094 		nfsm_postop_attr(vp, attrflag);
1095 	if (!error) {
1096 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1097 		nfsm_mtouio(uiop, len);
1098 	}
1099 
1100 	m_freem(info.nmi_mrep);
1101 
1102 nfsmout:
1103 	return (error);
1104 }
1105 
1106 /*
1107  * nfs read rpc call
1108  * Ditto above
1109  */
1110 int
1111 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1112 {
1113 	struct nfsm_info	info;
1114 	u_int32_t *tl;
1115 	int32_t t1;
1116 	caddr_t cp2;
1117 	struct nfsmount *nmp;
1118 	int error = 0, len, retlen, tsiz, eof, attrflag;
1119 
1120 	info.nmi_v3 = NFS_ISV3(vp);
1121 
1122 	eof = 0;
1123 
1124 	nmp = VFSTONFS(vp->v_mount);
1125 	tsiz = uiop->uio_resid;
1126 	if (uiop->uio_offset + tsiz > 0xffffffff && !info.nmi_v3)
1127 		return (EFBIG);
1128 	while (tsiz > 0) {
1129 		nfsstats.rpccnt[NFSPROC_READ]++;
1130 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1131 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1132 		    NFSX_UNSIGNED * 3);
1133 		nfsm_fhtom(&info, vp, info.nmi_v3);
1134 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED * 3);
1135 		if (info.nmi_v3) {
1136 			txdr_hyper(uiop->uio_offset, tl);
1137 			*(tl + 2) = txdr_unsigned(len);
1138 		} else {
1139 			*tl++ = txdr_unsigned(uiop->uio_offset);
1140 			*tl++ = txdr_unsigned(len);
1141 			*tl = 0;
1142 		}
1143 
1144 		info.nmi_procp = curproc;
1145 		info.nmi_cred = VTONFS(vp)->n_rcred;
1146 		error = nfs_request(vp, NFSPROC_READ, &info);
1147 		if (info.nmi_v3)
1148 			nfsm_postop_attr(vp, attrflag);
1149 		if (error) {
1150 			m_freem(info.nmi_mrep);
1151 			goto nfsmout;
1152 		}
1153 
1154 		if (info.nmi_v3) {
1155 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1156 			eof = fxdr_unsigned(int, *(tl + 1));
1157 		} else {
1158 			nfsm_loadattr(vp, NULL);
1159 		}
1160 
1161 		nfsm_strsiz(retlen, nmp->nm_rsize);
1162 		nfsm_mtouio(uiop, retlen);
1163 		m_freem(info.nmi_mrep);
1164 		tsiz -= retlen;
1165 		if (info.nmi_v3) {
1166 			if (eof || retlen == 0)
1167 				tsiz = 0;
1168 		} else if (retlen < len)
1169 			tsiz = 0;
1170 	}
1171 
1172 nfsmout:
1173 	return (error);
1174 }
1175 
1176 /*
1177  * nfs write call
1178  */
1179 int
1180 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1181 {
1182 	struct nfsm_info	info;
1183 	u_int32_t *tl;
1184 	int32_t t1, backup;
1185 	caddr_t cp2;
1186 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1187 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1188 	int committed = NFSV3WRITE_FILESYNC;
1189 
1190 	info.nmi_v3 = NFS_ISV3(vp);
1191 
1192 #ifdef DIAGNOSTIC
1193 	if (uiop->uio_iovcnt != 1)
1194 		panic("nfs: writerpc iovcnt > 1");
1195 #endif
1196 	*must_commit = 0;
1197 	tsiz = uiop->uio_resid;
1198 	if (uiop->uio_offset + tsiz > 0xffffffff && !info.nmi_v3)
1199 		return (EFBIG);
1200 	while (tsiz > 0) {
1201 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1202 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1203 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3)
1204 		    + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1205 		nfsm_fhtom(&info, vp, info.nmi_v3);
1206 		if (info.nmi_v3) {
1207 			tl = nfsm_build(&info.nmi_mb, 5 * NFSX_UNSIGNED);
1208 			txdr_hyper(uiop->uio_offset, tl);
1209 			tl += 2;
1210 			*tl++ = txdr_unsigned(len);
1211 			*tl++ = txdr_unsigned(*iomode);
1212 			*tl = txdr_unsigned(len);
1213 		} else {
1214 			u_int32_t x;
1215 
1216 			tl = nfsm_build(&info.nmi_mb, 4 * NFSX_UNSIGNED);
1217 			/* Set both "begin" and "current" to non-garbage. */
1218 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1219 			*tl++ = x;	/* "begin offset" */
1220 			*tl++ = x;	/* "current offset" */
1221 			x = txdr_unsigned(len);
1222 			*tl++ = x;	/* total to this offset */
1223 			*tl = x;	/* size of this write */
1224 
1225 		}
1226 		nfsm_uiotombuf(&info.nmi_mb, uiop, len);
1227 
1228 		info.nmi_procp = curproc;
1229 		info.nmi_cred = VTONFS(vp)->n_wcred;
1230 		error = nfs_request(vp, NFSPROC_WRITE, &info);
1231 		if (info.nmi_v3) {
1232 			wccflag = NFSV3_WCCCHK;
1233 			nfsm_wcc_data(vp, wccflag);
1234 		}
1235 
1236 		if (error) {
1237 			m_freem(info.nmi_mrep);
1238 			goto nfsmout;
1239 		}
1240 
1241 		if (info.nmi_v3) {
1242 			wccflag = NFSV3_WCCCHK;
1243 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1244 				+ NFSX_V3WRITEVERF);
1245 			rlen = fxdr_unsigned(int, *tl++);
1246 			if (rlen <= 0) {
1247 				error = NFSERR_IO;
1248 				break;
1249 			} else if (rlen < len) {
1250 				backup = len - rlen;
1251 				uiop->uio_iov->iov_base =
1252 				    (char *)uiop->uio_iov->iov_base -
1253 				    backup;
1254 				uiop->uio_iov->iov_len += backup;
1255 				uiop->uio_offset -= backup;
1256 				uiop->uio_resid += backup;
1257 				len = rlen;
1258 			}
1259 			commit = fxdr_unsigned(int, *tl++);
1260 
1261 			/*
1262 			 * Return the lowest commitment level
1263 			 * obtained by any of the RPCs.
1264 			 */
1265 			if (committed == NFSV3WRITE_FILESYNC)
1266 				committed = commit;
1267 			else if (committed == NFSV3WRITE_DATASYNC &&
1268 				commit == NFSV3WRITE_UNSTABLE)
1269 				committed = commit;
1270 			if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0) {
1271 				bcopy(tl, nmp->nm_verf,
1272 				    NFSX_V3WRITEVERF);
1273 				nmp->nm_flag |= NFSMNT_HASWRITEVERF;
1274 			} else if (bcmp(tl,
1275 			    nmp->nm_verf, NFSX_V3WRITEVERF)) {
1276 				*must_commit = 1;
1277 				bcopy(tl, nmp->nm_verf,
1278 				    NFSX_V3WRITEVERF);
1279 			}
1280 		} else {
1281 			nfsm_loadattr(vp, NULL);
1282 		}
1283 		if (wccflag)
1284 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1285 		m_freem(info.nmi_mrep);
1286 		tsiz -= len;
1287 	}
1288 nfsmout:
1289 	*iomode = committed;
1290 	if (error)
1291 		uiop->uio_resid = tsiz;
1292 	return (error);
1293 }
1294 
1295 /*
1296  * nfs mknod rpc
1297  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1298  * mode set to specify the file type and the size field for rdev.
1299  */
1300 int
1301 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1302     struct vattr *vap)
1303 {
1304 	struct nfsv2_sattr *sp;
1305 	struct nfsm_info	info;
1306 	u_int32_t *tl;
1307 	int32_t t1;
1308 	struct vnode *newvp = NULL;
1309 	struct nfsnode *np = NULL;
1310 	char *cp2;
1311 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1312 	u_int32_t rdev;
1313 
1314 	info.nmi_v3 = NFS_ISV3(dvp);
1315 
1316 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1317 		rdev = txdr_unsigned(vap->va_rdev);
1318 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1319 		rdev = nfs_xdrneg1;
1320 	else {
1321 		VOP_ABORTOP(dvp, cnp);
1322 		return (EOPNOTSUPP);
1323 	}
1324 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1325 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1326 	    4 * NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen) +
1327 	    NFSX_SATTR(info.nmi_v3));
1328 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1329 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1330 
1331 	if (info.nmi_v3) {
1332 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
1333 		*tl++ = vtonfsv3_type(vap->va_type);
1334 		nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1335 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1336 			tl = nfsm_build(&info.nmi_mb, 2 * NFSX_UNSIGNED);
1337 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1338 			*tl = txdr_unsigned(minor(vap->va_rdev));
1339 		}
1340 	} else {
1341 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1342 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1343 		sp->sa_uid = nfs_xdrneg1;
1344 		sp->sa_gid = nfs_xdrneg1;
1345 		sp->sa_size = rdev;
1346 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1347 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1348 	}
1349 
1350 	KASSERT(cnp->cn_proc == curproc);
1351 	info.nmi_procp = cnp->cn_proc;
1352 	info.nmi_cred = cnp->cn_cred;
1353 	error = nfs_request(dvp, NFSPROC_MKNOD, &info);
1354 	if (!error) {
1355 		nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1356 		if (!gotvp) {
1357 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1358 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1359 			if (!error)
1360 				newvp = NFSTOV(np);
1361 		}
1362 	}
1363 	if (info.nmi_v3)
1364 		nfsm_wcc_data(dvp, wccflag);
1365 	m_freem(info.nmi_mrep);
1366 
1367 nfsmout:
1368 	if (error) {
1369 		if (newvp)
1370 			vput(newvp);
1371 	} else {
1372 		if (cnp->cn_flags & MAKEENTRY)
1373 			nfs_cache_enter(dvp, newvp, cnp);
1374 		*vpp = newvp;
1375 	}
1376 	pool_put(&namei_pool, cnp->cn_pnbuf);
1377 	VTONFS(dvp)->n_flag |= NMODIFIED;
1378 	if (!wccflag)
1379 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1380 	return (error);
1381 }
1382 
1383 /*
1384  * nfs mknod vop
1385  * just call nfs_mknodrpc() to do the work.
1386  */
1387 int
1388 nfs_mknod(void *v)
1389 {
1390 	struct vop_mknod_args *ap = v;
1391 	struct vnode *newvp;
1392 	int error;
1393 
1394 	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1395 	if (!error)
1396 		vput(newvp);
1397 
1398 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1399 
1400 	return (error);
1401 }
1402 
1403 int
1404 nfs_create(void *v)
1405 {
1406 	struct vop_create_args *ap = v;
1407 	struct vnode *dvp = ap->a_dvp;
1408 	struct vattr *vap = ap->a_vap;
1409 	struct componentname *cnp = ap->a_cnp;
1410 	struct nfsv2_sattr *sp;
1411 	struct nfsm_info	info;
1412 	struct timespec ts;
1413 	u_int32_t *tl;
1414 	int32_t t1;
1415 	struct nfsnode *np = NULL;
1416 	struct vnode *newvp = NULL;
1417 	caddr_t cp2;
1418 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1419 
1420 	info.nmi_v3 = NFS_ISV3(dvp);
1421 
1422 	/*
1423 	 * Oops, not for me..
1424 	 */
1425 	if (vap->va_type == VSOCK)
1426 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1427 
1428 	if (vap->va_vaflags & VA_EXCLUSIVE)
1429 		fmode |= O_EXCL;
1430 
1431 again:
1432 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1433 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1434 	    2 * NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen) +
1435 	    NFSX_SATTR(info.nmi_v3));
1436 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1437 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1438 	if (info.nmi_v3) {
1439 		tl = nfsm_build(&info.nmi_mb, NFSX_UNSIGNED);
1440 		if (fmode & O_EXCL) {
1441 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1442 			tl = nfsm_build(&info.nmi_mb, NFSX_V3CREATEVERF);
1443 			arc4random_buf(tl, sizeof(*tl) * 2);
1444 		} else {
1445 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1446 			nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1447 		}
1448 	} else {
1449 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1450 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1451 		sp->sa_uid = nfs_xdrneg1;
1452 		sp->sa_gid = nfs_xdrneg1;
1453 		sp->sa_size = 0;
1454 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1455 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1456 	}
1457 
1458 	KASSERT(cnp->cn_proc == curproc);
1459 	info.nmi_procp = cnp->cn_proc;
1460 	info.nmi_cred = cnp->cn_cred;
1461 	error = nfs_request(dvp, NFSPROC_CREATE, &info);
1462 	if (!error) {
1463 		nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1464 		if (!gotvp) {
1465 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1466 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1467 			if (!error)
1468 				newvp = NFSTOV(np);
1469 		}
1470 	}
1471 	if (info.nmi_v3)
1472 		nfsm_wcc_data(dvp, wccflag);
1473 	m_freem(info.nmi_mrep);
1474 
1475 nfsmout:
1476 	if (error) {
1477 		if (newvp) {
1478 			vput(newvp);
1479 			newvp = NULL;
1480 		}
1481 		if (info.nmi_v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1482 			fmode &= ~O_EXCL;
1483 			goto again;
1484 		}
1485 	} else if (info.nmi_v3 && (fmode & O_EXCL)) {
1486 		getnanotime(&ts);
1487 		if (vap->va_atime.tv_nsec == VNOVAL)
1488 			vap->va_atime = ts;
1489 		if (vap->va_mtime.tv_nsec == VNOVAL)
1490 			vap->va_mtime = ts;
1491 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1492 	}
1493 	if (!error) {
1494 		if (cnp->cn_flags & MAKEENTRY)
1495 			nfs_cache_enter(dvp, newvp, cnp);
1496 		*ap->a_vpp = newvp;
1497 	}
1498 	pool_put(&namei_pool, cnp->cn_pnbuf);
1499 	VTONFS(dvp)->n_flag |= NMODIFIED;
1500 	if (!wccflag)
1501 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1502 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1503 	return (error);
1504 }
1505 
1506 /*
1507  * nfs file remove call
1508  * To try and make nfs semantics closer to ufs semantics, a file that has
1509  * other processes using the vnode is renamed instead of removed and then
1510  * removed later on the last close.
1511  * - If v_usecount > 1
1512  *	  If a rename is not already in the works
1513  *	     call nfs_sillyrename() to set it up
1514  *     else
1515  *	  do the remove rpc
1516  */
1517 int
1518 nfs_remove(void *v)
1519 {
1520 	struct vop_remove_args *ap = v;
1521 	struct vnode *vp = ap->a_vp;
1522 	struct vnode *dvp = ap->a_dvp;
1523 	struct componentname *cnp = ap->a_cnp;
1524 	struct nfsnode *np = VTONFS(vp);
1525 	int error = 0;
1526 	struct vattr vattr;
1527 
1528 #ifdef DIAGNOSTIC
1529 	if ((cnp->cn_flags & HASBUF) == 0)
1530 		panic("nfs_remove: no name");
1531 	if (vp->v_usecount < 1)
1532 		panic("nfs_remove: bad v_usecount");
1533 #endif
1534 	if (vp->v_type == VDIR)
1535 		error = EPERM;
1536 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1537 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1538 	    vattr.va_nlink > 1)) {
1539 		/*
1540 		 * Purge the name cache so that the chance of a lookup for
1541 		 * the name succeeding while the remove is in progress is
1542 		 * minimized. Without node locking it can still happen, such
1543 		 * that an I/O op returns ESTALE, but since you get this if
1544 		 * another host removes the file..
1545 		 */
1546 		cache_purge(vp);
1547 		/*
1548 		 * throw away biocache buffers, mainly to avoid
1549 		 * unnecessary delayed writes later.
1550 		 */
1551 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc);
1552 		/* Do the rpc */
1553 		if (error != EINTR)
1554 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1555 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1556 		/*
1557 		 * Kludge City: If the first reply to the remove rpc is lost..
1558 		 *   the reply to the retransmitted request will be ENOENT
1559 		 *   since the file was in fact removed
1560 		 *   Therefore, we cheat and return success.
1561 		 */
1562 		if (error == ENOENT)
1563 			error = 0;
1564 	} else if (!np->n_sillyrename)
1565 		error = nfs_sillyrename(dvp, vp, cnp);
1566 	pool_put(&namei_pool, cnp->cn_pnbuf);
1567 	NFS_INVALIDATE_ATTRCACHE(np);
1568 	VN_KNOTE(vp, NOTE_DELETE);
1569 	VN_KNOTE(dvp, NOTE_WRITE);
1570 	if (vp == dvp)
1571 		vrele(vp);
1572 	else
1573 		vput(vp);
1574 	vput(dvp);
1575 	return (error);
1576 }
1577 
1578 /*
1579  * nfs file remove rpc called from nfs_inactive
1580  */
1581 int
1582 nfs_removeit(struct sillyrename *sp)
1583 {
1584 	KASSERT(VOP_ISLOCKED(sp->s_dvp));
1585 	/*
1586 	 * Make sure that the directory vnode is still valid.
1587 	 *
1588 	 * NFS can potentially try to nuke a silly *after* the directory
1589 	 * has already been pushed out on a forced unmount. Since the silly
1590 	 * is going to go away anyway, this is fine.
1591 	 */
1592 	if (sp->s_dvp->v_type == VBAD)
1593 		return (0);
1594 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1595 		NULL));
1596 }
1597 
1598 /*
1599  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1600  */
1601 int
1602 nfs_removerpc(struct vnode *dvp, char *name, int namelen, struct ucred *cred,
1603     struct proc *proc)
1604 {
1605 	struct nfsm_info	info;
1606 	u_int32_t *tl;
1607 	int32_t t1;
1608 	caddr_t cp2;
1609 	int error = 0, wccflag = NFSV3_WCCRATTR;
1610 
1611 	info.nmi_v3 = NFS_ISV3(dvp);
1612 
1613 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1614 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1615 	     NFSX_UNSIGNED + nfsm_rndup(namelen));
1616 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1617 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1618 
1619 	info.nmi_procp = proc;
1620 	info.nmi_cred = cred;
1621 	error = nfs_request(dvp, NFSPROC_REMOVE, &info);
1622 	if (info.nmi_v3)
1623 		nfsm_wcc_data(dvp, wccflag);
1624 	m_freem(info.nmi_mrep);
1625 
1626 nfsmout:
1627 	VTONFS(dvp)->n_flag |= NMODIFIED;
1628 	if (!wccflag)
1629 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1630 	return (error);
1631 }
1632 
1633 /*
1634  * nfs file rename call
1635  */
1636 int
1637 nfs_rename(void *v)
1638 {
1639 	struct vop_rename_args	*ap = v;
1640 	struct vnode *fvp = ap->a_fvp;
1641 	struct vnode *tvp = ap->a_tvp;
1642 	struct vnode *fdvp = ap->a_fdvp;
1643 	struct vnode *tdvp = ap->a_tdvp;
1644 	struct componentname *tcnp = ap->a_tcnp;
1645 	struct componentname *fcnp = ap->a_fcnp;
1646 	int error;
1647 
1648 #ifdef DIAGNOSTIC
1649 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1650 	    (fcnp->cn_flags & HASBUF) == 0)
1651 		panic("nfs_rename: no name");
1652 #endif
1653 	/* Check for cross-device rename */
1654 	if ((fvp->v_mount != tdvp->v_mount) ||
1655 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1656 		error = EXDEV;
1657 		goto out;
1658 	}
1659 
1660 	/*
1661 	 * If the tvp exists and is in use, sillyrename it before doing the
1662 	 * rename of the new file over it.
1663 	 */
1664 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1665 	    tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1666 		VN_KNOTE(tvp, NOTE_DELETE);
1667 		vput(tvp);
1668 		tvp = NULL;
1669 	}
1670 
1671 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1672 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1673 		tcnp->cn_proc);
1674 
1675 	VN_KNOTE(fdvp, NOTE_WRITE);
1676 	VN_KNOTE(tdvp, NOTE_WRITE);
1677 
1678 	if (fvp->v_type == VDIR) {
1679 		if (tvp != NULL && tvp->v_type == VDIR)
1680 			cache_purge(tdvp);
1681 		cache_purge(fdvp);
1682 	}
1683 out:
1684 	if (tdvp == tvp)
1685 		vrele(tdvp);
1686 	else
1687 		vput(tdvp);
1688 	if (tvp)
1689 		vput(tvp);
1690 	vrele(fdvp);
1691 	vrele(fvp);
1692 	/*
1693 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1694 	 */
1695 	if (error == ENOENT)
1696 		error = 0;
1697 	return (error);
1698 }
1699 
1700 /*
1701  * nfs file rename rpc called from nfs_remove() above
1702  */
1703 int
1704 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1705     struct sillyrename *sp)
1706 {
1707 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1708 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curproc));
1709 }
1710 
1711 /*
1712  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1713  */
1714 int
1715 nfs_renamerpc(struct vnode *fdvp, char *fnameptr, int fnamelen,
1716     struct vnode *tdvp, char *tnameptr, int tnamelen, struct ucred *cred,
1717     struct proc *proc)
1718 {
1719 	struct nfsm_info	info;
1720 	u_int32_t *tl;
1721 	int32_t t1;
1722 	caddr_t cp2;
1723 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1724 
1725 	info.nmi_v3 = NFS_ISV3(fdvp);
1726 
1727 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1728 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead((NFSX_FH(info.nmi_v3) +
1729 	    NFSX_UNSIGNED) * 2 + nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1730 	nfsm_fhtom(&info, fdvp, info.nmi_v3);
1731 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1732 	nfsm_fhtom(&info, tdvp, info.nmi_v3);
1733 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1734 
1735 	info.nmi_procp = proc;
1736 	info.nmi_cred = cred;
1737 	error = nfs_request(fdvp, NFSPROC_RENAME, &info);
1738 	if (info.nmi_v3) {
1739 		nfsm_wcc_data(fdvp, fwccflag);
1740 		nfsm_wcc_data(tdvp, twccflag);
1741 	}
1742 	m_freem(info.nmi_mrep);
1743 
1744 nfsmout:
1745 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1746 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1747 	if (!fwccflag)
1748 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
1749 	if (!twccflag)
1750 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
1751 	return (error);
1752 }
1753 
1754 /*
1755  * nfs hard link create call
1756  */
1757 int
1758 nfs_link(void *v)
1759 {
1760 	struct vop_link_args *ap = v;
1761 	struct vnode *vp = ap->a_vp;
1762 	struct vnode *dvp = ap->a_dvp;
1763 	struct componentname *cnp = ap->a_cnp;
1764 	struct nfsm_info	info;
1765 	u_int32_t *tl;
1766 	int32_t t1;
1767 	caddr_t cp2;
1768 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1769 
1770 	info.nmi_v3 = NFS_ISV3(vp);
1771 
1772 	if (dvp->v_mount != vp->v_mount) {
1773 		pool_put(&namei_pool, cnp->cn_pnbuf);
1774 		vput(dvp);
1775 		return (EXDEV);
1776 	}
1777 	error = vn_lock(vp, LK_EXCLUSIVE);
1778 	if (error != 0) {
1779 		VOP_ABORTOP(dvp, cnp);
1780 		vput(dvp);
1781 		return (error);
1782 	}
1783 
1784 	/*
1785 	 * Push all writes to the server, so that the attribute cache
1786 	 * doesn't get "out of sync" with the server.
1787 	 * XXX There should be a better way!
1788 	 */
1789 	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1790 
1791 	nfsstats.rpccnt[NFSPROC_LINK]++;
1792 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(2 * NFSX_FH(info.nmi_v3) +
1793 	    NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1794 	nfsm_fhtom(&info, vp, info.nmi_v3);
1795 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1796 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1797 
1798 	info.nmi_procp = cnp->cn_proc;
1799 	info.nmi_cred = cnp->cn_cred;
1800 	error = nfs_request(vp, NFSPROC_LINK, &info);
1801 	if (info.nmi_v3) {
1802 		nfsm_postop_attr(vp, attrflag);
1803 		nfsm_wcc_data(dvp, wccflag);
1804 	}
1805 	m_freem(info.nmi_mrep);
1806 nfsmout:
1807 	pool_put(&namei_pool, cnp->cn_pnbuf);
1808 	VTONFS(dvp)->n_flag |= NMODIFIED;
1809 	if (!attrflag)
1810 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
1811 	if (!wccflag)
1812 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1813 
1814 	VN_KNOTE(vp, NOTE_LINK);
1815 	VN_KNOTE(dvp, NOTE_WRITE);
1816 	VOP_UNLOCK(vp);
1817 	vput(dvp);
1818 	return (error);
1819 }
1820 
1821 /*
1822  * nfs symbolic link create call
1823  */
1824 int
1825 nfs_symlink(void *v)
1826 {
1827 	struct vop_symlink_args *ap = v;
1828 	struct vnode *dvp = ap->a_dvp;
1829 	struct vattr *vap = ap->a_vap;
1830 	struct componentname *cnp = ap->a_cnp;
1831 	struct nfsv2_sattr *sp;
1832 	struct nfsm_info	info;
1833 	u_int32_t *tl;
1834 	int32_t t1;
1835 	caddr_t cp2;
1836 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1837 	struct vnode *newvp = NULL;
1838 
1839 	info.nmi_v3 = NFS_ISV3(dvp);
1840 
1841 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1842 	slen = strlen(ap->a_target);
1843 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1844 	    2 * NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) +
1845 	    NFSX_SATTR(info.nmi_v3));
1846 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1847 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1848 	if (info.nmi_v3)
1849 		nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1850 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1851 	if (!info.nmi_v3) {
1852 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1853 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1854 		sp->sa_uid = nfs_xdrneg1;
1855 		sp->sa_gid = nfs_xdrneg1;
1856 		sp->sa_size = nfs_xdrneg1;
1857 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1858 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1859 	}
1860 
1861 	info.nmi_procp = cnp->cn_proc;
1862 	info.nmi_cred = cnp->cn_cred;
1863 	error = nfs_request(dvp, NFSPROC_SYMLINK, &info);
1864 	if (info.nmi_v3) {
1865 		if (!error)
1866 			nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1867 		nfsm_wcc_data(dvp, wccflag);
1868 	}
1869 	m_freem(info.nmi_mrep);
1870 
1871 nfsmout:
1872 	if (newvp)
1873 		vput(newvp);
1874 	pool_put(&namei_pool, cnp->cn_pnbuf);
1875 	VTONFS(dvp)->n_flag |= NMODIFIED;
1876 	if (!wccflag)
1877 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1878 	VN_KNOTE(dvp, NOTE_WRITE);
1879 	vput(dvp);
1880 	return (error);
1881 }
1882 
1883 /*
1884  * nfs make dir call
1885  */
1886 int
1887 nfs_mkdir(void *v)
1888 {
1889 	struct vop_mkdir_args *ap = v;
1890 	struct vnode *dvp = ap->a_dvp;
1891 	struct vattr *vap = ap->a_vap;
1892 	struct componentname *cnp = ap->a_cnp;
1893 	struct nfsv2_sattr *sp;
1894 	struct nfsm_info	info;
1895 	u_int32_t *tl;
1896 	int32_t t1;
1897 	int len;
1898 	struct nfsnode *np = NULL;
1899 	struct vnode *newvp = NULL;
1900 	caddr_t cp2;
1901 	int error = 0, wccflag = NFSV3_WCCRATTR;
1902 	int gotvp = 0;
1903 
1904 	info.nmi_v3 = NFS_ISV3(dvp);
1905 
1906 	len = cnp->cn_namelen;
1907 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1908 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1909 	    NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(info.nmi_v3));
1910 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1911 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1912 
1913 	if (info.nmi_v3) {
1914 		nfsm_v3attrbuild(&info.nmi_mb, vap, 0);
1915 	} else {
1916 		sp = nfsm_build(&info.nmi_mb, NFSX_V2SATTR);
1917 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1918 		sp->sa_uid = nfs_xdrneg1;
1919 		sp->sa_gid = nfs_xdrneg1;
1920 		sp->sa_size = nfs_xdrneg1;
1921 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1922 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1923 	}
1924 
1925 	info.nmi_procp = cnp->cn_proc;
1926 	info.nmi_cred = cnp->cn_cred;
1927 	error = nfs_request(dvp, NFSPROC_MKDIR, &info);
1928 	if (!error)
1929 		nfsm_mtofh(dvp, newvp, info.nmi_v3, gotvp);
1930 	if (info.nmi_v3)
1931 		nfsm_wcc_data(dvp, wccflag);
1932 	m_freem(info.nmi_mrep);
1933 
1934 nfsmout:
1935 	VTONFS(dvp)->n_flag |= NMODIFIED;
1936 	if (!wccflag)
1937 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1938 
1939 	if (error == 0 && newvp == NULL) {
1940 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1941 			cnp->cn_proc, &np);
1942 		if (!error) {
1943 			newvp = NFSTOV(np);
1944 			if (newvp->v_type != VDIR)
1945 				error = EEXIST;
1946 		}
1947 	}
1948 	if (error) {
1949 		if (newvp)
1950 			vput(newvp);
1951 	} else {
1952 		VN_KNOTE(dvp, NOTE_WRITE|NOTE_LINK);
1953 		if (cnp->cn_flags & MAKEENTRY)
1954 			nfs_cache_enter(dvp, newvp, cnp);
1955 		*ap->a_vpp = newvp;
1956 	}
1957 	pool_put(&namei_pool, cnp->cn_pnbuf);
1958 	vput(dvp);
1959 	return (error);
1960 }
1961 
1962 /*
1963  * nfs remove directory call
1964  */
1965 int
1966 nfs_rmdir(void *v)
1967 {
1968 	struct vop_rmdir_args *ap = v;
1969 	struct vnode *vp = ap->a_vp;
1970 	struct vnode *dvp = ap->a_dvp;
1971 	struct componentname *cnp = ap->a_cnp;
1972 	struct nfsm_info	info;
1973 	u_int32_t *tl;
1974 	int32_t t1;
1975 	caddr_t cp2;
1976 	int error = 0, wccflag = NFSV3_WCCRATTR;
1977 
1978 	info.nmi_v3 = NFS_ISV3(dvp);
1979 
1980 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1981 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) +
1982 	    NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1983 	nfsm_fhtom(&info, dvp, info.nmi_v3);
1984 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1985 
1986 	info.nmi_procp = cnp->cn_proc;
1987 	info.nmi_cred = cnp->cn_cred;
1988 	error = nfs_request(dvp,  NFSPROC_RMDIR, &info);
1989 	if (info.nmi_v3)
1990 		nfsm_wcc_data(dvp, wccflag);
1991 	m_freem(info.nmi_mrep);
1992 
1993 nfsmout:
1994 	pool_put(&namei_pool, cnp->cn_pnbuf);
1995 	VTONFS(dvp)->n_flag |= NMODIFIED;
1996 	if (!wccflag)
1997 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1998 
1999 	VN_KNOTE(dvp, NOTE_WRITE|NOTE_LINK);
2000 	VN_KNOTE(vp, NOTE_DELETE);
2001 
2002 	cache_purge(vp);
2003 	vput(vp);
2004 	vput(dvp);
2005 	/*
2006 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2007 	 */
2008 	if (error == ENOENT)
2009 		error = 0;
2010 	return (error);
2011 }
2012 
2013 
2014 /*
2015  * The readdir logic below has a big design bug. It stores the NFS cookie in
2016  * the returned uio->uio_offset but does not store the verifier (it cannot).
2017  * Instead, the code stores the verifier in the nfsnode and applies that
2018  * verifies to all cookies, no matter what verifier was originally with
2019  * the cookie.
2020  *
2021  * From a practical standpoint, this is not a problem since almost all
2022  * NFS servers do not change the validity of cookies across deletes
2023  * and inserts.
2024  */
2025 
2026 struct nfs_dirent {
2027 	u_int32_t cookie[2];
2028 	struct dirent dirent;
2029 };
2030 
2031 #define	NFS_DIRHDSIZ	(sizeof (struct nfs_dirent) - (MAXNAMLEN + 1))
2032 #define NFS_DIRENT_OVERHEAD  offsetof(struct nfs_dirent, dirent)
2033 
2034 /*
2035  * nfs readdir call
2036  */
2037 int
2038 nfs_readdir(void *v)
2039 {
2040 	struct vop_readdir_args *ap = v;
2041 	struct vnode *vp = ap->a_vp;
2042 	struct nfsnode *np = VTONFS(vp);
2043 	struct uio *uio = ap->a_uio;
2044 	int tresid, error = 0;
2045 	struct vattr vattr;
2046 	int cnt;
2047 	u_int64_t  newoff = uio->uio_offset;
2048 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2049 	struct uio readdir_uio;
2050 	struct iovec readdir_iovec;
2051 	struct proc * p = uio->uio_procp;
2052 	int done = 0, eof = 0;
2053 	struct ucred *cred = ap->a_cred;
2054 	void *data;
2055 
2056 	if (vp->v_type != VDIR)
2057 		return (EPERM);
2058 	/*
2059 	 * First, check for hit on the EOF offset cache
2060 	 */
2061 	if (np->n_direofoffset != 0 &&
2062 	    uio->uio_offset == np->n_direofoffset) {
2063 		if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_procp) == 0 &&
2064 		    timespeccmp(&np->n_mtime, &vattr.va_mtime, ==)) {
2065 			nfsstats.direofcache_hits++;
2066 			*ap->a_eofflag = 1;
2067 			return (0);
2068 		}
2069 	}
2070 
2071 	if (uio->uio_resid < NFS_FABLKSIZE)
2072 		return (EINVAL);
2073 
2074 	tresid = uio->uio_resid;
2075 
2076 	if (uio->uio_rw != UIO_READ)
2077 		return (EINVAL);
2078 
2079 	if ((nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_GOTFSINFO)) == NFSMNT_NFSV3)
2080 		(void)nfs_fsinfo(nmp, vp, cred, p);
2081 
2082 	cnt = 5;
2083 
2084 	/* M_ZERO to avoid leaking kernel data in dirent padding */
2085 	data = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK|M_ZERO);
2086 	do {
2087 		struct nfs_dirent *ndp = data;
2088 
2089 		readdir_iovec.iov_len = NFS_DIRBLKSIZ;
2090 		readdir_iovec.iov_base = data;
2091 		readdir_uio.uio_offset = newoff;
2092 		readdir_uio.uio_iov = &readdir_iovec;
2093 		readdir_uio.uio_iovcnt = 1;
2094 		readdir_uio.uio_segflg = UIO_SYSSPACE;
2095 		readdir_uio.uio_rw = UIO_READ;
2096 		readdir_uio.uio_resid = NFS_DIRBLKSIZ;
2097 		readdir_uio.uio_procp = curproc;
2098 
2099 		if (nmp->nm_flag & NFSMNT_RDIRPLUS) {
2100 			error = nfs_readdirplusrpc(vp, &readdir_uio, cred,
2101 			    &eof, p);
2102 			if (error == NFSERR_NOTSUPP)
2103 				nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
2104 		}
2105 		if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0)
2106 			error = nfs_readdirrpc(vp, &readdir_uio, cred, &eof);
2107 
2108 		if (error == NFSERR_BAD_COOKIE)
2109 			error = EINVAL;
2110 
2111 		while (error == 0 &&
2112 		    ndp < (struct nfs_dirent *)readdir_iovec.iov_base) {
2113 			struct dirent *dp = &ndp->dirent;
2114 			int reclen = dp->d_reclen;
2115 
2116 			dp->d_reclen -= NFS_DIRENT_OVERHEAD;
2117 			dp->d_off = fxdr_hyper(&ndp->cookie[0]);
2118 
2119 			if (uio->uio_resid < dp->d_reclen) {
2120 				eof = 0;
2121 				done = 1;
2122 				break;
2123 			}
2124 
2125 			if ((error = uiomove(dp, dp->d_reclen, uio)))
2126 				break;
2127 
2128 			newoff = fxdr_hyper(&ndp->cookie[0]);
2129 
2130 			ndp = (struct nfs_dirent *)((u_int8_t *)ndp + reclen);
2131 		}
2132 	} while (!error && !done && !eof && cnt--);
2133 
2134 	free(data, M_TEMP, NFS_DIRBLKSIZ);
2135 	data = NULL;
2136 
2137 	uio->uio_offset = newoff;
2138 
2139 	if (!error && (eof || uio->uio_resid == tresid)) {
2140 		nfsstats.direofcache_misses++;
2141 		*ap->a_eofflag = 1;
2142 		return (0);
2143 	}
2144 
2145 	*ap->a_eofflag = 0;
2146 	return (error);
2147 }
2148 
2149 
2150 /*
2151  * The function below stuff the cookies in after the name
2152  */
2153 
2154 /*
2155  * Readdir rpc call.
2156  */
2157 int
2158 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2159     int *end_of_directory)
2160 {
2161 	int len, left;
2162 	struct nfs_dirent *ndp = NULL;
2163 	struct dirent *dp = NULL;
2164 	struct nfsm_info	info;
2165 	u_int32_t *tl;
2166 	caddr_t cp;
2167 	int32_t t1;
2168 	caddr_t cp2;
2169 	nfsuint64 cookie;
2170 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2171 	struct nfsnode *dnp = VTONFS(vp);
2172 	u_quad_t fileno;
2173 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2174 	int attrflag;
2175 
2176 	info.nmi_v3 = NFS_ISV3(vp);
2177 
2178 #ifdef DIAGNOSTIC
2179 	if (uiop->uio_iovcnt != 1 ||
2180 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2181 		panic("nfs readdirrpc bad uio");
2182 #endif
2183 
2184 	txdr_hyper(uiop->uio_offset, &cookie.nfsuquad[0]);
2185 
2186 	/*
2187 	 * Loop around doing readdir rpc's of size nm_readdirsize
2188 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2189 	 * The stopping criteria is EOF or buffer full.
2190 	 */
2191 	while (more_dirs && bigenough) {
2192 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2193 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3)
2194 		    + NFSX_READDIR(info.nmi_v3));
2195 		nfsm_fhtom(&info, vp, info.nmi_v3);
2196 		if (info.nmi_v3) {
2197 			tl = nfsm_build(&info.nmi_mb, 5 * NFSX_UNSIGNED);
2198 			*tl++ = cookie.nfsuquad[0];
2199 			*tl++ = cookie.nfsuquad[1];
2200 			if (cookie.nfsuquad[0] == 0 &&
2201 			    cookie.nfsuquad[1] == 0) {
2202 				*tl++ = 0;
2203 				*tl++ = 0;
2204 			} else {
2205 				*tl++ = dnp->n_cookieverf.nfsuquad[0];
2206 				*tl++ = dnp->n_cookieverf.nfsuquad[1];
2207 			}
2208 		} else {
2209 			tl = nfsm_build(&info.nmi_mb, 2 * NFSX_UNSIGNED);
2210 			*tl++ = cookie.nfsuquad[1];
2211 		}
2212 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2213 
2214 		info.nmi_procp = uiop->uio_procp;
2215 		info.nmi_cred = cred;
2216 		error = nfs_request(vp, NFSPROC_READDIR, &info);
2217 		if (info.nmi_v3)
2218 			nfsm_postop_attr(vp, attrflag);
2219 
2220 		if (error) {
2221 			m_freem(info.nmi_mrep);
2222 			goto nfsmout;
2223 		}
2224 
2225 		if (info.nmi_v3) {
2226 			nfsm_dissect(tl, u_int32_t *,
2227 			    2 * NFSX_UNSIGNED);
2228 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2229 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2230 		}
2231 
2232 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2233 		more_dirs = fxdr_unsigned(int, *tl);
2234 
2235 		/* loop thru the dir entries, doctoring them to dirent form */
2236 		while (more_dirs && bigenough) {
2237 			if (info.nmi_v3) {
2238 				nfsm_dissect(tl, u_int32_t *,
2239 				    3 * NFSX_UNSIGNED);
2240 				fileno = fxdr_hyper(tl);
2241 				len = fxdr_unsigned(int, *(tl + 2));
2242 			} else {
2243 				nfsm_dissect(tl, u_int32_t *,
2244 				    2 * NFSX_UNSIGNED);
2245 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2246 				len = fxdr_unsigned(int, *tl);
2247 			}
2248 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2249 				error = EBADRPC;
2250 				m_freem(info.nmi_mrep);
2251 				goto nfsmout;
2252 			}
2253 			tlen = DIRENT_RECSIZE(len) + NFS_DIRENT_OVERHEAD;
2254 			left = NFS_READDIRBLKSIZ - blksiz;
2255 			if (tlen > left) {
2256 				dp->d_reclen += left;
2257 				uiop->uio_iov->iov_base += left;
2258 				uiop->uio_iov->iov_len -= left;
2259 				uiop->uio_resid -= left;
2260 				blksiz = 0;
2261 			}
2262 			if (tlen > uiop->uio_resid)
2263 				bigenough = 0;
2264 			if (bigenough) {
2265 				ndp = (struct nfs_dirent *)
2266 				    uiop->uio_iov->iov_base;
2267 				dp = &ndp->dirent;
2268 				dp->d_fileno = fileno;
2269 				dp->d_namlen = len;
2270 				dp->d_reclen = tlen;
2271 				dp->d_type = DT_UNKNOWN;
2272 				blksiz += tlen;
2273 				if (blksiz == NFS_READDIRBLKSIZ)
2274 					blksiz = 0;
2275 				uiop->uio_resid -= NFS_DIRHDSIZ;
2276 				uiop->uio_iov->iov_base =
2277 				    (char *)uiop->uio_iov->iov_base +
2278 				    NFS_DIRHDSIZ;
2279 				uiop->uio_iov->iov_len -= NFS_DIRHDSIZ;
2280 				nfsm_mtouio(uiop, len);
2281 				cp = uiop->uio_iov->iov_base;
2282 				tlen -= NFS_DIRHDSIZ + len;
2283 				*cp = '\0';	/* null terminate */
2284 				uiop->uio_iov->iov_base += tlen;
2285 				uiop->uio_iov->iov_len -= tlen;
2286 				uiop->uio_resid -= tlen;
2287 			} else
2288 				nfsm_adv(nfsm_rndup(len));
2289 			if (info.nmi_v3) {
2290 				nfsm_dissect(tl, u_int32_t *,
2291 				    3 * NFSX_UNSIGNED);
2292 			} else {
2293 				nfsm_dissect(tl, u_int32_t *,
2294 				    2 * NFSX_UNSIGNED);
2295 			}
2296 			if (bigenough) {
2297 				if (info.nmi_v3) {
2298 					ndp->cookie[0] = cookie.nfsuquad[0] =
2299 					    *tl++;
2300 				} else
2301 					ndp->cookie[0] = 0;
2302 
2303 				ndp->cookie[1] = cookie.nfsuquad[1] = *tl++;
2304 			} else if (info.nmi_v3)
2305 				tl += 2;
2306 			else
2307 				tl++;
2308 			more_dirs = fxdr_unsigned(int, *tl);
2309 		}
2310 		/*
2311 		 * If at end of rpc data, get the eof boolean
2312 		 */
2313 		if (!more_dirs) {
2314 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2315 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2316 		}
2317 		m_freem(info.nmi_mrep);
2318 	}
2319 	/*
2320 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2321 	 * by increasing d_reclen for the last record.
2322 	 */
2323 	if (blksiz > 0) {
2324 		left = NFS_READDIRBLKSIZ - blksiz;
2325 		dp->d_reclen += left;
2326 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
2327 		    left;
2328 		uiop->uio_iov->iov_len -= left;
2329 		uiop->uio_resid -= left;
2330 	}
2331 
2332 	/*
2333 	 * We are now either at the end of the directory or have filled the
2334 	 * block.
2335 	 */
2336 	if (bigenough) {
2337 		dnp->n_direofoffset = fxdr_hyper(&cookie.nfsuquad[0]);
2338 		if (end_of_directory) *end_of_directory = 1;
2339 	} else {
2340 		if (uiop->uio_resid > 0)
2341 			printf("EEK! readdirrpc resid > 0\n");
2342 	}
2343 
2344 nfsmout:
2345 	return (error);
2346 }
2347 
2348 /*
2349  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2350  */
2351 int
2352 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2353     int *end_of_directory, struct proc *p)
2354 {
2355 	int len, left;
2356 	struct nfs_dirent *ndirp = NULL;
2357 	struct dirent *dp = NULL;
2358 	struct nfsm_info	info;
2359 	u_int32_t *tl;
2360 	caddr_t cp;
2361 	int32_t t1;
2362 	struct vnode *newvp;
2363 	caddr_t cp2, dpossav1, dpossav2;
2364 	struct mbuf *mdsav1, *mdsav2;
2365 	struct nameidata nami, *ndp = &nami;
2366 	struct componentname *cnp = &ndp->ni_cnd;
2367 	nfsuint64 cookie;
2368 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2369 	struct nfsnode *dnp = VTONFS(vp), *np;
2370 	nfsfh_t *fhp;
2371 	u_quad_t fileno;
2372 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2373 	int attrflag, fhsize;
2374 
2375 #ifdef DIAGNOSTIC
2376 	if (uiop->uio_iovcnt != 1 ||
2377 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2378 		panic("nfs readdirplusrpc bad uio");
2379 #endif
2380 	NDINIT(ndp, 0, 0, UIO_SYSSPACE, NULL, p);
2381 	ndp->ni_dvp = vp;
2382 	newvp = NULLVP;
2383 
2384 	txdr_hyper(uiop->uio_offset, &cookie.nfsuquad[0]);
2385 
2386 	/*
2387 	 * Loop around doing readdir rpc's of size nm_readdirsize
2388 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2389 	 * The stopping criteria is EOF or buffer full.
2390 	 */
2391 	while (more_dirs && bigenough) {
2392 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2393 		info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2394 		nfsm_fhtom(&info, vp, 1);
2395 		tl = nfsm_build(&info.nmi_mb, 6 * NFSX_UNSIGNED);
2396 		*tl++ = cookie.nfsuquad[0];
2397 		*tl++ = cookie.nfsuquad[1];
2398 		if (cookie.nfsuquad[0] == 0 &&
2399 		    cookie.nfsuquad[1] == 0) {
2400 			*tl++ = 0;
2401 			*tl++ = 0;
2402 		} else {
2403 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2404 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2405 		}
2406 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2407 		*tl = txdr_unsigned(nmp->nm_rsize);
2408 
2409 		info.nmi_procp = uiop->uio_procp;
2410 		info.nmi_cred = cred;
2411 		error = nfs_request(vp, NFSPROC_READDIRPLUS, &info);
2412 		nfsm_postop_attr(vp, attrflag);
2413 		if (error) {
2414 			m_freem(info.nmi_mrep);
2415 			goto nfsmout;
2416 		}
2417 
2418 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2419 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2420 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2421 		more_dirs = fxdr_unsigned(int, *tl);
2422 
2423 		/* loop thru the dir entries, doctoring them to 4bsd form */
2424 		while (more_dirs && bigenough) {
2425 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2426 			fileno = fxdr_hyper(tl);
2427 			len = fxdr_unsigned(int, *(tl + 2));
2428 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2429 				error = EBADRPC;
2430 				m_freem(info.nmi_mrep);
2431 				goto nfsmout;
2432 			}
2433 			tlen = DIRENT_RECSIZE(len) + NFS_DIRENT_OVERHEAD;
2434 			left = NFS_READDIRBLKSIZ - blksiz;
2435 			if (tlen > left) {
2436 				dp->d_reclen += left;
2437 				uiop->uio_iov->iov_base =
2438 				    (char *)uiop->uio_iov->iov_base + left;
2439 				uiop->uio_iov->iov_len -= left;
2440 				uiop->uio_resid -= left;
2441 				blksiz = 0;
2442 			}
2443 			if (tlen > uiop->uio_resid)
2444 				bigenough = 0;
2445 			if (bigenough) {
2446 				ndirp = (struct nfs_dirent *)
2447 				    uiop->uio_iov->iov_base;
2448 				dp = &ndirp->dirent;
2449 				dp->d_fileno = fileno;
2450 				dp->d_namlen = len;
2451 				dp->d_reclen = tlen;
2452 				dp->d_type = DT_UNKNOWN;
2453 				blksiz += tlen;
2454 				if (blksiz == NFS_READDIRBLKSIZ)
2455 					blksiz = 0;
2456 				uiop->uio_resid -= NFS_DIRHDSIZ;
2457 				uiop->uio_iov->iov_base =
2458 				    (char *)uiop->uio_iov->iov_base +
2459 				    NFS_DIRHDSIZ;
2460 				uiop->uio_iov->iov_len -= NFS_DIRHDSIZ;
2461 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2462 				cnp->cn_namelen = len;
2463 				nfsm_mtouio(uiop, len);
2464 				cp = uiop->uio_iov->iov_base;
2465 				tlen -= NFS_DIRHDSIZ + len;
2466 				*cp = '\0';
2467 				uiop->uio_iov->iov_base += tlen;
2468 				uiop->uio_iov->iov_len -= tlen;
2469 				uiop->uio_resid -= tlen;
2470 			} else
2471 				nfsm_adv(nfsm_rndup(len));
2472 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2473 			if (bigenough) {
2474 				ndirp->cookie[0] = cookie.nfsuquad[0] = *tl++;
2475 				ndirp->cookie[1] = cookie.nfsuquad[1] = *tl++;
2476 			} else
2477 				tl += 2;
2478 
2479 			/*
2480 			 * Since the attributes are before the file handle
2481 			 * (sigh), we must skip over the attributes and then
2482 			 * come back and get them.
2483 			 */
2484 			attrflag = fxdr_unsigned(int, *tl);
2485 			if (attrflag) {
2486 				dpossav1 = info.nmi_dpos;
2487 				mdsav1 = info.nmi_md;
2488 				nfsm_adv(NFSX_V3FATTR);
2489 				nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2490 				doit = fxdr_unsigned(int, *tl);
2491 				if (doit) {
2492 					nfsm_getfh(fhp, fhsize, 1);
2493 					if (NFS_CMPFH(dnp, fhp, fhsize)) {
2494 						vref(vp);
2495 						newvp = vp;
2496 						np = dnp;
2497 					} else {
2498 						error = nfs_nget(vp->v_mount,
2499 						    fhp, fhsize, &np);
2500 						if (error)
2501 							doit = 0;
2502 						else
2503 							newvp = NFSTOV(np);
2504 					}
2505 				}
2506 				if (doit && bigenough) {
2507 					dpossav2 = info.nmi_dpos;
2508 					info.nmi_dpos = dpossav1;
2509 					mdsav2 = info.nmi_md;
2510 					info.nmi_md = mdsav1;
2511 					nfsm_loadattr(newvp, NULL);
2512 					info.nmi_dpos = dpossav2;
2513 					info.nmi_md = mdsav2;
2514 					dp->d_type = IFTODT(
2515 						VTTOIF(np->n_vattr.va_type));
2516 					if (cnp->cn_namelen <=
2517 					    NAMECACHE_MAXLEN) {
2518 						ndp->ni_vp = newvp;
2519 						cache_purge(ndp->ni_dvp);
2520 						nfs_cache_enter(ndp->ni_dvp,
2521 						    ndp->ni_vp, cnp);
2522 					}
2523 				}
2524 			} else {
2525 				/* Just skip over the file handle */
2526 				nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2527 				i = fxdr_unsigned(int, *tl);
2528 				if (i > 0)
2529 					nfsm_adv(nfsm_rndup(i));
2530 			}
2531 			if (newvp != NULLVP) {
2532 				if (newvp == vp)
2533 					vrele(newvp);
2534 				else
2535 					vput(newvp);
2536 				newvp = NULLVP;
2537 			}
2538 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2539 			more_dirs = fxdr_unsigned(int, *tl);
2540 		}
2541 		/*
2542 		 * If at end of rpc data, get the eof boolean
2543 		 */
2544 		if (!more_dirs) {
2545 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2546 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2547 		}
2548 		m_freem(info.nmi_mrep);
2549 	}
2550 	/*
2551 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2552 	 * by increasing d_reclen for the last record.
2553 	 */
2554 	if (blksiz > 0) {
2555 		left = NFS_READDIRBLKSIZ - blksiz;
2556 		dp->d_reclen += left;
2557 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
2558 		    left;
2559 		uiop->uio_iov->iov_len -= left;
2560 		uiop->uio_resid -= left;
2561 	}
2562 
2563 	/*
2564 	 * We are now either at the end of the directory or have filled the
2565 	 * block.
2566 	 */
2567 	if (bigenough) {
2568 		dnp->n_direofoffset = fxdr_hyper(&cookie.nfsuquad[0]);
2569 		if (end_of_directory) *end_of_directory = 1;
2570 	} else {
2571 		if (uiop->uio_resid > 0)
2572 			printf("EEK! readdirplusrpc resid > 0\n");
2573 	}
2574 
2575 nfsmout:
2576 	if (newvp != NULLVP) {
2577 		if (newvp == vp)
2578 			vrele(newvp);
2579 		else
2580 			vput(newvp);
2581 	}
2582 	return (error);
2583 }
2584 
2585 /*
2586  * Silly rename. To make the NFS filesystem that is stateless look a little
2587  * more like the "ufs" a remove of an active vnode is translated to a rename
2588  * to a funny looking filename that is removed by nfs_inactive on the
2589  * nfsnode. There is the potential for another process on a different client
2590  * to create the same funny name between the nfs_lookitup() fails and the
2591  * nfs_rename() completes, but...
2592  */
2593 int
2594 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2595 {
2596 	struct sillyrename *sp;
2597 	struct nfsnode *np;
2598 	int error;
2599 
2600 	cache_purge(dvp);
2601 	np = VTONFS(vp);
2602 	sp = malloc(sizeof(*sp), M_NFSREQ, M_WAITOK);
2603 	sp->s_cred = crdup(cnp->cn_cred);
2604 	sp->s_dvp = dvp;
2605 	vref(dvp);
2606 
2607 	if (vp->v_type == VDIR) {
2608 #ifdef DIAGNOSTIC
2609 		printf("nfs: sillyrename dir\n");
2610 #endif
2611 		error = EINVAL;
2612 		goto bad;
2613 	}
2614 
2615 	/* Try lookitups until we get one that isn't there */
2616 	while (1) {
2617 		/* Fudge together a funny name */
2618 		u_int32_t rnd[2];
2619 
2620 		arc4random_buf(&rnd, sizeof rnd);
2621 		sp->s_namlen = snprintf(sp->s_name, sizeof sp->s_name,
2622 		    ".nfs%08X%08X", rnd[0], rnd[1]);
2623 		if (sp->s_namlen > sizeof sp->s_name)
2624 			sp->s_namlen = strlen(sp->s_name);
2625 
2626 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2627 		    cnp->cn_proc, NULL))
2628 			break;
2629 	}
2630 
2631 	error = nfs_renameit(dvp, cnp, sp);
2632 	if (error)
2633 		goto bad;
2634 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2635 		cnp->cn_proc, &np);
2636 	np->n_sillyrename = sp;
2637 	return (0);
2638 bad:
2639 	vrele(sp->s_dvp);
2640 	crfree(sp->s_cred);
2641 	free(sp, M_NFSREQ, sizeof(*sp));
2642 	return (error);
2643 }
2644 
2645 /*
2646  * Look up a file name and optionally either update the file handle or
2647  * allocate an nfsnode, depending on the value of npp.
2648  * npp == NULL	--> just do the lookup
2649  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2650  *			handled too
2651  * *npp != NULL --> update the file handle in the vnode
2652  */
2653 int
2654 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2655     struct proc *procp, struct nfsnode **npp)
2656 {
2657 	struct nfsm_info	info;
2658 	u_int32_t *tl;
2659 	int32_t t1;
2660 	struct vnode *newvp = NULL;
2661 	struct nfsnode *np, *dnp = VTONFS(dvp);
2662 	caddr_t cp2;
2663 	int error = 0, fhlen, attrflag = 0;
2664 	nfsfh_t *nfhp;
2665 
2666 	info.nmi_v3 = NFS_ISV3(dvp);
2667 
2668 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2669 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3) + NFSX_UNSIGNED +
2670 	    nfsm_rndup(len));
2671 	nfsm_fhtom(&info, dvp, info.nmi_v3);
2672 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2673 
2674 	info.nmi_procp = procp;
2675 	info.nmi_cred = cred;
2676 	error = nfs_request(dvp, NFSPROC_LOOKUP, &info);
2677 	if (error && !info.nmi_v3) {
2678 		m_freem(info.nmi_mrep);
2679 		goto nfsmout;
2680 	}
2681 
2682 	if (npp && !error) {
2683 		nfsm_getfh(nfhp, fhlen, info.nmi_v3);
2684 		if (*npp) {
2685 			np = *npp;
2686 			np->n_fhp = &np->n_fh;
2687 			bcopy(nfhp, np->n_fhp, fhlen);
2688 			np->n_fhsize = fhlen;
2689 			newvp = NFSTOV(np);
2690 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2691 			vref(dvp);
2692 			newvp = dvp;
2693 			np = dnp;
2694 		} else {
2695 			error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2696 			if (error) {
2697 				m_freem(info.nmi_mrep);
2698 				return (error);
2699 			}
2700 			newvp = NFSTOV(np);
2701 		}
2702 		if (info.nmi_v3) {
2703 			nfsm_postop_attr(newvp, attrflag);
2704 			if (!attrflag && *npp == NULL) {
2705 				m_freem(info.nmi_mrep);
2706 				if (newvp == dvp)
2707 					vrele(newvp);
2708 				else
2709 					vput(newvp);
2710 				return (ENOENT);
2711 			}
2712 		} else
2713 			nfsm_loadattr(newvp, NULL);
2714 	}
2715 	m_freem(info.nmi_mrep);
2716 nfsmout:
2717 	if (npp && *npp == NULL) {
2718 		if (error) {
2719 			if (newvp == dvp)
2720 				vrele(newvp);
2721 			else
2722 				vput(newvp);
2723 		} else
2724 			*npp = np;
2725 	}
2726 	return (error);
2727 }
2728 
2729 /*
2730  * Nfs Version 3 commit rpc
2731  */
2732 int
2733 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct proc *procp)
2734 {
2735 	struct nfsm_info	info;
2736 	u_int32_t *tl;
2737 	int32_t t1;
2738 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2739 	caddr_t cp2;
2740 	int error = 0, wccflag = NFSV3_WCCRATTR;
2741 
2742 	if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0)
2743 		return (0);
2744 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2745 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(1));
2746 	nfsm_fhtom(&info, vp, 1);
2747 
2748 	tl = nfsm_build(&info.nmi_mb, 3 * NFSX_UNSIGNED);
2749 	txdr_hyper(offset, tl);
2750 	tl += 2;
2751 	*tl = txdr_unsigned(cnt);
2752 
2753 	info.nmi_procp = procp;
2754 	info.nmi_cred = VTONFS(vp)->n_wcred;
2755 	error = nfs_request(vp, NFSPROC_COMMIT, &info);
2756 	nfsm_wcc_data(vp, wccflag);
2757 
2758 	if (!error) {
2759 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2760 		if (bcmp(nmp->nm_verf, tl,
2761 			NFSX_V3WRITEVERF)) {
2762 			bcopy(tl, nmp->nm_verf,
2763 				NFSX_V3WRITEVERF);
2764 			error = NFSERR_STALEWRITEVERF;
2765 		}
2766 	}
2767 	m_freem(info.nmi_mrep);
2768 
2769 nfsmout:
2770 	return (error);
2771 }
2772 
2773 /*
2774  * Kludge City..
2775  * - make nfs_bmap() essentially a no-op that does no translation
2776  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2777  *   (Maybe I could use the process's page mapping, but I was concerned that
2778  *    Kernel Write might not be enabled and also figured copyout() would do
2779  *    a lot more work than bcopy() and also it currently happens in the
2780  *    context of the swapper process (2).
2781  */
2782 int
2783 nfs_bmap(void *v)
2784 {
2785 	struct vop_bmap_args *ap = v;
2786 	struct vnode *vp = ap->a_vp;
2787 
2788 	if (ap->a_vpp != NULL)
2789 		*ap->a_vpp = vp;
2790 	if (ap->a_bnp != NULL)
2791 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2792 	return (0);
2793 }
2794 
2795 /*
2796  * Strategy routine.
2797  * For async requests when nfsiod(s) are running, queue the request by
2798  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2799  * request.
2800  */
2801 int
2802 nfs_strategy(void *v)
2803 {
2804 	struct vop_strategy_args *ap = v;
2805 	struct buf *bp = ap->a_bp;
2806 	struct proc *p;
2807 	int error = 0;
2808 
2809 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2810 		panic("nfs physio/async");
2811 	if (bp->b_flags & B_ASYNC)
2812 		p = NULL;
2813 	else
2814 		p = curproc;	/* XXX */
2815 	/*
2816 	 * If the op is asynchronous and an i/o daemon is waiting
2817 	 * queue the request, wake it up and wait for completion
2818 	 * otherwise just do it ourselves.
2819 	 */
2820 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp, 0))
2821 		error = nfs_doio(bp, p);
2822 	return (error);
2823 }
2824 
2825 /*
2826  * fsync vnode op. Just call nfs_flush() with commit == 1.
2827  */
2828 int
2829 nfs_fsync(void *v)
2830 {
2831 	struct vop_fsync_args *ap = v;
2832 
2833 	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2834 }
2835 
2836 /*
2837  * Flush all the blocks associated with a vnode.
2838  *	Walk through the buffer pool and push any dirty pages
2839  *	associated with the vnode.
2840  */
2841 int
2842 nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct proc *p,
2843     int commit)
2844 {
2845 	struct nfsnode *np = VTONFS(vp);
2846 	struct buf *bp;
2847 	int i;
2848 	struct buf *nbp;
2849 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2850 	uint64_t slptimeo = INFSLP;
2851 	int s, error = 0, slpflag = 0, retv, bvecpos;
2852 	int passone = 1;
2853 	u_quad_t off = (u_quad_t)-1, endoff = 0, toff;
2854 #ifndef NFS_COMMITBVECSIZ
2855 #define NFS_COMMITBVECSIZ	20
2856 #endif
2857 	struct buf *bvec[NFS_COMMITBVECSIZ];
2858 
2859 	if (nmp->nm_flag & NFSMNT_INT)
2860 		slpflag = PCATCH;
2861 	if (!commit)
2862 		passone = 0;
2863 	/*
2864 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2865 	 * server, but nas not been committed to stable storage on the server
2866 	 * yet. On the first pass, the byte range is worked out and the commit
2867 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2868 	 * job.
2869 	 */
2870 again:
2871 	bvecpos = 0;
2872 	if (NFS_ISV3(vp) && commit) {
2873 		s = splbio();
2874 		LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) {
2875 			if (bvecpos >= NFS_COMMITBVECSIZ)
2876 				break;
2877 			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2878 			    != (B_DELWRI | B_NEEDCOMMIT))
2879 				continue;
2880 			bremfree(bp);
2881 			bp->b_flags |= B_WRITEINPROG;
2882 			buf_acquire(bp);
2883 
2884 			/*
2885 			 * A list of these buffers is kept so that the
2886 			 * second loop knows which buffers have actually
2887 			 * been committed. This is necessary, since there
2888 			 * may be a race between the commit rpc and new
2889 			 * uncommitted writes on the file.
2890 			 */
2891 			bvec[bvecpos++] = bp;
2892 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2893 				bp->b_dirtyoff;
2894 			if (toff < off)
2895 				off = toff;
2896 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2897 			if (toff > endoff)
2898 				endoff = toff;
2899 		}
2900 		splx(s);
2901 	}
2902 	if (bvecpos > 0) {
2903 		/*
2904 		 * Commit data on the server, as required.
2905 		 */
2906 		bcstats.pendingwrites++;
2907 		bcstats.numwrites++;
2908 		retv = nfs_commit(vp, off, (int)(endoff - off), p);
2909 		if (retv == NFSERR_STALEWRITEVERF)
2910 			nfs_clearcommit(vp->v_mount);
2911 		/*
2912 		 * Now, either mark the blocks I/O done or mark the
2913 		 * blocks dirty, depending on whether the commit
2914 		 * succeeded.
2915 		 */
2916 		for (i = 0; i < bvecpos; i++) {
2917 			bp = bvec[i];
2918 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG);
2919 			if (retv) {
2920 				if (i == 0)
2921 					bcstats.pendingwrites--;
2922 				brelse(bp);
2923 			} else {
2924 				if (i > 0)
2925 					bcstats.pendingwrites++;
2926 				s = splbio();
2927 				buf_undirty(bp);
2928 				vp->v_numoutput++;
2929 				bp->b_flags |= B_ASYNC;
2930 				bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
2931 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2932 				biodone(bp);
2933 				splx(s);
2934 			}
2935 		}
2936 	}
2937 
2938 	/*
2939 	 * Start/do any write(s) that are required.
2940 	 */
2941 loop:
2942 	s = splbio();
2943 	LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) {
2944 		if (bp->b_flags & B_BUSY) {
2945 			if (waitfor != MNT_WAIT || passone)
2946 				continue;
2947 			bp->b_flags |= B_WANTED;
2948 			error = tsleep_nsec(bp, slpflag | (PRIBIO + 1),
2949 			    "nfsfsync", slptimeo);
2950 			splx(s);
2951 			if (error) {
2952 				if (nfs_sigintr(nmp, NULL, p))
2953 					return (EINTR);
2954 				if (slpflag == PCATCH) {
2955 					slpflag = 0;
2956 					slptimeo = SEC_TO_NSEC(2);
2957 				}
2958 			}
2959 			goto loop;
2960 		}
2961 		if ((bp->b_flags & B_DELWRI) == 0)
2962 			panic("nfs_fsync: not dirty");
2963 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT))
2964 			continue;
2965 		bremfree(bp);
2966 		if (passone || !commit) {
2967 			bp->b_flags |= B_ASYNC;
2968 		} else {
2969 			bp->b_flags |= (B_ASYNC|B_WRITEINPROG|B_NEEDCOMMIT);
2970 		}
2971 		buf_acquire(bp);
2972 		splx(s);
2973 		VOP_BWRITE(bp);
2974 		goto loop;
2975 	}
2976 	splx(s);
2977 	if (passone) {
2978 		passone = 0;
2979 		goto again;
2980 	}
2981 	if (waitfor == MNT_WAIT) {
2982  loop2:
2983 		s = splbio();
2984 		error = vwaitforio(vp, slpflag, "nfs_fsync", slptimeo);
2985 		splx(s);
2986 		if (error) {
2987 			if (nfs_sigintr(nmp, NULL, p))
2988 				return (EINTR);
2989 			if (slpflag == PCATCH) {
2990 				slpflag = 0;
2991 				slptimeo = SEC_TO_NSEC(2);
2992 			}
2993 			goto loop2;
2994 		}
2995 
2996 		if (!LIST_EMPTY(&vp->v_dirtyblkhd) && commit) {
2997 #if 0
2998 			vprint("nfs_fsync: dirty", vp);
2999 #endif
3000 			goto loop;
3001 		}
3002 	}
3003 	if (np->n_flag & NWRITEERR) {
3004 		error = np->n_error;
3005 		np->n_flag &= ~NWRITEERR;
3006 	}
3007 	return (error);
3008 }
3009 
3010 /*
3011  * Return POSIX pathconf information applicable to nfs.
3012  * Fake it. For v3 we could ask the server, but such code
3013  * hasn't been written yet.
3014  */
3015 /* ARGSUSED */
3016 int
3017 nfs_pathconf(void *v)
3018 {
3019 	struct vop_pathconf_args *ap = v;
3020 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3021 	int error = 0;
3022 
3023 	switch (ap->a_name) {
3024 	case _PC_LINK_MAX:
3025 		*ap->a_retval = LINK_MAX;
3026 		break;
3027 	case _PC_NAME_MAX:
3028 		*ap->a_retval = NAME_MAX;
3029 		break;
3030 	case _PC_CHOWN_RESTRICTED:
3031 		*ap->a_retval = 1;
3032 		break;
3033 	case _PC_NO_TRUNC:
3034 		*ap->a_retval = 1;
3035 		break;
3036 	case _PC_ALLOC_SIZE_MIN:
3037 		*ap->a_retval = NFS_FABLKSIZE;
3038 		break;
3039 	case _PC_FILESIZEBITS:
3040 		*ap->a_retval = 64;
3041 		break;
3042 	case _PC_REC_INCR_XFER_SIZE:
3043 		*ap->a_retval = min(nmp->nm_rsize, nmp->nm_wsize);
3044 		break;
3045 	case _PC_REC_MAX_XFER_SIZE:
3046 		*ap->a_retval = -1; /* means ``unlimited'' */
3047 		break;
3048 	case _PC_REC_MIN_XFER_SIZE:
3049 		*ap->a_retval = min(nmp->nm_rsize, nmp->nm_wsize);
3050 		break;
3051 	case _PC_REC_XFER_ALIGN:
3052 		*ap->a_retval = PAGE_SIZE;
3053 		break;
3054 	case _PC_SYMLINK_MAX:
3055 		*ap->a_retval = MAXPATHLEN;
3056 		break;
3057 	case _PC_2_SYMLINKS:
3058 		*ap->a_retval = 1;
3059 		break;
3060 	case _PC_TIMESTAMP_RESOLUTION:
3061 		*ap->a_retval = NFS_ISV3(ap->a_vp) ? 1 : 1000;
3062 		break;
3063 	default:
3064 		error = EINVAL;
3065 		break;
3066 	}
3067 
3068 	return (error);
3069 }
3070 
3071 /*
3072  * NFS advisory byte-level locks.
3073  */
3074 int
3075 nfs_advlock(void *v)
3076 {
3077 	struct vop_advlock_args *ap = v;
3078 	struct nfsnode *np = VTONFS(ap->a_vp);
3079 
3080 	return (lf_advlock(&np->n_lockf, np->n_size, ap->a_id, ap->a_op,
3081 	    ap->a_fl, ap->a_flags));
3082 }
3083 
3084 /*
3085  * Print out the contents of an nfsnode.
3086  */
3087 int
3088 nfs_print(void *v)
3089 {
3090 	struct vop_print_args *ap = v;
3091 	struct vnode *vp = ap->a_vp;
3092 	struct nfsnode *np = VTONFS(vp);
3093 
3094 	printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3095 		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3096 #ifdef FIFO
3097 	if (vp->v_type == VFIFO)
3098 		fifo_printinfo(vp);
3099 #endif
3100 	printf("\n");
3101 	return (0);
3102 }
3103 
3104 /*
3105  * Just call nfs_writebp() with the force argument set to 1.
3106  */
3107 int
3108 nfs_bwrite(void *v)
3109 {
3110 	struct vop_bwrite_args *ap = v;
3111 
3112 	return (nfs_writebp(ap->a_bp, 1));
3113 }
3114 
3115 /*
3116  * This is a clone of vop_generic_bwrite(), except that B_WRITEINPROG isn't set unless
3117  * the force flag is one and it also handles the B_NEEDCOMMIT flag.
3118  */
3119 int
3120 nfs_writebp(struct buf *bp, int force)
3121 {
3122 	int oldflags = bp->b_flags, retv = 1;
3123 	struct proc *p = curproc;	/* XXX */
3124 	off_t off;
3125 	size_t cnt;
3126 	int   s;
3127 	struct vnode *vp;
3128 	struct nfsnode *np;
3129 
3130 	if(!(bp->b_flags & B_BUSY))
3131 		panic("bwrite: buffer is not busy???");
3132 
3133 	vp = bp->b_vp;
3134 	np = VTONFS(vp);
3135 
3136 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3137 
3138 	s = splbio();
3139 	buf_undirty(bp);
3140 
3141 	if ((oldflags & B_ASYNC) && !(oldflags & B_DELWRI) && p)
3142 		++p->p_ru.ru_oublock;
3143 
3144 	bp->b_vp->v_numoutput++;
3145 	splx(s);
3146 
3147 	/*
3148 	 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not
3149 	 * an actual write will have to be scheduled via. VOP_STRATEGY().
3150 	 * If B_WRITEINPROG is already set, then push it with a write anyhow.
3151 	 */
3152 	if ((oldflags & (B_NEEDCOMMIT | B_WRITEINPROG)) == B_NEEDCOMMIT) {
3153 		off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
3154 		cnt = bp->b_dirtyend - bp->b_dirtyoff;
3155 
3156 		rw_enter_write(&np->n_commitlock);
3157 		if (!(bp->b_flags & B_NEEDCOMMIT)) {
3158 			rw_exit_write(&np->n_commitlock);
3159 			return (0);
3160 		}
3161 
3162 		/*
3163 		 * If it's already been committed by somebody else,
3164 		 * bail.
3165 		 */
3166 		if (!nfs_in_committed_range(vp, bp)) {
3167 			int pushedrange = 0;
3168 			/*
3169 			 * Since we're going to do this, push as much
3170 			 * as we can.
3171 			 */
3172 
3173 			if (nfs_in_tobecommitted_range(vp, bp)) {
3174 				pushedrange = 1;
3175 				off = np->n_pushlo;
3176 				cnt = np->n_pushhi - np->n_pushlo;
3177 			}
3178 
3179 			bp->b_flags |= B_WRITEINPROG;
3180 			bcstats.pendingwrites++;
3181 			bcstats.numwrites++;
3182 			retv = nfs_commit(bp->b_vp, off, cnt, curproc);
3183 			bp->b_flags &= ~B_WRITEINPROG;
3184 
3185 			if (retv == 0) {
3186 				if (pushedrange)
3187 					nfs_merge_commit_ranges(vp);
3188 				else
3189 					nfs_add_committed_range(vp, bp);
3190 			} else
3191 				bcstats.pendingwrites--;
3192 		} else
3193 			retv = 0; /* It has already been committed. */
3194 
3195 		rw_exit_write(&np->n_commitlock);
3196 		if (!retv) {
3197 			bp->b_dirtyoff = bp->b_dirtyend = 0;
3198 			bp->b_flags &= ~B_NEEDCOMMIT;
3199 			s = splbio();
3200 			biodone(bp);
3201 			splx(s);
3202 		} else if (retv == NFSERR_STALEWRITEVERF)
3203 			nfs_clearcommit(bp->b_vp->v_mount);
3204 	}
3205 	if (retv) {
3206 		buf_flip_dma(bp);
3207 		if (force)
3208 			bp->b_flags |= B_WRITEINPROG;
3209 		VOP_STRATEGY(bp->b_vp, bp);
3210 	}
3211 
3212 	if( (oldflags & B_ASYNC) == 0) {
3213 		int rtval;
3214 
3215 		bp->b_flags |= B_RAW;
3216 		rtval = biowait(bp);
3217 		if (!(oldflags & B_DELWRI) && p) {
3218 			++p->p_ru.ru_oublock;
3219 		}
3220 		brelse(bp);
3221 		return (rtval);
3222 	}
3223 
3224 	return (0);
3225 }
3226 
3227 /*
3228  * nfs special file access vnode op.
3229  * Essentially just get vattr and then imitate iaccess() since the device is
3230  * local to the client.
3231  */
3232 int
3233 nfsspec_access(void *v)
3234 {
3235 	struct vop_access_args *ap = v;
3236 	struct vattr va;
3237 	struct vnode *vp = ap->a_vp;
3238 	int error;
3239 
3240 	/*
3241 	 * Disallow write attempts on filesystems mounted read-only;
3242 	 * unless the file is a socket, fifo, or a block or character
3243 	 * device resident on the filesystem.
3244 	 */
3245 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3246 		switch (vp->v_type) {
3247 		case VREG:
3248 		case VDIR:
3249 		case VLNK:
3250 			return (EROFS);
3251 		default:
3252 			break;
3253 		}
3254 	}
3255 
3256 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3257 	if (error)
3258 		return (error);
3259 
3260 	return (vaccess(vp->v_type, va.va_mode, va.va_uid, va.va_gid,
3261 	    ap->a_mode, ap->a_cred));
3262 }
3263 
3264 /*
3265  * Read wrapper for special devices.
3266  */
3267 int
3268 nfsspec_read(void *v)
3269 {
3270 	struct vop_read_args *ap = v;
3271 	struct nfsnode *np = VTONFS(ap->a_vp);
3272 
3273 	/*
3274 	 * Set access flag.
3275 	 */
3276 	np->n_flag |= NACC;
3277 	getnanotime(&np->n_atim);
3278 	return (spec_read(ap));
3279 }
3280 
3281 /*
3282  * Write wrapper for special devices.
3283  */
3284 int
3285 nfsspec_write(void *v)
3286 {
3287 	struct vop_write_args *ap = v;
3288 	struct nfsnode *np = VTONFS(ap->a_vp);
3289 
3290 	/*
3291 	 * Set update flag.
3292 	 */
3293 	np->n_flag |= NUPD;
3294 	getnanotime(&np->n_mtim);
3295 	return (spec_write(ap));
3296 }
3297 
3298 /*
3299  * Close wrapper for special devices.
3300  *
3301  * Update the times on the nfsnode then do device close.
3302  */
3303 int
3304 nfsspec_close(void *v)
3305 {
3306 	struct vop_close_args *ap = v;
3307 	struct vnode *vp = ap->a_vp;
3308 	struct nfsnode *np = VTONFS(vp);
3309 	struct vattr vattr;
3310 
3311 	if (np->n_flag & (NACC | NUPD)) {
3312 		np->n_flag |= NCHG;
3313 		if (vp->v_usecount == 1 &&
3314 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3315 			VATTR_NULL(&vattr);
3316 			if (np->n_flag & NACC)
3317 				vattr.va_atime = np->n_atim;
3318 			if (np->n_flag & NUPD)
3319 				vattr.va_mtime = np->n_mtim;
3320 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3321 		}
3322 	}
3323 	return (spec_close(ap));
3324 }
3325 
3326 #ifdef FIFO
3327 /*
3328  * Read wrapper for fifos.
3329  */
3330 int
3331 nfsfifo_read(void *v)
3332 {
3333 	struct vop_read_args *ap = v;
3334 	struct nfsnode *np = VTONFS(ap->a_vp);
3335 
3336 	/*
3337 	 * Set access flag.
3338 	 */
3339 	np->n_flag |= NACC;
3340 	getnanotime(&np->n_atim);
3341 	return (fifo_read(ap));
3342 }
3343 
3344 /*
3345  * Write wrapper for fifos.
3346  */
3347 int
3348 nfsfifo_write(void *v)
3349 {
3350 	struct vop_write_args *ap = v;
3351 	struct nfsnode *np = VTONFS(ap->a_vp);
3352 
3353 	/*
3354 	 * Set update flag.
3355 	 */
3356 	np->n_flag |= NUPD;
3357 	getnanotime(&np->n_mtim);
3358 	return (fifo_write(ap));
3359 }
3360 
3361 /*
3362  * Close wrapper for fifos.
3363  *
3364  * Update the times on the nfsnode then do fifo close.
3365  */
3366 int
3367 nfsfifo_close(void *v)
3368 {
3369 	struct vop_close_args *ap = v;
3370 	struct vnode *vp = ap->a_vp;
3371 	struct nfsnode *np = VTONFS(vp);
3372 	struct vattr vattr;
3373 
3374 	if (np->n_flag & (NACC | NUPD)) {
3375 		if (np->n_flag & NACC) {
3376 			getnanotime(&np->n_atim);
3377 		}
3378 		if (np->n_flag & NUPD) {
3379 			getnanotime(&np->n_mtim);
3380 		}
3381 		np->n_flag |= NCHG;
3382 		if (vp->v_usecount == 1 &&
3383 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3384 			VATTR_NULL(&vattr);
3385 			if (np->n_flag & NACC)
3386 				vattr.va_atime = np->n_atim;
3387 			if (np->n_flag & NUPD)
3388 				vattr.va_mtime = np->n_mtim;
3389 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3390 		}
3391 	}
3392 	return (fifo_close(ap));
3393 }
3394 
3395 int
3396 nfsfifo_reclaim(void *v)
3397 {
3398 	fifo_reclaim(v);
3399 	return (nfs_reclaim(v));
3400 }
3401 #endif /* ! FIFO */
3402