xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision fd5cb0acea84d278e04e640d37ca2398f894991f)
1 /*	$NetBSD: nfs_vnops.c,v 1.219 2005/01/26 10:33:54 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_vnops.c	8.19 (Berkeley) 7/31/95
35  */
36 
37 /*
38  * vnode op calls for Sun NFS version 2 and 3
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.219 2005/01/26 10:33:54 yamt Exp $");
43 
44 #include "opt_inet.h"
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 
48 #include <sys/param.h>
49 #include <sys/proc.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/resourcevar.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/namei.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/hash.h>
63 #include <sys/lockf.h>
64 #include <sys/stat.h>
65 #include <sys/unistd.h>
66 
67 #include <uvm/uvm_extern.h>
68 #include <uvm/uvm.h>
69 
70 #include <miscfs/fifofs/fifo.h>
71 #include <miscfs/genfs/genfs.h>
72 #include <miscfs/specfs/specdev.h>
73 
74 #include <nfs/rpcv2.h>
75 #include <nfs/nfsproto.h>
76 #include <nfs/nfs.h>
77 #include <nfs/nfsnode.h>
78 #include <nfs/nfsmount.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfs/nfsm_subs.h>
81 #include <nfs/nqnfs.h>
82 #include <nfs/nfs_var.h>
83 
84 #include <net/if.h>
85 #include <netinet/in.h>
86 #include <netinet/in_var.h>
87 
88 /*
89  * Global vfs data structures for nfs
90  */
91 int (**nfsv2_vnodeop_p) __P((void *));
92 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
93 	{ &vop_default_desc, vn_default_error },
94 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
95 	{ &vop_create_desc, nfs_create },		/* create */
96 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
97 	{ &vop_open_desc, nfs_open },			/* open */
98 	{ &vop_close_desc, nfs_close },			/* close */
99 	{ &vop_access_desc, nfs_access },		/* access */
100 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
101 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
102 	{ &vop_read_desc, nfs_read },			/* read */
103 	{ &vop_write_desc, nfs_write },			/* write */
104 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
105 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
106 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
107 	{ &vop_poll_desc, nfs_poll },			/* poll */
108 	{ &vop_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
109 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
110 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
111 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
112 	{ &vop_seek_desc, nfs_seek },			/* seek */
113 	{ &vop_remove_desc, nfs_remove },		/* remove */
114 	{ &vop_link_desc, nfs_link },			/* link */
115 	{ &vop_rename_desc, nfs_rename },		/* rename */
116 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
117 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
118 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
119 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
120 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
121 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
122 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
123 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
124 	{ &vop_lock_desc, nfs_lock },			/* lock */
125 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
126 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
127 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
128 	{ &vop_print_desc, nfs_print },			/* print */
129 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
130 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
131 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
132 	{ &vop_blkatoff_desc, genfs_badop },		/* blkatoff */
133 	{ &vop_valloc_desc, genfs_badop },		/* valloc */
134 	{ &vop_reallocblks_desc, genfs_badop },		/* reallocblks */
135 	{ &vop_vfree_desc, genfs_badop },		/* vfree */
136 	{ &vop_truncate_desc, genfs_badop },		/* truncate */
137 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
138 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
139 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
140 	{ NULL, NULL }
141 };
142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
143 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
144 
145 /*
146  * Special device vnode ops
147  */
148 int (**spec_nfsv2nodeop_p) __P((void *));
149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
150 	{ &vop_default_desc, vn_default_error },
151 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
152 	{ &vop_create_desc, spec_create },		/* create */
153 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
154 	{ &vop_open_desc, spec_open },			/* open */
155 	{ &vop_close_desc, nfsspec_close },		/* close */
156 	{ &vop_access_desc, nfsspec_access },		/* access */
157 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
158 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
159 	{ &vop_read_desc, nfsspec_read },		/* read */
160 	{ &vop_write_desc, nfsspec_write },		/* write */
161 	{ &vop_lease_desc, spec_lease_check },		/* lease */
162 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
163 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
164 	{ &vop_poll_desc, spec_poll },			/* poll */
165 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
166 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
167 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
168 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
169 	{ &vop_seek_desc, spec_seek },			/* seek */
170 	{ &vop_remove_desc, spec_remove },		/* remove */
171 	{ &vop_link_desc, spec_link },			/* link */
172 	{ &vop_rename_desc, spec_rename },		/* rename */
173 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
174 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
175 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
176 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
177 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
178 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
179 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
180 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
181 	{ &vop_lock_desc, nfs_lock },			/* lock */
182 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
183 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
184 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
185 	{ &vop_print_desc, nfs_print },			/* print */
186 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
187 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
188 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
189 	{ &vop_blkatoff_desc, genfs_badop },		/* blkatoff */
190 	{ &vop_valloc_desc, genfs_badop },		/* valloc */
191 	{ &vop_reallocblks_desc, genfs_badop },		/* reallocblks */
192 	{ &vop_vfree_desc, genfs_badop },		/* vfree */
193 	{ &vop_truncate_desc, genfs_badop },		/* truncate */
194 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
195 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
196 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
197 	{ NULL, NULL }
198 };
199 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
200 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
201 
202 int (**fifo_nfsv2nodeop_p) __P((void *));
203 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
204 	{ &vop_default_desc, vn_default_error },
205 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
206 	{ &vop_create_desc, fifo_create },		/* create */
207 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
208 	{ &vop_open_desc, fifo_open },			/* open */
209 	{ &vop_close_desc, nfsfifo_close },		/* close */
210 	{ &vop_access_desc, nfsspec_access },		/* access */
211 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
212 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
213 	{ &vop_read_desc, nfsfifo_read },		/* read */
214 	{ &vop_write_desc, nfsfifo_write },		/* write */
215 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
216 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
217 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
218 	{ &vop_poll_desc, fifo_poll },			/* poll */
219 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
220 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
221 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
222 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
223 	{ &vop_seek_desc, fifo_seek },			/* seek */
224 	{ &vop_remove_desc, fifo_remove },		/* remove */
225 	{ &vop_link_desc, fifo_link },			/* link */
226 	{ &vop_rename_desc, fifo_rename },		/* rename */
227 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
228 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
229 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
230 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
231 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
232 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
233 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
234 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
235 	{ &vop_lock_desc, nfs_lock },			/* lock */
236 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
237 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
238 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
239 	{ &vop_print_desc, nfs_print },			/* print */
240 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
241 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
242 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
243 	{ &vop_blkatoff_desc, genfs_badop },		/* blkatoff */
244 	{ &vop_valloc_desc, genfs_badop },		/* valloc */
245 	{ &vop_reallocblks_desc, genfs_badop },	/* reallocblks */
246 	{ &vop_vfree_desc, genfs_badop },		/* vfree */
247 	{ &vop_truncate_desc, genfs_badop },		/* truncate */
248 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
249 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
250 	{ NULL, NULL }
251 };
252 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
253 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
254 
255 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
256 
257 /*
258  * Global variables
259  */
260 extern u_int32_t nfs_true, nfs_false;
261 extern u_int32_t nfs_xdrneg1;
262 extern const nfstype nfsv3_type[9];
263 
264 int nfs_numasync = 0;
265 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
266 
267 static void nfs_cache_enter(struct vnode *, struct vnode *,
268     struct componentname *);
269 
270 static void
271 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
272     struct componentname *cnp)
273 {
274 	struct nfsnode *dnp = VTONFS(dvp);
275 
276 	if (vp != NULL) {
277 		struct nfsnode *np = VTONFS(vp);
278 
279 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
280 	}
281 
282 	if (!timespecisset(&dnp->n_nctime))
283 		dnp->n_nctime = dnp->n_vattr->va_mtime;
284 
285 	cache_enter(dvp, vp, cnp);
286 }
287 
288 /*
289  * nfs null call from vfs.
290  */
291 int
292 nfs_null(vp, cred, procp)
293 	struct vnode *vp;
294 	struct ucred *cred;
295 	struct proc *procp;
296 {
297 	caddr_t bpos, dpos;
298 	int error = 0;
299 	struct mbuf *mreq, *mrep, *md, *mb;
300 	struct nfsnode *np = VTONFS(vp);
301 
302 	nfsm_reqhead(np, NFSPROC_NULL, 0);
303 	nfsm_request(np, NFSPROC_NULL, procp, cred);
304 	nfsm_reqdone;
305 	return (error);
306 }
307 
308 /*
309  * nfs access vnode op.
310  * For nfs version 2, just return ok. File accesses may fail later.
311  * For nfs version 3, use the access rpc to check accessibility. If file modes
312  * are changed on the server, accesses might still fail later.
313  */
314 int
315 nfs_access(v)
316 	void *v;
317 {
318 	struct vop_access_args /* {
319 		struct vnode *a_vp;
320 		int  a_mode;
321 		struct ucred *a_cred;
322 		struct proc *a_p;
323 	} */ *ap = v;
324 	struct vnode *vp = ap->a_vp;
325 #ifndef NFS_V2_ONLY
326 	u_int32_t *tl;
327 	caddr_t cp;
328 	int32_t t1, t2;
329 	caddr_t bpos, dpos, cp2;
330 	int error = 0, attrflag;
331 	struct mbuf *mreq, *mrep, *md, *mb;
332 	u_int32_t mode, rmode;
333 	const int v3 = NFS_ISV3(vp);
334 #endif
335 	int cachevalid;
336 	struct nfsnode *np = VTONFS(vp);
337 
338 	cachevalid = (np->n_accstamp != -1 &&
339 	    (mono_time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
340 	    np->n_accuid == ap->a_cred->cr_uid);
341 
342 	/*
343 	 * Check access cache first. If this request has been made for this
344 	 * uid shortly before, use the cached result.
345 	 */
346 	if (cachevalid) {
347 		if (!np->n_accerror) {
348 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
349 				return np->n_accerror;
350 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
351 			return np->n_accerror;
352 	}
353 
354 #ifndef NFS_V2_ONLY
355 	/*
356 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
357 	 * ufs_access() locally using the vattr. This may not be correct,
358 	 * since the server may apply other access criteria such as
359 	 * client uid-->server uid mapping that we do not know about, but
360 	 * this is better than just returning anything that is lying about
361 	 * in the cache.
362 	 */
363 	if (v3) {
364 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
365 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
366 		nfsm_fhtom(np, v3);
367 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
368 		if (ap->a_mode & VREAD)
369 			mode = NFSV3ACCESS_READ;
370 		else
371 			mode = 0;
372 		if (vp->v_type != VDIR) {
373 			if (ap->a_mode & VWRITE)
374 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
375 			if (ap->a_mode & VEXEC)
376 				mode |= NFSV3ACCESS_EXECUTE;
377 		} else {
378 			if (ap->a_mode & VWRITE)
379 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
380 					 NFSV3ACCESS_DELETE);
381 			if (ap->a_mode & VEXEC)
382 				mode |= NFSV3ACCESS_LOOKUP;
383 		}
384 		*tl = txdr_unsigned(mode);
385 		nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
386 		nfsm_postop_attr(vp, attrflag, 0);
387 		if (!error) {
388 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
389 			rmode = fxdr_unsigned(u_int32_t, *tl);
390 			/*
391 			 * The NFS V3 spec does not clarify whether or not
392 			 * the returned access bits can be a superset of
393 			 * the ones requested, so...
394 			 */
395 			if ((rmode & mode) != mode)
396 				error = EACCES;
397 		}
398 		nfsm_reqdone;
399 	} else
400 #endif
401 		return (nfsspec_access(ap));
402 #ifndef NFS_V2_ONLY
403 	/*
404 	 * Disallow write attempts on filesystems mounted read-only;
405 	 * unless the file is a socket, fifo, or a block or character
406 	 * device resident on the filesystem.
407 	 */
408 	if (!error && (ap->a_mode & VWRITE) &&
409 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
410 		switch (vp->v_type) {
411 		case VREG:
412 		case VDIR:
413 		case VLNK:
414 			error = EROFS;
415 		default:
416 			break;
417 		}
418 	}
419 
420 	if (!error || error == EACCES) {
421 		/*
422 		 * If we got the same result as for a previous,
423 		 * different request, OR it in. Don't update
424 		 * the timestamp in that case.
425 		 */
426 		if (cachevalid && np->n_accstamp != -1 &&
427 		    error == np->n_accerror) {
428 			if (!error)
429 				np->n_accmode |= ap->a_mode;
430 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
431 				np->n_accmode = ap->a_mode;
432 		} else {
433 			np->n_accstamp = mono_time.tv_sec;
434 			np->n_accuid = ap->a_cred->cr_uid;
435 			np->n_accmode = ap->a_mode;
436 			np->n_accerror = error;
437 		}
438 	}
439 
440 	return (error);
441 #endif
442 }
443 
444 /*
445  * nfs open vnode op
446  * Check to see if the type is ok
447  * and that deletion is not in progress.
448  * For paged in text files, you will need to flush the page cache
449  * if consistency is lost.
450  */
451 /* ARGSUSED */
452 int
453 nfs_open(v)
454 	void *v;
455 {
456 	struct vop_open_args /* {
457 		struct vnode *a_vp;
458 		int  a_mode;
459 		struct ucred *a_cred;
460 		struct proc *a_p;
461 	} */ *ap = v;
462 	struct vnode *vp = ap->a_vp;
463 	struct nfsnode *np = VTONFS(vp);
464 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
465 	int error;
466 
467 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
468 		return (EACCES);
469 	}
470 
471 	if (ap->a_mode & FREAD) {
472 		if (np->n_rcred != NULL)
473 			crfree(np->n_rcred);
474 		np->n_rcred = ap->a_cred;
475 		crhold(np->n_rcred);
476 	}
477 	if (ap->a_mode & FWRITE) {
478 		if (np->n_wcred != NULL)
479 			crfree(np->n_wcred);
480 		np->n_wcred = ap->a_cred;
481 		crhold(np->n_wcred);
482 	}
483 
484 #ifndef NFS_V2_ONLY
485 	/*
486 	 * Get a valid lease. If cached data is stale, flush it.
487 	 */
488 	if (nmp->nm_flag & NFSMNT_NQNFS) {
489 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
490 		    do {
491 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
492 			    ap->a_p);
493 		    } while (error == NQNFS_EXPIRED);
494 		    if (error)
495 			return (error);
496 		    if (np->n_lrev != np->n_brev ||
497 			(np->n_flag & NQNFSNONCACHE)) {
498 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
499 				ap->a_p, 1)) == EINTR)
500 				return (error);
501 			np->n_brev = np->n_lrev;
502 		    }
503 		}
504 	} else
505 #endif
506 	{
507 		error = nfs_flushstalebuf(vp, ap->a_cred, ap->a_p, 0);
508 		if (error)
509 			return error;
510 	}
511 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
512 		NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
513 	return (0);
514 }
515 
516 /*
517  * nfs close vnode op
518  * What an NFS client should do upon close after writing is a debatable issue.
519  * Most NFS clients push delayed writes to the server upon close, basically for
520  * two reasons:
521  * 1 - So that any write errors may be reported back to the client process
522  *     doing the close system call. By far the two most likely errors are
523  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
524  * 2 - To put a worst case upper bound on cache inconsistency between
525  *     multiple clients for the file.
526  * There is also a consistency problem for Version 2 of the protocol w.r.t.
527  * not being able to tell if other clients are writing a file concurrently,
528  * since there is no way of knowing if the changed modify time in the reply
529  * is only due to the write for this client.
530  * (NFS Version 3 provides weak cache consistency data in the reply that
531  *  should be sufficient to detect and handle this case.)
532  *
533  * The current code does the following:
534  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
535  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
536  *                     or commit them (this satisfies 1 and 2 except for the
537  *                     case where the server crashes after this close but
538  *                     before the commit RPC, which is felt to be "good
539  *                     enough". Changing the last argument to nfs_flush() to
540  *                     a 1 would force a commit operation, if it is felt a
541  *                     commit is necessary now.
542  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
543  *                     1 should be dealt with via an fsync() system call for
544  *                     cases where write errors are important.
545  */
546 /* ARGSUSED */
547 int
548 nfs_close(v)
549 	void *v;
550 {
551 	struct vop_close_args /* {
552 		struct vnodeop_desc *a_desc;
553 		struct vnode *a_vp;
554 		int  a_fflag;
555 		struct ucred *a_cred;
556 		struct proc *a_p;
557 	} */ *ap = v;
558 	struct vnode *vp = ap->a_vp;
559 	struct nfsnode *np = VTONFS(vp);
560 	int error = 0;
561 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
562 
563 	if (vp->v_type == VREG) {
564 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
565 		(np->n_flag & NMODIFIED)) {
566 #ifndef NFS_V2_ONLY
567 		if (NFS_ISV3(vp)) {
568 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
569 		    np->n_flag &= ~NMODIFIED;
570 		} else
571 #endif
572 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
573 		NFS_INVALIDATE_ATTRCACHE(np);
574 	    }
575 	    if (np->n_flag & NWRITEERR) {
576 		np->n_flag &= ~NWRITEERR;
577 		error = np->n_error;
578 	    }
579 	}
580 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
581 	return (error);
582 }
583 
584 /*
585  * nfs getattr call from vfs.
586  */
587 int
588 nfs_getattr(v)
589 	void *v;
590 {
591 	struct vop_getattr_args /* {
592 		struct vnode *a_vp;
593 		struct vattr *a_vap;
594 		struct ucred *a_cred;
595 		struct proc *a_p;
596 	} */ *ap = v;
597 	struct vnode *vp = ap->a_vp;
598 	struct nfsnode *np = VTONFS(vp);
599 	caddr_t cp;
600 	u_int32_t *tl;
601 	int32_t t1, t2;
602 	caddr_t bpos, dpos;
603 	int error = 0;
604 	struct mbuf *mreq, *mrep, *md, *mb;
605 	const int v3 = NFS_ISV3(vp);
606 
607 	/*
608 	 * Update local times for special files.
609 	 */
610 	if (np->n_flag & (NACC | NUPD))
611 		np->n_flag |= NCHG;
612 
613 	/*
614 	 * if we have delayed truncation, do it now.
615 	 */
616 	nfs_delayedtruncate(vp);
617 
618 	/*
619 	 * First look in the cache.
620 	 */
621 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
622 		return (0);
623 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
624 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
625 	nfsm_fhtom(np, v3);
626 	nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
627 	if (!error) {
628 		nfsm_loadattr(vp, ap->a_vap, 0);
629 		if (vp->v_type == VDIR &&
630 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
631 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
632 	}
633 	nfsm_reqdone;
634 	return (error);
635 }
636 
637 /*
638  * nfs setattr call.
639  */
640 int
641 nfs_setattr(v)
642 	void *v;
643 {
644 	struct vop_setattr_args /* {
645 		struct vnodeop_desc *a_desc;
646 		struct vnode *a_vp;
647 		struct vattr *a_vap;
648 		struct ucred *a_cred;
649 		struct proc *a_p;
650 	} */ *ap = v;
651 	struct vnode *vp = ap->a_vp;
652 	struct nfsnode *np = VTONFS(vp);
653 	struct vattr *vap = ap->a_vap;
654 	int error = 0;
655 	u_quad_t tsize = 0;
656 
657 	/*
658 	 * Setting of flags is not supported.
659 	 */
660 	if (vap->va_flags != VNOVAL)
661 		return (EOPNOTSUPP);
662 
663 	/*
664 	 * Disallow write attempts if the filesystem is mounted read-only.
665 	 */
666   	if ((vap->va_uid != (uid_t)VNOVAL ||
667 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
668 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
669 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
670 		return (EROFS);
671 	if (vap->va_size != VNOVAL) {
672  		switch (vp->v_type) {
673  		case VDIR:
674  			return (EISDIR);
675  		case VCHR:
676  		case VBLK:
677  		case VSOCK:
678  		case VFIFO:
679 			if (vap->va_mtime.tv_sec == VNOVAL &&
680 			    vap->va_atime.tv_sec == VNOVAL &&
681 			    vap->va_mode == (mode_t)VNOVAL &&
682 			    vap->va_uid == (uid_t)VNOVAL &&
683 			    vap->va_gid == (gid_t)VNOVAL)
684 				return (0);
685  			vap->va_size = VNOVAL;
686  			break;
687  		default:
688 			/*
689 			 * Disallow write attempts if the filesystem is
690 			 * mounted read-only.
691 			 */
692 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
693 				return (EROFS);
694  			uvm_vnp_setsize(vp, vap->va_size);
695  			tsize = np->n_size;
696 			np->n_size = vap->va_size;
697  			if (vap->va_size == 0)
698  				error = nfs_vinvalbuf(vp, 0,
699  				     ap->a_cred, ap->a_p, 1);
700 			else
701 				error = nfs_vinvalbuf(vp, V_SAVE,
702 				     ap->a_cred, ap->a_p, 1);
703 			if (error) {
704 				uvm_vnp_setsize(vp, tsize);
705 				return (error);
706 			}
707  			np->n_vattr->va_size = vap->va_size;
708   		}
709   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
710 		vap->va_atime.tv_sec != VNOVAL) &&
711 		vp->v_type == VREG &&
712   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
713 		 ap->a_p, 1)) == EINTR)
714 		return (error);
715 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
716 	if (error && vap->va_size != VNOVAL) {
717 		np->n_size = np->n_vattr->va_size = tsize;
718 		uvm_vnp_setsize(vp, np->n_size);
719 	}
720 	VN_KNOTE(vp, NOTE_ATTRIB);
721 	return (error);
722 }
723 
724 /*
725  * Do an nfs setattr rpc.
726  */
727 int
728 nfs_setattrrpc(vp, vap, cred, procp)
729 	struct vnode *vp;
730 	struct vattr *vap;
731 	struct ucred *cred;
732 	struct proc *procp;
733 {
734 	struct nfsv2_sattr *sp;
735 	caddr_t cp;
736 	int32_t t1, t2;
737 	caddr_t bpos, dpos;
738 	u_int32_t *tl;
739 	int error = 0;
740 	struct mbuf *mreq, *mrep, *md, *mb;
741 	const int v3 = NFS_ISV3(vp);
742 	struct nfsnode *np = VTONFS(vp);
743 #ifndef NFS_V2_ONLY
744 	int wccflag = NFSV3_WCCRATTR;
745 	caddr_t cp2;
746 #endif
747 
748 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
749 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
750 	nfsm_fhtom(np, v3);
751 #ifndef NFS_V2_ONLY
752 	if (v3) {
753 		nfsm_v3attrbuild(vap, TRUE);
754 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
755 		*tl = nfs_false;
756 	} else {
757 #endif
758 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
759 		if (vap->va_mode == (mode_t)VNOVAL)
760 			sp->sa_mode = nfs_xdrneg1;
761 		else
762 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
763 		if (vap->va_uid == (uid_t)VNOVAL)
764 			sp->sa_uid = nfs_xdrneg1;
765 		else
766 			sp->sa_uid = txdr_unsigned(vap->va_uid);
767 		if (vap->va_gid == (gid_t)VNOVAL)
768 			sp->sa_gid = nfs_xdrneg1;
769 		else
770 			sp->sa_gid = txdr_unsigned(vap->va_gid);
771 		sp->sa_size = txdr_unsigned(vap->va_size);
772 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
773 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
774 #ifndef NFS_V2_ONLY
775 	}
776 #endif
777 	nfsm_request(np, NFSPROC_SETATTR, procp, cred);
778 #ifndef NFS_V2_ONLY
779 	if (v3) {
780 		nfsm_wcc_data(vp, wccflag, 0, FALSE);
781 	} else
782 #endif
783 		nfsm_loadattr(vp, (struct vattr *)0, 0);
784 	nfsm_reqdone;
785 	return (error);
786 }
787 
788 /*
789  * nfs lookup call, one step at a time...
790  * First look in cache
791  * If not found, unlock the directory nfsnode and do the rpc
792  *
793  * This code is full of lock/unlock statements and checks, because
794  * we continue after cache_lookup has finished (we need to check
795  * with the attr cache and do an rpc if it has timed out). This means
796  * that the locking effects of cache_lookup have to be taken into
797  * account.
798  */
799 int
800 nfs_lookup(v)
801 	void *v;
802 {
803 	struct vop_lookup_args /* {
804 		struct vnodeop_desc *a_desc;
805 		struct vnode *a_dvp;
806 		struct vnode **a_vpp;
807 		struct componentname *a_cnp;
808 	} */ *ap = v;
809 	struct componentname *cnp = ap->a_cnp;
810 	struct vnode *dvp = ap->a_dvp;
811 	struct vnode **vpp = ap->a_vpp;
812 	int flags;
813 	struct vnode *newvp;
814 	u_int32_t *tl;
815 	caddr_t cp;
816 	int32_t t1, t2;
817 	caddr_t bpos, dpos, cp2;
818 	struct mbuf *mreq, *mrep, *md, *mb;
819 	long len;
820 	nfsfh_t *fhp;
821 	struct nfsnode *np;
822 	int lockparent, wantparent, error = 0, attrflag, fhsize;
823 	const int v3 = NFS_ISV3(dvp);
824 
825 	cnp->cn_flags &= ~PDIRUNLOCK;
826 	flags = cnp->cn_flags;
827 
828 	*vpp = NULLVP;
829 	newvp = NULLVP;
830 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
831 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
832 		return (EROFS);
833 	if (dvp->v_type != VDIR)
834 		return (ENOTDIR);
835 
836 	/*
837 	 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
838 	 */
839 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
840 		error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
841 		if (error)
842 			return error;
843 		if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
844 			return EISDIR;
845 		VREF(dvp);
846 		*vpp = dvp;
847 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
848 				cnp->cn_flags |= SAVENAME;
849 		return 0;
850 	}
851 
852 	lockparent = flags & LOCKPARENT;
853 	wantparent = flags & (LOCKPARENT|WANTPARENT);
854 	np = VTONFS(dvp);
855 
856 	/*
857 	 * Before tediously performing a linear scan of the directory,
858 	 * check the name cache to see if the directory/name pair
859 	 * we are looking for is known already.
860 	 * If the directory/name pair is found in the name cache,
861 	 * we have to ensure the directory has not changed from
862 	 * the time the cache entry has been created. If it has,
863 	 * the cache entry has to be ignored.
864 	 */
865 	error = cache_lookup_raw(dvp, vpp, cnp);
866 	KASSERT(dvp != *vpp);
867 	if (error >= 0) {
868 		struct vattr vattr;
869 		int err2;
870 
871 		if (error && error != ENOENT) {
872 			*vpp = NULLVP;
873 			return error;
874 		}
875 
876 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
877 		if (err2 != 0) {
878 			if (error == 0)
879 				vrele(*vpp);
880 			*vpp = NULLVP;
881 			return err2;
882 		}
883 
884 		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
885 		    cnp->cn_proc) || timespeccmp(&vattr.va_mtime,
886 		    &VTONFS(dvp)->n_nctime, !=)) {
887 			if (error == 0) {
888 				vrele(*vpp);
889 				*vpp = NULLVP;
890 			}
891 			cache_purge1(dvp, NULL, PURGE_CHILDREN);
892 			timespecclear(&np->n_nctime);
893 			goto dorpc;
894 		}
895 
896 		if (error == ENOENT) {
897 			goto noentry;
898 		}
899 
900 		newvp = *vpp;
901 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
902 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
903 			nfsstats.lookupcache_hits++;
904 			if ((flags & ISDOTDOT) != 0 ||
905 			    (~flags & (LOCKPARENT|ISLASTCN)) != 0) {
906 				VOP_UNLOCK(dvp, 0);
907 				cnp->cn_flags |= PDIRUNLOCK;
908 			}
909 			error = vn_lock(newvp, LK_EXCLUSIVE);
910 			if (error) {
911 				/* newvp has been revoked. */
912 				vrele(newvp);
913 				*vpp = NULL;
914 				return error;
915 			}
916 			if ((~flags & (LOCKPARENT|ISLASTCN)) == 0
917 			    && (cnp->cn_flags & PDIRUNLOCK)) {
918 				KASSERT(flags & ISDOTDOT);
919 				error = vn_lock(dvp, LK_EXCLUSIVE);
920 				if (error) {
921 					vput(newvp);
922 					*vpp = NULL;
923 					return error;
924 				}
925 				cnp->cn_flags &= ~PDIRUNLOCK;
926 			}
927 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
928 				cnp->cn_flags |= SAVENAME;
929 			KASSERT(newvp->v_type != VNON);
930 			return (0);
931 		}
932 		cache_purge1(newvp, NULL, PURGE_PARENTS);
933 		vrele(newvp);
934 		*vpp = NULLVP;
935 	}
936 dorpc:
937 #if 0
938 	/*
939 	 * because nfsv3 has the same CREATE semantics as ours,
940 	 * we don't have to perform LOOKUPs beforehand.
941 	 *
942 	 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
943 	 * XXX although we have no way to know if O_EXCL is requested or not.
944 	 */
945 
946 	if (v3 && cnp->cn_nameiop == CREATE &&
947 	    (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
948 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
949 		KASSERT(lockparent);
950 		cnp->cn_flags |= SAVENAME;
951 		return (EJUSTRETURN);
952 	}
953 #endif /* 0 */
954 
955 	error = 0;
956 	newvp = NULLVP;
957 	nfsstats.lookupcache_misses++;
958 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
959 	len = cnp->cn_namelen;
960 	nfsm_reqhead(np, NFSPROC_LOOKUP,
961 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
962 	nfsm_fhtom(np, v3);
963 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
964 	nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
965 	if (error) {
966 		nfsm_postop_attr(dvp, attrflag, 0);
967 		m_freem(mrep);
968 		goto nfsmout;
969 	}
970 	nfsm_getfh(fhp, fhsize, v3);
971 
972 	/*
973 	 * Handle RENAME case...
974 	 */
975 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
976 		if (NFS_CMPFH(np, fhp, fhsize)) {
977 			m_freem(mrep);
978 			return (EISDIR);
979 		}
980 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
981 		if (error) {
982 			m_freem(mrep);
983 			return error;
984 		}
985 		newvp = NFSTOV(np);
986 #ifndef NFS_V2_ONLY
987 		if (v3) {
988 			nfsm_postop_attr(newvp, attrflag, 0);
989 			nfsm_postop_attr(dvp, attrflag, 0);
990 		} else
991 #endif
992 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
993 		*vpp = newvp;
994 		m_freem(mrep);
995 		cnp->cn_flags |= SAVENAME;
996 		if (!lockparent) {
997 			VOP_UNLOCK(dvp, 0);
998 			cnp->cn_flags |= PDIRUNLOCK;
999 		}
1000 		goto validate;
1001 	}
1002 
1003 	/*
1004 	 * The postop attr handling is duplicated for each if case,
1005 	 * because it should be done while dvp is locked (unlocking
1006 	 * dvp is different for each case).
1007 	 */
1008 
1009 	if (NFS_CMPFH(np, fhp, fhsize)) {
1010 		/*
1011 		 * "." lookup
1012 		 */
1013 		VREF(dvp);
1014 		newvp = dvp;
1015 #ifndef NFS_V2_ONLY
1016 		if (v3) {
1017 			nfsm_postop_attr(newvp, attrflag, 0);
1018 			nfsm_postop_attr(dvp, attrflag, 0);
1019 		} else
1020 #endif
1021 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1022 	} else if (flags & ISDOTDOT) {
1023 		/*
1024 		 * ".." lookup
1025 		 */
1026 		VOP_UNLOCK(dvp, 0);
1027 		cnp->cn_flags |= PDIRUNLOCK;
1028 
1029 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1030 		if (error) {
1031 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
1032 				cnp->cn_flags &= ~PDIRUNLOCK;
1033 			m_freem(mrep);
1034 			return error;
1035 		}
1036 		newvp = NFSTOV(np);
1037 
1038 #ifndef NFS_V2_ONLY
1039 		if (v3) {
1040 			nfsm_postop_attr(newvp, attrflag, 0);
1041 			nfsm_postop_attr(dvp, attrflag, 0);
1042 		} else
1043 #endif
1044 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1045 
1046 		if (lockparent && (flags & ISLASTCN)) {
1047 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
1048 				m_freem(mrep);
1049 				vput(newvp);
1050 				return error;
1051 			}
1052 			cnp->cn_flags &= ~PDIRUNLOCK;
1053 		}
1054 	} else {
1055 		/*
1056 		 * Other lookups.
1057 		 */
1058 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1059 		if (error) {
1060 			m_freem(mrep);
1061 			return error;
1062 		}
1063 		newvp = NFSTOV(np);
1064 #ifndef NFS_V2_ONLY
1065 		if (v3) {
1066 			nfsm_postop_attr(newvp, attrflag, 0);
1067 			nfsm_postop_attr(dvp, attrflag, 0);
1068 		} else
1069 #endif
1070 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1071 		if (!lockparent || !(flags & ISLASTCN)) {
1072 			VOP_UNLOCK(dvp, 0);
1073 			cnp->cn_flags |= PDIRUNLOCK;
1074 		}
1075 	}
1076 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1077 		cnp->cn_flags |= SAVENAME;
1078 	if ((cnp->cn_flags & MAKEENTRY) &&
1079 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1080 		nfs_cache_enter(dvp, newvp, cnp);
1081 	}
1082 	*vpp = newvp;
1083 	nfsm_reqdone;
1084 	if (error) {
1085 		/*
1086 		 * We get here only because of errors returned by
1087 		 * the RPC. Otherwise we'll have returned above
1088 		 * (the nfsm_* macros will jump to nfsm_reqdone
1089 		 * on error).
1090 		 */
1091 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1092 		    cnp->cn_nameiop != CREATE) {
1093 			nfs_cache_enter(dvp, NULL, cnp);
1094 		}
1095 		if (newvp != NULLVP) {
1096 			vrele(newvp);
1097 			if (newvp != dvp)
1098 				VOP_UNLOCK(newvp, 0);
1099 		}
1100 noentry:
1101 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1102 		    (flags & ISLASTCN) && error == ENOENT) {
1103 			if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1104 				error = EROFS;
1105 			} else {
1106 				error = EJUSTRETURN;
1107 				cnp->cn_flags |= SAVENAME;
1108 			}
1109 		}
1110 		*vpp = NULL;
1111 		return error;
1112 	}
1113 
1114 validate:
1115 	/*
1116 	 * make sure we have valid type and size.
1117 	 */
1118 
1119 	newvp = *vpp;
1120 	if (newvp->v_type == VNON) {
1121 		struct vattr vattr; /* dummy */
1122 
1123 		KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1124 		error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc);
1125 		if (error) {
1126 			vput(newvp);
1127 			*vpp = NULL;
1128 		}
1129 	}
1130 
1131 	return error;
1132 }
1133 
1134 /*
1135  * nfs read call.
1136  * Just call nfs_bioread() to do the work.
1137  */
1138 int
1139 nfs_read(v)
1140 	void *v;
1141 {
1142 	struct vop_read_args /* {
1143 		struct vnode *a_vp;
1144 		struct uio *a_uio;
1145 		int  a_ioflag;
1146 		struct ucred *a_cred;
1147 	} */ *ap = v;
1148 	struct vnode *vp = ap->a_vp;
1149 
1150 	if (vp->v_type != VREG)
1151 		return (EPERM);
1152 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1153 }
1154 
1155 /*
1156  * nfs readlink call
1157  */
1158 int
1159 nfs_readlink(v)
1160 	void *v;
1161 {
1162 	struct vop_readlink_args /* {
1163 		struct vnode *a_vp;
1164 		struct uio *a_uio;
1165 		struct ucred *a_cred;
1166 	} */ *ap = v;
1167 	struct vnode *vp = ap->a_vp;
1168 
1169 	if (vp->v_type != VLNK)
1170 		return (EPERM);
1171 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1172 }
1173 
1174 /*
1175  * Do a readlink rpc.
1176  * Called by nfs_doio() from below the buffer cache.
1177  */
1178 int
1179 nfs_readlinkrpc(vp, uiop, cred)
1180 	struct vnode *vp;
1181 	struct uio *uiop;
1182 	struct ucred *cred;
1183 {
1184 	u_int32_t *tl;
1185 	caddr_t cp;
1186 	int32_t t1, t2;
1187 	caddr_t bpos, dpos, cp2;
1188 	int error = 0;
1189 	uint32_t len;
1190 	struct mbuf *mreq, *mrep, *md, *mb;
1191 	const int v3 = NFS_ISV3(vp);
1192 	struct nfsnode *np = VTONFS(vp);
1193 #ifndef NFS_V2_ONLY
1194 	int attrflag;
1195 #endif
1196 
1197 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1198 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1199 	nfsm_fhtom(np, v3);
1200 	nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1201 #ifndef NFS_V2_ONLY
1202 	if (v3)
1203 		nfsm_postop_attr(vp, attrflag, 0);
1204 #endif
1205 	if (!error) {
1206 #ifndef NFS_V2_ONLY
1207 		if (v3) {
1208 			nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1209 			len = fxdr_unsigned(uint32_t, *tl);
1210 			if (len > MAXPATHLEN) {
1211 				/*
1212 				 * this pathname is too long for us.
1213 				 */
1214 				m_freem(mrep);
1215 				/* Solaris returns EINVAL. should we follow? */
1216 				error = ENAMETOOLONG;
1217 				goto nfsmout;
1218 			}
1219 		} else
1220 #endif
1221 		{
1222 			nfsm_strsiz(len, NFS_MAXPATHLEN);
1223 		}
1224 		nfsm_mtouio(uiop, len);
1225 	}
1226 	nfsm_reqdone;
1227 	return (error);
1228 }
1229 
1230 /*
1231  * nfs read rpc call
1232  * Ditto above
1233  */
1234 int
1235 nfs_readrpc(vp, uiop)
1236 	struct vnode *vp;
1237 	struct uio *uiop;
1238 {
1239 	u_int32_t *tl;
1240 	caddr_t cp;
1241 	int32_t t1, t2;
1242 	caddr_t bpos, dpos, cp2;
1243 	struct mbuf *mreq, *mrep, *md, *mb;
1244 	struct nfsmount *nmp;
1245 	int error = 0, len, retlen, tsiz, eof;
1246 	const int v3 = NFS_ISV3(vp);
1247 	struct nfsnode *np = VTONFS(vp);
1248 #ifndef NFS_V2_ONLY
1249 	int attrflag;
1250 #endif
1251 
1252 #ifndef nolint
1253 	eof = 0;
1254 #endif
1255 	nmp = VFSTONFS(vp->v_mount);
1256 	tsiz = uiop->uio_resid;
1257 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1258 		return (EFBIG);
1259 	while (tsiz > 0) {
1260 		nfsstats.rpccnt[NFSPROC_READ]++;
1261 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1262 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1263 		nfsm_fhtom(np, v3);
1264 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1265 #ifndef NFS_V2_ONLY
1266 		if (v3) {
1267 			txdr_hyper(uiop->uio_offset, tl);
1268 			*(tl + 2) = txdr_unsigned(len);
1269 		} else
1270 #endif
1271 		{
1272 			*tl++ = txdr_unsigned(uiop->uio_offset);
1273 			*tl++ = txdr_unsigned(len);
1274 			*tl = 0;
1275 		}
1276 		nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred);
1277 #ifndef NFS_V2_ONLY
1278 		if (v3) {
1279 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1280 			if (error) {
1281 				m_freem(mrep);
1282 				goto nfsmout;
1283 			}
1284 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1285 			eof = fxdr_unsigned(int, *(tl + 1));
1286 		} else
1287 #endif
1288 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1289 		nfsm_strsiz(retlen, nmp->nm_rsize);
1290 		nfsm_mtouio(uiop, retlen);
1291 		m_freem(mrep);
1292 		tsiz -= retlen;
1293 #ifndef NFS_V2_ONLY
1294 		if (v3) {
1295 			if (eof || retlen == 0)
1296 				tsiz = 0;
1297 		} else
1298 #endif
1299 		if (retlen < len)
1300 			tsiz = 0;
1301 	}
1302 nfsmout:
1303 	return (error);
1304 }
1305 
1306 struct nfs_writerpc_context {
1307 	struct simplelock nwc_slock;
1308 	volatile int nwc_mbufcount;
1309 };
1310 
1311 /*
1312  * free mbuf used to refer protected pages while write rpc call.
1313  * called at splvm.
1314  */
1315 static void
1316 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
1317 {
1318 	struct nfs_writerpc_context *ctx = arg;
1319 
1320 	KASSERT(m != NULL);
1321 	KASSERT(ctx != NULL);
1322 	pool_cache_put(&mbpool_cache, m);
1323 	simple_lock(&ctx->nwc_slock);
1324 	if (--ctx->nwc_mbufcount == 0) {
1325 		wakeup(ctx);
1326 	}
1327 	simple_unlock(&ctx->nwc_slock);
1328 }
1329 
1330 /*
1331  * nfs write call
1332  */
1333 int
1334 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1335 	struct vnode *vp;
1336 	struct uio *uiop;
1337 	int *iomode;
1338 	boolean_t pageprotected;
1339 	boolean_t *stalewriteverfp;
1340 {
1341 	u_int32_t *tl;
1342 	caddr_t cp;
1343 	int32_t t1, t2;
1344 	caddr_t bpos, dpos;
1345 	struct mbuf *mreq, *mrep, *md, *mb;
1346 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1347 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1348 	const int v3 = NFS_ISV3(vp);
1349 	int committed = NFSV3WRITE_FILESYNC;
1350 	struct nfsnode *np = VTONFS(vp);
1351 	struct nfs_writerpc_context ctx;
1352 	int s;
1353 	struct lwp *l = NULL;
1354 	size_t origresid;
1355 #ifndef NFS_V2_ONLY
1356 	caddr_t cp2;
1357 	int rlen, commit;
1358 #endif
1359 
1360 	simple_lock_init(&ctx.nwc_slock);
1361 	ctx.nwc_mbufcount = 1;
1362 
1363 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1364 		panic("writerpc readonly vp %p", vp);
1365 	}
1366 
1367 #ifdef DIAGNOSTIC
1368 	if (uiop->uio_iovcnt != 1)
1369 		panic("nfs: writerpc iovcnt > 1");
1370 #endif
1371 	tsiz = uiop->uio_resid;
1372 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1373 		return (EFBIG);
1374 	if (pageprotected) {
1375 		l = curlwp;
1376 		PHOLD(l);
1377 	}
1378 retry:
1379 	origresid = uiop->uio_resid;
1380 	KASSERT(origresid == uiop->uio_iov->iov_len);
1381 	while (tsiz > 0) {
1382 		uint32_t datalen; /* data bytes need to be allocated in mbuf */
1383 		uint32_t backup;
1384 		boolean_t stalewriteverf = FALSE;
1385 
1386 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1387 		len = min(tsiz, nmp->nm_wsize);
1388 		datalen = pageprotected ? 0 : nfsm_rndup(len);
1389 		nfsm_reqhead(np, NFSPROC_WRITE,
1390 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1391 		nfsm_fhtom(np, v3);
1392 #ifndef NFS_V2_ONLY
1393 		if (v3) {
1394 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1395 			txdr_hyper(uiop->uio_offset, tl);
1396 			tl += 2;
1397 			*tl++ = txdr_unsigned(len);
1398 			*tl++ = txdr_unsigned(*iomode);
1399 			*tl = txdr_unsigned(len);
1400 		} else
1401 #endif
1402 		{
1403 			u_int32_t x;
1404 
1405 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1406 			/* Set both "begin" and "current" to non-garbage. */
1407 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1408 			*tl++ = x;      /* "begin offset" */
1409 			*tl++ = x;      /* "current offset" */
1410 			x = txdr_unsigned(len);
1411 			*tl++ = x;      /* total to this offset */
1412 			*tl = x;        /* size of this write */
1413 
1414 		}
1415 		if (pageprotected) {
1416 			/*
1417 			 * since we know pages can't be modified during i/o,
1418 			 * no need to copy them for us.
1419 			 */
1420 			struct mbuf *m;
1421 			struct iovec *iovp = uiop->uio_iov;
1422 
1423 			m = m_get(M_WAIT, MT_DATA);
1424 			MCLAIM(m, &nfs_mowner);
1425 			MEXTADD(m, iovp->iov_base, len, M_MBUF,
1426 			    nfs_writerpc_extfree, &ctx);
1427 			m->m_flags |= M_EXT_ROMAP;
1428 			m->m_len = len;
1429 			mb->m_next = m;
1430 			/*
1431 			 * no need to maintain mb and bpos here
1432 			 * because no one care them later.
1433 			 */
1434 #if 0
1435 			mb = m;
1436 			bpos = mtod(caddr_t, mb) + mb->m_len;
1437 #endif
1438 			iovp->iov_base = (char *)iovp->iov_base + len;
1439 			iovp->iov_len -= len;
1440 			uiop->uio_offset += len;
1441 			uiop->uio_resid -= len;
1442 			s = splvm();
1443 			simple_lock(&ctx.nwc_slock);
1444 			ctx.nwc_mbufcount++;
1445 			simple_unlock(&ctx.nwc_slock);
1446 			splx(s);
1447 			nfs_zeropad(mb, 0, nfsm_padlen(len));
1448 		} else {
1449 			nfsm_uiotom(uiop, len);
1450 		}
1451 		nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred);
1452 #ifndef NFS_V2_ONLY
1453 		if (v3) {
1454 			wccflag = NFSV3_WCCCHK;
1455 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1456 			if (!error) {
1457 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1458 					+ NFSX_V3WRITEVERF);
1459 				rlen = fxdr_unsigned(int, *tl++);
1460 				if (rlen == 0) {
1461 					error = NFSERR_IO;
1462 					m_freem(mrep);
1463 					break;
1464 				} else if (rlen < len) {
1465 					backup = len - rlen;
1466 					uiop->uio_iov->iov_base =
1467 					    (caddr_t)uiop->uio_iov->iov_base -
1468 					    backup;
1469 					uiop->uio_iov->iov_len += backup;
1470 					uiop->uio_offset -= backup;
1471 					uiop->uio_resid += backup;
1472 					len = rlen;
1473 				}
1474 				commit = fxdr_unsigned(int, *tl++);
1475 
1476 				/*
1477 				 * Return the lowest committment level
1478 				 * obtained by any of the RPCs.
1479 				 */
1480 				if (committed == NFSV3WRITE_FILESYNC)
1481 					committed = commit;
1482 				else if (committed == NFSV3WRITE_DATASYNC &&
1483 					commit == NFSV3WRITE_UNSTABLE)
1484 					committed = commit;
1485 				simple_lock(&nmp->nm_slock);
1486 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1487 					memcpy(nmp->nm_writeverf, tl,
1488 					    NFSX_V3WRITEVERF);
1489 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1490 				} else if ((nmp->nm_iflag &
1491 				    NFSMNT_STALEWRITEVERF) ||
1492 				    memcmp(tl, nmp->nm_writeverf,
1493 				    NFSX_V3WRITEVERF)) {
1494 					memcpy(nmp->nm_writeverf, tl,
1495 					    NFSX_V3WRITEVERF);
1496 					/*
1497 					 * note NFSMNT_STALEWRITEVERF
1498 					 * if we're the first thread to
1499 					 * notice it.
1500 					 */
1501 					if ((nmp->nm_iflag &
1502 					    NFSMNT_STALEWRITEVERF) == 0) {
1503 						stalewriteverf = TRUE;
1504 						nmp->nm_iflag |=
1505 						    NFSMNT_STALEWRITEVERF;
1506 					}
1507 				}
1508 				simple_unlock(&nmp->nm_slock);
1509 			}
1510 		} else
1511 #endif
1512 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1513 		if (wccflag)
1514 			VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1515 		m_freem(mrep);
1516 		if (error)
1517 			break;
1518 		tsiz -= len;
1519 		if (stalewriteverf) {
1520 			*stalewriteverfp = TRUE;
1521 			stalewriteverf = FALSE;
1522 			if (committed == NFSV3WRITE_UNSTABLE &&
1523 			    len != origresid) {
1524 				/*
1525 				 * if our write requests weren't atomic but
1526 				 * unstable, datas in previous iterations
1527 				 * might have already been lost now.
1528 				 * then, we should resend them to nfsd.
1529 				 */
1530 				backup = origresid - tsiz;
1531 				uiop->uio_iov->iov_base =
1532 				    (caddr_t)uiop->uio_iov->iov_base - backup;
1533 				uiop->uio_iov->iov_len += backup;
1534 				uiop->uio_offset -= backup;
1535 				uiop->uio_resid += backup;
1536 				tsiz = origresid;
1537 				goto retry;
1538 			}
1539 		}
1540 	}
1541 nfsmout:
1542 	if (pageprotected) {
1543 		/*
1544 		 * wait until mbufs go away.
1545 		 * retransmitted mbufs can survive longer than rpc requests
1546 		 * themselves.
1547 		 */
1548 		s = splvm();
1549 		simple_lock(&ctx.nwc_slock);
1550 		ctx.nwc_mbufcount--;
1551 		while (ctx.nwc_mbufcount > 0) {
1552 			ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1553 		}
1554 		simple_unlock(&ctx.nwc_slock);
1555 		splx(s);
1556 		PRELE(l);
1557 	}
1558 	*iomode = committed;
1559 	if (error)
1560 		uiop->uio_resid = tsiz;
1561 	return (error);
1562 }
1563 
1564 /*
1565  * nfs mknod rpc
1566  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1567  * mode set to specify the file type and the size field for rdev.
1568  */
1569 int
1570 nfs_mknodrpc(dvp, vpp, cnp, vap)
1571 	struct vnode *dvp;
1572 	struct vnode **vpp;
1573 	struct componentname *cnp;
1574 	struct vattr *vap;
1575 {
1576 	struct nfsv2_sattr *sp;
1577 	u_int32_t *tl;
1578 	caddr_t cp;
1579 	int32_t t1, t2;
1580 	struct vnode *newvp = (struct vnode *)0;
1581 	struct nfsnode *dnp, *np;
1582 	char *cp2;
1583 	caddr_t bpos, dpos;
1584 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1585 	struct mbuf *mreq, *mrep, *md, *mb;
1586 	u_int32_t rdev;
1587 	const int v3 = NFS_ISV3(dvp);
1588 
1589 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1590 		rdev = txdr_unsigned(vap->va_rdev);
1591 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1592 		rdev = nfs_xdrneg1;
1593 	else {
1594 		VOP_ABORTOP(dvp, cnp);
1595 		vput(dvp);
1596 		return (EOPNOTSUPP);
1597 	}
1598 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1599 	dnp = VTONFS(dvp);
1600 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1601 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1602 	nfsm_fhtom(dnp, v3);
1603 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1604 #ifndef NFS_V2_ONLY
1605 	if (v3) {
1606 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1607 		*tl++ = vtonfsv3_type(vap->va_type);
1608 		nfsm_v3attrbuild(vap, FALSE);
1609 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1610 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1611 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1612 			*tl = txdr_unsigned(minor(vap->va_rdev));
1613 		}
1614 	} else
1615 #endif
1616 	{
1617 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1618 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1619 		sp->sa_uid = nfs_xdrneg1;
1620 		sp->sa_gid = nfs_xdrneg1;
1621 		sp->sa_size = rdev;
1622 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1623 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1624 	}
1625 	nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1626 	if (!error) {
1627 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1628 		if (!gotvp) {
1629 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1630 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1631 			if (!error)
1632 				newvp = NFSTOV(np);
1633 		}
1634 	}
1635 #ifndef NFS_V2_ONLY
1636 	if (v3)
1637 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1638 #endif
1639 	nfsm_reqdone;
1640 	if (error) {
1641 		if (newvp)
1642 			vput(newvp);
1643 	} else {
1644 		if (cnp->cn_flags & MAKEENTRY)
1645 			nfs_cache_enter(dvp, newvp, cnp);
1646 		*vpp = newvp;
1647 	}
1648 	PNBUF_PUT(cnp->cn_pnbuf);
1649 	VTONFS(dvp)->n_flag |= NMODIFIED;
1650 	if (!wccflag)
1651 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1652 	vput(dvp);
1653 	return (error);
1654 }
1655 
1656 /*
1657  * nfs mknod vop
1658  * just call nfs_mknodrpc() to do the work.
1659  */
1660 /* ARGSUSED */
1661 int
1662 nfs_mknod(v)
1663 	void *v;
1664 {
1665 	struct vop_mknod_args /* {
1666 		struct vnode *a_dvp;
1667 		struct vnode **a_vpp;
1668 		struct componentname *a_cnp;
1669 		struct vattr *a_vap;
1670 	} */ *ap = v;
1671 	struct vnode *dvp = ap->a_dvp;
1672 	struct componentname *cnp = ap->a_cnp;
1673 	int error;
1674 
1675 	error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1676 	VN_KNOTE(dvp, NOTE_WRITE);
1677 	if (error == 0 || error == EEXIST)
1678 		cache_purge1(dvp, cnp, 0);
1679 	return (error);
1680 }
1681 
1682 #ifndef NFS_V2_ONLY
1683 static u_long create_verf;
1684 #endif
1685 /*
1686  * nfs file create call
1687  */
1688 int
1689 nfs_create(v)
1690 	void *v;
1691 {
1692 	struct vop_create_args /* {
1693 		struct vnode *a_dvp;
1694 		struct vnode **a_vpp;
1695 		struct componentname *a_cnp;
1696 		struct vattr *a_vap;
1697 	} */ *ap = v;
1698 	struct vnode *dvp = ap->a_dvp;
1699 	struct vattr *vap = ap->a_vap;
1700 	struct componentname *cnp = ap->a_cnp;
1701 	struct nfsv2_sattr *sp;
1702 	u_int32_t *tl;
1703 	caddr_t cp;
1704 	int32_t t1, t2;
1705 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1706 	struct vnode *newvp = (struct vnode *)0;
1707 	caddr_t bpos, dpos, cp2;
1708 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1709 	struct mbuf *mreq, *mrep, *md, *mb;
1710 	const int v3 = NFS_ISV3(dvp);
1711 
1712 	/*
1713 	 * Oops, not for me..
1714 	 */
1715 	if (vap->va_type == VSOCK)
1716 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1717 
1718 	KASSERT(vap->va_type == VREG);
1719 
1720 #ifdef VA_EXCLUSIVE
1721 	if (vap->va_vaflags & VA_EXCLUSIVE)
1722 		fmode |= O_EXCL;
1723 #endif
1724 again:
1725 	error = 0;
1726 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1727 	dnp = VTONFS(dvp);
1728 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1729 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1730 	nfsm_fhtom(dnp, v3);
1731 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1732 #ifndef NFS_V2_ONLY
1733 	if (v3) {
1734 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1735 		if (fmode & O_EXCL) {
1736 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1737 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1738 #ifdef INET
1739 			if (TAILQ_FIRST(&in_ifaddrhead))
1740 				*tl++ = TAILQ_FIRST(&in_ifaddrhead)->
1741 				    ia_addr.sin_addr.s_addr;
1742 			else
1743 				*tl++ = create_verf;
1744 #else
1745 			*tl++ = create_verf;
1746 #endif
1747 			*tl = ++create_verf;
1748 		} else {
1749 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1750 			nfsm_v3attrbuild(vap, FALSE);
1751 		}
1752 	} else
1753 #endif
1754 	{
1755 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1756 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1757 		sp->sa_uid = nfs_xdrneg1;
1758 		sp->sa_gid = nfs_xdrneg1;
1759 		sp->sa_size = 0;
1760 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1761 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1762 	}
1763 	nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1764 	if (!error) {
1765 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1766 		if (!gotvp) {
1767 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1768 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1769 			if (!error)
1770 				newvp = NFSTOV(np);
1771 		}
1772 	}
1773 #ifndef NFS_V2_ONLY
1774 	if (v3)
1775 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1776 #endif
1777 	nfsm_reqdone;
1778 	if (error) {
1779 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1780 			fmode &= ~O_EXCL;
1781 			goto again;
1782 		}
1783 	} else if (v3 && (fmode & O_EXCL)) {
1784 		struct timeval tm = time;
1785 
1786 		/*
1787 		 * make sure that we'll update timestamps as
1788 		 * most server implementations use them to store
1789 		 * the create verifier.
1790 		 *
1791 		 * XXX it's better to use TOSERVER always.
1792 		 */
1793 
1794 		if (vap->va_atime.tv_sec == VNOVAL) {
1795 			vap->va_atime.tv_sec = tm.tv_sec;
1796 			vap->va_atime.tv_nsec = tm.tv_usec * 1000;
1797 		}
1798 		if (vap->va_mtime.tv_sec == VNOVAL) {
1799 			vap->va_mtime.tv_sec = tm.tv_sec;
1800 			vap->va_mtime.tv_nsec = tm.tv_usec * 1000;
1801 		}
1802 
1803 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1804 	}
1805 	if (error == 0) {
1806 		if (cnp->cn_flags & MAKEENTRY)
1807 			nfs_cache_enter(dvp, newvp, cnp);
1808 		else
1809 			cache_purge1(dvp, cnp, 0);
1810 		*ap->a_vpp = newvp;
1811 	} else {
1812 		if (newvp)
1813 			vput(newvp);
1814 		if (error == EEXIST)
1815 			cache_purge1(dvp, cnp, 0);
1816 	}
1817 	PNBUF_PUT(cnp->cn_pnbuf);
1818 	VTONFS(dvp)->n_flag |= NMODIFIED;
1819 	if (!wccflag)
1820 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1821 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1822 	vput(dvp);
1823 	return (error);
1824 }
1825 
1826 /*
1827  * nfs file remove call
1828  * To try and make nfs semantics closer to ufs semantics, a file that has
1829  * other processes using the vnode is renamed instead of removed and then
1830  * removed later on the last close.
1831  * - If v_usecount > 1
1832  *	  If a rename is not already in the works
1833  *	     call nfs_sillyrename() to set it up
1834  *     else
1835  *	  do the remove rpc
1836  */
1837 int
1838 nfs_remove(v)
1839 	void *v;
1840 {
1841 	struct vop_remove_args /* {
1842 		struct vnodeop_desc *a_desc;
1843 		struct vnode * a_dvp;
1844 		struct vnode * a_vp;
1845 		struct componentname * a_cnp;
1846 	} */ *ap = v;
1847 	struct vnode *vp = ap->a_vp;
1848 	struct vnode *dvp = ap->a_dvp;
1849 	struct componentname *cnp = ap->a_cnp;
1850 	struct nfsnode *np = VTONFS(vp);
1851 	int error = 0;
1852 	struct vattr vattr;
1853 
1854 #ifndef DIAGNOSTIC
1855 	if ((cnp->cn_flags & HASBUF) == 0)
1856 		panic("nfs_remove: no name");
1857 	if (vp->v_usecount < 1)
1858 		panic("nfs_remove: bad v_usecount");
1859 #endif
1860 	if (vp->v_type == VDIR)
1861 		error = EPERM;
1862 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1863 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1864 	    vattr.va_nlink > 1)) {
1865 		/*
1866 		 * Purge the name cache so that the chance of a lookup for
1867 		 * the name succeeding while the remove is in progress is
1868 		 * minimized. Without node locking it can still happen, such
1869 		 * that an I/O op returns ESTALE, but since you get this if
1870 		 * another host removes the file..
1871 		 */
1872 		cache_purge(vp);
1873 		/*
1874 		 * throw away biocache buffers, mainly to avoid
1875 		 * unnecessary delayed writes later.
1876 		 */
1877 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1878 		/* Do the rpc */
1879 		if (error != EINTR)
1880 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1881 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1882 	} else if (!np->n_sillyrename)
1883 		error = nfs_sillyrename(dvp, vp, cnp);
1884 	PNBUF_PUT(cnp->cn_pnbuf);
1885 	if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1886 	    vattr.va_nlink == 1) {
1887 		np->n_flag |= NREMOVED;
1888 	}
1889 	NFS_INVALIDATE_ATTRCACHE(np);
1890 	VN_KNOTE(vp, NOTE_DELETE);
1891 	VN_KNOTE(dvp, NOTE_WRITE);
1892 	if (dvp == vp)
1893 		vrele(vp);
1894 	else
1895 		vput(vp);
1896 	vput(dvp);
1897 	return (error);
1898 }
1899 
1900 /*
1901  * nfs file remove rpc called from nfs_inactive
1902  */
1903 int
1904 nfs_removeit(sp)
1905 	struct sillyrename *sp;
1906 {
1907 
1908 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1909 		(struct proc *)0));
1910 }
1911 
1912 /*
1913  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1914  */
1915 int
1916 nfs_removerpc(dvp, name, namelen, cred, proc)
1917 	struct vnode *dvp;
1918 	const char *name;
1919 	int namelen;
1920 	struct ucred *cred;
1921 	struct proc *proc;
1922 {
1923 	u_int32_t *tl;
1924 	caddr_t cp;
1925 #ifndef NFS_V2_ONLY
1926 	int32_t t1;
1927 	caddr_t cp2;
1928 #endif
1929 	int32_t t2;
1930 	caddr_t bpos, dpos;
1931 	int error = 0, wccflag = NFSV3_WCCRATTR;
1932 	struct mbuf *mreq, *mrep, *md, *mb;
1933 	const int v3 = NFS_ISV3(dvp);
1934 	int rexmit;
1935 	struct nfsnode *dnp = VTONFS(dvp);
1936 
1937 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1938 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1939 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1940 	nfsm_fhtom(dnp, v3);
1941 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1942 	nfsm_request1(dnp, NFSPROC_REMOVE, proc, cred, &rexmit);
1943 #ifndef NFS_V2_ONLY
1944 	if (v3)
1945 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1946 #endif
1947 	nfsm_reqdone;
1948 	VTONFS(dvp)->n_flag |= NMODIFIED;
1949 	if (!wccflag)
1950 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1951 	/*
1952 	 * Kludge City: If the first reply to the remove rpc is lost..
1953 	 *   the reply to the retransmitted request will be ENOENT
1954 	 *   since the file was in fact removed
1955 	 *   Therefore, we cheat and return success.
1956 	 */
1957 	if (rexmit && error == ENOENT)
1958 		error = 0;
1959 	return (error);
1960 }
1961 
1962 /*
1963  * nfs file rename call
1964  */
1965 int
1966 nfs_rename(v)
1967 	void *v;
1968 {
1969 	struct vop_rename_args  /* {
1970 		struct vnode *a_fdvp;
1971 		struct vnode *a_fvp;
1972 		struct componentname *a_fcnp;
1973 		struct vnode *a_tdvp;
1974 		struct vnode *a_tvp;
1975 		struct componentname *a_tcnp;
1976 	} */ *ap = v;
1977 	struct vnode *fvp = ap->a_fvp;
1978 	struct vnode *tvp = ap->a_tvp;
1979 	struct vnode *fdvp = ap->a_fdvp;
1980 	struct vnode *tdvp = ap->a_tdvp;
1981 	struct componentname *tcnp = ap->a_tcnp;
1982 	struct componentname *fcnp = ap->a_fcnp;
1983 	int error;
1984 
1985 #ifndef DIAGNOSTIC
1986 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1987 	    (fcnp->cn_flags & HASBUF) == 0)
1988 		panic("nfs_rename: no name");
1989 #endif
1990 	/* Check for cross-device rename */
1991 	if ((fvp->v_mount != tdvp->v_mount) ||
1992 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1993 		error = EXDEV;
1994 		goto out;
1995 	}
1996 
1997 	/*
1998 	 * If the tvp exists and is in use, sillyrename it before doing the
1999 	 * rename of the new file over it.
2000 	 */
2001 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
2002 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2003 		VN_KNOTE(tvp, NOTE_DELETE);
2004 		vput(tvp);
2005 		tvp = NULL;
2006 	}
2007 
2008 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2009 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2010 		tcnp->cn_proc);
2011 
2012 	VN_KNOTE(fdvp, NOTE_WRITE);
2013 	VN_KNOTE(tdvp, NOTE_WRITE);
2014 	if (error == 0 || error == EEXIST) {
2015 		if (fvp->v_type == VDIR)
2016 			cache_purge(fvp);
2017 		else
2018 			cache_purge1(fdvp, fcnp, 0);
2019 		if (tvp != NULL && tvp->v_type == VDIR)
2020 			cache_purge(tvp);
2021 		else
2022 			cache_purge1(tdvp, tcnp, 0);
2023 	}
2024 out:
2025 	if (tdvp == tvp)
2026 		vrele(tdvp);
2027 	else
2028 		vput(tdvp);
2029 	if (tvp)
2030 		vput(tvp);
2031 	vrele(fdvp);
2032 	vrele(fvp);
2033 	return (error);
2034 }
2035 
2036 /*
2037  * nfs file rename rpc called from nfs_remove() above
2038  */
2039 int
2040 nfs_renameit(sdvp, scnp, sp)
2041 	struct vnode *sdvp;
2042 	struct componentname *scnp;
2043 	struct sillyrename *sp;
2044 {
2045 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2046 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
2047 }
2048 
2049 /*
2050  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2051  */
2052 int
2053 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
2054 	struct vnode *fdvp;
2055 	const char *fnameptr;
2056 	int fnamelen;
2057 	struct vnode *tdvp;
2058 	const char *tnameptr;
2059 	int tnamelen;
2060 	struct ucred *cred;
2061 	struct proc *proc;
2062 {
2063 	u_int32_t *tl;
2064 	caddr_t cp;
2065 #ifndef NFS_V2_ONLY
2066 	int32_t t1;
2067 	caddr_t cp2;
2068 #endif
2069 	int32_t t2;
2070 	caddr_t bpos, dpos;
2071 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2072 	struct mbuf *mreq, *mrep, *md, *mb;
2073 	const int v3 = NFS_ISV3(fdvp);
2074 	int rexmit;
2075 	struct nfsnode *fdnp = VTONFS(fdvp);
2076 
2077 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2078 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
2079 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2080 		nfsm_rndup(tnamelen));
2081 	nfsm_fhtom(fdnp, v3);
2082 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2083 	nfsm_fhtom(VTONFS(tdvp), v3);
2084 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2085 	nfsm_request1(fdnp, NFSPROC_RENAME, proc, cred, &rexmit);
2086 #ifndef NFS_V2_ONLY
2087 	if (v3) {
2088 		nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2089 		nfsm_wcc_data(tdvp, twccflag, 0, !error);
2090 	}
2091 #endif
2092 	nfsm_reqdone;
2093 	VTONFS(fdvp)->n_flag |= NMODIFIED;
2094 	VTONFS(tdvp)->n_flag |= NMODIFIED;
2095 	if (!fwccflag)
2096 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2097 	if (!twccflag)
2098 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2099 	/*
2100 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2101 	 */
2102 	if (rexmit && error == ENOENT)
2103 		error = 0;
2104 	return (error);
2105 }
2106 
2107 /*
2108  * nfs hard link create call
2109  */
2110 int
2111 nfs_link(v)
2112 	void *v;
2113 {
2114 	struct vop_link_args /* {
2115 		struct vnode *a_dvp;
2116 		struct vnode *a_vp;
2117 		struct componentname *a_cnp;
2118 	} */ *ap = v;
2119 	struct vnode *vp = ap->a_vp;
2120 	struct vnode *dvp = ap->a_dvp;
2121 	struct componentname *cnp = ap->a_cnp;
2122 	u_int32_t *tl;
2123 	caddr_t cp;
2124 #ifndef NFS_V2_ONLY
2125 	int32_t t1;
2126 	caddr_t cp2;
2127 #endif
2128 	int32_t t2;
2129 	caddr_t bpos, dpos;
2130 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2131 	struct mbuf *mreq, *mrep, *md, *mb;
2132 	/* XXX Should be const and initialised? */
2133 	int v3;
2134 	int rexmit;
2135 	struct nfsnode *np;
2136 
2137 	if (dvp->v_mount != vp->v_mount) {
2138 		VOP_ABORTOP(dvp, cnp);
2139 		vput(dvp);
2140 		return (EXDEV);
2141 	}
2142 	if (dvp != vp) {
2143 		error = vn_lock(vp, LK_EXCLUSIVE);
2144 		if (error != 0) {
2145 			VOP_ABORTOP(dvp, cnp);
2146 			vput(dvp);
2147 			return error;
2148 		}
2149 	}
2150 
2151 	/*
2152 	 * Push all writes to the server, so that the attribute cache
2153 	 * doesn't get "out of sync" with the server.
2154 	 * XXX There should be a better way!
2155 	 */
2156 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
2157 
2158 	v3 = NFS_ISV3(vp);
2159 	nfsstats.rpccnt[NFSPROC_LINK]++;
2160 	np = VTONFS(vp);
2161 	nfsm_reqhead(np, NFSPROC_LINK,
2162 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2163 	nfsm_fhtom(np, v3);
2164 	nfsm_fhtom(VTONFS(dvp), v3);
2165 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2166 	nfsm_request1(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred, &rexmit);
2167 #ifndef NFS_V2_ONLY
2168 	if (v3) {
2169 		nfsm_postop_attr(vp, attrflag, 0);
2170 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2171 	}
2172 #endif
2173 	nfsm_reqdone;
2174 	if (error == 0 || error == EEXIST)
2175 		cache_purge1(dvp, cnp, 0);
2176 	PNBUF_PUT(cnp->cn_pnbuf);
2177 	VTONFS(dvp)->n_flag |= NMODIFIED;
2178 	if (!attrflag)
2179 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2180 	if (!wccflag)
2181 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2182 	if (dvp != vp)
2183 		VOP_UNLOCK(vp, 0);
2184 	VN_KNOTE(vp, NOTE_LINK);
2185 	VN_KNOTE(dvp, NOTE_WRITE);
2186 	vput(dvp);
2187 	/*
2188 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2189 	 */
2190 	if (rexmit && error == EEXIST)
2191 		error = 0;
2192 	return (error);
2193 }
2194 
2195 /*
2196  * nfs symbolic link create call
2197  */
2198 int
2199 nfs_symlink(v)
2200 	void *v;
2201 {
2202 	struct vop_symlink_args /* {
2203 		struct vnode *a_dvp;
2204 		struct vnode **a_vpp;
2205 		struct componentname *a_cnp;
2206 		struct vattr *a_vap;
2207 		char *a_target;
2208 	} */ *ap = v;
2209 	struct vnode *dvp = ap->a_dvp;
2210 	struct vattr *vap = ap->a_vap;
2211 	struct componentname *cnp = ap->a_cnp;
2212 	struct nfsv2_sattr *sp;
2213 	u_int32_t *tl;
2214 	caddr_t cp;
2215 	int32_t t1, t2;
2216 	caddr_t bpos, dpos, cp2;
2217 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2218 	struct mbuf *mreq, *mrep, *md, *mb;
2219 	struct vnode *newvp = (struct vnode *)0;
2220 	const int v3 = NFS_ISV3(dvp);
2221 	int rexmit;
2222 	struct nfsnode *dnp = VTONFS(dvp);
2223 
2224 	*ap->a_vpp = NULL;
2225 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2226 	slen = strlen(ap->a_target);
2227 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2228 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2229 	nfsm_fhtom(dnp, v3);
2230 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2231 #ifndef NFS_V2_ONlY
2232 	if (v3)
2233 		nfsm_v3attrbuild(vap, FALSE);
2234 #endif
2235 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2236 #ifndef NFS_V2_ONlY
2237 	if (!v3) {
2238 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2239 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2240 		sp->sa_uid = nfs_xdrneg1;
2241 		sp->sa_gid = nfs_xdrneg1;
2242 		sp->sa_size = nfs_xdrneg1;
2243 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2244 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2245 	}
2246 #endif
2247 	nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred,
2248 	    &rexmit);
2249 #ifndef NFS_V2_ONlY
2250 	if (v3) {
2251 		if (!error)
2252 			nfsm_mtofh(dvp, newvp, v3, gotvp);
2253 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2254 	}
2255 #endif
2256 	nfsm_reqdone;
2257 	/*
2258 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2259 	 */
2260 	if (rexmit && error == EEXIST)
2261 		error = 0;
2262 	if (error == 0 || error == EEXIST)
2263 		cache_purge1(dvp, cnp, 0);
2264 	if (error == 0 && newvp == NULL) {
2265 		struct nfsnode *np = NULL;
2266 
2267 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2268 		    cnp->cn_cred, cnp->cn_proc, &np);
2269 		if (error == 0)
2270 			newvp = NFSTOV(np);
2271 	}
2272 	if (error) {
2273 		if (newvp != NULL)
2274 			vput(newvp);
2275 	} else {
2276 		*ap->a_vpp = newvp;
2277 	}
2278 	PNBUF_PUT(cnp->cn_pnbuf);
2279 	VTONFS(dvp)->n_flag |= NMODIFIED;
2280 	if (!wccflag)
2281 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2282 	VN_KNOTE(dvp, NOTE_WRITE);
2283 	vput(dvp);
2284 	return (error);
2285 }
2286 
2287 /*
2288  * nfs make dir call
2289  */
2290 int
2291 nfs_mkdir(v)
2292 	void *v;
2293 {
2294 	struct vop_mkdir_args /* {
2295 		struct vnode *a_dvp;
2296 		struct vnode **a_vpp;
2297 		struct componentname *a_cnp;
2298 		struct vattr *a_vap;
2299 	} */ *ap = v;
2300 	struct vnode *dvp = ap->a_dvp;
2301 	struct vattr *vap = ap->a_vap;
2302 	struct componentname *cnp = ap->a_cnp;
2303 	struct nfsv2_sattr *sp;
2304 	u_int32_t *tl;
2305 	caddr_t cp;
2306 	int32_t t1, t2;
2307 	int len;
2308 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2309 	struct vnode *newvp = (struct vnode *)0;
2310 	caddr_t bpos, dpos, cp2;
2311 	int error = 0, wccflag = NFSV3_WCCRATTR;
2312 	int gotvp = 0;
2313 	int rexmit;
2314 	struct mbuf *mreq, *mrep, *md, *mb;
2315 	const int v3 = NFS_ISV3(dvp);
2316 
2317 	len = cnp->cn_namelen;
2318 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2319 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
2320 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2321 	nfsm_fhtom(dnp, v3);
2322 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2323 #ifndef NFS_V2_ONLY
2324 	if (v3) {
2325 		nfsm_v3attrbuild(vap, FALSE);
2326 	} else
2327 #endif
2328 	{
2329 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2330 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2331 		sp->sa_uid = nfs_xdrneg1;
2332 		sp->sa_gid = nfs_xdrneg1;
2333 		sp->sa_size = nfs_xdrneg1;
2334 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2335 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2336 	}
2337 	nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred, &rexmit);
2338 	if (!error)
2339 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2340 	if (v3)
2341 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2342 	nfsm_reqdone;
2343 	VTONFS(dvp)->n_flag |= NMODIFIED;
2344 	if (!wccflag)
2345 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2346 	/*
2347 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2348 	 * if we can succeed in looking up the directory.
2349 	 */
2350 	if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2351 		if (newvp) {
2352 			vput(newvp);
2353 			newvp = (struct vnode *)0;
2354 		}
2355 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2356 			cnp->cn_proc, &np);
2357 		if (!error) {
2358 			newvp = NFSTOV(np);
2359 			if (newvp->v_type != VDIR || newvp == dvp)
2360 				error = EEXIST;
2361 		}
2362 	}
2363 	if (error) {
2364 		if (newvp) {
2365 			if (dvp != newvp)
2366 				vput(newvp);
2367 			else
2368 				vrele(newvp);
2369 		}
2370 	} else {
2371 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2372 		if (cnp->cn_flags & MAKEENTRY)
2373 			nfs_cache_enter(dvp, newvp, cnp);
2374 		*ap->a_vpp = newvp;
2375 	}
2376 	PNBUF_PUT(cnp->cn_pnbuf);
2377 	vput(dvp);
2378 	return (error);
2379 }
2380 
2381 /*
2382  * nfs remove directory call
2383  */
2384 int
2385 nfs_rmdir(v)
2386 	void *v;
2387 {
2388 	struct vop_rmdir_args /* {
2389 		struct vnode *a_dvp;
2390 		struct vnode *a_vp;
2391 		struct componentname *a_cnp;
2392 	} */ *ap = v;
2393 	struct vnode *vp = ap->a_vp;
2394 	struct vnode *dvp = ap->a_dvp;
2395 	struct componentname *cnp = ap->a_cnp;
2396 	u_int32_t *tl;
2397 	caddr_t cp;
2398 #ifndef NFS_V2_ONLY
2399 	int32_t t1;
2400 	caddr_t cp2;
2401 #endif
2402 	int32_t t2;
2403 	caddr_t bpos, dpos;
2404 	int error = 0, wccflag = NFSV3_WCCRATTR;
2405 	int rexmit;
2406 	struct mbuf *mreq, *mrep, *md, *mb;
2407 	const int v3 = NFS_ISV3(dvp);
2408 	struct nfsnode *dnp;
2409 
2410 	if (dvp == vp) {
2411 		vrele(dvp);
2412 		vput(dvp);
2413 		PNBUF_PUT(cnp->cn_pnbuf);
2414 		return (EINVAL);
2415 	}
2416 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2417 	dnp = VTONFS(dvp);
2418 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2419 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2420 	nfsm_fhtom(dnp, v3);
2421 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2422 	nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred, &rexmit);
2423 #ifndef NFS_V2_ONLY
2424 	if (v3)
2425 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2426 #endif
2427 	nfsm_reqdone;
2428 	PNBUF_PUT(cnp->cn_pnbuf);
2429 	VTONFS(dvp)->n_flag |= NMODIFIED;
2430 	if (!wccflag)
2431 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2432 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2433 	VN_KNOTE(vp, NOTE_DELETE);
2434 	cache_purge(vp);
2435 	vput(vp);
2436 	vput(dvp);
2437 	/*
2438 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2439 	 */
2440 	if (rexmit && error == ENOENT)
2441 		error = 0;
2442 	return (error);
2443 }
2444 
2445 /*
2446  * nfs readdir call
2447  */
2448 int
2449 nfs_readdir(v)
2450 	void *v;
2451 {
2452 	struct vop_readdir_args /* {
2453 		struct vnode *a_vp;
2454 		struct uio *a_uio;
2455 		struct ucred *a_cred;
2456 		int *a_eofflag;
2457 		off_t **a_cookies;
2458 		int *a_ncookies;
2459 	} */ *ap = v;
2460 	struct vnode *vp = ap->a_vp;
2461 	struct uio *uio = ap->a_uio;
2462 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2463 	char *base = uio->uio_iov->iov_base;
2464 	int tresid, error;
2465 	size_t count, lost;
2466 	struct dirent *dp;
2467 	off_t *cookies = NULL;
2468 	int ncookies = 0, nc;
2469 
2470 	if (vp->v_type != VDIR)
2471 		return (EPERM);
2472 
2473 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2474 	count = uio->uio_resid - lost;
2475 	if (count <= 0)
2476 		return (EINVAL);
2477 
2478 	/*
2479 	 * Call nfs_bioread() to do the real work.
2480 	 */
2481 	tresid = uio->uio_resid = count;
2482 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2483 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2484 
2485 	if (!error && ap->a_cookies) {
2486 		ncookies = count / 16;
2487 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2488 		*ap->a_cookies = cookies;
2489 	}
2490 
2491 	if (!error && uio->uio_resid == tresid) {
2492 		uio->uio_resid += lost;
2493 		nfsstats.direofcache_misses++;
2494 		if (ap->a_cookies)
2495 			*ap->a_ncookies = 0;
2496 		*ap->a_eofflag = 1;
2497 		return (0);
2498 	}
2499 
2500 	if (!error && ap->a_cookies) {
2501 		/*
2502 		 * Only the NFS server and emulations use cookies, and they
2503 		 * load the directory block into system space, so we can
2504 		 * just look at it directly.
2505 		 */
2506 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2507 			panic("nfs_readdir: lost in space");
2508 		for (nc = 0; ncookies-- &&
2509 		     base < (char *)uio->uio_iov->iov_base; nc++){
2510 			dp = (struct dirent *) base;
2511 			if (dp->d_reclen == 0)
2512 				break;
2513 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2514 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2515 			else
2516 				*(cookies++) = NFS_GETCOOKIE(dp);
2517 			base += dp->d_reclen;
2518 		}
2519 		uio->uio_resid +=
2520 		    ((caddr_t)uio->uio_iov->iov_base - base);
2521 		uio->uio_iov->iov_len +=
2522 		    ((caddr_t)uio->uio_iov->iov_base - base);
2523 		uio->uio_iov->iov_base = base;
2524 		*ap->a_ncookies = nc;
2525 	}
2526 
2527 	uio->uio_resid += lost;
2528 	*ap->a_eofflag = 0;
2529 	return (error);
2530 }
2531 
2532 /*
2533  * Readdir rpc call.
2534  * Called from below the buffer cache by nfs_doio().
2535  */
2536 int
2537 nfs_readdirrpc(vp, uiop, cred)
2538 	struct vnode *vp;
2539 	struct uio *uiop;
2540 	struct ucred *cred;
2541 {
2542 	int len, left;
2543 	struct dirent *dp = NULL;
2544 	u_int32_t *tl;
2545 	caddr_t cp;
2546 	int32_t t1, t2;
2547 	caddr_t bpos, dpos, cp2;
2548 	struct mbuf *mreq, *mrep, *md, *mb;
2549 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2550 	struct nfsnode *dnp = VTONFS(vp);
2551 	u_quad_t fileno;
2552 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2553 #ifndef NFS_V2_ONLY
2554 	int attrflag;
2555 #endif
2556 	int nrpcs = 0, reclen;
2557 	const int v3 = NFS_ISV3(vp);
2558 
2559 #ifdef DIAGNOSTIC
2560 	/*
2561 	 * Should be called from buffer cache, so only amount of
2562 	 * NFS_DIRBLKSIZ will be requested.
2563 	 */
2564 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2565 		panic("nfs readdirrpc bad uio");
2566 #endif
2567 
2568 	/*
2569 	 * Loop around doing readdir rpc's of size nm_readdirsize
2570 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2571 	 * The stopping criteria is EOF or buffer full.
2572 	 */
2573 	while (more_dirs && bigenough) {
2574 		/*
2575 		 * Heuristic: don't bother to do another RPC to further
2576 		 * fill up this block if there is not much room left. (< 50%
2577 		 * of the readdir RPC size). This wastes some buffer space
2578 		 * but can save up to 50% in RPC calls.
2579 		 */
2580 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2581 			bigenough = 0;
2582 			break;
2583 		}
2584 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2585 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2586 			NFSX_READDIR(v3));
2587 		nfsm_fhtom(dnp, v3);
2588 #ifndef NFS_V2_ONLY
2589 		if (v3) {
2590 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2591 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2592 				txdr_swapcookie3(uiop->uio_offset, tl);
2593 			} else {
2594 				txdr_cookie3(uiop->uio_offset, tl);
2595 			}
2596 			tl += 2;
2597 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2598 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2599 		} else
2600 #endif
2601 		{
2602 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2603 			*tl++ = txdr_unsigned(uiop->uio_offset);
2604 		}
2605 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2606 		nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2607 		nrpcs++;
2608 #ifndef NFS_V2_ONLY
2609 		if (v3) {
2610 			nfsm_postop_attr(vp, attrflag, 0);
2611 			if (!error) {
2612 				nfsm_dissect(tl, u_int32_t *,
2613 				    2 * NFSX_UNSIGNED);
2614 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2615 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2616 			} else {
2617 				m_freem(mrep);
2618 				goto nfsmout;
2619 			}
2620 		}
2621 #endif
2622 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2623 		more_dirs = fxdr_unsigned(int, *tl);
2624 
2625 		/* loop thru the dir entries, doctoring them to 4bsd form */
2626 		while (more_dirs && bigenough) {
2627 #ifndef NFS_V2_ONLY
2628 			if (v3) {
2629 				nfsm_dissect(tl, u_int32_t *,
2630 				    3 * NFSX_UNSIGNED);
2631 				fileno = fxdr_hyper(tl);
2632 				len = fxdr_unsigned(int, *(tl + 2));
2633 			} else
2634 #endif
2635 			{
2636 				nfsm_dissect(tl, u_int32_t *,
2637 				    2 * NFSX_UNSIGNED);
2638 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2639 				len = fxdr_unsigned(int, *tl);
2640 			}
2641 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2642 				error = EBADRPC;
2643 				m_freem(mrep);
2644 				goto nfsmout;
2645 			}
2646 			tlen = nfsm_rndup(len);
2647 			if (tlen == len)
2648 				tlen += 4;	/* To ensure null termination */
2649 			tlen += sizeof (off_t) + sizeof (int);
2650 			reclen = ALIGN(tlen + DIRHDSIZ);
2651 			tlen = reclen - DIRHDSIZ;
2652 			left = NFS_DIRFRAGSIZ - blksiz;
2653 			if (reclen > left) {
2654 				memset(uiop->uio_iov->iov_base, 0, left);
2655 				dp->d_reclen += left;
2656 				uiop->uio_iov->iov_base =
2657 				    (caddr_t)uiop->uio_iov->iov_base + left;
2658 				uiop->uio_iov->iov_len -= left;
2659 				uiop->uio_resid -= left;
2660 				blksiz = 0;
2661 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2662 			}
2663 			if (reclen > uiop->uio_resid)
2664 				bigenough = 0;
2665 			if (bigenough) {
2666 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2667 				dp->d_fileno = (int)fileno;
2668 				dp->d_namlen = len;
2669 				dp->d_reclen = reclen;
2670 				dp->d_type = DT_UNKNOWN;
2671 				blksiz += dp->d_reclen;
2672 				if (blksiz == NFS_DIRFRAGSIZ)
2673 					blksiz = 0;
2674 				uiop->uio_resid -= DIRHDSIZ;
2675 				uiop->uio_iov->iov_base =
2676 				    (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2677 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2678 				nfsm_mtouio(uiop, len);
2679 				tlen -= len;
2680 				/* null terminate */
2681 				KDASSERT(tlen > 0);
2682 				memset(uiop->uio_iov->iov_base, 0, tlen);
2683 				uiop->uio_iov->iov_base =
2684 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2685 				uiop->uio_iov->iov_len -= tlen;
2686 				uiop->uio_resid -= tlen;
2687 			} else
2688 				nfsm_adv(nfsm_rndup(len));
2689 #ifndef NFS_V2_ONLY
2690 			if (v3) {
2691 				nfsm_dissect(tl, u_int32_t *,
2692 				    3 * NFSX_UNSIGNED);
2693 			} else
2694 #endif
2695 			{
2696 				nfsm_dissect(tl, u_int32_t *,
2697 				    2 * NFSX_UNSIGNED);
2698 			}
2699 			if (bigenough) {
2700 #ifndef NFS_V2_ONLY
2701 				if (v3) {
2702 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2703 						uiop->uio_offset =
2704 						    fxdr_swapcookie3(tl);
2705 					else
2706 						uiop->uio_offset =
2707 						    fxdr_cookie3(tl);
2708 				}
2709 				else
2710 #endif
2711 				{
2712 					uiop->uio_offset =
2713 					    fxdr_unsigned(off_t, *tl);
2714 				}
2715 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2716 			}
2717 			if (v3)
2718 				tl += 2;
2719 			else
2720 				tl++;
2721 			more_dirs = fxdr_unsigned(int, *tl);
2722 		}
2723 		/*
2724 		 * If at end of rpc data, get the eof boolean
2725 		 */
2726 		if (!more_dirs) {
2727 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2728 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2729 
2730 			/*
2731 			 * kludge: if we got no entries, treat it as EOF.
2732 			 * some server sometimes send a reply without any
2733 			 * entries or EOF.
2734 			 * although it might mean the server has very long name,
2735 			 * we can't handle such entries anyway.
2736 			 */
2737 
2738 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2739 				more_dirs = 0;
2740 		}
2741 		m_freem(mrep);
2742 	}
2743 	/*
2744 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2745 	 * by increasing d_reclen for the last record.
2746 	 */
2747 	if (blksiz > 0) {
2748 		left = NFS_DIRFRAGSIZ - blksiz;
2749 		memset(uiop->uio_iov->iov_base, 0, left);
2750 		dp->d_reclen += left;
2751 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2752 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2753 		    left;
2754 		uiop->uio_iov->iov_len -= left;
2755 		uiop->uio_resid -= left;
2756 	}
2757 
2758 	/*
2759 	 * We are now either at the end of the directory or have filled the
2760 	 * block.
2761 	 */
2762 	if (bigenough) {
2763 		dnp->n_direofoffset = uiop->uio_offset;
2764 		dnp->n_flag |= NEOFVALID;
2765 	}
2766 nfsmout:
2767 	return (error);
2768 }
2769 
2770 #ifndef NFS_V2_ONLY
2771 /*
2772  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2773  */
2774 int
2775 nfs_readdirplusrpc(vp, uiop, cred)
2776 	struct vnode *vp;
2777 	struct uio *uiop;
2778 	struct ucred *cred;
2779 {
2780 	int len, left;
2781 	struct dirent *dp = NULL;
2782 	u_int32_t *tl;
2783 	caddr_t cp;
2784 	int32_t t1, t2;
2785 	struct vnode *newvp;
2786 	caddr_t bpos, dpos, cp2;
2787 	struct mbuf *mreq, *mrep, *md, *mb;
2788 	struct nameidata nami, *ndp = &nami;
2789 	struct componentname *cnp = &ndp->ni_cnd;
2790 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2791 	struct nfsnode *dnp = VTONFS(vp), *np;
2792 	nfsfh_t *fhp;
2793 	u_quad_t fileno;
2794 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2795 	int attrflag, fhsize, nrpcs = 0, reclen;
2796 	struct nfs_fattr fattr, *fp;
2797 
2798 #ifdef DIAGNOSTIC
2799 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2800 		panic("nfs readdirplusrpc bad uio");
2801 #endif
2802 	ndp->ni_dvp = vp;
2803 	newvp = NULLVP;
2804 
2805 	/*
2806 	 * Loop around doing readdir rpc's of size nm_readdirsize
2807 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2808 	 * The stopping criteria is EOF or buffer full.
2809 	 */
2810 	while (more_dirs && bigenough) {
2811 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2812 			bigenough = 0;
2813 			break;
2814 		}
2815 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2816 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2817 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2818 		nfsm_fhtom(dnp, 1);
2819  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2820 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2821 			txdr_swapcookie3(uiop->uio_offset, tl);
2822 		} else {
2823 			txdr_cookie3(uiop->uio_offset, tl);
2824 		}
2825 		tl += 2;
2826 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2827 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2828 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2829 		*tl = txdr_unsigned(nmp->nm_rsize);
2830 		nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2831 		nfsm_postop_attr(vp, attrflag, 0);
2832 		if (error) {
2833 			m_freem(mrep);
2834 			goto nfsmout;
2835 		}
2836 		nrpcs++;
2837 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2838 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2839 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2840 		more_dirs = fxdr_unsigned(int, *tl);
2841 
2842 		/* loop thru the dir entries, doctoring them to 4bsd form */
2843 		while (more_dirs && bigenough) {
2844 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2845 			fileno = fxdr_hyper(tl);
2846 			len = fxdr_unsigned(int, *(tl + 2));
2847 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2848 				error = EBADRPC;
2849 				m_freem(mrep);
2850 				goto nfsmout;
2851 			}
2852 			tlen = nfsm_rndup(len);
2853 			if (tlen == len)
2854 				tlen += 4;	/* To ensure null termination*/
2855 			tlen += sizeof (off_t) + sizeof (int);
2856 			reclen = ALIGN(tlen + DIRHDSIZ);
2857 			tlen = reclen - DIRHDSIZ;
2858 			left = NFS_DIRFRAGSIZ - blksiz;
2859 			if (reclen > left) {
2860 				/*
2861 				 * DIRFRAGSIZ is aligned, no need to align
2862 				 * again here.
2863 				 */
2864 				memset(uiop->uio_iov->iov_base, 0, left);
2865 				dp->d_reclen += left;
2866 				uiop->uio_iov->iov_base =
2867 				    (caddr_t)uiop->uio_iov->iov_base + left;
2868 				uiop->uio_iov->iov_len -= left;
2869 				uiop->uio_resid -= left;
2870 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2871 				blksiz = 0;
2872 			}
2873 			if (reclen > uiop->uio_resid)
2874 				bigenough = 0;
2875 			if (bigenough) {
2876 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2877 				dp->d_fileno = (int)fileno;
2878 				dp->d_namlen = len;
2879 				dp->d_reclen = reclen;
2880 				dp->d_type = DT_UNKNOWN;
2881 				blksiz += dp->d_reclen;
2882 				if (blksiz == NFS_DIRFRAGSIZ)
2883 					blksiz = 0;
2884 				uiop->uio_resid -= DIRHDSIZ;
2885 				uiop->uio_iov->iov_base =
2886 				    (caddr_t)uiop->uio_iov->iov_base +
2887 				    DIRHDSIZ;
2888 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2889 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2890 				cnp->cn_namelen = len;
2891 				nfsm_mtouio(uiop, len);
2892 				tlen -= len;
2893 				/* null terminate */
2894 				KDASSERT(tlen > 0);
2895 				memset(uiop->uio_iov->iov_base, 0, tlen);
2896 				uiop->uio_iov->iov_base =
2897 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2898 				uiop->uio_iov->iov_len -= tlen;
2899 				uiop->uio_resid -= tlen;
2900 			} else
2901 				nfsm_adv(nfsm_rndup(len));
2902 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2903 			if (bigenough) {
2904 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2905 					uiop->uio_offset =
2906 						fxdr_swapcookie3(tl);
2907 				else
2908 					uiop->uio_offset =
2909 						fxdr_cookie3(tl);
2910 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2911 			}
2912 			tl += 2;
2913 
2914 			/*
2915 			 * Since the attributes are before the file handle
2916 			 * (sigh), we must skip over the attributes and then
2917 			 * come back and get them.
2918 			 */
2919 			attrflag = fxdr_unsigned(int, *tl);
2920 			if (attrflag) {
2921 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2922 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2923 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2924 			    doit = fxdr_unsigned(int, *tl);
2925 			    if (doit) {
2926 				nfsm_getfh(fhp, fhsize, 1);
2927 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2928 				    VREF(vp);
2929 				    newvp = vp;
2930 				    np = dnp;
2931 				} else {
2932 				    error = nfs_nget1(vp->v_mount, fhp,
2933 					fhsize, &np, LK_NOWAIT);
2934 				    if (!error)
2935 					newvp = NFSTOV(np);
2936 				}
2937 				if (!error) {
2938 				    const char *cp;
2939 
2940 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2941 				    if (bigenough) {
2942 					dp->d_type =
2943 					   IFTODT(VTTOIF(np->n_vattr->va_type));
2944 					if (cnp->cn_namelen <= NCHNAMLEN) {
2945 					    ndp->ni_vp = newvp;
2946 					    cp = cnp->cn_nameptr +
2947 						cnp->cn_namelen;
2948 					    cnp->cn_hash =
2949 					       namei_hash(cnp->cn_nameptr, &cp);
2950 					    nfs_cache_enter(ndp->ni_dvp,
2951 						ndp->ni_vp, cnp);
2952 					}
2953 				    }
2954 				}
2955 				error = 0;
2956 			   }
2957 			} else {
2958 			    /* Just skip over the file handle */
2959 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2960 			    i = fxdr_unsigned(int, *tl);
2961 			    nfsm_adv(nfsm_rndup(i));
2962 			}
2963 			if (newvp != NULLVP) {
2964 			    if (newvp == vp)
2965 				vrele(newvp);
2966 			    else
2967 				vput(newvp);
2968 			    newvp = NULLVP;
2969 			}
2970 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2971 			more_dirs = fxdr_unsigned(int, *tl);
2972 		}
2973 		/*
2974 		 * If at end of rpc data, get the eof boolean
2975 		 */
2976 		if (!more_dirs) {
2977 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2978 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2979 
2980 			/*
2981 			 * kludge: see a comment in nfs_readdirrpc.
2982 			 */
2983 
2984 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2985 				more_dirs = 0;
2986 		}
2987 		m_freem(mrep);
2988 	}
2989 	/*
2990 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2991 	 * by increasing d_reclen for the last record.
2992 	 */
2993 	if (blksiz > 0) {
2994 		left = NFS_DIRFRAGSIZ - blksiz;
2995 		memset(uiop->uio_iov->iov_base, 0, left);
2996 		dp->d_reclen += left;
2997 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2998 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2999 		    left;
3000 		uiop->uio_iov->iov_len -= left;
3001 		uiop->uio_resid -= left;
3002 	}
3003 
3004 	/*
3005 	 * We are now either at the end of the directory or have filled the
3006 	 * block.
3007 	 */
3008 	if (bigenough) {
3009 		dnp->n_direofoffset = uiop->uio_offset;
3010 		dnp->n_flag |= NEOFVALID;
3011 	}
3012 nfsmout:
3013 	if (newvp != NULLVP) {
3014 		if(newvp == vp)
3015 		    vrele(newvp);
3016 		else
3017 		    vput(newvp);
3018 	}
3019 	return (error);
3020 }
3021 #endif
3022 
3023 static char hextoasc[] = "0123456789abcdef";
3024 
3025 /*
3026  * Silly rename. To make the NFS filesystem that is stateless look a little
3027  * more like the "ufs" a remove of an active vnode is translated to a rename
3028  * to a funny looking filename that is removed by nfs_inactive on the
3029  * nfsnode. There is the potential for another process on a different client
3030  * to create the same funny name between the nfs_lookitup() fails and the
3031  * nfs_rename() completes, but...
3032  */
3033 int
3034 nfs_sillyrename(dvp, vp, cnp)
3035 	struct vnode *dvp, *vp;
3036 	struct componentname *cnp;
3037 {
3038 	struct sillyrename *sp;
3039 	struct nfsnode *np;
3040 	int error;
3041 	short pid;
3042 
3043 	cache_purge(dvp);
3044 	np = VTONFS(vp);
3045 #ifndef DIAGNOSTIC
3046 	if (vp->v_type == VDIR)
3047 		panic("nfs: sillyrename dir");
3048 #endif
3049 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
3050 		M_NFSREQ, M_WAITOK);
3051 	sp->s_cred = crdup(cnp->cn_cred);
3052 	sp->s_dvp = dvp;
3053 	VREF(dvp);
3054 
3055 	/* Fudge together a funny name */
3056 	pid = cnp->cn_proc->p_pid;
3057 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
3058 	sp->s_namlen = 12;
3059 	sp->s_name[8] = hextoasc[pid & 0xf];
3060 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
3061 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
3062 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
3063 
3064 	/* Try lookitups until we get one that isn't there */
3065 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3066 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
3067 		sp->s_name[4]++;
3068 		if (sp->s_name[4] > 'z') {
3069 			error = EINVAL;
3070 			goto bad;
3071 		}
3072 	}
3073 	error = nfs_renameit(dvp, cnp, sp);
3074 	if (error)
3075 		goto bad;
3076 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3077 		cnp->cn_proc, &np);
3078 	np->n_sillyrename = sp;
3079 	return (0);
3080 bad:
3081 	vrele(sp->s_dvp);
3082 	crfree(sp->s_cred);
3083 	free((caddr_t)sp, M_NFSREQ);
3084 	return (error);
3085 }
3086 
3087 /*
3088  * Look up a file name and optionally either update the file handle or
3089  * allocate an nfsnode, depending on the value of npp.
3090  * npp == NULL	--> just do the lookup
3091  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3092  *			handled too
3093  * *npp != NULL --> update the file handle in the vnode
3094  */
3095 int
3096 nfs_lookitup(dvp, name, len, cred, procp, npp)
3097 	struct vnode *dvp;
3098 	const char *name;
3099 	int len;
3100 	struct ucred *cred;
3101 	struct proc *procp;
3102 	struct nfsnode **npp;
3103 {
3104 	u_int32_t *tl;
3105 	caddr_t cp;
3106 	int32_t t1, t2;
3107 	struct vnode *newvp = (struct vnode *)0;
3108 	struct nfsnode *np, *dnp = VTONFS(dvp);
3109 	caddr_t bpos, dpos, cp2;
3110 	int error = 0, fhlen;
3111 #ifndef NFS_V2_ONLY
3112 	int attrflag;
3113 #endif
3114 	struct mbuf *mreq, *mrep, *md, *mb;
3115 	nfsfh_t *nfhp;
3116 	const int v3 = NFS_ISV3(dvp);
3117 
3118 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3119 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3120 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3121 	nfsm_fhtom(dnp, v3);
3122 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
3123 	nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
3124 	if (npp && !error) {
3125 		nfsm_getfh(nfhp, fhlen, v3);
3126 		if (*npp) {
3127 		    np = *npp;
3128 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3129 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
3130 			np->n_fhp = &np->n_fh;
3131 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
3132 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
3133 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
3134 		    np->n_fhsize = fhlen;
3135 		    newvp = NFSTOV(np);
3136 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3137 		    VREF(dvp);
3138 		    newvp = dvp;
3139 		    np = dnp;
3140 		} else {
3141 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3142 		    if (error) {
3143 			m_freem(mrep);
3144 			return (error);
3145 		    }
3146 		    newvp = NFSTOV(np);
3147 		}
3148 #ifndef NFS_V2_ONLY
3149 		if (v3) {
3150 			nfsm_postop_attr(newvp, attrflag, 0);
3151 			if (!attrflag && *npp == NULL) {
3152 				m_freem(mrep);
3153 				vput(newvp);
3154 				return (ENOENT);
3155 			}
3156 		} else
3157 #endif
3158 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
3159 	}
3160 	nfsm_reqdone;
3161 	if (npp && *npp == NULL) {
3162 		if (error) {
3163 			if (newvp)
3164 				vput(newvp);
3165 		} else
3166 			*npp = np;
3167 	}
3168 	return (error);
3169 }
3170 
3171 #ifndef NFS_V2_ONLY
3172 /*
3173  * Nfs Version 3 commit rpc
3174  */
3175 int
3176 nfs_commit(vp, offset, cnt, procp)
3177 	struct vnode *vp;
3178 	off_t offset;
3179 	uint32_t cnt;
3180 	struct proc *procp;
3181 {
3182 	caddr_t cp;
3183 	u_int32_t *tl;
3184 	int32_t t1, t2;
3185 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3186 	caddr_t bpos, dpos, cp2;
3187 	int error = 0, wccflag = NFSV3_WCCRATTR;
3188 	struct mbuf *mreq, *mrep, *md, *mb;
3189 	struct nfsnode *np;
3190 
3191 	KASSERT(NFS_ISV3(vp));
3192 
3193 #ifdef NFS_DEBUG_COMMIT
3194 	printf("commit %lu - %lu\n", (unsigned long)offset,
3195 	    (unsigned long)(offset + cnt));
3196 #endif
3197 
3198 	simple_lock(&nmp->nm_slock);
3199 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3200 		simple_unlock(&nmp->nm_slock);
3201 		return (0);
3202 	}
3203 	simple_unlock(&nmp->nm_slock);
3204 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3205 	np = VTONFS(vp);
3206 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3207 	nfsm_fhtom(np, 1);
3208 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3209 	txdr_hyper(offset, tl);
3210 	tl += 2;
3211 	*tl = txdr_unsigned(cnt);
3212 	nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred);
3213 	nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, FALSE);
3214 	if (!error) {
3215 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3216 		simple_lock(&nmp->nm_slock);
3217 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3218 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3219 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3220 			error = NFSERR_STALEWRITEVERF;
3221 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3222 		}
3223 		simple_unlock(&nmp->nm_slock);
3224 	}
3225 	nfsm_reqdone;
3226 	return (error);
3227 }
3228 #endif
3229 
3230 /*
3231  * Kludge City..
3232  * - make nfs_bmap() essentially a no-op that does no translation
3233  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3234  *   (Maybe I could use the process's page mapping, but I was concerned that
3235  *    Kernel Write might not be enabled and also figured copyout() would do
3236  *    a lot more work than memcpy() and also it currently happens in the
3237  *    context of the swapper process (2).
3238  */
3239 int
3240 nfs_bmap(v)
3241 	void *v;
3242 {
3243 	struct vop_bmap_args /* {
3244 		struct vnode *a_vp;
3245 		daddr_t  a_bn;
3246 		struct vnode **a_vpp;
3247 		daddr_t *a_bnp;
3248 		int *a_runp;
3249 	} */ *ap = v;
3250 	struct vnode *vp = ap->a_vp;
3251 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3252 
3253 	if (ap->a_vpp != NULL)
3254 		*ap->a_vpp = vp;
3255 	if (ap->a_bnp != NULL)
3256 		*ap->a_bnp = ap->a_bn << bshift;
3257 	if (ap->a_runp != NULL)
3258 		*ap->a_runp = 1024 * 1024; /* XXX */
3259 	return (0);
3260 }
3261 
3262 /*
3263  * Strategy routine.
3264  * For async requests when nfsiod(s) are running, queue the request by
3265  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3266  * request.
3267  */
3268 int
3269 nfs_strategy(v)
3270 	void *v;
3271 {
3272 	struct vop_strategy_args *ap = v;
3273 	struct buf *bp = ap->a_bp;
3274 	struct proc *p;
3275 	int error = 0;
3276 
3277 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3278 		panic("nfs physio/async");
3279 	if (bp->b_flags & B_ASYNC)
3280 		p = NULL;
3281 	else
3282 		p = curproc;	/* XXX */
3283 
3284 	/*
3285 	 * If the op is asynchronous and an i/o daemon is waiting
3286 	 * queue the request, wake it up and wait for completion
3287 	 * otherwise just do it ourselves.
3288 	 */
3289 
3290 	if ((bp->b_flags & B_ASYNC) == 0 ||
3291 	    nfs_asyncio(bp))
3292 		error = nfs_doio(bp, p);
3293 	return (error);
3294 }
3295 
3296 /*
3297  * fsync vnode op. Just call nfs_flush() with commit == 1.
3298  */
3299 /* ARGSUSED */
3300 int
3301 nfs_fsync(v)
3302 	void *v;
3303 {
3304 	struct vop_fsync_args /* {
3305 		struct vnodeop_desc *a_desc;
3306 		struct vnode * a_vp;
3307 		struct ucred * a_cred;
3308 		int  a_flags;
3309 		off_t offlo;
3310 		off_t offhi;
3311 		struct proc * a_p;
3312 	} */ *ap = v;
3313 
3314 	struct vnode *vp = ap->a_vp;
3315 
3316 	if (vp->v_type != VREG)
3317 		return 0;
3318 
3319 	return (nfs_flush(vp, ap->a_cred,
3320 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3321 }
3322 
3323 /*
3324  * Flush all the data associated with a vnode.
3325  */
3326 int
3327 nfs_flush(vp, cred, waitfor, p, commit)
3328 	struct vnode *vp;
3329 	struct ucred *cred;
3330 	int waitfor;
3331 	struct proc *p;
3332 	int commit;
3333 {
3334 	struct nfsnode *np = VTONFS(vp);
3335 	int error;
3336 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3337 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3338 
3339 	simple_lock(&vp->v_interlock);
3340 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3341 	if (np->n_flag & NWRITEERR) {
3342 		error = np->n_error;
3343 		np->n_flag &= ~NWRITEERR;
3344 	}
3345 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3346 	return (error);
3347 }
3348 
3349 /*
3350  * Return POSIX pathconf information applicable to nfs.
3351  *
3352  * N.B. The NFS V2 protocol doesn't support this RPC.
3353  */
3354 /* ARGSUSED */
3355 int
3356 nfs_pathconf(v)
3357 	void *v;
3358 {
3359 	struct vop_pathconf_args /* {
3360 		struct vnode *a_vp;
3361 		int a_name;
3362 		register_t *a_retval;
3363 	} */ *ap = v;
3364 	struct nfsv3_pathconf *pcp;
3365 	struct vnode *vp = ap->a_vp;
3366 	struct mbuf *mreq, *mrep, *md, *mb;
3367 	int32_t t1, t2;
3368 	u_int32_t *tl;
3369 	caddr_t bpos, dpos, cp, cp2;
3370 	int error = 0, attrflag;
3371 #ifndef NFS_V2_ONLY
3372 	struct nfsmount *nmp;
3373 	unsigned int l;
3374 	u_int64_t maxsize;
3375 #endif
3376 	const int v3 = NFS_ISV3(vp);
3377 	struct nfsnode *np = VTONFS(vp);
3378 
3379 	switch (ap->a_name) {
3380 		/* Names that can be resolved locally. */
3381 	case _PC_PIPE_BUF:
3382 		*ap->a_retval = PIPE_BUF;
3383 		break;
3384 	case _PC_SYNC_IO:
3385 		*ap->a_retval = 1;
3386 		break;
3387 	/* Names that cannot be resolved locally; do an RPC, if possible. */
3388 	case _PC_LINK_MAX:
3389 	case _PC_NAME_MAX:
3390 	case _PC_CHOWN_RESTRICTED:
3391 	case _PC_NO_TRUNC:
3392 		if (!v3) {
3393 			error = EINVAL;
3394 			break;
3395 		}
3396 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3397 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3398 		nfsm_fhtom(np, 1);
3399 		nfsm_request(np, NFSPROC_PATHCONF,
3400 		    curproc, curproc->p_ucred);	/* XXX */
3401 		nfsm_postop_attr(vp, attrflag, 0);
3402 		if (!error) {
3403 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
3404 			    NFSX_V3PATHCONF);
3405 			switch (ap->a_name) {
3406 			case _PC_LINK_MAX:
3407 				*ap->a_retval =
3408 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3409 				break;
3410 			case _PC_NAME_MAX:
3411 				*ap->a_retval =
3412 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3413 				break;
3414 			case _PC_CHOWN_RESTRICTED:
3415 				*ap->a_retval =
3416 				    (pcp->pc_chownrestricted == nfs_true);
3417 				break;
3418 			case _PC_NO_TRUNC:
3419 				*ap->a_retval =
3420 				    (pcp->pc_notrunc == nfs_true);
3421 				break;
3422 			}
3423 		}
3424 		nfsm_reqdone;
3425 		break;
3426 	case _PC_FILESIZEBITS:
3427 #ifndef NFS_V2_ONLY
3428 		if (v3) {
3429 			nmp = VFSTONFS(vp->v_mount);
3430 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3431 				if ((error = nfs_fsinfo(nmp, vp,
3432 				    curproc->p_ucred, curproc)) != 0) /* XXX */
3433 					break;
3434 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3435 			    (maxsize >> l) > 0; l++)
3436 				;
3437 			*ap->a_retval = l + 1;
3438 		} else
3439 #endif
3440 		{
3441 			*ap->a_retval = 32;	/* NFS V2 limitation */
3442 		}
3443 		break;
3444 	default:
3445 		error = EINVAL;
3446 		break;
3447 	}
3448 
3449 	return (error);
3450 }
3451 
3452 /*
3453  * NFS advisory byte-level locks.
3454  */
3455 int
3456 nfs_advlock(v)
3457 	void *v;
3458 {
3459 	struct vop_advlock_args /* {
3460 		struct vnode *a_vp;
3461 		caddr_t  a_id;
3462 		int  a_op;
3463 		struct flock *a_fl;
3464 		int  a_flags;
3465 	} */ *ap = v;
3466 	struct nfsnode *np = VTONFS(ap->a_vp);
3467 
3468 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3469 }
3470 
3471 /*
3472  * Print out the contents of an nfsnode.
3473  */
3474 int
3475 nfs_print(v)
3476 	void *v;
3477 {
3478 	struct vop_print_args /* {
3479 		struct vnode *a_vp;
3480 	} */ *ap = v;
3481 	struct vnode *vp = ap->a_vp;
3482 	struct nfsnode *np = VTONFS(vp);
3483 
3484 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3485 	    np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3486 	if (vp->v_type == VFIFO)
3487 		fifo_printinfo(vp);
3488 	printf("\n");
3489 	return (0);
3490 }
3491 
3492 /*
3493  * nfs unlock wrapper.
3494  */
3495 int
3496 nfs_unlock(void *v)
3497 {
3498 	struct vop_unlock_args /* {
3499 		struct vnode *a_vp;
3500 		int a_flags;
3501 	} */ *ap = v;
3502 	struct vnode *vp = ap->a_vp;
3503 
3504 	/*
3505 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3506 	 * with v_data == 0.
3507 	 */
3508 	if (VTONFS(vp)) {
3509 		nfs_delayedtruncate(vp);
3510 	}
3511 
3512 	return genfs_unlock(v);
3513 }
3514 
3515 /*
3516  * nfs special file access vnode op.
3517  * Essentially just get vattr and then imitate iaccess() since the device is
3518  * local to the client.
3519  */
3520 int
3521 nfsspec_access(v)
3522 	void *v;
3523 {
3524 	struct vop_access_args /* {
3525 		struct vnode *a_vp;
3526 		int  a_mode;
3527 		struct ucred *a_cred;
3528 		struct proc *a_p;
3529 	} */ *ap = v;
3530 	struct vattr va;
3531 	struct vnode *vp = ap->a_vp;
3532 	int error;
3533 
3534 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3535 	if (error)
3536 		return (error);
3537 
3538         /*
3539 	 * Disallow write attempts on filesystems mounted read-only;
3540 	 * unless the file is a socket, fifo, or a block or character
3541 	 * device resident on the filesystem.
3542 	 */
3543 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3544 		switch (vp->v_type) {
3545 		case VREG:
3546 		case VDIR:
3547 		case VLNK:
3548 			return (EROFS);
3549 		default:
3550 			break;
3551 		}
3552 	}
3553 
3554 	return (vaccess(va.va_type, va.va_mode,
3555 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3556 }
3557 
3558 /*
3559  * Read wrapper for special devices.
3560  */
3561 int
3562 nfsspec_read(v)
3563 	void *v;
3564 {
3565 	struct vop_read_args /* {
3566 		struct vnode *a_vp;
3567 		struct uio *a_uio;
3568 		int  a_ioflag;
3569 		struct ucred *a_cred;
3570 	} */ *ap = v;
3571 	struct nfsnode *np = VTONFS(ap->a_vp);
3572 
3573 	/*
3574 	 * Set access flag.
3575 	 */
3576 	np->n_flag |= NACC;
3577 	np->n_atim.tv_sec = time.tv_sec;
3578 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3579 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3580 }
3581 
3582 /*
3583  * Write wrapper for special devices.
3584  */
3585 int
3586 nfsspec_write(v)
3587 	void *v;
3588 {
3589 	struct vop_write_args /* {
3590 		struct vnode *a_vp;
3591 		struct uio *a_uio;
3592 		int  a_ioflag;
3593 		struct ucred *a_cred;
3594 	} */ *ap = v;
3595 	struct nfsnode *np = VTONFS(ap->a_vp);
3596 
3597 	/*
3598 	 * Set update flag.
3599 	 */
3600 	np->n_flag |= NUPD;
3601 	np->n_mtim.tv_sec = time.tv_sec;
3602 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3603 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3604 }
3605 
3606 /*
3607  * Close wrapper for special devices.
3608  *
3609  * Update the times on the nfsnode then do device close.
3610  */
3611 int
3612 nfsspec_close(v)
3613 	void *v;
3614 {
3615 	struct vop_close_args /* {
3616 		struct vnode *a_vp;
3617 		int  a_fflag;
3618 		struct ucred *a_cred;
3619 		struct proc *a_p;
3620 	} */ *ap = v;
3621 	struct vnode *vp = ap->a_vp;
3622 	struct nfsnode *np = VTONFS(vp);
3623 	struct vattr vattr;
3624 
3625 	if (np->n_flag & (NACC | NUPD)) {
3626 		np->n_flag |= NCHG;
3627 		if (vp->v_usecount == 1 &&
3628 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3629 			VATTR_NULL(&vattr);
3630 			if (np->n_flag & NACC)
3631 				vattr.va_atime = np->n_atim;
3632 			if (np->n_flag & NUPD)
3633 				vattr.va_mtime = np->n_mtim;
3634 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3635 		}
3636 	}
3637 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3638 }
3639 
3640 /*
3641  * Read wrapper for fifos.
3642  */
3643 int
3644 nfsfifo_read(v)
3645 	void *v;
3646 {
3647 	struct vop_read_args /* {
3648 		struct vnode *a_vp;
3649 		struct uio *a_uio;
3650 		int  a_ioflag;
3651 		struct ucred *a_cred;
3652 	} */ *ap = v;
3653 	struct nfsnode *np = VTONFS(ap->a_vp);
3654 
3655 	/*
3656 	 * Set access flag.
3657 	 */
3658 	np->n_flag |= NACC;
3659 	np->n_atim.tv_sec = time.tv_sec;
3660 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3661 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3662 }
3663 
3664 /*
3665  * Write wrapper for fifos.
3666  */
3667 int
3668 nfsfifo_write(v)
3669 	void *v;
3670 {
3671 	struct vop_write_args /* {
3672 		struct vnode *a_vp;
3673 		struct uio *a_uio;
3674 		int  a_ioflag;
3675 		struct ucred *a_cred;
3676 	} */ *ap = v;
3677 	struct nfsnode *np = VTONFS(ap->a_vp);
3678 
3679 	/*
3680 	 * Set update flag.
3681 	 */
3682 	np->n_flag |= NUPD;
3683 	np->n_mtim.tv_sec = time.tv_sec;
3684 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3685 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3686 }
3687 
3688 /*
3689  * Close wrapper for fifos.
3690  *
3691  * Update the times on the nfsnode then do fifo close.
3692  */
3693 int
3694 nfsfifo_close(v)
3695 	void *v;
3696 {
3697 	struct vop_close_args /* {
3698 		struct vnode *a_vp;
3699 		int  a_fflag;
3700 		struct ucred *a_cred;
3701 		struct proc *a_p;
3702 	} */ *ap = v;
3703 	struct vnode *vp = ap->a_vp;
3704 	struct nfsnode *np = VTONFS(vp);
3705 	struct vattr vattr;
3706 
3707 	if (np->n_flag & (NACC | NUPD)) {
3708 		if (np->n_flag & NACC) {
3709 			np->n_atim.tv_sec = time.tv_sec;
3710 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3711 		}
3712 		if (np->n_flag & NUPD) {
3713 			np->n_mtim.tv_sec = time.tv_sec;
3714 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3715 		}
3716 		np->n_flag |= NCHG;
3717 		if (vp->v_usecount == 1 &&
3718 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3719 			VATTR_NULL(&vattr);
3720 			if (np->n_flag & NACC)
3721 				vattr.va_atime = np->n_atim;
3722 			if (np->n_flag & NUPD)
3723 				vattr.va_mtime = np->n_mtim;
3724 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3725 		}
3726 	}
3727 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3728 }
3729