xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision fad4c9f71477ae11cea2ee75ec82151ac770a534)
1 /*	$NetBSD: nfs_vnops.c,v 1.240 2006/07/01 11:30:44 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 1.240 2006/07/01 11:30:44 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/disk.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/hash.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/unistd.h>
67 #include <sys/kauth.h>
68 
69 #include <uvm/uvm_extern.h>
70 #include <uvm/uvm.h>
71 
72 #include <miscfs/fifofs/fifo.h>
73 #include <miscfs/genfs/genfs.h>
74 #include <miscfs/specfs/specdev.h>
75 
76 #include <nfs/rpcv2.h>
77 #include <nfs/nfsproto.h>
78 #include <nfs/nfs.h>
79 #include <nfs/nfsnode.h>
80 #include <nfs/nfsmount.h>
81 #include <nfs/xdr_subs.h>
82 #include <nfs/nfsm_subs.h>
83 #include <nfs/nqnfs.h>
84 #include <nfs/nfs_var.h>
85 
86 #include <net/if.h>
87 #include <netinet/in.h>
88 #include <netinet/in_var.h>
89 
90 /*
91  * Global vfs data structures for nfs
92  */
93 int (**nfsv2_vnodeop_p) __P((void *));
94 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
95 	{ &vop_default_desc, vn_default_error },
96 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
97 	{ &vop_create_desc, nfs_create },		/* create */
98 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
99 	{ &vop_open_desc, nfs_open },			/* open */
100 	{ &vop_close_desc, nfs_close },			/* close */
101 	{ &vop_access_desc, nfs_access },		/* access */
102 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
103 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
104 	{ &vop_read_desc, nfs_read },			/* read */
105 	{ &vop_write_desc, nfs_write },			/* write */
106 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
107 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
108 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
109 	{ &vop_poll_desc, nfs_poll },			/* poll */
110 	{ &vop_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
111 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
112 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
113 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
114 	{ &vop_seek_desc, nfs_seek },			/* seek */
115 	{ &vop_remove_desc, nfs_remove },		/* remove */
116 	{ &vop_link_desc, nfs_link },			/* link */
117 	{ &vop_rename_desc, nfs_rename },		/* rename */
118 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
119 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
120 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
121 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
122 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
123 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
124 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
125 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
126 	{ &vop_lock_desc, nfs_lock },			/* lock */
127 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
128 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
129 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
130 	{ &vop_print_desc, nfs_print },			/* print */
131 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
132 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
133 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
134 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
135 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
136 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
137 	{ NULL, NULL }
138 };
139 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
140 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
141 
142 /*
143  * Special device vnode ops
144  */
145 int (**spec_nfsv2nodeop_p) __P((void *));
146 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
147 	{ &vop_default_desc, vn_default_error },
148 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
149 	{ &vop_create_desc, spec_create },		/* create */
150 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
151 	{ &vop_open_desc, spec_open },			/* open */
152 	{ &vop_close_desc, nfsspec_close },		/* close */
153 	{ &vop_access_desc, nfsspec_access },		/* access */
154 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
155 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
156 	{ &vop_read_desc, nfsspec_read },		/* read */
157 	{ &vop_write_desc, nfsspec_write },		/* write */
158 	{ &vop_lease_desc, spec_lease_check },		/* lease */
159 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
160 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
161 	{ &vop_poll_desc, spec_poll },			/* poll */
162 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
163 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
164 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
165 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
166 	{ &vop_seek_desc, spec_seek },			/* seek */
167 	{ &vop_remove_desc, spec_remove },		/* remove */
168 	{ &vop_link_desc, spec_link },			/* link */
169 	{ &vop_rename_desc, spec_rename },		/* rename */
170 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
171 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
172 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
173 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
174 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
175 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
176 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
177 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
178 	{ &vop_lock_desc, nfs_lock },			/* lock */
179 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
180 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
181 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
182 	{ &vop_print_desc, nfs_print },			/* print */
183 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
184 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
185 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
186 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
187 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
188 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
189 	{ NULL, NULL }
190 };
191 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
192 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
193 
194 int (**fifo_nfsv2nodeop_p) __P((void *));
195 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
196 	{ &vop_default_desc, vn_default_error },
197 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
198 	{ &vop_create_desc, fifo_create },		/* create */
199 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
200 	{ &vop_open_desc, fifo_open },			/* open */
201 	{ &vop_close_desc, nfsfifo_close },		/* close */
202 	{ &vop_access_desc, nfsspec_access },		/* access */
203 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
204 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
205 	{ &vop_read_desc, nfsfifo_read },		/* read */
206 	{ &vop_write_desc, nfsfifo_write },		/* write */
207 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
208 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
209 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
210 	{ &vop_poll_desc, fifo_poll },			/* poll */
211 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
212 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
213 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
214 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
215 	{ &vop_seek_desc, fifo_seek },			/* seek */
216 	{ &vop_remove_desc, fifo_remove },		/* remove */
217 	{ &vop_link_desc, fifo_link },			/* link */
218 	{ &vop_rename_desc, fifo_rename },		/* rename */
219 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
220 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
221 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
222 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
223 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
224 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
225 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
226 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
227 	{ &vop_lock_desc, nfs_lock },			/* lock */
228 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
229 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
230 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
231 	{ &vop_print_desc, nfs_print },			/* print */
232 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
233 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
234 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
235 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
236 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
237 	{ NULL, NULL }
238 };
239 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
240 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
241 
242 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
243     size_t, kauth_cred_t, struct lwp *);
244 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
245 
246 /*
247  * Global variables
248  */
249 extern u_int32_t nfs_true, nfs_false;
250 extern u_int32_t nfs_xdrneg1;
251 extern const nfstype nfsv3_type[9];
252 
253 int nfs_numasync = 0;
254 #define	DIRHDSIZ	_DIRENT_NAMEOFF(dp)
255 #define UIO_ADVANCE(uio, siz) \
256     (void)((uio)->uio_resid -= (siz), \
257     (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
258     (uio)->uio_iov->iov_len -= (siz))
259 
260 static void nfs_cache_enter(struct vnode *, struct vnode *,
261     struct componentname *);
262 
263 static void
264 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
265     struct componentname *cnp)
266 {
267 	struct nfsnode *dnp = VTONFS(dvp);
268 
269 	if (vp != NULL) {
270 		struct nfsnode *np = VTONFS(vp);
271 
272 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
273 	}
274 
275 	if (!timespecisset(&dnp->n_nctime))
276 		dnp->n_nctime = dnp->n_vattr->va_mtime;
277 
278 	cache_enter(dvp, vp, cnp);
279 }
280 
281 /*
282  * nfs null call from vfs.
283  */
284 int
285 nfs_null(vp, cred, l)
286 	struct vnode *vp;
287 	kauth_cred_t cred;
288 	struct lwp *l;
289 {
290 	caddr_t bpos, dpos;
291 	int error = 0;
292 	struct mbuf *mreq, *mrep, *md, *mb;
293 	struct nfsnode *np = VTONFS(vp);
294 
295 	nfsm_reqhead(np, NFSPROC_NULL, 0);
296 	nfsm_request(np, NFSPROC_NULL, l, cred);
297 	nfsm_reqdone;
298 	return (error);
299 }
300 
301 /*
302  * nfs access vnode op.
303  * For nfs version 2, just return ok. File accesses may fail later.
304  * For nfs version 3, use the access rpc to check accessibility. If file modes
305  * are changed on the server, accesses might still fail later.
306  */
307 int
308 nfs_access(v)
309 	void *v;
310 {
311 	struct vop_access_args /* {
312 		struct vnode *a_vp;
313 		int  a_mode;
314 		kauth_cred_t a_cred;
315 		struct lwp *a_l;
316 	} */ *ap = v;
317 	struct vnode *vp = ap->a_vp;
318 #ifndef NFS_V2_ONLY
319 	u_int32_t *tl;
320 	caddr_t cp;
321 	int32_t t1, t2;
322 	caddr_t bpos, dpos, cp2;
323 	int error = 0, attrflag;
324 	struct mbuf *mreq, *mrep, *md, *mb;
325 	u_int32_t mode, rmode;
326 	const int v3 = NFS_ISV3(vp);
327 #endif
328 	int cachevalid;
329 	struct nfsnode *np = VTONFS(vp);
330 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
331 
332 	cachevalid = (np->n_accstamp != -1 &&
333 	    (time_uptime - np->n_accstamp) < NFS_ATTRTIMEO(nmp, np) &&
334 	    np->n_accuid == kauth_cred_geteuid(ap->a_cred));
335 
336 	/*
337 	 * Check access cache first. If this request has been made for this
338 	 * uid shortly before, use the cached result.
339 	 */
340 	if (cachevalid) {
341 		if (!np->n_accerror) {
342 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
343 				return np->n_accerror;
344 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
345 			return np->n_accerror;
346 	}
347 
348 #ifndef NFS_V2_ONLY
349 	/*
350 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
351 	 * ufs_access() locally using the vattr. This may not be correct,
352 	 * since the server may apply other access criteria such as
353 	 * client uid-->server uid mapping that we do not know about, but
354 	 * this is better than just returning anything that is lying about
355 	 * in the cache.
356 	 */
357 	if (v3) {
358 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
359 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
360 		nfsm_fhtom(np, v3);
361 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
362 		if (ap->a_mode & VREAD)
363 			mode = NFSV3ACCESS_READ;
364 		else
365 			mode = 0;
366 		if (vp->v_type != VDIR) {
367 			if (ap->a_mode & VWRITE)
368 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
369 			if (ap->a_mode & VEXEC)
370 				mode |= NFSV3ACCESS_EXECUTE;
371 		} else {
372 			if (ap->a_mode & VWRITE)
373 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
374 					 NFSV3ACCESS_DELETE);
375 			if (ap->a_mode & VEXEC)
376 				mode |= NFSV3ACCESS_LOOKUP;
377 		}
378 		*tl = txdr_unsigned(mode);
379 		nfsm_request(np, NFSPROC_ACCESS, ap->a_l, ap->a_cred);
380 		nfsm_postop_attr(vp, attrflag, 0);
381 		if (!error) {
382 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
383 			rmode = fxdr_unsigned(u_int32_t, *tl);
384 			/*
385 			 * The NFS V3 spec does not clarify whether or not
386 			 * the returned access bits can be a superset of
387 			 * the ones requested, so...
388 			 */
389 			if ((rmode & mode) != mode)
390 				error = EACCES;
391 		}
392 		nfsm_reqdone;
393 	} else
394 #endif
395 		return (nfsspec_access(ap));
396 #ifndef NFS_V2_ONLY
397 	/*
398 	 * Disallow write attempts on filesystems mounted read-only;
399 	 * unless the file is a socket, fifo, or a block or character
400 	 * device resident on the filesystem.
401 	 */
402 	if (!error && (ap->a_mode & VWRITE) &&
403 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
404 		switch (vp->v_type) {
405 		case VREG:
406 		case VDIR:
407 		case VLNK:
408 			error = EROFS;
409 		default:
410 			break;
411 		}
412 	}
413 
414 	if (!error || error == EACCES) {
415 		/*
416 		 * If we got the same result as for a previous,
417 		 * different request, OR it in. Don't update
418 		 * the timestamp in that case.
419 		 */
420 		if (cachevalid && np->n_accstamp != -1 &&
421 		    error == np->n_accerror) {
422 			if (!error)
423 				np->n_accmode |= ap->a_mode;
424 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
425 				np->n_accmode = ap->a_mode;
426 		} else {
427 			np->n_accstamp = time_uptime;
428 			np->n_accuid = kauth_cred_geteuid(ap->a_cred);
429 			np->n_accmode = ap->a_mode;
430 			np->n_accerror = error;
431 		}
432 	}
433 
434 	return (error);
435 #endif
436 }
437 
438 /*
439  * nfs open vnode op
440  * Check to see if the type is ok
441  * and that deletion is not in progress.
442  * For paged in text files, you will need to flush the page cache
443  * if consistency is lost.
444  */
445 /* ARGSUSED */
446 int
447 nfs_open(v)
448 	void *v;
449 {
450 	struct vop_open_args /* {
451 		struct vnode *a_vp;
452 		int  a_mode;
453 		kauth_cred_t a_cred;
454 		struct lwp *a_l;
455 	} */ *ap = v;
456 	struct vnode *vp = ap->a_vp;
457 	struct nfsnode *np = VTONFS(vp);
458 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
459 	int error;
460 
461 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
462 		return (EACCES);
463 	}
464 
465 	if (ap->a_mode & FREAD) {
466 		if (np->n_rcred != NULL)
467 			kauth_cred_free(np->n_rcred);
468 		np->n_rcred = ap->a_cred;
469 		kauth_cred_hold(np->n_rcred);
470 	}
471 	if (ap->a_mode & FWRITE) {
472 		if (np->n_wcred != NULL)
473 			kauth_cred_free(np->n_wcred);
474 		np->n_wcred = ap->a_cred;
475 		kauth_cred_hold(np->n_wcred);
476 	}
477 
478 #ifndef NFS_V2_ONLY
479 	/*
480 	 * Get a valid lease. If cached data is stale, flush it.
481 	 */
482 	if (nmp->nm_flag & NFSMNT_NQNFS) {
483 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
484 		    do {
485 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
486 			    ap->a_l);
487 		    } while (error == NQNFS_EXPIRED);
488 		    if (error)
489 			return (error);
490 		    if (np->n_lrev != np->n_brev ||
491 			(np->n_flag & NQNFSNONCACHE)) {
492 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
493 				ap->a_l, 1)) == EINTR)
494 				return (error);
495 			np->n_brev = np->n_lrev;
496 		    }
497 		}
498 	} else
499 #endif
500 	{
501 		error = nfs_flushstalebuf(vp, ap->a_cred, ap->a_l, 0);
502 		if (error)
503 			return error;
504 	}
505 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
506 		NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
507 	return (0);
508 }
509 
510 /*
511  * nfs close vnode op
512  * What an NFS client should do upon close after writing is a debatable issue.
513  * Most NFS clients push delayed writes to the server upon close, basically for
514  * two reasons:
515  * 1 - So that any write errors may be reported back to the client process
516  *     doing the close system call. By far the two most likely errors are
517  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
518  * 2 - To put a worst case upper bound on cache inconsistency between
519  *     multiple clients for the file.
520  * There is also a consistency problem for Version 2 of the protocol w.r.t.
521  * not being able to tell if other clients are writing a file concurrently,
522  * since there is no way of knowing if the changed modify time in the reply
523  * is only due to the write for this client.
524  * (NFS Version 3 provides weak cache consistency data in the reply that
525  *  should be sufficient to detect and handle this case.)
526  *
527  * The current code does the following:
528  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
529  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
530  *                     or commit them (this satisfies 1 and 2 except for the
531  *                     case where the server crashes after this close but
532  *                     before the commit RPC, which is felt to be "good
533  *                     enough". Changing the last argument to nfs_flush() to
534  *                     a 1 would force a commit operation, if it is felt a
535  *                     commit is necessary now.
536  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
537  *                     1 should be dealt with via an fsync() system call for
538  *                     cases where write errors are important.
539  */
540 /* ARGSUSED */
541 int
542 nfs_close(v)
543 	void *v;
544 {
545 	struct vop_close_args /* {
546 		struct vnodeop_desc *a_desc;
547 		struct vnode *a_vp;
548 		int  a_fflag;
549 		kauth_cred_t a_cred;
550 		struct lwp *a_l;
551 	} */ *ap = v;
552 	struct vnode *vp = ap->a_vp;
553 	struct nfsnode *np = VTONFS(vp);
554 	int error = 0;
555 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
556 
557 	if (vp->v_type == VREG) {
558 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
559 		(np->n_flag & NMODIFIED)) {
560 #ifndef NFS_V2_ONLY
561 		if (NFS_ISV3(vp)) {
562 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_l, 0);
563 		    np->n_flag &= ~NMODIFIED;
564 		} else
565 #endif
566 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 1);
567 		NFS_INVALIDATE_ATTRCACHE(np);
568 	    }
569 	    if (np->n_flag & NWRITEERR) {
570 		np->n_flag &= ~NWRITEERR;
571 		error = np->n_error;
572 	    }
573 	}
574 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
575 	return (error);
576 }
577 
578 /*
579  * nfs getattr call from vfs.
580  */
581 int
582 nfs_getattr(v)
583 	void *v;
584 {
585 	struct vop_getattr_args /* {
586 		struct vnode *a_vp;
587 		struct vattr *a_vap;
588 		kauth_cred_t a_cred;
589 		struct lwp *a_l;
590 	} */ *ap = v;
591 	struct vnode *vp = ap->a_vp;
592 	struct nfsnode *np = VTONFS(vp);
593 	caddr_t cp;
594 	u_int32_t *tl;
595 	int32_t t1, t2;
596 	caddr_t bpos, dpos;
597 	int error = 0;
598 	struct mbuf *mreq, *mrep, *md, *mb;
599 	const int v3 = NFS_ISV3(vp);
600 
601 	/*
602 	 * Update local times for special files.
603 	 */
604 	if (np->n_flag & (NACC | NUPD))
605 		np->n_flag |= NCHG;
606 
607 	/*
608 	 * if we have delayed truncation, do it now.
609 	 */
610 	nfs_delayedtruncate(vp);
611 
612 	/*
613 	 * First look in the cache.
614 	 */
615 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
616 		return (0);
617 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
618 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
619 	nfsm_fhtom(np, v3);
620 	nfsm_request(np, NFSPROC_GETATTR, ap->a_l, ap->a_cred);
621 	if (!error) {
622 		nfsm_loadattr(vp, ap->a_vap, 0);
623 		if (vp->v_type == VDIR &&
624 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
625 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
626 	}
627 	nfsm_reqdone;
628 	return (error);
629 }
630 
631 /*
632  * nfs setattr call.
633  */
634 int
635 nfs_setattr(v)
636 	void *v;
637 {
638 	struct vop_setattr_args /* {
639 		struct vnodeop_desc *a_desc;
640 		struct vnode *a_vp;
641 		struct vattr *a_vap;
642 		kauth_cred_t a_cred;
643 		struct lwp *a_l;
644 	} */ *ap = v;
645 	struct vnode *vp = ap->a_vp;
646 	struct nfsnode *np = VTONFS(vp);
647 	struct vattr *vap = ap->a_vap;
648 	int error = 0;
649 	u_quad_t tsize = 0;
650 
651 	/*
652 	 * Setting of flags is not supported.
653 	 */
654 	if (vap->va_flags != VNOVAL)
655 		return (EOPNOTSUPP);
656 
657 	/*
658 	 * Disallow write attempts if the filesystem is mounted read-only.
659 	 */
660   	if ((vap->va_uid != (uid_t)VNOVAL ||
661 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
662 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
663 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
664 		return (EROFS);
665 	if (vap->va_size != VNOVAL) {
666  		switch (vp->v_type) {
667  		case VDIR:
668  			return (EISDIR);
669  		case VCHR:
670  		case VBLK:
671  		case VSOCK:
672  		case VFIFO:
673 			if (vap->va_mtime.tv_sec == VNOVAL &&
674 			    vap->va_atime.tv_sec == VNOVAL &&
675 			    vap->va_mode == (mode_t)VNOVAL &&
676 			    vap->va_uid == (uid_t)VNOVAL &&
677 			    vap->va_gid == (gid_t)VNOVAL)
678 				return (0);
679  			vap->va_size = VNOVAL;
680  			break;
681  		default:
682 			/*
683 			 * Disallow write attempts if the filesystem is
684 			 * mounted read-only.
685 			 */
686 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
687 				return (EROFS);
688  			uvm_vnp_setsize(vp, vap->va_size);
689  			tsize = np->n_size;
690 			np->n_size = vap->va_size;
691  			if (vap->va_size == 0)
692  				error = nfs_vinvalbuf(vp, 0,
693  				     ap->a_cred, ap->a_l, 1);
694 			else
695 				error = nfs_vinvalbuf(vp, V_SAVE,
696 				     ap->a_cred, ap->a_l, 1);
697 			if (error) {
698 				uvm_vnp_setsize(vp, tsize);
699 				return (error);
700 			}
701  			np->n_vattr->va_size = vap->va_size;
702   		}
703   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
704 		vap->va_atime.tv_sec != VNOVAL) &&
705 		vp->v_type == VREG &&
706   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
707 		 ap->a_l, 1)) == EINTR)
708 		return (error);
709 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_l);
710 	if (error && vap->va_size != VNOVAL) {
711 		np->n_size = np->n_vattr->va_size = tsize;
712 		uvm_vnp_setsize(vp, np->n_size);
713 	}
714 	VN_KNOTE(vp, NOTE_ATTRIB);
715 	return (error);
716 }
717 
718 /*
719  * Do an nfs setattr rpc.
720  */
721 int
722 nfs_setattrrpc(vp, vap, cred, l)
723 	struct vnode *vp;
724 	struct vattr *vap;
725 	kauth_cred_t cred;
726 	struct lwp *l;
727 {
728 	struct nfsv2_sattr *sp;
729 	caddr_t cp;
730 	int32_t t1, t2;
731 	caddr_t bpos, dpos;
732 	u_int32_t *tl;
733 	int error = 0;
734 	struct mbuf *mreq, *mrep, *md, *mb;
735 	const int v3 = NFS_ISV3(vp);
736 	struct nfsnode *np = VTONFS(vp);
737 #ifndef NFS_V2_ONLY
738 	int wccflag = NFSV3_WCCRATTR;
739 	caddr_t cp2;
740 #endif
741 
742 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
743 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
744 	nfsm_fhtom(np, v3);
745 #ifndef NFS_V2_ONLY
746 	if (v3) {
747 		nfsm_v3attrbuild(vap, TRUE);
748 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
749 		*tl = nfs_false;
750 	} else {
751 #endif
752 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
753 		if (vap->va_mode == (mode_t)VNOVAL)
754 			sp->sa_mode = nfs_xdrneg1;
755 		else
756 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
757 		if (vap->va_uid == (uid_t)VNOVAL)
758 			sp->sa_uid = nfs_xdrneg1;
759 		else
760 			sp->sa_uid = txdr_unsigned(vap->va_uid);
761 		if (vap->va_gid == (gid_t)VNOVAL)
762 			sp->sa_gid = nfs_xdrneg1;
763 		else
764 			sp->sa_gid = txdr_unsigned(vap->va_gid);
765 		sp->sa_size = txdr_unsigned(vap->va_size);
766 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
767 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
768 #ifndef NFS_V2_ONLY
769 	}
770 #endif
771 	nfsm_request(np, NFSPROC_SETATTR, l, cred);
772 #ifndef NFS_V2_ONLY
773 	if (v3) {
774 		nfsm_wcc_data(vp, wccflag, 0, FALSE);
775 	} else
776 #endif
777 		nfsm_loadattr(vp, (struct vattr *)0, 0);
778 	nfsm_reqdone;
779 	return (error);
780 }
781 
782 /*
783  * nfs lookup call, one step at a time...
784  * First look in cache
785  * If not found, unlock the directory nfsnode and do the rpc
786  *
787  * This code is full of lock/unlock statements and checks, because
788  * we continue after cache_lookup has finished (we need to check
789  * with the attr cache and do an rpc if it has timed out). This means
790  * that the locking effects of cache_lookup have to be taken into
791  * account.
792  */
793 int
794 nfs_lookup(v)
795 	void *v;
796 {
797 	struct vop_lookup_args /* {
798 		struct vnodeop_desc *a_desc;
799 		struct vnode *a_dvp;
800 		struct vnode **a_vpp;
801 		struct componentname *a_cnp;
802 	} */ *ap = v;
803 	struct componentname *cnp = ap->a_cnp;
804 	struct vnode *dvp = ap->a_dvp;
805 	struct vnode **vpp = ap->a_vpp;
806 	int flags;
807 	struct vnode *newvp;
808 	u_int32_t *tl;
809 	caddr_t cp;
810 	int32_t t1, t2;
811 	caddr_t bpos, dpos, cp2;
812 	struct mbuf *mreq, *mrep, *md, *mb;
813 	long len;
814 	nfsfh_t *fhp;
815 	struct nfsnode *np;
816 	int lockparent, wantparent, error = 0, attrflag, fhsize;
817 	const int v3 = NFS_ISV3(dvp);
818 
819 	cnp->cn_flags &= ~PDIRUNLOCK;
820 	flags = cnp->cn_flags;
821 
822 	*vpp = NULLVP;
823 	newvp = NULLVP;
824 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
825 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
826 		return (EROFS);
827 	if (dvp->v_type != VDIR)
828 		return (ENOTDIR);
829 
830 	/*
831 	 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
832 	 */
833 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
834 		error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp);
835 		if (error)
836 			return error;
837 		if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
838 			return EISDIR;
839 		VREF(dvp);
840 		*vpp = dvp;
841 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
842 				cnp->cn_flags |= SAVENAME;
843 		return 0;
844 	}
845 
846 	lockparent = flags & LOCKPARENT;
847 	wantparent = flags & (LOCKPARENT|WANTPARENT);
848 	np = VTONFS(dvp);
849 
850 	/*
851 	 * Before tediously performing a linear scan of the directory,
852 	 * check the name cache to see if the directory/name pair
853 	 * we are looking for is known already.
854 	 * If the directory/name pair is found in the name cache,
855 	 * we have to ensure the directory has not changed from
856 	 * the time the cache entry has been created. If it has,
857 	 * the cache entry has to be ignored.
858 	 */
859 	error = cache_lookup_raw(dvp, vpp, cnp);
860 	KASSERT(dvp != *vpp);
861 	if (error >= 0) {
862 		struct vattr vattr;
863 		int err2;
864 
865 		if (error && error != ENOENT) {
866 			*vpp = NULLVP;
867 			return error;
868 		}
869 
870 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp);
871 		if (err2 != 0) {
872 			if (error == 0)
873 				vrele(*vpp);
874 			*vpp = NULLVP;
875 			return err2;
876 		}
877 
878 		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
879 		    cnp->cn_lwp) || timespeccmp(&vattr.va_mtime,
880 		    &VTONFS(dvp)->n_nctime, !=)) {
881 			if (error == 0) {
882 				vrele(*vpp);
883 				*vpp = NULLVP;
884 			}
885 			cache_purge1(dvp, NULL, PURGE_CHILDREN);
886 			timespecclear(&np->n_nctime);
887 			goto dorpc;
888 		}
889 
890 		if (error == ENOENT) {
891 			goto noentry;
892 		}
893 
894 		newvp = *vpp;
895 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp)
896 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
897 			nfsstats.lookupcache_hits++;
898 			if ((flags & ISDOTDOT) != 0 ||
899 			    (~flags & (LOCKPARENT|ISLASTCN)) != 0) {
900 				VOP_UNLOCK(dvp, 0);
901 				cnp->cn_flags |= PDIRUNLOCK;
902 			}
903 			error = vn_lock(newvp, LK_EXCLUSIVE);
904 			if (error) {
905 				/* newvp has been revoked. */
906 				vrele(newvp);
907 				*vpp = NULL;
908 				return error;
909 			}
910 			if ((~flags & (LOCKPARENT|ISLASTCN)) == 0
911 			    && (cnp->cn_flags & PDIRUNLOCK)) {
912 				KASSERT(flags & ISDOTDOT);
913 				error = vn_lock(dvp, LK_EXCLUSIVE);
914 				if (error) {
915 					vput(newvp);
916 					*vpp = NULL;
917 					return error;
918 				}
919 				cnp->cn_flags &= ~PDIRUNLOCK;
920 			}
921 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
922 				cnp->cn_flags |= SAVENAME;
923 			KASSERT(newvp->v_type != VNON);
924 			return (0);
925 		}
926 		cache_purge1(newvp, NULL, PURGE_PARENTS);
927 		vrele(newvp);
928 		*vpp = NULLVP;
929 	}
930 dorpc:
931 #if 0
932 	/*
933 	 * because nfsv3 has the same CREATE semantics as ours,
934 	 * we don't have to perform LOOKUPs beforehand.
935 	 *
936 	 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
937 	 * XXX although we have no way to know if O_EXCL is requested or not.
938 	 */
939 
940 	if (v3 && cnp->cn_nameiop == CREATE &&
941 	    (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
942 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
943 		KASSERT(lockparent);
944 		cnp->cn_flags |= SAVENAME;
945 		return (EJUSTRETURN);
946 	}
947 #endif /* 0 */
948 
949 	error = 0;
950 	newvp = NULLVP;
951 	nfsstats.lookupcache_misses++;
952 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
953 	len = cnp->cn_namelen;
954 	nfsm_reqhead(np, NFSPROC_LOOKUP,
955 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
956 	nfsm_fhtom(np, v3);
957 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
958 	nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_lwp, cnp->cn_cred);
959 	if (error) {
960 		nfsm_postop_attr(dvp, attrflag, 0);
961 		m_freem(mrep);
962 		goto nfsmout;
963 	}
964 	nfsm_getfh(fhp, fhsize, v3);
965 
966 	/*
967 	 * Handle RENAME case...
968 	 */
969 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
970 		if (NFS_CMPFH(np, fhp, fhsize)) {
971 			m_freem(mrep);
972 			return (EISDIR);
973 		}
974 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
975 		if (error) {
976 			m_freem(mrep);
977 			return error;
978 		}
979 		newvp = NFSTOV(np);
980 #ifndef NFS_V2_ONLY
981 		if (v3) {
982 			nfsm_postop_attr(newvp, attrflag, 0);
983 			nfsm_postop_attr(dvp, attrflag, 0);
984 		} else
985 #endif
986 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
987 		*vpp = newvp;
988 		m_freem(mrep);
989 		cnp->cn_flags |= SAVENAME;
990 		if (!lockparent) {
991 			VOP_UNLOCK(dvp, 0);
992 			cnp->cn_flags |= PDIRUNLOCK;
993 		}
994 		goto validate;
995 	}
996 
997 	/*
998 	 * The postop attr handling is duplicated for each if case,
999 	 * because it should be done while dvp is locked (unlocking
1000 	 * dvp is different for each case).
1001 	 */
1002 
1003 	if (NFS_CMPFH(np, fhp, fhsize)) {
1004 		/*
1005 		 * "." lookup
1006 		 */
1007 		VREF(dvp);
1008 		newvp = dvp;
1009 #ifndef NFS_V2_ONLY
1010 		if (v3) {
1011 			nfsm_postop_attr(newvp, attrflag, 0);
1012 			nfsm_postop_attr(dvp, attrflag, 0);
1013 		} else
1014 #endif
1015 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1016 	} else if (flags & ISDOTDOT) {
1017 		/*
1018 		 * ".." lookup
1019 		 */
1020 		VOP_UNLOCK(dvp, 0);
1021 		cnp->cn_flags |= PDIRUNLOCK;
1022 
1023 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1024 		if (error) {
1025 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
1026 				cnp->cn_flags &= ~PDIRUNLOCK;
1027 			m_freem(mrep);
1028 			return error;
1029 		}
1030 		newvp = NFSTOV(np);
1031 
1032 #ifndef NFS_V2_ONLY
1033 		if (v3) {
1034 			nfsm_postop_attr(newvp, attrflag, 0);
1035 			nfsm_postop_attr(dvp, attrflag, 0);
1036 		} else
1037 #endif
1038 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1039 
1040 		if (lockparent && (flags & ISLASTCN)) {
1041 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
1042 				m_freem(mrep);
1043 				vput(newvp);
1044 				return error;
1045 			}
1046 			cnp->cn_flags &= ~PDIRUNLOCK;
1047 		}
1048 	} else {
1049 		/*
1050 		 * Other lookups.
1051 		 */
1052 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1053 		if (error) {
1054 			m_freem(mrep);
1055 			return error;
1056 		}
1057 		newvp = NFSTOV(np);
1058 #ifndef NFS_V2_ONLY
1059 		if (v3) {
1060 			nfsm_postop_attr(newvp, attrflag, 0);
1061 			nfsm_postop_attr(dvp, attrflag, 0);
1062 		} else
1063 #endif
1064 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1065 		if (!lockparent || !(flags & ISLASTCN)) {
1066 			VOP_UNLOCK(dvp, 0);
1067 			cnp->cn_flags |= PDIRUNLOCK;
1068 		}
1069 	}
1070 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1071 		cnp->cn_flags |= SAVENAME;
1072 	if ((cnp->cn_flags & MAKEENTRY) &&
1073 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1074 		nfs_cache_enter(dvp, newvp, cnp);
1075 	}
1076 	*vpp = newvp;
1077 	nfsm_reqdone;
1078 	if (error) {
1079 		/*
1080 		 * We get here only because of errors returned by
1081 		 * the RPC. Otherwise we'll have returned above
1082 		 * (the nfsm_* macros will jump to nfsm_reqdone
1083 		 * on error).
1084 		 */
1085 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1086 		    cnp->cn_nameiop != CREATE) {
1087 			nfs_cache_enter(dvp, NULL, cnp);
1088 		}
1089 		if (newvp != NULLVP) {
1090 			vrele(newvp);
1091 			if (newvp != dvp)
1092 				VOP_UNLOCK(newvp, 0);
1093 		}
1094 noentry:
1095 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1096 		    (flags & ISLASTCN) && error == ENOENT) {
1097 			if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1098 				error = EROFS;
1099 			} else {
1100 				error = EJUSTRETURN;
1101 				cnp->cn_flags |= SAVENAME;
1102 			}
1103 		}
1104 		*vpp = NULL;
1105 		return error;
1106 	}
1107 
1108 validate:
1109 	/*
1110 	 * make sure we have valid type and size.
1111 	 */
1112 
1113 	newvp = *vpp;
1114 	if (newvp->v_type == VNON) {
1115 		struct vattr vattr; /* dummy */
1116 
1117 		KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1118 		error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp);
1119 		if (error) {
1120 			vput(newvp);
1121 			*vpp = NULL;
1122 		}
1123 	}
1124 
1125 	return error;
1126 }
1127 
1128 /*
1129  * nfs read call.
1130  * Just call nfs_bioread() to do the work.
1131  */
1132 int
1133 nfs_read(v)
1134 	void *v;
1135 {
1136 	struct vop_read_args /* {
1137 		struct vnode *a_vp;
1138 		struct uio *a_uio;
1139 		int  a_ioflag;
1140 		kauth_cred_t a_cred;
1141 	} */ *ap = v;
1142 	struct vnode *vp = ap->a_vp;
1143 
1144 	if (vp->v_type != VREG)
1145 		return EISDIR;
1146 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1147 }
1148 
1149 /*
1150  * nfs readlink call
1151  */
1152 int
1153 nfs_readlink(v)
1154 	void *v;
1155 {
1156 	struct vop_readlink_args /* {
1157 		struct vnode *a_vp;
1158 		struct uio *a_uio;
1159 		kauth_cred_t a_cred;
1160 	} */ *ap = v;
1161 	struct vnode *vp = ap->a_vp;
1162 	struct nfsnode *np = VTONFS(vp);
1163 
1164 	if (vp->v_type != VLNK)
1165 		return (EPERM);
1166 
1167 	if (np->n_rcred != NULL) {
1168 		kauth_cred_free(np->n_rcred);
1169 	}
1170 	np->n_rcred = ap->a_cred;
1171 	kauth_cred_hold(np->n_rcred);
1172 
1173 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1174 }
1175 
1176 /*
1177  * Do a readlink rpc.
1178  * Called by nfs_doio() from below the buffer cache.
1179  */
1180 int
1181 nfs_readlinkrpc(vp, uiop, cred)
1182 	struct vnode *vp;
1183 	struct uio *uiop;
1184 	kauth_cred_t cred;
1185 {
1186 	u_int32_t *tl;
1187 	caddr_t cp;
1188 	int32_t t1, t2;
1189 	caddr_t bpos, dpos, cp2;
1190 	int error = 0;
1191 	uint32_t len;
1192 	struct mbuf *mreq, *mrep, *md, *mb;
1193 	const int v3 = NFS_ISV3(vp);
1194 	struct nfsnode *np = VTONFS(vp);
1195 #ifndef NFS_V2_ONLY
1196 	int attrflag;
1197 #endif
1198 
1199 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1200 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1201 	nfsm_fhtom(np, v3);
1202 	nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1203 #ifndef NFS_V2_ONLY
1204 	if (v3)
1205 		nfsm_postop_attr(vp, attrflag, 0);
1206 #endif
1207 	if (!error) {
1208 #ifndef NFS_V2_ONLY
1209 		if (v3) {
1210 			nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1211 			len = fxdr_unsigned(uint32_t, *tl);
1212 			if (len > MAXPATHLEN) {
1213 				/*
1214 				 * this pathname is too long for us.
1215 				 */
1216 				m_freem(mrep);
1217 				/* Solaris returns EINVAL. should we follow? */
1218 				error = ENAMETOOLONG;
1219 				goto nfsmout;
1220 			}
1221 		} else
1222 #endif
1223 		{
1224 			nfsm_strsiz(len, NFS_MAXPATHLEN);
1225 		}
1226 		nfsm_mtouio(uiop, len);
1227 	}
1228 	nfsm_reqdone;
1229 	return (error);
1230 }
1231 
1232 /*
1233  * nfs read rpc call
1234  * Ditto above
1235  */
1236 int
1237 nfs_readrpc(vp, uiop)
1238 	struct vnode *vp;
1239 	struct uio *uiop;
1240 {
1241 	u_int32_t *tl;
1242 	caddr_t cp;
1243 	int32_t t1, t2;
1244 	caddr_t bpos, dpos, cp2;
1245 	struct mbuf *mreq, *mrep, *md, *mb;
1246 	struct nfsmount *nmp;
1247 	int error = 0, len, retlen, tsiz, eof, byte_count;
1248 	const int v3 = NFS_ISV3(vp);
1249 	struct nfsnode *np = VTONFS(vp);
1250 #ifndef NFS_V2_ONLY
1251 	int attrflag;
1252 #endif
1253 
1254 #ifndef nolint
1255 	eof = 0;
1256 #endif
1257 	nmp = VFSTONFS(vp->v_mount);
1258 	tsiz = uiop->uio_resid;
1259 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1260 		return (EFBIG);
1261 	iostat_busy(nmp->nm_stats);
1262 	byte_count = 0; /* count bytes actually transferred */
1263 	while (tsiz > 0) {
1264 		nfsstats.rpccnt[NFSPROC_READ]++;
1265 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1266 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1267 		nfsm_fhtom(np, v3);
1268 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1269 #ifndef NFS_V2_ONLY
1270 		if (v3) {
1271 			txdr_hyper(uiop->uio_offset, tl);
1272 			*(tl + 2) = txdr_unsigned(len);
1273 		} else
1274 #endif
1275 		{
1276 			*tl++ = txdr_unsigned(uiop->uio_offset);
1277 			*tl++ = txdr_unsigned(len);
1278 			*tl = 0;
1279 		}
1280 		nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1281 #ifndef NFS_V2_ONLY
1282 		if (v3) {
1283 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1284 			if (error) {
1285 				m_freem(mrep);
1286 				goto nfsmout;
1287 			}
1288 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1289 			eof = fxdr_unsigned(int, *(tl + 1));
1290 		} else
1291 #endif
1292 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1293 		nfsm_strsiz(retlen, nmp->nm_rsize);
1294 		nfsm_mtouio(uiop, retlen);
1295 		m_freem(mrep);
1296 		tsiz -= retlen;
1297 		byte_count += retlen;
1298 #ifndef NFS_V2_ONLY
1299 		if (v3) {
1300 			if (eof || retlen == 0)
1301 				tsiz = 0;
1302 		} else
1303 #endif
1304 		if (retlen < len)
1305 			tsiz = 0;
1306 	}
1307 nfsmout:
1308 	iostat_unbusy(nmp->nm_stats, byte_count, 1);
1309 	return (error);
1310 }
1311 
1312 struct nfs_writerpc_context {
1313 	struct simplelock nwc_slock;
1314 	volatile int nwc_mbufcount;
1315 };
1316 
1317 /*
1318  * free mbuf used to refer protected pages while write rpc call.
1319  * called at splvm.
1320  */
1321 static void
1322 nfs_writerpc_extfree(struct mbuf *m, caddr_t tbuf, size_t size, void *arg)
1323 {
1324 	struct nfs_writerpc_context *ctx = arg;
1325 
1326 	KASSERT(m != NULL);
1327 	KASSERT(ctx != NULL);
1328 	pool_cache_put(&mbpool_cache, m);
1329 	simple_lock(&ctx->nwc_slock);
1330 	if (--ctx->nwc_mbufcount == 0) {
1331 		wakeup(ctx);
1332 	}
1333 	simple_unlock(&ctx->nwc_slock);
1334 }
1335 
1336 /*
1337  * nfs write call
1338  */
1339 int
1340 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1341 	struct vnode *vp;
1342 	struct uio *uiop;
1343 	int *iomode;
1344 	boolean_t pageprotected;
1345 	boolean_t *stalewriteverfp;
1346 {
1347 	u_int32_t *tl;
1348 	caddr_t cp;
1349 	int32_t t1, t2;
1350 	caddr_t bpos, dpos;
1351 	struct mbuf *mreq, *mrep, *md, *mb;
1352 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1353 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1354 	const int v3 = NFS_ISV3(vp);
1355 	int committed = NFSV3WRITE_FILESYNC;
1356 	struct nfsnode *np = VTONFS(vp);
1357 	struct nfs_writerpc_context ctx;
1358 	int s, byte_count;
1359 	struct lwp *l = NULL;
1360 	size_t origresid;
1361 #ifndef NFS_V2_ONLY
1362 	caddr_t cp2;
1363 	int rlen, commit;
1364 #endif
1365 
1366 	simple_lock_init(&ctx.nwc_slock);
1367 	ctx.nwc_mbufcount = 1;
1368 
1369 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1370 		panic("writerpc readonly vp %p", vp);
1371 	}
1372 
1373 #ifdef DIAGNOSTIC
1374 	if (uiop->uio_iovcnt != 1)
1375 		panic("nfs: writerpc iovcnt > 1");
1376 #endif
1377 	tsiz = uiop->uio_resid;
1378 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1379 		return (EFBIG);
1380 	if (pageprotected) {
1381 		l = curlwp;
1382 		PHOLD(l);
1383 	}
1384 retry:
1385 	origresid = uiop->uio_resid;
1386 	KASSERT(origresid == uiop->uio_iov->iov_len);
1387 	iostat_busy(nmp->nm_stats);
1388 	byte_count = 0; /* count of bytes actually written */
1389 	while (tsiz > 0) {
1390 		uint32_t datalen; /* data bytes need to be allocated in mbuf */
1391 		uint32_t backup;
1392 		boolean_t stalewriteverf = FALSE;
1393 
1394 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1395 		len = min(tsiz, nmp->nm_wsize);
1396 		datalen = pageprotected ? 0 : nfsm_rndup(len);
1397 		nfsm_reqhead(np, NFSPROC_WRITE,
1398 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1399 		nfsm_fhtom(np, v3);
1400 #ifndef NFS_V2_ONLY
1401 		if (v3) {
1402 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1403 			txdr_hyper(uiop->uio_offset, tl);
1404 			tl += 2;
1405 			*tl++ = txdr_unsigned(len);
1406 			*tl++ = txdr_unsigned(*iomode);
1407 			*tl = txdr_unsigned(len);
1408 		} else
1409 #endif
1410 		{
1411 			u_int32_t x;
1412 
1413 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1414 			/* Set both "begin" and "current" to non-garbage. */
1415 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1416 			*tl++ = x;      /* "begin offset" */
1417 			*tl++ = x;      /* "current offset" */
1418 			x = txdr_unsigned(len);
1419 			*tl++ = x;      /* total to this offset */
1420 			*tl = x;        /* size of this write */
1421 
1422 		}
1423 		if (pageprotected) {
1424 			/*
1425 			 * since we know pages can't be modified during i/o,
1426 			 * no need to copy them for us.
1427 			 */
1428 			struct mbuf *m;
1429 			struct iovec *iovp = uiop->uio_iov;
1430 
1431 			m = m_get(M_WAIT, MT_DATA);
1432 			MCLAIM(m, &nfs_mowner);
1433 			MEXTADD(m, iovp->iov_base, len, M_MBUF,
1434 			    nfs_writerpc_extfree, &ctx);
1435 			m->m_flags |= M_EXT_ROMAP;
1436 			m->m_len = len;
1437 			mb->m_next = m;
1438 			/*
1439 			 * no need to maintain mb and bpos here
1440 			 * because no one care them later.
1441 			 */
1442 #if 0
1443 			mb = m;
1444 			bpos = mtod(caddr_t, mb) + mb->m_len;
1445 #endif
1446 			UIO_ADVANCE(uiop, len);
1447 			uiop->uio_offset += len;
1448 			s = splvm();
1449 			simple_lock(&ctx.nwc_slock);
1450 			ctx.nwc_mbufcount++;
1451 			simple_unlock(&ctx.nwc_slock);
1452 			splx(s);
1453 			nfs_zeropad(mb, 0, nfsm_padlen(len));
1454 		} else {
1455 			nfsm_uiotom(uiop, len);
1456 		}
1457 		nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1458 #ifndef NFS_V2_ONLY
1459 		if (v3) {
1460 			wccflag = NFSV3_WCCCHK;
1461 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1462 			if (!error) {
1463 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1464 					+ NFSX_V3WRITEVERF);
1465 				rlen = fxdr_unsigned(int, *tl++);
1466 				if (rlen == 0) {
1467 					error = NFSERR_IO;
1468 					m_freem(mrep);
1469 					break;
1470 				} else if (rlen < len) {
1471 					backup = len - rlen;
1472 					UIO_ADVANCE(uiop, -backup);
1473 					uiop->uio_offset -= backup;
1474 					len = rlen;
1475 				}
1476 				commit = fxdr_unsigned(int, *tl++);
1477 
1478 				/*
1479 				 * Return the lowest committment level
1480 				 * obtained by any of the RPCs.
1481 				 */
1482 				if (committed == NFSV3WRITE_FILESYNC)
1483 					committed = commit;
1484 				else if (committed == NFSV3WRITE_DATASYNC &&
1485 					commit == NFSV3WRITE_UNSTABLE)
1486 					committed = commit;
1487 				simple_lock(&nmp->nm_slock);
1488 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1489 					memcpy(nmp->nm_writeverf, tl,
1490 					    NFSX_V3WRITEVERF);
1491 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1492 				} else if ((nmp->nm_iflag &
1493 				    NFSMNT_STALEWRITEVERF) ||
1494 				    memcmp(tl, nmp->nm_writeverf,
1495 				    NFSX_V3WRITEVERF)) {
1496 					memcpy(nmp->nm_writeverf, tl,
1497 					    NFSX_V3WRITEVERF);
1498 					/*
1499 					 * note NFSMNT_STALEWRITEVERF
1500 					 * if we're the first thread to
1501 					 * notice it.
1502 					 */
1503 					if ((nmp->nm_iflag &
1504 					    NFSMNT_STALEWRITEVERF) == 0) {
1505 						stalewriteverf = TRUE;
1506 						nmp->nm_iflag |=
1507 						    NFSMNT_STALEWRITEVERF;
1508 					}
1509 				}
1510 				simple_unlock(&nmp->nm_slock);
1511 			}
1512 		} else
1513 #endif
1514 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1515 		if (wccflag)
1516 			VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1517 		m_freem(mrep);
1518 		if (error)
1519 			break;
1520 		tsiz -= len;
1521 		byte_count += len;
1522 		if (stalewriteverf) {
1523 			*stalewriteverfp = TRUE;
1524 			stalewriteverf = FALSE;
1525 			if (committed == NFSV3WRITE_UNSTABLE &&
1526 			    len != origresid) {
1527 				/*
1528 				 * if our write requests weren't atomic but
1529 				 * unstable, datas in previous iterations
1530 				 * might have already been lost now.
1531 				 * then, we should resend them to nfsd.
1532 				 */
1533 				backup = origresid - tsiz;
1534 				UIO_ADVANCE(uiop, -backup);
1535 				uiop->uio_offset -= backup;
1536 				tsiz = origresid;
1537 				goto retry;
1538 			}
1539 		}
1540 	}
1541 nfsmout:
1542 	iostat_unbusy(nmp->nm_stats, byte_count, 0);
1543 	if (pageprotected) {
1544 		/*
1545 		 * wait until mbufs go away.
1546 		 * retransmitted mbufs can survive longer than rpc requests
1547 		 * themselves.
1548 		 */
1549 		s = splvm();
1550 		simple_lock(&ctx.nwc_slock);
1551 		ctx.nwc_mbufcount--;
1552 		while (ctx.nwc_mbufcount > 0) {
1553 			ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1554 		}
1555 		simple_unlock(&ctx.nwc_slock);
1556 		splx(s);
1557 		PRELE(l);
1558 	}
1559 	*iomode = committed;
1560 	if (error)
1561 		uiop->uio_resid = tsiz;
1562 	return (error);
1563 }
1564 
1565 /*
1566  * nfs mknod rpc
1567  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1568  * mode set to specify the file type and the size field for rdev.
1569  */
1570 int
1571 nfs_mknodrpc(dvp, vpp, cnp, vap)
1572 	struct vnode *dvp;
1573 	struct vnode **vpp;
1574 	struct componentname *cnp;
1575 	struct vattr *vap;
1576 {
1577 	struct nfsv2_sattr *sp;
1578 	u_int32_t *tl;
1579 	caddr_t cp;
1580 	int32_t t1, t2;
1581 	struct vnode *newvp = (struct vnode *)0;
1582 	struct nfsnode *dnp, *np;
1583 	char *cp2;
1584 	caddr_t bpos, dpos;
1585 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1586 	struct mbuf *mreq, *mrep, *md, *mb;
1587 	u_int32_t rdev;
1588 	const int v3 = NFS_ISV3(dvp);
1589 
1590 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1591 		rdev = txdr_unsigned(vap->va_rdev);
1592 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1593 		rdev = nfs_xdrneg1;
1594 	else {
1595 		VOP_ABORTOP(dvp, cnp);
1596 		vput(dvp);
1597 		return (EOPNOTSUPP);
1598 	}
1599 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1600 	dnp = VTONFS(dvp);
1601 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1602 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1603 	nfsm_fhtom(dnp, v3);
1604 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1605 #ifndef NFS_V2_ONLY
1606 	if (v3) {
1607 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1608 		*tl++ = vtonfsv3_type(vap->va_type);
1609 		nfsm_v3attrbuild(vap, FALSE);
1610 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1611 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1612 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1613 			*tl = txdr_unsigned(minor(vap->va_rdev));
1614 		}
1615 	} else
1616 #endif
1617 	{
1618 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1619 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1620 		sp->sa_uid = nfs_xdrneg1;
1621 		sp->sa_gid = nfs_xdrneg1;
1622 		sp->sa_size = rdev;
1623 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1624 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1625 	}
1626 	nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_lwp, cnp->cn_cred);
1627 	if (!error) {
1628 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1629 		if (!gotvp) {
1630 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1631 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np);
1632 			if (!error)
1633 				newvp = NFSTOV(np);
1634 		}
1635 	}
1636 #ifndef NFS_V2_ONLY
1637 	if (v3)
1638 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1639 #endif
1640 	nfsm_reqdone;
1641 	if (error) {
1642 		if (newvp)
1643 			vput(newvp);
1644 	} else {
1645 		if (cnp->cn_flags & MAKEENTRY)
1646 			nfs_cache_enter(dvp, newvp, cnp);
1647 		*vpp = newvp;
1648 	}
1649 	PNBUF_PUT(cnp->cn_pnbuf);
1650 	VTONFS(dvp)->n_flag |= NMODIFIED;
1651 	if (!wccflag)
1652 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1653 	vput(dvp);
1654 	return (error);
1655 }
1656 
1657 /*
1658  * nfs mknod vop
1659  * just call nfs_mknodrpc() to do the work.
1660  */
1661 /* ARGSUSED */
1662 int
1663 nfs_mknod(v)
1664 	void *v;
1665 {
1666 	struct vop_mknod_args /* {
1667 		struct vnode *a_dvp;
1668 		struct vnode **a_vpp;
1669 		struct componentname *a_cnp;
1670 		struct vattr *a_vap;
1671 	} */ *ap = v;
1672 	struct vnode *dvp = ap->a_dvp;
1673 	struct componentname *cnp = ap->a_cnp;
1674 	int error;
1675 
1676 	error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1677 	VN_KNOTE(dvp, NOTE_WRITE);
1678 	if (error == 0 || error == EEXIST)
1679 		cache_purge1(dvp, cnp, 0);
1680 	return (error);
1681 }
1682 
1683 #ifndef NFS_V2_ONLY
1684 static u_long create_verf;
1685 #endif
1686 /*
1687  * nfs file create call
1688  */
1689 int
1690 nfs_create(v)
1691 	void *v;
1692 {
1693 	struct vop_create_args /* {
1694 		struct vnode *a_dvp;
1695 		struct vnode **a_vpp;
1696 		struct componentname *a_cnp;
1697 		struct vattr *a_vap;
1698 	} */ *ap = v;
1699 	struct vnode *dvp = ap->a_dvp;
1700 	struct vattr *vap = ap->a_vap;
1701 	struct componentname *cnp = ap->a_cnp;
1702 	struct nfsv2_sattr *sp;
1703 	u_int32_t *tl;
1704 	caddr_t cp;
1705 	int32_t t1, t2;
1706 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1707 	struct vnode *newvp = (struct vnode *)0;
1708 	caddr_t bpos, dpos, cp2;
1709 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1710 	struct mbuf *mreq, *mrep, *md, *mb;
1711 	const int v3 = NFS_ISV3(dvp);
1712 
1713 	/*
1714 	 * Oops, not for me..
1715 	 */
1716 	if (vap->va_type == VSOCK)
1717 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1718 
1719 	KASSERT(vap->va_type == VREG);
1720 
1721 #ifdef VA_EXCLUSIVE
1722 	if (vap->va_vaflags & VA_EXCLUSIVE)
1723 		fmode |= O_EXCL;
1724 #endif
1725 again:
1726 	error = 0;
1727 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1728 	dnp = VTONFS(dvp);
1729 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1730 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1731 	nfsm_fhtom(dnp, v3);
1732 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1733 #ifndef NFS_V2_ONLY
1734 	if (v3) {
1735 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1736 		if (fmode & O_EXCL) {
1737 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1738 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1739 #ifdef INET
1740 			if (TAILQ_FIRST(&in_ifaddrhead))
1741 				*tl++ = TAILQ_FIRST(&in_ifaddrhead)->
1742 				    ia_addr.sin_addr.s_addr;
1743 			else
1744 				*tl++ = create_verf;
1745 #else
1746 			*tl++ = create_verf;
1747 #endif
1748 			*tl = ++create_verf;
1749 		} else {
1750 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1751 			nfsm_v3attrbuild(vap, FALSE);
1752 		}
1753 	} else
1754 #endif
1755 	{
1756 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1757 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1758 		sp->sa_uid = nfs_xdrneg1;
1759 		sp->sa_gid = nfs_xdrneg1;
1760 		sp->sa_size = 0;
1761 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1762 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1763 	}
1764 	nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_lwp, cnp->cn_cred);
1765 	if (!error) {
1766 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1767 		if (!gotvp) {
1768 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1769 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np);
1770 			if (!error)
1771 				newvp = NFSTOV(np);
1772 		}
1773 	}
1774 #ifndef NFS_V2_ONLY
1775 	if (v3)
1776 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1777 #endif
1778 	nfsm_reqdone;
1779 	if (error) {
1780 		/*
1781 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1782 		 */
1783 		if (v3 && (fmode & O_EXCL) && error == ENOTSUP) {
1784 			fmode &= ~O_EXCL;
1785 			goto again;
1786 		}
1787 	} else if (v3 && (fmode & O_EXCL)) {
1788 		struct timespec ts;
1789 
1790 		getnanotime(&ts);
1791 
1792 		/*
1793 		 * make sure that we'll update timestamps as
1794 		 * most server implementations use them to store
1795 		 * the create verifier.
1796 		 *
1797 		 * XXX it's better to use TOSERVER always.
1798 		 */
1799 
1800 		if (vap->va_atime.tv_sec == VNOVAL)
1801 			vap->va_atime = ts;
1802 		if (vap->va_mtime.tv_sec == VNOVAL)
1803 			vap->va_mtime = ts;
1804 
1805 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_lwp);
1806 	}
1807 	if (error == 0) {
1808 		if (cnp->cn_flags & MAKEENTRY)
1809 			nfs_cache_enter(dvp, newvp, cnp);
1810 		else
1811 			cache_purge1(dvp, cnp, 0);
1812 		*ap->a_vpp = newvp;
1813 	} else {
1814 		if (newvp)
1815 			vput(newvp);
1816 		if (error == EEXIST)
1817 			cache_purge1(dvp, cnp, 0);
1818 	}
1819 	PNBUF_PUT(cnp->cn_pnbuf);
1820 	VTONFS(dvp)->n_flag |= NMODIFIED;
1821 	if (!wccflag)
1822 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1823 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1824 	vput(dvp);
1825 	return (error);
1826 }
1827 
1828 /*
1829  * nfs file remove call
1830  * To try and make nfs semantics closer to ufs semantics, a file that has
1831  * other processes using the vnode is renamed instead of removed and then
1832  * removed later on the last close.
1833  * - If v_usecount > 1
1834  *	  If a rename is not already in the works
1835  *	     call nfs_sillyrename() to set it up
1836  *     else
1837  *	  do the remove rpc
1838  */
1839 int
1840 nfs_remove(v)
1841 	void *v;
1842 {
1843 	struct vop_remove_args /* {
1844 		struct vnodeop_desc *a_desc;
1845 		struct vnode * a_dvp;
1846 		struct vnode * a_vp;
1847 		struct componentname * a_cnp;
1848 	} */ *ap = v;
1849 	struct vnode *vp = ap->a_vp;
1850 	struct vnode *dvp = ap->a_dvp;
1851 	struct componentname *cnp = ap->a_cnp;
1852 	struct nfsnode *np = VTONFS(vp);
1853 	int error = 0;
1854 	struct vattr vattr;
1855 
1856 #ifndef DIAGNOSTIC
1857 	if ((cnp->cn_flags & HASBUF) == 0)
1858 		panic("nfs_remove: no name");
1859 	if (vp->v_usecount < 1)
1860 		panic("nfs_remove: bad v_usecount");
1861 #endif
1862 	if (vp->v_type == VDIR)
1863 		error = EPERM;
1864 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1865 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_lwp) == 0 &&
1866 	    vattr.va_nlink > 1)) {
1867 		/*
1868 		 * Purge the name cache so that the chance of a lookup for
1869 		 * the name succeeding while the remove is in progress is
1870 		 * minimized. Without node locking it can still happen, such
1871 		 * that an I/O op returns ESTALE, but since you get this if
1872 		 * another host removes the file..
1873 		 */
1874 		cache_purge(vp);
1875 		/*
1876 		 * throw away biocache buffers, mainly to avoid
1877 		 * unnecessary delayed writes later.
1878 		 */
1879 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_lwp, 1);
1880 		/* Do the rpc */
1881 		if (error != EINTR)
1882 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1883 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp);
1884 	} else if (!np->n_sillyrename)
1885 		error = nfs_sillyrename(dvp, vp, cnp, FALSE);
1886 	PNBUF_PUT(cnp->cn_pnbuf);
1887 	if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1888 	    vattr.va_nlink == 1) {
1889 		np->n_flag |= NREMOVED;
1890 	}
1891 	NFS_INVALIDATE_ATTRCACHE(np);
1892 	VN_KNOTE(vp, NOTE_DELETE);
1893 	VN_KNOTE(dvp, NOTE_WRITE);
1894 	if (dvp == vp)
1895 		vrele(vp);
1896 	else
1897 		vput(vp);
1898 	vput(dvp);
1899 	return (error);
1900 }
1901 
1902 /*
1903  * nfs file remove rpc called from nfs_inactive
1904  */
1905 int
1906 nfs_removeit(sp)
1907 	struct sillyrename *sp;
1908 {
1909 
1910 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1911 		(struct lwp *)0));
1912 }
1913 
1914 /*
1915  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1916  */
1917 int
1918 nfs_removerpc(dvp, name, namelen, cred, l)
1919 	struct vnode *dvp;
1920 	const char *name;
1921 	int namelen;
1922 	kauth_cred_t cred;
1923 	struct lwp *l;
1924 {
1925 	u_int32_t *tl;
1926 	caddr_t cp;
1927 #ifndef NFS_V2_ONLY
1928 	int32_t t1;
1929 	caddr_t cp2;
1930 #endif
1931 	int32_t t2;
1932 	caddr_t bpos, dpos;
1933 	int error = 0, wccflag = NFSV3_WCCRATTR;
1934 	struct mbuf *mreq, *mrep, *md, *mb;
1935 	const int v3 = NFS_ISV3(dvp);
1936 	int rexmit = 0;
1937 	struct nfsnode *dnp = VTONFS(dvp);
1938 
1939 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1940 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1941 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1942 	nfsm_fhtom(dnp, v3);
1943 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1944 	nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1945 #ifndef NFS_V2_ONLY
1946 	if (v3)
1947 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1948 #endif
1949 	nfsm_reqdone;
1950 	VTONFS(dvp)->n_flag |= NMODIFIED;
1951 	if (!wccflag)
1952 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1953 	/*
1954 	 * Kludge City: If the first reply to the remove rpc is lost..
1955 	 *   the reply to the retransmitted request will be ENOENT
1956 	 *   since the file was in fact removed
1957 	 *   Therefore, we cheat and return success.
1958 	 */
1959 	if (rexmit && error == ENOENT)
1960 		error = 0;
1961 	return (error);
1962 }
1963 
1964 /*
1965  * nfs file rename call
1966  */
1967 int
1968 nfs_rename(v)
1969 	void *v;
1970 {
1971 	struct vop_rename_args  /* {
1972 		struct vnode *a_fdvp;
1973 		struct vnode *a_fvp;
1974 		struct componentname *a_fcnp;
1975 		struct vnode *a_tdvp;
1976 		struct vnode *a_tvp;
1977 		struct componentname *a_tcnp;
1978 	} */ *ap = v;
1979 	struct vnode *fvp = ap->a_fvp;
1980 	struct vnode *tvp = ap->a_tvp;
1981 	struct vnode *fdvp = ap->a_fdvp;
1982 	struct vnode *tdvp = ap->a_tdvp;
1983 	struct componentname *tcnp = ap->a_tcnp;
1984 	struct componentname *fcnp = ap->a_fcnp;
1985 	int error;
1986 
1987 #ifndef DIAGNOSTIC
1988 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1989 	    (fcnp->cn_flags & HASBUF) == 0)
1990 		panic("nfs_rename: no name");
1991 #endif
1992 	/* Check for cross-device rename */
1993 	if ((fvp->v_mount != tdvp->v_mount) ||
1994 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1995 		error = EXDEV;
1996 		goto out;
1997 	}
1998 
1999 	/*
2000 	 * If the tvp exists and is in use, sillyrename it before doing the
2001 	 * rename of the new file over it.
2002 	 *
2003 	 * Have sillyrename use link instead of rename if possible,
2004 	 * so that we don't lose the file if the rename fails, and so
2005 	 * that there's no window when the "to" file doesn't exist.
2006 	 */
2007 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
2008 	    tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, TRUE)) {
2009 		VN_KNOTE(tvp, NOTE_DELETE);
2010 		vput(tvp);
2011 		tvp = NULL;
2012 	}
2013 
2014 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2015 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2016 		tcnp->cn_lwp);
2017 
2018 	VN_KNOTE(fdvp, NOTE_WRITE);
2019 	VN_KNOTE(tdvp, NOTE_WRITE);
2020 	if (error == 0 || error == EEXIST) {
2021 		if (fvp->v_type == VDIR)
2022 			cache_purge(fvp);
2023 		else
2024 			cache_purge1(fdvp, fcnp, 0);
2025 		if (tvp != NULL && tvp->v_type == VDIR)
2026 			cache_purge(tvp);
2027 		else
2028 			cache_purge1(tdvp, tcnp, 0);
2029 	}
2030 out:
2031 	if (tdvp == tvp)
2032 		vrele(tdvp);
2033 	else
2034 		vput(tdvp);
2035 	if (tvp)
2036 		vput(tvp);
2037 	vrele(fdvp);
2038 	vrele(fvp);
2039 	return (error);
2040 }
2041 
2042 /*
2043  * nfs file rename rpc called from nfs_remove() above
2044  */
2045 int
2046 nfs_renameit(sdvp, scnp, sp)
2047 	struct vnode *sdvp;
2048 	struct componentname *scnp;
2049 	struct sillyrename *sp;
2050 {
2051 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2052 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_lwp));
2053 }
2054 
2055 /*
2056  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2057  */
2058 int
2059 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l)
2060 	struct vnode *fdvp;
2061 	const char *fnameptr;
2062 	int fnamelen;
2063 	struct vnode *tdvp;
2064 	const char *tnameptr;
2065 	int tnamelen;
2066 	kauth_cred_t cred;
2067 	struct lwp *l;
2068 {
2069 	u_int32_t *tl;
2070 	caddr_t cp;
2071 #ifndef NFS_V2_ONLY
2072 	int32_t t1;
2073 	caddr_t cp2;
2074 #endif
2075 	int32_t t2;
2076 	caddr_t bpos, dpos;
2077 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2078 	struct mbuf *mreq, *mrep, *md, *mb;
2079 	const int v3 = NFS_ISV3(fdvp);
2080 	int rexmit = 0;
2081 	struct nfsnode *fdnp = VTONFS(fdvp);
2082 
2083 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2084 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
2085 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2086 		nfsm_rndup(tnamelen));
2087 	nfsm_fhtom(fdnp, v3);
2088 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2089 	nfsm_fhtom(VTONFS(tdvp), v3);
2090 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2091 	nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
2092 #ifndef NFS_V2_ONLY
2093 	if (v3) {
2094 		nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2095 		nfsm_wcc_data(tdvp, twccflag, 0, !error);
2096 	}
2097 #endif
2098 	nfsm_reqdone;
2099 	VTONFS(fdvp)->n_flag |= NMODIFIED;
2100 	VTONFS(tdvp)->n_flag |= NMODIFIED;
2101 	if (!fwccflag)
2102 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2103 	if (!twccflag)
2104 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2105 	/*
2106 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2107 	 */
2108 	if (rexmit && error == ENOENT)
2109 		error = 0;
2110 	return (error);
2111 }
2112 
2113 /*
2114  * NFS link RPC, called from nfs_link.
2115  * Assumes dvp and vp locked, and leaves them that way.
2116  */
2117 
2118 static int
2119 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2120     size_t namelen, kauth_cred_t cred, struct lwp *l)
2121 {
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 	const int v3 = NFS_ISV3(dvp);
2133 	int rexmit = 0;
2134 	struct nfsnode *np = VTONFS(vp);
2135 
2136 	nfsstats.rpccnt[NFSPROC_LINK]++;
2137 	nfsm_reqhead(np, NFSPROC_LINK,
2138 	    NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2139 	nfsm_fhtom(np, v3);
2140 	nfsm_fhtom(VTONFS(dvp), v3);
2141 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2142 	nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2143 #ifndef NFS_V2_ONLY
2144 	if (v3) {
2145 		nfsm_postop_attr(vp, attrflag, 0);
2146 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2147 	}
2148 #endif
2149 	nfsm_reqdone;
2150 
2151 	VTONFS(dvp)->n_flag |= NMODIFIED;
2152 	if (!attrflag)
2153 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2154 	if (!wccflag)
2155 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2156 
2157 	/*
2158 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2159 	 */
2160 	if (rexmit && error == EEXIST)
2161 		error = 0;
2162 
2163 	return error;
2164 }
2165 
2166 /*
2167  * nfs hard link create call
2168  */
2169 int
2170 nfs_link(v)
2171 	void *v;
2172 {
2173 	struct vop_link_args /* {
2174 		struct vnode *a_dvp;
2175 		struct vnode *a_vp;
2176 		struct componentname *a_cnp;
2177 	} */ *ap = v;
2178 	struct vnode *vp = ap->a_vp;
2179 	struct vnode *dvp = ap->a_dvp;
2180 	struct componentname *cnp = ap->a_cnp;
2181 	int error = 0;
2182 
2183 	if (dvp->v_mount != vp->v_mount) {
2184 		VOP_ABORTOP(dvp, cnp);
2185 		vput(dvp);
2186 		return (EXDEV);
2187 	}
2188 	if (dvp != vp) {
2189 		error = vn_lock(vp, LK_EXCLUSIVE);
2190 		if (error != 0) {
2191 			VOP_ABORTOP(dvp, cnp);
2192 			vput(dvp);
2193 			return error;
2194 		}
2195 	}
2196 
2197 	/*
2198 	 * Push all writes to the server, so that the attribute cache
2199 	 * doesn't get "out of sync" with the server.
2200 	 * XXX There should be a better way!
2201 	 */
2202 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_lwp);
2203 
2204 	error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2205 	    cnp->cn_cred, cnp->cn_lwp);
2206 
2207 	if (error == 0)
2208 		cache_purge1(dvp, cnp, 0);
2209 	PNBUF_PUT(cnp->cn_pnbuf);
2210 	if (dvp != vp)
2211 		VOP_UNLOCK(vp, 0);
2212 	VN_KNOTE(vp, NOTE_LINK);
2213 	VN_KNOTE(dvp, NOTE_WRITE);
2214 	vput(dvp);
2215 	return (error);
2216 }
2217 
2218 /*
2219  * nfs symbolic link create call
2220  */
2221 int
2222 nfs_symlink(v)
2223 	void *v;
2224 {
2225 	struct vop_symlink_args /* {
2226 		struct vnode *a_dvp;
2227 		struct vnode **a_vpp;
2228 		struct componentname *a_cnp;
2229 		struct vattr *a_vap;
2230 		char *a_target;
2231 	} */ *ap = v;
2232 	struct vnode *dvp = ap->a_dvp;
2233 	struct vattr *vap = ap->a_vap;
2234 	struct componentname *cnp = ap->a_cnp;
2235 	struct nfsv2_sattr *sp;
2236 	u_int32_t *tl;
2237 	caddr_t cp;
2238 	int32_t t1, t2;
2239 	caddr_t bpos, dpos, cp2;
2240 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2241 	struct mbuf *mreq, *mrep, *md, *mb;
2242 	struct vnode *newvp = (struct vnode *)0;
2243 	const int v3 = NFS_ISV3(dvp);
2244 	int rexmit = 0;
2245 	struct nfsnode *dnp = VTONFS(dvp);
2246 
2247 	*ap->a_vpp = NULL;
2248 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2249 	slen = strlen(ap->a_target);
2250 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2251 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2252 	nfsm_fhtom(dnp, v3);
2253 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2254 #ifndef NFS_V2_ONlY
2255 	if (v3)
2256 		nfsm_v3attrbuild(vap, FALSE);
2257 #endif
2258 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2259 #ifndef NFS_V2_ONlY
2260 	if (!v3) {
2261 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2262 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2263 		sp->sa_uid = nfs_xdrneg1;
2264 		sp->sa_gid = nfs_xdrneg1;
2265 		sp->sa_size = nfs_xdrneg1;
2266 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2267 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2268 	}
2269 #endif
2270 	nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_lwp, cnp->cn_cred,
2271 	    &rexmit);
2272 #ifndef NFS_V2_ONlY
2273 	if (v3) {
2274 		if (!error)
2275 			nfsm_mtofh(dvp, newvp, v3, gotvp);
2276 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2277 	}
2278 #endif
2279 	nfsm_reqdone;
2280 	/*
2281 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2282 	 */
2283 	if (rexmit && error == EEXIST)
2284 		error = 0;
2285 	if (error == 0 || error == EEXIST)
2286 		cache_purge1(dvp, cnp, 0);
2287 	if (error == 0 && newvp == NULL) {
2288 		struct nfsnode *np = NULL;
2289 
2290 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2291 		    cnp->cn_cred, cnp->cn_lwp, &np);
2292 		if (error == 0)
2293 			newvp = NFSTOV(np);
2294 	}
2295 	if (error) {
2296 		if (newvp != NULL)
2297 			vput(newvp);
2298 	} else {
2299 		*ap->a_vpp = newvp;
2300 	}
2301 	PNBUF_PUT(cnp->cn_pnbuf);
2302 	VTONFS(dvp)->n_flag |= NMODIFIED;
2303 	if (!wccflag)
2304 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2305 	VN_KNOTE(dvp, NOTE_WRITE);
2306 	vput(dvp);
2307 	return (error);
2308 }
2309 
2310 /*
2311  * nfs make dir call
2312  */
2313 int
2314 nfs_mkdir(v)
2315 	void *v;
2316 {
2317 	struct vop_mkdir_args /* {
2318 		struct vnode *a_dvp;
2319 		struct vnode **a_vpp;
2320 		struct componentname *a_cnp;
2321 		struct vattr *a_vap;
2322 	} */ *ap = v;
2323 	struct vnode *dvp = ap->a_dvp;
2324 	struct vattr *vap = ap->a_vap;
2325 	struct componentname *cnp = ap->a_cnp;
2326 	struct nfsv2_sattr *sp;
2327 	u_int32_t *tl;
2328 	caddr_t cp;
2329 	int32_t t1, t2;
2330 	int len;
2331 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2332 	struct vnode *newvp = (struct vnode *)0;
2333 	caddr_t bpos, dpos, cp2;
2334 	int error = 0, wccflag = NFSV3_WCCRATTR;
2335 	int gotvp = 0;
2336 	int rexmit = 0;
2337 	struct mbuf *mreq, *mrep, *md, *mb;
2338 	const int v3 = NFS_ISV3(dvp);
2339 
2340 	len = cnp->cn_namelen;
2341 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2342 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
2343 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2344 	nfsm_fhtom(dnp, v3);
2345 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2346 #ifndef NFS_V2_ONLY
2347 	if (v3) {
2348 		nfsm_v3attrbuild(vap, FALSE);
2349 	} else
2350 #endif
2351 	{
2352 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2353 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2354 		sp->sa_uid = nfs_xdrneg1;
2355 		sp->sa_gid = nfs_xdrneg1;
2356 		sp->sa_size = nfs_xdrneg1;
2357 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2358 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2359 	}
2360 	nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit);
2361 	if (!error)
2362 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2363 	if (v3)
2364 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2365 	nfsm_reqdone;
2366 	VTONFS(dvp)->n_flag |= NMODIFIED;
2367 	if (!wccflag)
2368 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2369 	/*
2370 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2371 	 * if we can succeed in looking up the directory.
2372 	 */
2373 	if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2374 		if (newvp) {
2375 			vput(newvp);
2376 			newvp = (struct vnode *)0;
2377 		}
2378 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2379 			cnp->cn_lwp, &np);
2380 		if (!error) {
2381 			newvp = NFSTOV(np);
2382 			if (newvp->v_type != VDIR || newvp == dvp)
2383 				error = EEXIST;
2384 		}
2385 	}
2386 	if (error) {
2387 		if (newvp) {
2388 			if (dvp != newvp)
2389 				vput(newvp);
2390 			else
2391 				vrele(newvp);
2392 		}
2393 	} else {
2394 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2395 		if (cnp->cn_flags & MAKEENTRY)
2396 			nfs_cache_enter(dvp, newvp, cnp);
2397 		*ap->a_vpp = newvp;
2398 	}
2399 	PNBUF_PUT(cnp->cn_pnbuf);
2400 	vput(dvp);
2401 	return (error);
2402 }
2403 
2404 /*
2405  * nfs remove directory call
2406  */
2407 int
2408 nfs_rmdir(v)
2409 	void *v;
2410 {
2411 	struct vop_rmdir_args /* {
2412 		struct vnode *a_dvp;
2413 		struct vnode *a_vp;
2414 		struct componentname *a_cnp;
2415 	} */ *ap = v;
2416 	struct vnode *vp = ap->a_vp;
2417 	struct vnode *dvp = ap->a_dvp;
2418 	struct componentname *cnp = ap->a_cnp;
2419 	u_int32_t *tl;
2420 	caddr_t cp;
2421 #ifndef NFS_V2_ONLY
2422 	int32_t t1;
2423 	caddr_t cp2;
2424 #endif
2425 	int32_t t2;
2426 	caddr_t bpos, dpos;
2427 	int error = 0, wccflag = NFSV3_WCCRATTR;
2428 	int rexmit = 0;
2429 	struct mbuf *mreq, *mrep, *md, *mb;
2430 	const int v3 = NFS_ISV3(dvp);
2431 	struct nfsnode *dnp;
2432 
2433 	if (dvp == vp) {
2434 		vrele(dvp);
2435 		vput(dvp);
2436 		PNBUF_PUT(cnp->cn_pnbuf);
2437 		return (EINVAL);
2438 	}
2439 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2440 	dnp = VTONFS(dvp);
2441 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2442 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2443 	nfsm_fhtom(dnp, v3);
2444 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2445 	nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit);
2446 #ifndef NFS_V2_ONLY
2447 	if (v3)
2448 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2449 #endif
2450 	nfsm_reqdone;
2451 	PNBUF_PUT(cnp->cn_pnbuf);
2452 	VTONFS(dvp)->n_flag |= NMODIFIED;
2453 	if (!wccflag)
2454 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2455 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2456 	VN_KNOTE(vp, NOTE_DELETE);
2457 	cache_purge(vp);
2458 	vput(vp);
2459 	vput(dvp);
2460 	/*
2461 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2462 	 */
2463 	if (rexmit && error == ENOENT)
2464 		error = 0;
2465 	return (error);
2466 }
2467 
2468 /*
2469  * nfs readdir call
2470  */
2471 int
2472 nfs_readdir(v)
2473 	void *v;
2474 {
2475 	struct vop_readdir_args /* {
2476 		struct vnode *a_vp;
2477 		struct uio *a_uio;
2478 		kauth_cred_t a_cred;
2479 		int *a_eofflag;
2480 		off_t **a_cookies;
2481 		int *a_ncookies;
2482 	} */ *ap = v;
2483 	struct vnode *vp = ap->a_vp;
2484 	struct uio *uio = ap->a_uio;
2485 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2486 	char *base = uio->uio_iov->iov_base;
2487 	int tresid, error;
2488 	size_t count, lost;
2489 	struct dirent *dp;
2490 	off_t *cookies = NULL;
2491 	int ncookies = 0, nc;
2492 
2493 	if (vp->v_type != VDIR)
2494 		return (EPERM);
2495 
2496 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2497 	count = uio->uio_resid - lost;
2498 	if (count <= 0)
2499 		return (EINVAL);
2500 
2501 	/*
2502 	 * Call nfs_bioread() to do the real work.
2503 	 */
2504 	tresid = uio->uio_resid = count;
2505 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2506 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2507 
2508 	if (!error && ap->a_cookies) {
2509 		ncookies = count / 16;
2510 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2511 		*ap->a_cookies = cookies;
2512 	}
2513 
2514 	if (!error && uio->uio_resid == tresid) {
2515 		uio->uio_resid += lost;
2516 		nfsstats.direofcache_misses++;
2517 		if (ap->a_cookies)
2518 			*ap->a_ncookies = 0;
2519 		*ap->a_eofflag = 1;
2520 		return (0);
2521 	}
2522 
2523 	if (!error && ap->a_cookies) {
2524 		/*
2525 		 * Only the NFS server and emulations use cookies, and they
2526 		 * load the directory block into system space, so we can
2527 		 * just look at it directly.
2528 		 */
2529 		if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2530 		    uio->uio_iovcnt != 1)
2531 			panic("nfs_readdir: lost in space");
2532 		for (nc = 0; ncookies-- &&
2533 		     base < (char *)uio->uio_iov->iov_base; nc++){
2534 			dp = (struct dirent *) base;
2535 			if (dp->d_reclen == 0)
2536 				break;
2537 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2538 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2539 			else
2540 				*(cookies++) = NFS_GETCOOKIE(dp);
2541 			base += dp->d_reclen;
2542 		}
2543 		uio->uio_resid +=
2544 		    ((caddr_t)uio->uio_iov->iov_base - base);
2545 		uio->uio_iov->iov_len +=
2546 		    ((caddr_t)uio->uio_iov->iov_base - base);
2547 		uio->uio_iov->iov_base = base;
2548 		*ap->a_ncookies = nc;
2549 	}
2550 
2551 	uio->uio_resid += lost;
2552 	*ap->a_eofflag = 0;
2553 	return (error);
2554 }
2555 
2556 /*
2557  * Readdir rpc call.
2558  * Called from below the buffer cache by nfs_doio().
2559  */
2560 int
2561 nfs_readdirrpc(vp, uiop, cred)
2562 	struct vnode *vp;
2563 	struct uio *uiop;
2564 	kauth_cred_t cred;
2565 {
2566 	int len, left;
2567 	struct dirent *dp = NULL;
2568 	u_int32_t *tl;
2569 	caddr_t cp;
2570 	int32_t t1, t2;
2571 	caddr_t bpos, dpos, cp2;
2572 	struct mbuf *mreq, *mrep, *md, *mb;
2573 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2574 	struct nfsnode *dnp = VTONFS(vp);
2575 	u_quad_t fileno;
2576 	int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2577 #ifndef NFS_V2_ONLY
2578 	int attrflag;
2579 #endif
2580 	int nrpcs = 0, reclen;
2581 	const int v3 = NFS_ISV3(vp);
2582 
2583 #ifdef DIAGNOSTIC
2584 	/*
2585 	 * Should be called from buffer cache, so only amount of
2586 	 * NFS_DIRBLKSIZ will be requested.
2587 	 */
2588 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2589 		panic("nfs readdirrpc bad uio");
2590 #endif
2591 
2592 	/*
2593 	 * Loop around doing readdir rpc's of size nm_readdirsize
2594 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2595 	 * The stopping criteria is EOF or buffer full.
2596 	 */
2597 	while (more_dirs && bigenough) {
2598 		/*
2599 		 * Heuristic: don't bother to do another RPC to further
2600 		 * fill up this block if there is not much room left. (< 50%
2601 		 * of the readdir RPC size). This wastes some buffer space
2602 		 * but can save up to 50% in RPC calls.
2603 		 */
2604 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2605 			bigenough = 0;
2606 			break;
2607 		}
2608 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2609 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2610 			NFSX_READDIR(v3));
2611 		nfsm_fhtom(dnp, v3);
2612 #ifndef NFS_V2_ONLY
2613 		if (v3) {
2614 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2615 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2616 				txdr_swapcookie3(uiop->uio_offset, tl);
2617 			} else {
2618 				txdr_cookie3(uiop->uio_offset, tl);
2619 			}
2620 			tl += 2;
2621 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2622 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2623 		} else
2624 #endif
2625 		{
2626 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2627 			*tl++ = txdr_unsigned(uiop->uio_offset);
2628 		}
2629 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2630 		nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2631 		nrpcs++;
2632 #ifndef NFS_V2_ONLY
2633 		if (v3) {
2634 			nfsm_postop_attr(vp, attrflag, 0);
2635 			if (!error) {
2636 				nfsm_dissect(tl, u_int32_t *,
2637 				    2 * NFSX_UNSIGNED);
2638 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2639 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2640 			} else {
2641 				m_freem(mrep);
2642 				goto nfsmout;
2643 			}
2644 		}
2645 #endif
2646 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2647 		more_dirs = fxdr_unsigned(int, *tl);
2648 
2649 		/* loop thru the dir entries, doctoring them to 4bsd form */
2650 		while (more_dirs && bigenough) {
2651 #ifndef NFS_V2_ONLY
2652 			if (v3) {
2653 				nfsm_dissect(tl, u_int32_t *,
2654 				    3 * NFSX_UNSIGNED);
2655 				fileno = fxdr_hyper(tl);
2656 				len = fxdr_unsigned(int, *(tl + 2));
2657 			} else
2658 #endif
2659 			{
2660 				nfsm_dissect(tl, u_int32_t *,
2661 				    2 * NFSX_UNSIGNED);
2662 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2663 				len = fxdr_unsigned(int, *tl);
2664 			}
2665 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2666 				error = EBADRPC;
2667 				m_freem(mrep);
2668 				goto nfsmout;
2669 			}
2670 			/* for cookie stashing */
2671 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2672 			left = NFS_DIRFRAGSIZ - blksiz;
2673 			if (reclen > left) {
2674 				memset(uiop->uio_iov->iov_base, 0, left);
2675 				dp->d_reclen += left;
2676 				UIO_ADVANCE(uiop, left);
2677 				blksiz = 0;
2678 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2679 			}
2680 			if (reclen > uiop->uio_resid)
2681 				bigenough = 0;
2682 			if (bigenough) {
2683 				int tlen;
2684 
2685 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2686 				dp->d_fileno = fileno;
2687 				dp->d_namlen = len;
2688 				dp->d_reclen = reclen;
2689 				dp->d_type = DT_UNKNOWN;
2690 				blksiz += reclen;
2691 				if (blksiz == NFS_DIRFRAGSIZ)
2692 					blksiz = 0;
2693 				UIO_ADVANCE(uiop, DIRHDSIZ);
2694 				nfsm_mtouio(uiop, len);
2695 				tlen = reclen - (DIRHDSIZ + len);
2696 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2697 				UIO_ADVANCE(uiop, tlen);
2698 			} else
2699 				nfsm_adv(nfsm_rndup(len));
2700 #ifndef NFS_V2_ONLY
2701 			if (v3) {
2702 				nfsm_dissect(tl, u_int32_t *,
2703 				    3 * NFSX_UNSIGNED);
2704 			} else
2705 #endif
2706 			{
2707 				nfsm_dissect(tl, u_int32_t *,
2708 				    2 * NFSX_UNSIGNED);
2709 			}
2710 			if (bigenough) {
2711 #ifndef NFS_V2_ONLY
2712 				if (v3) {
2713 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2714 						uiop->uio_offset =
2715 						    fxdr_swapcookie3(tl);
2716 					else
2717 						uiop->uio_offset =
2718 						    fxdr_cookie3(tl);
2719 				}
2720 				else
2721 #endif
2722 				{
2723 					uiop->uio_offset =
2724 					    fxdr_unsigned(off_t, *tl);
2725 				}
2726 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2727 			}
2728 			if (v3)
2729 				tl += 2;
2730 			else
2731 				tl++;
2732 			more_dirs = fxdr_unsigned(int, *tl);
2733 		}
2734 		/*
2735 		 * If at end of rpc data, get the eof boolean
2736 		 */
2737 		if (!more_dirs) {
2738 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2739 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2740 
2741 			/*
2742 			 * kludge: if we got no entries, treat it as EOF.
2743 			 * some server sometimes send a reply without any
2744 			 * entries or EOF.
2745 			 * although it might mean the server has very long name,
2746 			 * we can't handle such entries anyway.
2747 			 */
2748 
2749 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2750 				more_dirs = 0;
2751 		}
2752 		m_freem(mrep);
2753 	}
2754 	/*
2755 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2756 	 * by increasing d_reclen for the last record.
2757 	 */
2758 	if (blksiz > 0) {
2759 		left = NFS_DIRFRAGSIZ - blksiz;
2760 		memset(uiop->uio_iov->iov_base, 0, left);
2761 		dp->d_reclen += left;
2762 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2763 		UIO_ADVANCE(uiop, left);
2764 	}
2765 
2766 	/*
2767 	 * We are now either at the end of the directory or have filled the
2768 	 * block.
2769 	 */
2770 	if (bigenough) {
2771 		dnp->n_direofoffset = uiop->uio_offset;
2772 		dnp->n_flag |= NEOFVALID;
2773 	}
2774 nfsmout:
2775 	return (error);
2776 }
2777 
2778 #ifndef NFS_V2_ONLY
2779 /*
2780  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2781  */
2782 int
2783 nfs_readdirplusrpc(vp, uiop, cred)
2784 	struct vnode *vp;
2785 	struct uio *uiop;
2786 	kauth_cred_t cred;
2787 {
2788 	int len, left;
2789 	struct dirent *dp = NULL;
2790 	u_int32_t *tl;
2791 	caddr_t cp;
2792 	int32_t t1, t2;
2793 	struct vnode *newvp;
2794 	caddr_t bpos, dpos, cp2;
2795 	struct mbuf *mreq, *mrep, *md, *mb;
2796 	struct nameidata nami, *ndp = &nami;
2797 	struct componentname *cnp = &ndp->ni_cnd;
2798 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2799 	struct nfsnode *dnp = VTONFS(vp), *np;
2800 	nfsfh_t *fhp;
2801 	u_quad_t fileno;
2802 	int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2803 	int attrflag, fhsize, nrpcs = 0, reclen;
2804 	struct nfs_fattr fattr, *fp;
2805 
2806 #ifdef DIAGNOSTIC
2807 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2808 		panic("nfs readdirplusrpc bad uio");
2809 #endif
2810 	ndp->ni_dvp = vp;
2811 	newvp = NULLVP;
2812 
2813 	/*
2814 	 * Loop around doing readdir rpc's of size nm_readdirsize
2815 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2816 	 * The stopping criteria is EOF or buffer full.
2817 	 */
2818 	while (more_dirs && bigenough) {
2819 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2820 			bigenough = 0;
2821 			break;
2822 		}
2823 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2824 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2825 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2826 		nfsm_fhtom(dnp, 1);
2827  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2828 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2829 			txdr_swapcookie3(uiop->uio_offset, tl);
2830 		} else {
2831 			txdr_cookie3(uiop->uio_offset, tl);
2832 		}
2833 		tl += 2;
2834 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2835 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2836 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2837 		*tl = txdr_unsigned(nmp->nm_rsize);
2838 		nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2839 		nfsm_postop_attr(vp, attrflag, 0);
2840 		if (error) {
2841 			m_freem(mrep);
2842 			goto nfsmout;
2843 		}
2844 		nrpcs++;
2845 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2846 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2847 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2848 		more_dirs = fxdr_unsigned(int, *tl);
2849 
2850 		/* loop thru the dir entries, doctoring them to 4bsd form */
2851 		while (more_dirs && bigenough) {
2852 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2853 			fileno = fxdr_hyper(tl);
2854 			len = fxdr_unsigned(int, *(tl + 2));
2855 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2856 				error = EBADRPC;
2857 				m_freem(mrep);
2858 				goto nfsmout;
2859 			}
2860 			/* for cookie stashing */
2861 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2862 			left = NFS_DIRFRAGSIZ - blksiz;
2863 			if (reclen > left) {
2864 				/*
2865 				 * DIRFRAGSIZ is aligned, no need to align
2866 				 * again here.
2867 				 */
2868 				memset(uiop->uio_iov->iov_base, 0, left);
2869 				dp->d_reclen += left;
2870 				UIO_ADVANCE(uiop, left);
2871 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2872 				blksiz = 0;
2873 			}
2874 			if (reclen > uiop->uio_resid)
2875 				bigenough = 0;
2876 			if (bigenough) {
2877 				int tlen;
2878 
2879 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2880 				dp->d_fileno = fileno;
2881 				dp->d_namlen = len;
2882 				dp->d_reclen = reclen;
2883 				dp->d_type = DT_UNKNOWN;
2884 				blksiz += reclen;
2885 				if (blksiz == NFS_DIRFRAGSIZ)
2886 					blksiz = 0;
2887 				UIO_ADVANCE(uiop, DIRHDSIZ);
2888 				nfsm_mtouio(uiop, len);
2889 				tlen = reclen - (DIRHDSIZ + len);
2890 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2891 				UIO_ADVANCE(uiop, tlen);
2892 				cnp->cn_nameptr = dp->d_name;
2893 				cnp->cn_namelen = dp->d_namlen;
2894 			} else
2895 				nfsm_adv(nfsm_rndup(len));
2896 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2897 			if (bigenough) {
2898 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2899 					uiop->uio_offset =
2900 						fxdr_swapcookie3(tl);
2901 				else
2902 					uiop->uio_offset =
2903 						fxdr_cookie3(tl);
2904 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2905 			}
2906 			tl += 2;
2907 
2908 			/*
2909 			 * Since the attributes are before the file handle
2910 			 * (sigh), we must skip over the attributes and then
2911 			 * come back and get them.
2912 			 */
2913 			attrflag = fxdr_unsigned(int, *tl);
2914 			if (attrflag) {
2915 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2916 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2917 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2918 			    doit = fxdr_unsigned(int, *tl);
2919 			    if (doit) {
2920 				nfsm_getfh(fhp, fhsize, 1);
2921 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2922 				    VREF(vp);
2923 				    newvp = vp;
2924 				    np = dnp;
2925 				} else {
2926 				    error = nfs_nget1(vp->v_mount, fhp,
2927 					fhsize, &np, LK_NOWAIT);
2928 				    if (!error)
2929 					newvp = NFSTOV(np);
2930 				}
2931 				if (!error) {
2932 				    const char *xcp;
2933 
2934 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2935 				    if (bigenough) {
2936 					dp->d_type =
2937 					   IFTODT(VTTOIF(np->n_vattr->va_type));
2938 					if (cnp->cn_namelen <= NCHNAMLEN) {
2939 					    ndp->ni_vp = newvp;
2940 					    xcp = cnp->cn_nameptr +
2941 						cnp->cn_namelen;
2942 					    cnp->cn_hash =
2943 					       namei_hash(cnp->cn_nameptr, &xcp);
2944 					    nfs_cache_enter(ndp->ni_dvp,
2945 						ndp->ni_vp, cnp);
2946 					}
2947 				    }
2948 				}
2949 				error = 0;
2950 			   }
2951 			} else {
2952 			    /* Just skip over the file handle */
2953 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2954 			    i = fxdr_unsigned(int, *tl);
2955 			    nfsm_adv(nfsm_rndup(i));
2956 			}
2957 			if (newvp != NULLVP) {
2958 			    if (newvp == vp)
2959 				vrele(newvp);
2960 			    else
2961 				vput(newvp);
2962 			    newvp = NULLVP;
2963 			}
2964 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2965 			more_dirs = fxdr_unsigned(int, *tl);
2966 		}
2967 		/*
2968 		 * If at end of rpc data, get the eof boolean
2969 		 */
2970 		if (!more_dirs) {
2971 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2972 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2973 
2974 			/*
2975 			 * kludge: see a comment in nfs_readdirrpc.
2976 			 */
2977 
2978 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2979 				more_dirs = 0;
2980 		}
2981 		m_freem(mrep);
2982 	}
2983 	/*
2984 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2985 	 * by increasing d_reclen for the last record.
2986 	 */
2987 	if (blksiz > 0) {
2988 		left = NFS_DIRFRAGSIZ - blksiz;
2989 		memset(uiop->uio_iov->iov_base, 0, left);
2990 		dp->d_reclen += left;
2991 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2992 		UIO_ADVANCE(uiop, left);
2993 	}
2994 
2995 	/*
2996 	 * We are now either at the end of the directory or have filled the
2997 	 * block.
2998 	 */
2999 	if (bigenough) {
3000 		dnp->n_direofoffset = uiop->uio_offset;
3001 		dnp->n_flag |= NEOFVALID;
3002 	}
3003 nfsmout:
3004 	if (newvp != NULLVP) {
3005 		if(newvp == vp)
3006 		    vrele(newvp);
3007 		else
3008 		    vput(newvp);
3009 	}
3010 	return (error);
3011 }
3012 #endif
3013 
3014 /*
3015  * Silly rename. To make the NFS filesystem that is stateless look a little
3016  * more like the "ufs" a remove of an active vnode is translated to a rename
3017  * to a funny looking filename that is removed by nfs_inactive on the
3018  * nfsnode. There is the potential for another process on a different client
3019  * to create the same funny name between the nfs_lookitup() fails and the
3020  * nfs_rename() completes, but...
3021  */
3022 int
3023 nfs_sillyrename(dvp, vp, cnp, dolink)
3024 	struct vnode *dvp, *vp;
3025 	struct componentname *cnp;
3026 	boolean_t dolink;
3027 {
3028 	struct sillyrename *sp;
3029 	struct nfsnode *np;
3030 	int error;
3031 	short pid;
3032 
3033 	cache_purge(dvp);
3034 	np = VTONFS(vp);
3035 #ifndef DIAGNOSTIC
3036 	if (vp->v_type == VDIR)
3037 		panic("nfs: sillyrename dir");
3038 #endif
3039 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
3040 		M_NFSREQ, M_WAITOK);
3041 	sp->s_cred = kauth_cred_dup(cnp->cn_cred);
3042 	sp->s_dvp = dvp;
3043 	VREF(dvp);
3044 
3045 	/* Fudge together a funny name */
3046 	pid = cnp->cn_lwp->l_proc->p_pid;
3047 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
3048 	sp->s_namlen = 12;
3049 	sp->s_name[8] = hexdigits[pid & 0xf];
3050 	sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
3051 	sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
3052 	sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
3053 
3054 	/* Try lookitups until we get one that isn't there */
3055 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3056 		cnp->cn_lwp, (struct nfsnode **)0) == 0) {
3057 		sp->s_name[4]++;
3058 		if (sp->s_name[4] > 'z') {
3059 			error = EINVAL;
3060 			goto bad;
3061 		}
3062 	}
3063 	if (dolink) {
3064 		error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
3065 		    sp->s_cred, cnp->cn_lwp);
3066 		/*
3067 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
3068 		 */
3069 		if (error == ENOTSUP) {
3070 			error = nfs_renameit(dvp, cnp, sp);
3071 		}
3072 	} else {
3073 		error = nfs_renameit(dvp, cnp, sp);
3074 	}
3075 	if (error)
3076 		goto bad;
3077 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3078 		cnp->cn_lwp, &np);
3079 	np->n_sillyrename = sp;
3080 	return (0);
3081 bad:
3082 	vrele(sp->s_dvp);
3083 	kauth_cred_free(sp->s_cred);
3084 	free((caddr_t)sp, M_NFSREQ);
3085 	return (error);
3086 }
3087 
3088 /*
3089  * Look up a file name and optionally either update the file handle or
3090  * allocate an nfsnode, depending on the value of npp.
3091  * npp == NULL	--> just do the lookup
3092  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3093  *			handled too
3094  * *npp != NULL --> update the file handle in the vnode
3095  */
3096 int
3097 nfs_lookitup(dvp, name, len, cred, l, npp)
3098 	struct vnode *dvp;
3099 	const char *name;
3100 	int len;
3101 	kauth_cred_t cred;
3102 	struct lwp *l;
3103 	struct nfsnode **npp;
3104 {
3105 	u_int32_t *tl;
3106 	caddr_t cp;
3107 	int32_t t1, t2;
3108 	struct vnode *newvp = (struct vnode *)0;
3109 	struct nfsnode *np, *dnp = VTONFS(dvp);
3110 	caddr_t bpos, dpos, cp2;
3111 	int error = 0, fhlen;
3112 #ifndef NFS_V2_ONLY
3113 	int attrflag;
3114 #endif
3115 	struct mbuf *mreq, *mrep, *md, *mb;
3116 	nfsfh_t *nfhp;
3117 	const int v3 = NFS_ISV3(dvp);
3118 
3119 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3120 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3121 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3122 	nfsm_fhtom(dnp, v3);
3123 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
3124 	nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
3125 	if (npp && !error) {
3126 		nfsm_getfh(nfhp, fhlen, v3);
3127 		if (*npp) {
3128 		    np = *npp;
3129 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3130 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
3131 			np->n_fhp = &np->n_fh;
3132 		    }
3133 #if NFS_SMALLFH < NFSX_V3FHMAX
3134 		    else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
3135 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
3136 #endif
3137 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
3138 		    np->n_fhsize = fhlen;
3139 		    newvp = NFSTOV(np);
3140 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3141 		    VREF(dvp);
3142 		    newvp = dvp;
3143 		    np = dnp;
3144 		} else {
3145 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3146 		    if (error) {
3147 			m_freem(mrep);
3148 			return (error);
3149 		    }
3150 		    newvp = NFSTOV(np);
3151 		}
3152 #ifndef NFS_V2_ONLY
3153 		if (v3) {
3154 			nfsm_postop_attr(newvp, attrflag, 0);
3155 			if (!attrflag && *npp == NULL) {
3156 				m_freem(mrep);
3157 				vput(newvp);
3158 				return (ENOENT);
3159 			}
3160 		} else
3161 #endif
3162 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
3163 	}
3164 	nfsm_reqdone;
3165 	if (npp && *npp == NULL) {
3166 		if (error) {
3167 			if (newvp)
3168 				vput(newvp);
3169 		} else
3170 			*npp = np;
3171 	}
3172 	return (error);
3173 }
3174 
3175 #ifndef NFS_V2_ONLY
3176 /*
3177  * Nfs Version 3 commit rpc
3178  */
3179 int
3180 nfs_commit(vp, offset, cnt, l)
3181 	struct vnode *vp;
3182 	off_t offset;
3183 	uint32_t cnt;
3184 	struct lwp *l;
3185 {
3186 	caddr_t cp;
3187 	u_int32_t *tl;
3188 	int32_t t1, t2;
3189 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3190 	caddr_t bpos, dpos, cp2;
3191 	int error = 0, wccflag = NFSV3_WCCRATTR;
3192 	struct mbuf *mreq, *mrep, *md, *mb;
3193 	struct nfsnode *np;
3194 
3195 	KASSERT(NFS_ISV3(vp));
3196 
3197 #ifdef NFS_DEBUG_COMMIT
3198 	printf("commit %lu - %lu\n", (unsigned long)offset,
3199 	    (unsigned long)(offset + cnt));
3200 #endif
3201 
3202 	simple_lock(&nmp->nm_slock);
3203 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3204 		simple_unlock(&nmp->nm_slock);
3205 		return (0);
3206 	}
3207 	simple_unlock(&nmp->nm_slock);
3208 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3209 	np = VTONFS(vp);
3210 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3211 	nfsm_fhtom(np, 1);
3212 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3213 	txdr_hyper(offset, tl);
3214 	tl += 2;
3215 	*tl = txdr_unsigned(cnt);
3216 	nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3217 	nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, FALSE);
3218 	if (!error) {
3219 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3220 		simple_lock(&nmp->nm_slock);
3221 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3222 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3223 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3224 			error = NFSERR_STALEWRITEVERF;
3225 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3226 		}
3227 		simple_unlock(&nmp->nm_slock);
3228 	}
3229 	nfsm_reqdone;
3230 	return (error);
3231 }
3232 #endif
3233 
3234 /*
3235  * Kludge City..
3236  * - make nfs_bmap() essentially a no-op that does no translation
3237  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3238  *   (Maybe I could use the process's page mapping, but I was concerned that
3239  *    Kernel Write might not be enabled and also figured copyout() would do
3240  *    a lot more work than memcpy() and also it currently happens in the
3241  *    context of the swapper process (2).
3242  */
3243 int
3244 nfs_bmap(v)
3245 	void *v;
3246 {
3247 	struct vop_bmap_args /* {
3248 		struct vnode *a_vp;
3249 		daddr_t  a_bn;
3250 		struct vnode **a_vpp;
3251 		daddr_t *a_bnp;
3252 		int *a_runp;
3253 	} */ *ap = v;
3254 	struct vnode *vp = ap->a_vp;
3255 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3256 
3257 	if (ap->a_vpp != NULL)
3258 		*ap->a_vpp = vp;
3259 	if (ap->a_bnp != NULL)
3260 		*ap->a_bnp = ap->a_bn << bshift;
3261 	if (ap->a_runp != NULL)
3262 		*ap->a_runp = 1024 * 1024; /* XXX */
3263 	return (0);
3264 }
3265 
3266 /*
3267  * Strategy routine.
3268  * For async requests when nfsiod(s) are running, queue the request by
3269  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3270  * request.
3271  */
3272 int
3273 nfs_strategy(v)
3274 	void *v;
3275 {
3276 	struct vop_strategy_args *ap = v;
3277 	struct buf *bp = ap->a_bp;
3278 	int error = 0;
3279 
3280 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3281 		panic("nfs physio/async");
3282 
3283 	/*
3284 	 * If the op is asynchronous and an i/o daemon is waiting
3285 	 * queue the request, wake it up and wait for completion
3286 	 * otherwise just do it ourselves.
3287 	 */
3288 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3289 		error = nfs_doio(bp);
3290 	return (error);
3291 }
3292 
3293 /*
3294  * fsync vnode op. Just call nfs_flush() with commit == 1.
3295  */
3296 /* ARGSUSED */
3297 int
3298 nfs_fsync(v)
3299 	void *v;
3300 {
3301 	struct vop_fsync_args /* {
3302 		struct vnodeop_desc *a_desc;
3303 		struct vnode * a_vp;
3304 		kauth_cred_t  a_cred;
3305 		int  a_flags;
3306 		off_t offlo;
3307 		off_t offhi;
3308 		struct lwp * a_l;
3309 	} */ *ap = v;
3310 
3311 	struct vnode *vp = ap->a_vp;
3312 
3313 	if (vp->v_type != VREG)
3314 		return 0;
3315 
3316 	return (nfs_flush(vp, ap->a_cred,
3317 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_l, 1));
3318 }
3319 
3320 /*
3321  * Flush all the data associated with a vnode.
3322  */
3323 int
3324 nfs_flush(vp, cred, waitfor, l, commit)
3325 	struct vnode *vp;
3326 	kauth_cred_t cred;
3327 	int waitfor;
3328 	struct lwp *l;
3329 	int commit;
3330 {
3331 	struct nfsnode *np = VTONFS(vp);
3332 	int error;
3333 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3334 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3335 
3336 	simple_lock(&vp->v_interlock);
3337 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3338 	if (np->n_flag & NWRITEERR) {
3339 		error = np->n_error;
3340 		np->n_flag &= ~NWRITEERR;
3341 	}
3342 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3343 	return (error);
3344 }
3345 
3346 /*
3347  * Return POSIX pathconf information applicable to nfs.
3348  *
3349  * N.B. The NFS V2 protocol doesn't support this RPC.
3350  */
3351 /* ARGSUSED */
3352 int
3353 nfs_pathconf(v)
3354 	void *v;
3355 {
3356 	struct vop_pathconf_args /* {
3357 		struct vnode *a_vp;
3358 		int a_name;
3359 		register_t *a_retval;
3360 	} */ *ap = v;
3361 	struct nfsv3_pathconf *pcp;
3362 	struct vnode *vp = ap->a_vp;
3363 	struct mbuf *mreq, *mrep, *md, *mb;
3364 	int32_t t1, t2;
3365 	u_int32_t *tl;
3366 	caddr_t bpos, dpos, cp, cp2;
3367 	int error = 0, attrflag;
3368 #ifndef NFS_V2_ONLY
3369 	struct nfsmount *nmp;
3370 	unsigned int l;
3371 	u_int64_t maxsize;
3372 #endif
3373 	const int v3 = NFS_ISV3(vp);
3374 	struct nfsnode *np = VTONFS(vp);
3375 
3376 	switch (ap->a_name) {
3377 		/* Names that can be resolved locally. */
3378 	case _PC_PIPE_BUF:
3379 		*ap->a_retval = PIPE_BUF;
3380 		break;
3381 	case _PC_SYNC_IO:
3382 		*ap->a_retval = 1;
3383 		break;
3384 	/* Names that cannot be resolved locally; do an RPC, if possible. */
3385 	case _PC_LINK_MAX:
3386 	case _PC_NAME_MAX:
3387 	case _PC_CHOWN_RESTRICTED:
3388 	case _PC_NO_TRUNC:
3389 		if (!v3) {
3390 			error = EINVAL;
3391 			break;
3392 		}
3393 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3394 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3395 		nfsm_fhtom(np, 1);
3396 		nfsm_request(np, NFSPROC_PATHCONF,
3397 		    curlwp, curlwp->l_proc->p_cred);	/* XXX */
3398 		nfsm_postop_attr(vp, attrflag, 0);
3399 		if (!error) {
3400 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
3401 			    NFSX_V3PATHCONF);
3402 			switch (ap->a_name) {
3403 			case _PC_LINK_MAX:
3404 				*ap->a_retval =
3405 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3406 				break;
3407 			case _PC_NAME_MAX:
3408 				*ap->a_retval =
3409 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3410 				break;
3411 			case _PC_CHOWN_RESTRICTED:
3412 				*ap->a_retval =
3413 				    (pcp->pc_chownrestricted == nfs_true);
3414 				break;
3415 			case _PC_NO_TRUNC:
3416 				*ap->a_retval =
3417 				    (pcp->pc_notrunc == nfs_true);
3418 				break;
3419 			}
3420 		}
3421 		nfsm_reqdone;
3422 		break;
3423 	case _PC_FILESIZEBITS:
3424 #ifndef NFS_V2_ONLY
3425 		if (v3) {
3426 			nmp = VFSTONFS(vp->v_mount);
3427 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3428 				if ((error = nfs_fsinfo(nmp, vp,
3429 				    curproc->p_cred, curlwp)) != 0) /* XXX */
3430 					break;
3431 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3432 			    (maxsize >> l) > 0; l++)
3433 				;
3434 			*ap->a_retval = l + 1;
3435 		} else
3436 #endif
3437 		{
3438 			*ap->a_retval = 32;	/* NFS V2 limitation */
3439 		}
3440 		break;
3441 	default:
3442 		error = EINVAL;
3443 		break;
3444 	}
3445 
3446 	return (error);
3447 }
3448 
3449 /*
3450  * NFS advisory byte-level locks.
3451  */
3452 int
3453 nfs_advlock(v)
3454 	void *v;
3455 {
3456 	struct vop_advlock_args /* {
3457 		struct vnode *a_vp;
3458 		caddr_t  a_id;
3459 		int  a_op;
3460 		struct flock *a_fl;
3461 		int  a_flags;
3462 	} */ *ap = v;
3463 	struct nfsnode *np = VTONFS(ap->a_vp);
3464 
3465 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3466 }
3467 
3468 /*
3469  * Print out the contents of an nfsnode.
3470  */
3471 int
3472 nfs_print(v)
3473 	void *v;
3474 {
3475 	struct vop_print_args /* {
3476 		struct vnode *a_vp;
3477 	} */ *ap = v;
3478 	struct vnode *vp = ap->a_vp;
3479 	struct nfsnode *np = VTONFS(vp);
3480 
3481 	printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3482 	    (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3483 	if (vp->v_type == VFIFO)
3484 		fifo_printinfo(vp);
3485 	printf("\n");
3486 	return (0);
3487 }
3488 
3489 /*
3490  * nfs unlock wrapper.
3491  */
3492 int
3493 nfs_unlock(void *v)
3494 {
3495 	struct vop_unlock_args /* {
3496 		struct vnode *a_vp;
3497 		int a_flags;
3498 	} */ *ap = v;
3499 	struct vnode *vp = ap->a_vp;
3500 
3501 	/*
3502 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3503 	 * with v_data == 0.
3504 	 */
3505 	if (VTONFS(vp)) {
3506 		nfs_delayedtruncate(vp);
3507 	}
3508 
3509 	return genfs_unlock(v);
3510 }
3511 
3512 /*
3513  * nfs special file access vnode op.
3514  * Essentially just get vattr and then imitate iaccess() since the device is
3515  * local to the client.
3516  */
3517 int
3518 nfsspec_access(v)
3519 	void *v;
3520 {
3521 	struct vop_access_args /* {
3522 		struct vnode *a_vp;
3523 		int  a_mode;
3524 		kauth_cred_t a_cred;
3525 		struct lwp *a_l;
3526 	} */ *ap = v;
3527 	struct vattr va;
3528 	struct vnode *vp = ap->a_vp;
3529 	int error;
3530 
3531 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_l);
3532 	if (error)
3533 		return (error);
3534 
3535         /*
3536 	 * Disallow write attempts on filesystems mounted read-only;
3537 	 * unless the file is a socket, fifo, or a block or character
3538 	 * device resident on the filesystem.
3539 	 */
3540 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3541 		switch (vp->v_type) {
3542 		case VREG:
3543 		case VDIR:
3544 		case VLNK:
3545 			return (EROFS);
3546 		default:
3547 			break;
3548 		}
3549 	}
3550 
3551 	return (vaccess(va.va_type, va.va_mode,
3552 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3553 }
3554 
3555 /*
3556  * Read wrapper for special devices.
3557  */
3558 int
3559 nfsspec_read(v)
3560 	void *v;
3561 {
3562 	struct vop_read_args /* {
3563 		struct vnode *a_vp;
3564 		struct uio *a_uio;
3565 		int  a_ioflag;
3566 		kauth_cred_t a_cred;
3567 	} */ *ap = v;
3568 	struct nfsnode *np = VTONFS(ap->a_vp);
3569 
3570 	/*
3571 	 * Set access flag.
3572 	 */
3573 	np->n_flag |= NACC;
3574 	getnanotime(&np->n_atim);
3575 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3576 }
3577 
3578 /*
3579  * Write wrapper for special devices.
3580  */
3581 int
3582 nfsspec_write(v)
3583 	void *v;
3584 {
3585 	struct vop_write_args /* {
3586 		struct vnode *a_vp;
3587 		struct uio *a_uio;
3588 		int  a_ioflag;
3589 		kauth_cred_t a_cred;
3590 	} */ *ap = v;
3591 	struct nfsnode *np = VTONFS(ap->a_vp);
3592 
3593 	/*
3594 	 * Set update flag.
3595 	 */
3596 	np->n_flag |= NUPD;
3597 	getnanotime(&np->n_mtim);
3598 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3599 }
3600 
3601 /*
3602  * Close wrapper for special devices.
3603  *
3604  * Update the times on the nfsnode then do device close.
3605  */
3606 int
3607 nfsspec_close(v)
3608 	void *v;
3609 {
3610 	struct vop_close_args /* {
3611 		struct vnode *a_vp;
3612 		int  a_fflag;
3613 		kauth_cred_t a_cred;
3614 		struct lwp *a_l;
3615 	} */ *ap = v;
3616 	struct vnode *vp = ap->a_vp;
3617 	struct nfsnode *np = VTONFS(vp);
3618 	struct vattr vattr;
3619 
3620 	if (np->n_flag & (NACC | NUPD)) {
3621 		np->n_flag |= NCHG;
3622 		if (vp->v_usecount == 1 &&
3623 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3624 			VATTR_NULL(&vattr);
3625 			if (np->n_flag & NACC)
3626 				vattr.va_atime = np->n_atim;
3627 			if (np->n_flag & NUPD)
3628 				vattr.va_mtime = np->n_mtim;
3629 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l);
3630 		}
3631 	}
3632 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3633 }
3634 
3635 /*
3636  * Read wrapper for fifos.
3637  */
3638 int
3639 nfsfifo_read(v)
3640 	void *v;
3641 {
3642 	struct vop_read_args /* {
3643 		struct vnode *a_vp;
3644 		struct uio *a_uio;
3645 		int  a_ioflag;
3646 		kauth_cred_t a_cred;
3647 	} */ *ap = v;
3648 	struct nfsnode *np = VTONFS(ap->a_vp);
3649 
3650 	/*
3651 	 * Set access flag.
3652 	 */
3653 	np->n_flag |= NACC;
3654 	getnanotime(&np->n_atim);
3655 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3656 }
3657 
3658 /*
3659  * Write wrapper for fifos.
3660  */
3661 int
3662 nfsfifo_write(v)
3663 	void *v;
3664 {
3665 	struct vop_write_args /* {
3666 		struct vnode *a_vp;
3667 		struct uio *a_uio;
3668 		int  a_ioflag;
3669 		kauth_cred_t a_cred;
3670 	} */ *ap = v;
3671 	struct nfsnode *np = VTONFS(ap->a_vp);
3672 
3673 	/*
3674 	 * Set update flag.
3675 	 */
3676 	np->n_flag |= NUPD;
3677 	getnanotime(&np->n_mtim);
3678 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3679 }
3680 
3681 /*
3682  * Close wrapper for fifos.
3683  *
3684  * Update the times on the nfsnode then do fifo close.
3685  */
3686 int
3687 nfsfifo_close(v)
3688 	void *v;
3689 {
3690 	struct vop_close_args /* {
3691 		struct vnode *a_vp;
3692 		int  a_fflag;
3693 		kauth_cred_t a_cred;
3694 		struct lwp *a_l;
3695 	} */ *ap = v;
3696 	struct vnode *vp = ap->a_vp;
3697 	struct nfsnode *np = VTONFS(vp);
3698 	struct vattr vattr;
3699 
3700 	if (np->n_flag & (NACC | NUPD)) {
3701 		struct timespec ts;
3702 
3703 		getnanotime(&ts);
3704 		if (np->n_flag & NACC)
3705 			np->n_atim = ts;
3706 		if (np->n_flag & NUPD)
3707 			np->n_mtim = ts;
3708 		np->n_flag |= NCHG;
3709 		if (vp->v_usecount == 1 &&
3710 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3711 			VATTR_NULL(&vattr);
3712 			if (np->n_flag & NACC)
3713 				vattr.va_atime = np->n_atim;
3714 			if (np->n_flag & NUPD)
3715 				vattr.va_mtime = np->n_mtim;
3716 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l);
3717 		}
3718 	}
3719 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3720 }
3721