xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1 /*	$NetBSD: nfs_vnops.c,v 1.269 2009/01/11 02:45:54 christos 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 1.269 2009/01/11 02:45:54 christos Exp $");
43 
44 #ifdef _KERNEL_OPT
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 #endif
48 
49 #include <sys/param.h>
50 #include <sys/proc.h>
51 #include <sys/kernel.h>
52 #include <sys/systm.h>
53 #include <sys/resourcevar.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/condvar.h>
57 #include <sys/disk.h>
58 #include <sys/malloc.h>
59 #include <sys/kmem.h>
60 #include <sys/mbuf.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/vnode.h>
64 #include <sys/dirent.h>
65 #include <sys/fcntl.h>
66 #include <sys/hash.h>
67 #include <sys/lockf.h>
68 #include <sys/stat.h>
69 #include <sys/unistd.h>
70 #include <sys/kauth.h>
71 
72 #include <uvm/uvm_extern.h>
73 #include <uvm/uvm.h>
74 
75 #include <miscfs/fifofs/fifo.h>
76 #include <miscfs/genfs/genfs.h>
77 #include <miscfs/genfs/genfs_node.h>
78 #include <miscfs/specfs/specdev.h>
79 
80 #include <nfs/rpcv2.h>
81 #include <nfs/nfsproto.h>
82 #include <nfs/nfs.h>
83 #include <nfs/nfsnode.h>
84 #include <nfs/nfsmount.h>
85 #include <nfs/xdr_subs.h>
86 #include <nfs/nfsm_subs.h>
87 #include <nfs/nfs_var.h>
88 
89 #include <net/if.h>
90 #include <netinet/in.h>
91 #include <netinet/in_var.h>
92 
93 /*
94  * Global vfs data structures for nfs
95  */
96 int (**nfsv2_vnodeop_p) __P((void *));
97 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
98 	{ &vop_default_desc, vn_default_error },
99 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
100 	{ &vop_create_desc, nfs_create },		/* create */
101 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
102 	{ &vop_open_desc, nfs_open },			/* open */
103 	{ &vop_close_desc, nfs_close },			/* close */
104 	{ &vop_access_desc, nfs_access },		/* access */
105 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
106 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
107 	{ &vop_read_desc, nfs_read },			/* read */
108 	{ &vop_write_desc, nfs_write },			/* write */
109 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
110 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
111 	{ &vop_poll_desc, nfs_poll },			/* poll */
112 	{ &vop_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
113 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
114 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
115 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
116 	{ &vop_seek_desc, nfs_seek },			/* seek */
117 	{ &vop_remove_desc, nfs_remove },		/* remove */
118 	{ &vop_link_desc, nfs_link },			/* link */
119 	{ &vop_rename_desc, nfs_rename },		/* rename */
120 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
121 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
122 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
123 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
124 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
125 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
126 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
127 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
128 	{ &vop_lock_desc, nfs_lock },			/* lock */
129 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
130 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
131 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
132 	{ &vop_print_desc, nfs_print },			/* print */
133 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
134 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
135 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
136 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
137 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
138 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
139 	{ NULL, NULL }
140 };
141 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
142 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
143 
144 /*
145  * Special device vnode ops
146  */
147 int (**spec_nfsv2nodeop_p) __P((void *));
148 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
149 	{ &vop_default_desc, vn_default_error },
150 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
151 	{ &vop_create_desc, spec_create },		/* create */
152 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
153 	{ &vop_open_desc, spec_open },			/* open */
154 	{ &vop_close_desc, nfsspec_close },		/* close */
155 	{ &vop_access_desc, nfsspec_access },		/* access */
156 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
157 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
158 	{ &vop_read_desc, nfsspec_read },		/* read */
159 	{ &vop_write_desc, nfsspec_write },		/* write */
160 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
161 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
162 	{ &vop_poll_desc, spec_poll },			/* poll */
163 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
164 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
165 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
166 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
167 	{ &vop_seek_desc, spec_seek },			/* seek */
168 	{ &vop_remove_desc, spec_remove },		/* remove */
169 	{ &vop_link_desc, spec_link },			/* link */
170 	{ &vop_rename_desc, spec_rename },		/* rename */
171 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
172 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
173 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
174 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
175 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
176 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
177 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
178 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
179 	{ &vop_lock_desc, nfs_lock },			/* lock */
180 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
181 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
182 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
183 	{ &vop_print_desc, nfs_print },			/* print */
184 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
185 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
186 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
187 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
188 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
189 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
190 	{ NULL, NULL }
191 };
192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
193 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
194 
195 int (**fifo_nfsv2nodeop_p) __P((void *));
196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
197 	{ &vop_default_desc, vn_default_error },
198 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
199 	{ &vop_create_desc, fifo_create },		/* create */
200 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
201 	{ &vop_open_desc, fifo_open },			/* open */
202 	{ &vop_close_desc, nfsfifo_close },		/* close */
203 	{ &vop_access_desc, nfsspec_access },		/* access */
204 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
205 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
206 	{ &vop_read_desc, nfsfifo_read },		/* read */
207 	{ &vop_write_desc, nfsfifo_write },		/* write */
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 *, void *, 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 	char *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 	} */ *ap = v;
316 	struct vnode *vp = ap->a_vp;
317 #ifndef NFS_V2_ONLY
318 	u_int32_t *tl;
319 	char *cp;
320 	int32_t t1, t2;
321 	char *bpos, *dpos, *cp2;
322 	int error = 0, attrflag;
323 	struct mbuf *mreq, *mrep, *md, *mb;
324 	u_int32_t mode, rmode;
325 	const int v3 = NFS_ISV3(vp);
326 #endif
327 	int cachevalid;
328 	struct nfsnode *np = VTONFS(vp);
329 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
330 
331 	cachevalid = (np->n_accstamp != -1 &&
332 	    (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
333 	    np->n_accuid == kauth_cred_geteuid(ap->a_cred));
334 
335 	/*
336 	 * Check access cache first. If this request has been made for this
337 	 * uid shortly before, use the cached result.
338 	 */
339 	if (cachevalid) {
340 		if (!np->n_accerror) {
341 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
342 				return np->n_accerror;
343 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
344 			return np->n_accerror;
345 	}
346 
347 #ifndef NFS_V2_ONLY
348 	/*
349 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
350 	 * ufs_access() locally using the vattr. This may not be correct,
351 	 * since the server may apply other access criteria such as
352 	 * client uid-->server uid mapping that we do not know about, but
353 	 * this is better than just returning anything that is lying about
354 	 * in the cache.
355 	 */
356 	if (v3) {
357 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
358 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
359 		nfsm_fhtom(np, v3);
360 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
361 		if (ap->a_mode & VREAD)
362 			mode = NFSV3ACCESS_READ;
363 		else
364 			mode = 0;
365 		if (vp->v_type != VDIR) {
366 			if (ap->a_mode & VWRITE)
367 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
368 			if (ap->a_mode & VEXEC)
369 				mode |= NFSV3ACCESS_EXECUTE;
370 		} else {
371 			if (ap->a_mode & VWRITE)
372 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
373 					 NFSV3ACCESS_DELETE);
374 			if (ap->a_mode & VEXEC)
375 				mode |= NFSV3ACCESS_LOOKUP;
376 		}
377 		*tl = txdr_unsigned(mode);
378 		nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
379 		nfsm_postop_attr(vp, attrflag, 0);
380 		if (!error) {
381 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
382 			rmode = fxdr_unsigned(u_int32_t, *tl);
383 			/*
384 			 * The NFS V3 spec does not clarify whether or not
385 			 * the returned access bits can be a superset of
386 			 * the ones requested, so...
387 			 */
388 			if ((rmode & mode) != mode)
389 				error = EACCES;
390 		}
391 		nfsm_reqdone;
392 	} else
393 #endif
394 		return (nfsspec_access(ap));
395 #ifndef NFS_V2_ONLY
396 	/*
397 	 * Disallow write attempts on filesystems mounted read-only;
398 	 * unless the file is a socket, fifo, or a block or character
399 	 * device resident on the filesystem.
400 	 */
401 	if (!error && (ap->a_mode & VWRITE) &&
402 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
403 		switch (vp->v_type) {
404 		case VREG:
405 		case VDIR:
406 		case VLNK:
407 			error = EROFS;
408 		default:
409 			break;
410 		}
411 	}
412 
413 	if (!error || error == EACCES) {
414 		/*
415 		 * If we got the same result as for a previous,
416 		 * different request, OR it in. Don't update
417 		 * the timestamp in that case.
418 		 */
419 		if (cachevalid && np->n_accstamp != -1 &&
420 		    error == np->n_accerror) {
421 			if (!error)
422 				np->n_accmode |= ap->a_mode;
423 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
424 				np->n_accmode = ap->a_mode;
425 		} else {
426 			np->n_accstamp = time_uptime;
427 			np->n_accuid = kauth_cred_geteuid(ap->a_cred);
428 			np->n_accmode = ap->a_mode;
429 			np->n_accerror = error;
430 		}
431 	}
432 
433 	return (error);
434 #endif
435 }
436 
437 /*
438  * nfs open vnode op
439  * Check to see if the type is ok
440  * and that deletion is not in progress.
441  * For paged in text files, you will need to flush the page cache
442  * if consistency is lost.
443  */
444 /* ARGSUSED */
445 int
446 nfs_open(v)
447 	void *v;
448 {
449 	struct vop_open_args /* {
450 		struct vnode *a_vp;
451 		int  a_mode;
452 		kauth_cred_t a_cred;
453 	} */ *ap = v;
454 	struct vnode *vp = ap->a_vp;
455 	struct nfsnode *np = VTONFS(vp);
456 	int error;
457 
458 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
459 		return (EACCES);
460 	}
461 
462 	if (ap->a_mode & FREAD) {
463 		if (np->n_rcred != NULL)
464 			kauth_cred_free(np->n_rcred);
465 		np->n_rcred = ap->a_cred;
466 		kauth_cred_hold(np->n_rcred);
467 	}
468 	if (ap->a_mode & FWRITE) {
469 		if (np->n_wcred != NULL)
470 			kauth_cred_free(np->n_wcred);
471 		np->n_wcred = ap->a_cred;
472 		kauth_cred_hold(np->n_wcred);
473 	}
474 
475 	error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
476 	if (error)
477 		return error;
478 
479 	NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
480 
481 	return (0);
482 }
483 
484 /*
485  * nfs close vnode op
486  * What an NFS client should do upon close after writing is a debatable issue.
487  * Most NFS clients push delayed writes to the server upon close, basically for
488  * two reasons:
489  * 1 - So that any write errors may be reported back to the client process
490  *     doing the close system call. By far the two most likely errors are
491  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
492  * 2 - To put a worst case upper bound on cache inconsistency between
493  *     multiple clients for the file.
494  * There is also a consistency problem for Version 2 of the protocol w.r.t.
495  * not being able to tell if other clients are writing a file concurrently,
496  * since there is no way of knowing if the changed modify time in the reply
497  * is only due to the write for this client.
498  * (NFS Version 3 provides weak cache consistency data in the reply that
499  *  should be sufficient to detect and handle this case.)
500  *
501  * The current code does the following:
502  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
503  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
504  *                     or commit them (this satisfies 1 and 2 except for the
505  *                     case where the server crashes after this close but
506  *                     before the commit RPC, which is felt to be "good
507  *                     enough". Changing the last argument to nfs_flush() to
508  *                     a 1 would force a commit operation, if it is felt a
509  *                     commit is necessary now.
510  */
511 /* ARGSUSED */
512 int
513 nfs_close(v)
514 	void *v;
515 {
516 	struct vop_close_args /* {
517 		struct vnodeop_desc *a_desc;
518 		struct vnode *a_vp;
519 		int  a_fflag;
520 		kauth_cred_t a_cred;
521 	} */ *ap = v;
522 	struct vnode *vp = ap->a_vp;
523 	struct nfsnode *np = VTONFS(vp);
524 	int error = 0;
525 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
526 
527 	if (vp->v_type == VREG) {
528 	    if (np->n_flag & NMODIFIED) {
529 #ifndef NFS_V2_ONLY
530 		if (NFS_ISV3(vp)) {
531 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
532 		    np->n_flag &= ~NMODIFIED;
533 		} else
534 #endif
535 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
536 		NFS_INVALIDATE_ATTRCACHE(np);
537 	    }
538 	    if (np->n_flag & NWRITEERR) {
539 		np->n_flag &= ~NWRITEERR;
540 		error = np->n_error;
541 	    }
542 	}
543 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
544 	return (error);
545 }
546 
547 /*
548  * nfs getattr call from vfs.
549  */
550 int
551 nfs_getattr(v)
552 	void *v;
553 {
554 	struct vop_getattr_args /* {
555 		struct vnode *a_vp;
556 		struct vattr *a_vap;
557 		kauth_cred_t a_cred;
558 	} */ *ap = v;
559 	struct vnode *vp = ap->a_vp;
560 	struct nfsnode *np = VTONFS(vp);
561 	char *cp;
562 	u_int32_t *tl;
563 	int32_t t1, t2;
564 	char *bpos, *dpos;
565 	int error = 0;
566 	struct mbuf *mreq, *mrep, *md, *mb;
567 	const int v3 = NFS_ISV3(vp);
568 
569 	/*
570 	 * Update local times for special files.
571 	 */
572 	if (np->n_flag & (NACC | NUPD))
573 		np->n_flag |= NCHG;
574 
575 	/*
576 	 * if we have delayed truncation, do it now.
577 	 */
578 	nfs_delayedtruncate(vp);
579 
580 	/*
581 	 * First look in the cache.
582 	 */
583 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
584 		return (0);
585 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
586 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
587 	nfsm_fhtom(np, v3);
588 	nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
589 	if (!error) {
590 		nfsm_loadattr(vp, ap->a_vap, 0);
591 		if (vp->v_type == VDIR &&
592 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
593 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
594 	}
595 	nfsm_reqdone;
596 	return (error);
597 }
598 
599 /*
600  * nfs setattr call.
601  */
602 int
603 nfs_setattr(v)
604 	void *v;
605 {
606 	struct vop_setattr_args /* {
607 		struct vnodeop_desc *a_desc;
608 		struct vnode *a_vp;
609 		struct vattr *a_vap;
610 		kauth_cred_t a_cred;
611 	} */ *ap = v;
612 	struct vnode *vp = ap->a_vp;
613 	struct nfsnode *np = VTONFS(vp);
614 	struct vattr *vap = ap->a_vap;
615 	int error = 0;
616 	u_quad_t tsize = 0;
617 
618 	/*
619 	 * Setting of flags is not supported.
620 	 */
621 	if (vap->va_flags != VNOVAL)
622 		return (EOPNOTSUPP);
623 
624 	/*
625 	 * Disallow write attempts if the filesystem is mounted read-only.
626 	 */
627   	if ((vap->va_uid != (uid_t)VNOVAL ||
628 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
629 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
630 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
631 		return (EROFS);
632 	if (vap->va_size != VNOVAL) {
633 		if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
634 			return EFBIG;
635 		}
636  		switch (vp->v_type) {
637  		case VDIR:
638  			return (EISDIR);
639  		case VCHR:
640  		case VBLK:
641  		case VSOCK:
642  		case VFIFO:
643 			if (vap->va_mtime.tv_sec == VNOVAL &&
644 			    vap->va_atime.tv_sec == VNOVAL &&
645 			    vap->va_mode == (mode_t)VNOVAL &&
646 			    vap->va_uid == (uid_t)VNOVAL &&
647 			    vap->va_gid == (gid_t)VNOVAL)
648 				return (0);
649  			vap->va_size = VNOVAL;
650  			break;
651  		default:
652 			/*
653 			 * Disallow write attempts if the filesystem is
654 			 * mounted read-only.
655 			 */
656 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
657 				return (EROFS);
658 			genfs_node_wrlock(vp);
659  			uvm_vnp_setsize(vp, vap->va_size);
660  			tsize = np->n_size;
661 			np->n_size = vap->va_size;
662  			if (vap->va_size == 0)
663  				error = nfs_vinvalbuf(vp, 0,
664  				     ap->a_cred, curlwp, 1);
665 			else
666 				error = nfs_vinvalbuf(vp, V_SAVE,
667 				     ap->a_cred, curlwp, 1);
668 			if (error) {
669 				uvm_vnp_setsize(vp, tsize);
670 				genfs_node_unlock(vp);
671 				return (error);
672 			}
673  			np->n_vattr->va_size = vap->va_size;
674   		}
675   	} else {
676 		/*
677 		 * flush files before setattr because a later write of
678 		 * cached data might change timestamps or reset sugid bits
679 		 */
680 		if ((vap->va_mtime.tv_sec != VNOVAL ||
681 		     vap->va_atime.tv_sec != VNOVAL ||
682 		     vap->va_mode != VNOVAL) &&
683 		    vp->v_type == VREG &&
684   		    (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
685 		 			   curlwp, 1)) == EINTR)
686 			return (error);
687 	}
688 	error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
689 	if (vap->va_size != VNOVAL) {
690 		if (error) {
691 			np->n_size = np->n_vattr->va_size = tsize;
692 			uvm_vnp_setsize(vp, np->n_size);
693 		}
694 		genfs_node_unlock(vp);
695 	}
696 	VN_KNOTE(vp, NOTE_ATTRIB);
697 	return (error);
698 }
699 
700 /*
701  * Do an nfs setattr rpc.
702  */
703 int
704 nfs_setattrrpc(vp, vap, cred, l)
705 	struct vnode *vp;
706 	struct vattr *vap;
707 	kauth_cred_t cred;
708 	struct lwp *l;
709 {
710 	struct nfsv2_sattr *sp;
711 	char *cp;
712 	int32_t t1, t2;
713 	char *bpos, *dpos;
714 	u_int32_t *tl;
715 	int error = 0;
716 	struct mbuf *mreq, *mrep, *md, *mb;
717 	const int v3 = NFS_ISV3(vp);
718 	struct nfsnode *np = VTONFS(vp);
719 #ifndef NFS_V2_ONLY
720 	int wccflag = NFSV3_WCCRATTR;
721 	char *cp2;
722 #endif
723 
724 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
725 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
726 	nfsm_fhtom(np, v3);
727 #ifndef NFS_V2_ONLY
728 	if (v3) {
729 		nfsm_v3attrbuild(vap, true);
730 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
731 		*tl = nfs_false;
732 	} else {
733 #endif
734 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
735 		if (vap->va_mode == (mode_t)VNOVAL)
736 			sp->sa_mode = nfs_xdrneg1;
737 		else
738 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
739 		if (vap->va_uid == (uid_t)VNOVAL)
740 			sp->sa_uid = nfs_xdrneg1;
741 		else
742 			sp->sa_uid = txdr_unsigned(vap->va_uid);
743 		if (vap->va_gid == (gid_t)VNOVAL)
744 			sp->sa_gid = nfs_xdrneg1;
745 		else
746 			sp->sa_gid = txdr_unsigned(vap->va_gid);
747 		sp->sa_size = txdr_unsigned(vap->va_size);
748 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
749 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
750 #ifndef NFS_V2_ONLY
751 	}
752 #endif
753 	nfsm_request(np, NFSPROC_SETATTR, l, cred);
754 #ifndef NFS_V2_ONLY
755 	if (v3) {
756 		nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
757 	} else
758 #endif
759 		nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
760 	nfsm_reqdone;
761 	return (error);
762 }
763 
764 /*
765  * nfs lookup call, one step at a time...
766  * First look in cache
767  * If not found, unlock the directory nfsnode and do the rpc
768  *
769  * This code is full of lock/unlock statements and checks, because
770  * we continue after cache_lookup has finished (we need to check
771  * with the attr cache and do an rpc if it has timed out). This means
772  * that the locking effects of cache_lookup have to be taken into
773  * account.
774  */
775 int
776 nfs_lookup(v)
777 	void *v;
778 {
779 	struct vop_lookup_args /* {
780 		struct vnodeop_desc *a_desc;
781 		struct vnode *a_dvp;
782 		struct vnode **a_vpp;
783 		struct componentname *a_cnp;
784 	} */ *ap = v;
785 	struct componentname *cnp = ap->a_cnp;
786 	struct vnode *dvp = ap->a_dvp;
787 	struct vnode **vpp = ap->a_vpp;
788 	int flags;
789 	struct vnode *newvp;
790 	u_int32_t *tl;
791 	char *cp;
792 	int32_t t1, t2;
793 	char *bpos, *dpos, *cp2;
794 	struct mbuf *mreq, *mrep, *md, *mb;
795 	long len;
796 	nfsfh_t *fhp;
797 	struct nfsnode *np;
798 	int error = 0, attrflag, fhsize;
799 	const int v3 = NFS_ISV3(dvp);
800 
801 	flags = cnp->cn_flags;
802 
803 	*vpp = NULLVP;
804 	newvp = NULLVP;
805 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
806 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
807 		return (EROFS);
808 	if (dvp->v_type != VDIR)
809 		return (ENOTDIR);
810 
811 	/*
812 	 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
813 	 */
814 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
815 		error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
816 		if (error)
817 			return error;
818 		if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
819 			return EISDIR;
820 		VREF(dvp);
821 		*vpp = dvp;
822 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
823 			cnp->cn_flags |= SAVENAME;
824 		return 0;
825 	}
826 
827 	np = VTONFS(dvp);
828 
829 	/*
830 	 * Before tediously performing a linear scan of the directory,
831 	 * check the name cache to see if the directory/name pair
832 	 * we are looking for is known already.
833 	 * If the directory/name pair is found in the name cache,
834 	 * we have to ensure the directory has not changed from
835 	 * the time the cache entry has been created. If it has,
836 	 * the cache entry has to be ignored.
837 	 */
838 	error = cache_lookup_raw(dvp, vpp, cnp);
839 	KASSERT(dvp != *vpp);
840 	if (error >= 0) {
841 		struct vattr vattr;
842 		int err2;
843 
844 		if (error && error != ENOENT) {
845 			*vpp = NULLVP;
846 			return error;
847 		}
848 
849 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
850 		if (err2 != 0) {
851 			if (error == 0)
852 				vrele(*vpp);
853 			*vpp = NULLVP;
854 			return err2;
855 		}
856 
857 		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
858 		    || timespeccmp(&vattr.va_mtime,
859 		    &VTONFS(dvp)->n_nctime, !=)) {
860 			if (error == 0) {
861 				vrele(*vpp);
862 				*vpp = NULLVP;
863 			}
864 			cache_purge1(dvp, NULL, PURGE_CHILDREN);
865 			timespecclear(&np->n_nctime);
866 			goto dorpc;
867 		}
868 
869 		if (error == ENOENT) {
870 			goto noentry;
871 		}
872 
873 		newvp = *vpp;
874 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
875 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
876 			nfsstats.lookupcache_hits++;
877 			if ((flags & ISDOTDOT) != 0) {
878 				VOP_UNLOCK(dvp, 0);
879 			}
880 			error = vn_lock(newvp, LK_EXCLUSIVE);
881 			if ((flags & ISDOTDOT) != 0) {
882 				vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
883 			}
884 			if (error) {
885 				/* newvp has been revoked. */
886 				vrele(newvp);
887 				*vpp = NULL;
888 				return error;
889 			}
890 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
891 				cnp->cn_flags |= SAVENAME;
892 			KASSERT(newvp->v_type != VNON);
893 			return (0);
894 		}
895 		cache_purge1(newvp, NULL, PURGE_PARENTS);
896 		vrele(newvp);
897 		*vpp = NULLVP;
898 	}
899 dorpc:
900 #if 0
901 	/*
902 	 * because nfsv3 has the same CREATE semantics as ours,
903 	 * we don't have to perform LOOKUPs beforehand.
904 	 *
905 	 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
906 	 * XXX although we have no way to know if O_EXCL is requested or not.
907 	 */
908 
909 	if (v3 && cnp->cn_nameiop == CREATE &&
910 	    (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
911 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
912 		cnp->cn_flags |= SAVENAME;
913 		return (EJUSTRETURN);
914 	}
915 #endif /* 0 */
916 
917 	error = 0;
918 	newvp = NULLVP;
919 	nfsstats.lookupcache_misses++;
920 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
921 	len = cnp->cn_namelen;
922 	nfsm_reqhead(np, NFSPROC_LOOKUP,
923 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
924 	nfsm_fhtom(np, v3);
925 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
926 	nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
927 	if (error) {
928 		nfsm_postop_attr(dvp, attrflag, 0);
929 		m_freem(mrep);
930 		goto nfsmout;
931 	}
932 	nfsm_getfh(fhp, fhsize, v3);
933 
934 	/*
935 	 * Handle RENAME case...
936 	 */
937 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
938 		if (NFS_CMPFH(np, fhp, fhsize)) {
939 			m_freem(mrep);
940 			return (EISDIR);
941 		}
942 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
943 		if (error) {
944 			m_freem(mrep);
945 			return error;
946 		}
947 		newvp = NFSTOV(np);
948 #ifndef NFS_V2_ONLY
949 		if (v3) {
950 			nfsm_postop_attr(newvp, attrflag, 0);
951 			nfsm_postop_attr(dvp, attrflag, 0);
952 		} else
953 #endif
954 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
955 		*vpp = newvp;
956 		m_freem(mrep);
957 		cnp->cn_flags |= SAVENAME;
958 		goto validate;
959 	}
960 
961 	/*
962 	 * The postop attr handling is duplicated for each if case,
963 	 * because it should be done while dvp is locked (unlocking
964 	 * dvp is different for each case).
965 	 */
966 
967 	if (NFS_CMPFH(np, fhp, fhsize)) {
968 		/*
969 		 * "." lookup
970 		 */
971 		VREF(dvp);
972 		newvp = dvp;
973 #ifndef NFS_V2_ONLY
974 		if (v3) {
975 			nfsm_postop_attr(newvp, attrflag, 0);
976 			nfsm_postop_attr(dvp, attrflag, 0);
977 		} else
978 #endif
979 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
980 	} else if (flags & ISDOTDOT) {
981 		/*
982 		 * ".." lookup
983 		 */
984 		VOP_UNLOCK(dvp, 0);
985 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
986 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
987 		if (error) {
988 			m_freem(mrep);
989 			return error;
990 		}
991 		newvp = NFSTOV(np);
992 
993 #ifndef NFS_V2_ONLY
994 		if (v3) {
995 			nfsm_postop_attr(newvp, attrflag, 0);
996 			nfsm_postop_attr(dvp, attrflag, 0);
997 		} else
998 #endif
999 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1000 	} else {
1001 		/*
1002 		 * Other lookups.
1003 		 */
1004 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1005 		if (error) {
1006 			m_freem(mrep);
1007 			return error;
1008 		}
1009 		newvp = NFSTOV(np);
1010 #ifndef NFS_V2_ONLY
1011 		if (v3) {
1012 			nfsm_postop_attr(newvp, attrflag, 0);
1013 			nfsm_postop_attr(dvp, attrflag, 0);
1014 		} else
1015 #endif
1016 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1017 	}
1018 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1019 		cnp->cn_flags |= SAVENAME;
1020 	if ((cnp->cn_flags & MAKEENTRY) &&
1021 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1022 		nfs_cache_enter(dvp, newvp, cnp);
1023 	}
1024 	*vpp = newvp;
1025 	nfsm_reqdone;
1026 	if (error) {
1027 		/*
1028 		 * We get here only because of errors returned by
1029 		 * the RPC. Otherwise we'll have returned above
1030 		 * (the nfsm_* macros will jump to nfsm_reqdone
1031 		 * on error).
1032 		 */
1033 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1034 		    cnp->cn_nameiop != CREATE) {
1035 			nfs_cache_enter(dvp, NULL, cnp);
1036 		}
1037 		if (newvp != NULLVP) {
1038 			if (newvp == dvp) {
1039 				vrele(newvp);
1040 			} else {
1041 				vput(newvp);
1042 			}
1043 		}
1044 noentry:
1045 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1046 		    (flags & ISLASTCN) && error == ENOENT) {
1047 			if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1048 				error = EROFS;
1049 			} else {
1050 				error = EJUSTRETURN;
1051 				cnp->cn_flags |= SAVENAME;
1052 			}
1053 		}
1054 		*vpp = NULL;
1055 		return error;
1056 	}
1057 
1058 validate:
1059 	/*
1060 	 * make sure we have valid type and size.
1061 	 */
1062 
1063 	newvp = *vpp;
1064 	if (newvp->v_type == VNON) {
1065 		struct vattr vattr; /* dummy */
1066 
1067 		KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1068 		error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1069 		if (error) {
1070 			vput(newvp);
1071 			*vpp = NULL;
1072 		}
1073 	}
1074 
1075 	return error;
1076 }
1077 
1078 /*
1079  * nfs read call.
1080  * Just call nfs_bioread() to do the work.
1081  */
1082 int
1083 nfs_read(v)
1084 	void *v;
1085 {
1086 	struct vop_read_args /* {
1087 		struct vnode *a_vp;
1088 		struct uio *a_uio;
1089 		int  a_ioflag;
1090 		kauth_cred_t a_cred;
1091 	} */ *ap = v;
1092 	struct vnode *vp = ap->a_vp;
1093 
1094 	if (vp->v_type != VREG)
1095 		return EISDIR;
1096 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1097 }
1098 
1099 /*
1100  * nfs readlink call
1101  */
1102 int
1103 nfs_readlink(v)
1104 	void *v;
1105 {
1106 	struct vop_readlink_args /* {
1107 		struct vnode *a_vp;
1108 		struct uio *a_uio;
1109 		kauth_cred_t a_cred;
1110 	} */ *ap = v;
1111 	struct vnode *vp = ap->a_vp;
1112 	struct nfsnode *np = VTONFS(vp);
1113 
1114 	if (vp->v_type != VLNK)
1115 		return (EPERM);
1116 
1117 	if (np->n_rcred != NULL) {
1118 		kauth_cred_free(np->n_rcred);
1119 	}
1120 	np->n_rcred = ap->a_cred;
1121 	kauth_cred_hold(np->n_rcred);
1122 
1123 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1124 }
1125 
1126 /*
1127  * Do a readlink rpc.
1128  * Called by nfs_doio() from below the buffer cache.
1129  */
1130 int
1131 nfs_readlinkrpc(vp, uiop, cred)
1132 	struct vnode *vp;
1133 	struct uio *uiop;
1134 	kauth_cred_t cred;
1135 {
1136 	u_int32_t *tl;
1137 	char *cp;
1138 	int32_t t1, t2;
1139 	char *bpos, *dpos, *cp2;
1140 	int error = 0;
1141 	uint32_t len;
1142 	struct mbuf *mreq, *mrep, *md, *mb;
1143 	const int v3 = NFS_ISV3(vp);
1144 	struct nfsnode *np = VTONFS(vp);
1145 #ifndef NFS_V2_ONLY
1146 	int attrflag;
1147 #endif
1148 
1149 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1150 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1151 	nfsm_fhtom(np, v3);
1152 	nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1153 #ifndef NFS_V2_ONLY
1154 	if (v3)
1155 		nfsm_postop_attr(vp, attrflag, 0);
1156 #endif
1157 	if (!error) {
1158 #ifndef NFS_V2_ONLY
1159 		if (v3) {
1160 			nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1161 			len = fxdr_unsigned(uint32_t, *tl);
1162 			if (len > MAXPATHLEN) {
1163 				/*
1164 				 * this pathname is too long for us.
1165 				 */
1166 				m_freem(mrep);
1167 				/* Solaris returns EINVAL. should we follow? */
1168 				error = ENAMETOOLONG;
1169 				goto nfsmout;
1170 			}
1171 		} else
1172 #endif
1173 		{
1174 			nfsm_strsiz(len, NFS_MAXPATHLEN);
1175 		}
1176 		nfsm_mtouio(uiop, len);
1177 	}
1178 	nfsm_reqdone;
1179 	return (error);
1180 }
1181 
1182 /*
1183  * nfs read rpc call
1184  * Ditto above
1185  */
1186 int
1187 nfs_readrpc(vp, uiop)
1188 	struct vnode *vp;
1189 	struct uio *uiop;
1190 {
1191 	u_int32_t *tl;
1192 	char *cp;
1193 	int32_t t1, t2;
1194 	char *bpos, *dpos, *cp2;
1195 	struct mbuf *mreq, *mrep, *md, *mb;
1196 	struct nfsmount *nmp;
1197 	int error = 0, len, retlen, tsiz, eof, byte_count;
1198 	const int v3 = NFS_ISV3(vp);
1199 	struct nfsnode *np = VTONFS(vp);
1200 #ifndef NFS_V2_ONLY
1201 	int attrflag;
1202 #endif
1203 
1204 #ifndef nolint
1205 	eof = 0;
1206 #endif
1207 	nmp = VFSTONFS(vp->v_mount);
1208 	tsiz = uiop->uio_resid;
1209 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1210 		return (EFBIG);
1211 	iostat_busy(nmp->nm_stats);
1212 	byte_count = 0; /* count bytes actually transferred */
1213 	while (tsiz > 0) {
1214 		nfsstats.rpccnt[NFSPROC_READ]++;
1215 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1216 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1217 		nfsm_fhtom(np, v3);
1218 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1219 #ifndef NFS_V2_ONLY
1220 		if (v3) {
1221 			txdr_hyper(uiop->uio_offset, tl);
1222 			*(tl + 2) = txdr_unsigned(len);
1223 		} else
1224 #endif
1225 		{
1226 			*tl++ = txdr_unsigned(uiop->uio_offset);
1227 			*tl++ = txdr_unsigned(len);
1228 			*tl = 0;
1229 		}
1230 		nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1231 #ifndef NFS_V2_ONLY
1232 		if (v3) {
1233 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1234 			if (error) {
1235 				m_freem(mrep);
1236 				goto nfsmout;
1237 			}
1238 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1239 			eof = fxdr_unsigned(int, *(tl + 1));
1240 		} else
1241 #endif
1242 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1243 		nfsm_strsiz(retlen, nmp->nm_rsize);
1244 		nfsm_mtouio(uiop, retlen);
1245 		m_freem(mrep);
1246 		tsiz -= retlen;
1247 		byte_count += retlen;
1248 #ifndef NFS_V2_ONLY
1249 		if (v3) {
1250 			if (eof || retlen == 0)
1251 				tsiz = 0;
1252 		} else
1253 #endif
1254 		if (retlen < len)
1255 			tsiz = 0;
1256 	}
1257 nfsmout:
1258 	iostat_unbusy(nmp->nm_stats, byte_count, 1);
1259 	return (error);
1260 }
1261 
1262 struct nfs_writerpc_context {
1263 	kmutex_t nwc_lock;
1264 	kcondvar_t nwc_cv;
1265 	int nwc_mbufcount;
1266 };
1267 
1268 /*
1269  * free mbuf used to refer protected pages while write rpc call.
1270  * called at splvm.
1271  */
1272 static void
1273 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1274 {
1275 	struct nfs_writerpc_context *ctx = arg;
1276 
1277 	KASSERT(m != NULL);
1278 	KASSERT(ctx != NULL);
1279 	pool_cache_put(mb_cache, m);
1280 	mutex_enter(&ctx->nwc_lock);
1281 	if (--ctx->nwc_mbufcount == 0) {
1282 		cv_signal(&ctx->nwc_cv);
1283 	}
1284 	mutex_exit(&ctx->nwc_lock);
1285 }
1286 
1287 /*
1288  * nfs write call
1289  */
1290 int
1291 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1292 	struct vnode *vp;
1293 	struct uio *uiop;
1294 	int *iomode;
1295 	bool pageprotected;
1296 	bool *stalewriteverfp;
1297 {
1298 	u_int32_t *tl;
1299 	char *cp;
1300 	int32_t t1, t2;
1301 	char *bpos, *dpos;
1302 	struct mbuf *mreq, *mrep, *md, *mb;
1303 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1304 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1305 	const int v3 = NFS_ISV3(vp);
1306 	int committed = NFSV3WRITE_FILESYNC;
1307 	struct nfsnode *np = VTONFS(vp);
1308 	struct nfs_writerpc_context ctx;
1309 	int byte_count;
1310 	struct lwp *l = NULL;
1311 	size_t origresid;
1312 #ifndef NFS_V2_ONLY
1313 	char *cp2;
1314 	int rlen, commit;
1315 #endif
1316 
1317 	mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1318 	cv_init(&ctx.nwc_cv, "nfsmblk");
1319 	ctx.nwc_mbufcount = 1;
1320 
1321 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1322 		panic("writerpc readonly vp %p", vp);
1323 	}
1324 
1325 #ifdef DIAGNOSTIC
1326 	if (uiop->uio_iovcnt != 1)
1327 		panic("nfs: writerpc iovcnt > 1");
1328 #endif
1329 	tsiz = uiop->uio_resid;
1330 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1331 		return (EFBIG);
1332 	if (pageprotected) {
1333 		l = curlwp;
1334 		uvm_lwp_hold(l);
1335 	}
1336 retry:
1337 	origresid = uiop->uio_resid;
1338 	KASSERT(origresid == uiop->uio_iov->iov_len);
1339 	iostat_busy(nmp->nm_stats);
1340 	byte_count = 0; /* count of bytes actually written */
1341 	while (tsiz > 0) {
1342 		uint32_t datalen; /* data bytes need to be allocated in mbuf */
1343 		uint32_t backup;
1344 		bool stalewriteverf = false;
1345 
1346 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1347 		len = min(tsiz, nmp->nm_wsize);
1348 		datalen = pageprotected ? 0 : nfsm_rndup(len);
1349 		nfsm_reqhead(np, NFSPROC_WRITE,
1350 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1351 		nfsm_fhtom(np, v3);
1352 #ifndef NFS_V2_ONLY
1353 		if (v3) {
1354 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1355 			txdr_hyper(uiop->uio_offset, tl);
1356 			tl += 2;
1357 			*tl++ = txdr_unsigned(len);
1358 			*tl++ = txdr_unsigned(*iomode);
1359 			*tl = txdr_unsigned(len);
1360 		} else
1361 #endif
1362 		{
1363 			u_int32_t x;
1364 
1365 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1366 			/* Set both "begin" and "current" to non-garbage. */
1367 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1368 			*tl++ = x;      /* "begin offset" */
1369 			*tl++ = x;      /* "current offset" */
1370 			x = txdr_unsigned(len);
1371 			*tl++ = x;      /* total to this offset */
1372 			*tl = x;        /* size of this write */
1373 
1374 		}
1375 		if (pageprotected) {
1376 			/*
1377 			 * since we know pages can't be modified during i/o,
1378 			 * no need to copy them for us.
1379 			 */
1380 			struct mbuf *m;
1381 			struct iovec *iovp = uiop->uio_iov;
1382 
1383 			m = m_get(M_WAIT, MT_DATA);
1384 			MCLAIM(m, &nfs_mowner);
1385 			MEXTADD(m, iovp->iov_base, len, M_MBUF,
1386 			    nfs_writerpc_extfree, &ctx);
1387 			m->m_flags |= M_EXT_ROMAP;
1388 			m->m_len = len;
1389 			mb->m_next = m;
1390 			/*
1391 			 * no need to maintain mb and bpos here
1392 			 * because no one care them later.
1393 			 */
1394 #if 0
1395 			mb = m;
1396 			bpos = mtod(void *, mb) + mb->m_len;
1397 #endif
1398 			UIO_ADVANCE(uiop, len);
1399 			uiop->uio_offset += len;
1400 			mutex_enter(&ctx.nwc_lock);
1401 			ctx.nwc_mbufcount++;
1402 			mutex_exit(&ctx.nwc_lock);
1403 			nfs_zeropad(mb, 0, nfsm_padlen(len));
1404 		} else {
1405 			nfsm_uiotom(uiop, len);
1406 		}
1407 		nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1408 #ifndef NFS_V2_ONLY
1409 		if (v3) {
1410 			wccflag = NFSV3_WCCCHK;
1411 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1412 			if (!error) {
1413 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1414 					+ NFSX_V3WRITEVERF);
1415 				rlen = fxdr_unsigned(int, *tl++);
1416 				if (rlen == 0) {
1417 					error = NFSERR_IO;
1418 					m_freem(mrep);
1419 					break;
1420 				} else if (rlen < len) {
1421 					backup = len - rlen;
1422 					UIO_ADVANCE(uiop, -backup);
1423 					uiop->uio_offset -= backup;
1424 					len = rlen;
1425 				}
1426 				commit = fxdr_unsigned(int, *tl++);
1427 
1428 				/*
1429 				 * Return the lowest committment level
1430 				 * obtained by any of the RPCs.
1431 				 */
1432 				if (committed == NFSV3WRITE_FILESYNC)
1433 					committed = commit;
1434 				else if (committed == NFSV3WRITE_DATASYNC &&
1435 					commit == NFSV3WRITE_UNSTABLE)
1436 					committed = commit;
1437 				mutex_enter(&nmp->nm_lock);
1438 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1439 					memcpy(nmp->nm_writeverf, tl,
1440 					    NFSX_V3WRITEVERF);
1441 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1442 				} else if ((nmp->nm_iflag &
1443 				    NFSMNT_STALEWRITEVERF) ||
1444 				    memcmp(tl, nmp->nm_writeverf,
1445 				    NFSX_V3WRITEVERF)) {
1446 					memcpy(nmp->nm_writeverf, tl,
1447 					    NFSX_V3WRITEVERF);
1448 					/*
1449 					 * note NFSMNT_STALEWRITEVERF
1450 					 * if we're the first thread to
1451 					 * notice it.
1452 					 */
1453 					if ((nmp->nm_iflag &
1454 					    NFSMNT_STALEWRITEVERF) == 0) {
1455 						stalewriteverf = true;
1456 						nmp->nm_iflag |=
1457 						    NFSMNT_STALEWRITEVERF;
1458 					}
1459 				}
1460 				mutex_exit(&nmp->nm_lock);
1461 			}
1462 		} else
1463 #endif
1464 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1465 		if (wccflag)
1466 			VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1467 		m_freem(mrep);
1468 		if (error)
1469 			break;
1470 		tsiz -= len;
1471 		byte_count += len;
1472 		if (stalewriteverf) {
1473 			*stalewriteverfp = true;
1474 			stalewriteverf = false;
1475 			if (committed == NFSV3WRITE_UNSTABLE &&
1476 			    len != origresid) {
1477 				/*
1478 				 * if our write requests weren't atomic but
1479 				 * unstable, datas in previous iterations
1480 				 * might have already been lost now.
1481 				 * then, we should resend them to nfsd.
1482 				 */
1483 				backup = origresid - tsiz;
1484 				UIO_ADVANCE(uiop, -backup);
1485 				uiop->uio_offset -= backup;
1486 				tsiz = origresid;
1487 				goto retry;
1488 			}
1489 		}
1490 	}
1491 nfsmout:
1492 	iostat_unbusy(nmp->nm_stats, byte_count, 0);
1493 	if (pageprotected) {
1494 		/*
1495 		 * wait until mbufs go away.
1496 		 * retransmitted mbufs can survive longer than rpc requests
1497 		 * themselves.
1498 		 */
1499 		mutex_enter(&ctx.nwc_lock);
1500 		ctx.nwc_mbufcount--;
1501 		while (ctx.nwc_mbufcount > 0) {
1502 			cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1503 		}
1504 		mutex_exit(&ctx.nwc_lock);
1505 		uvm_lwp_rele(l);
1506 	}
1507 	mutex_destroy(&ctx.nwc_lock);
1508 	cv_destroy(&ctx.nwc_cv);
1509 	*iomode = committed;
1510 	if (error)
1511 		uiop->uio_resid = tsiz;
1512 	return (error);
1513 }
1514 
1515 /*
1516  * nfs mknod rpc
1517  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1518  * mode set to specify the file type and the size field for rdev.
1519  */
1520 int
1521 nfs_mknodrpc(dvp, vpp, cnp, vap)
1522 	struct vnode *dvp;
1523 	struct vnode **vpp;
1524 	struct componentname *cnp;
1525 	struct vattr *vap;
1526 {
1527 	struct nfsv2_sattr *sp;
1528 	u_int32_t *tl;
1529 	char *cp;
1530 	int32_t t1, t2;
1531 	struct vnode *newvp = (struct vnode *)0;
1532 	struct nfsnode *dnp, *np;
1533 	char *cp2;
1534 	char *bpos, *dpos;
1535 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1536 	struct mbuf *mreq, *mrep, *md, *mb;
1537 	u_int32_t rdev;
1538 	const int v3 = NFS_ISV3(dvp);
1539 
1540 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1541 		rdev = txdr_unsigned(vap->va_rdev);
1542 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1543 		rdev = nfs_xdrneg1;
1544 	else {
1545 		VOP_ABORTOP(dvp, cnp);
1546 		vput(dvp);
1547 		return (EOPNOTSUPP);
1548 	}
1549 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1550 	dnp = VTONFS(dvp);
1551 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1552 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1553 	nfsm_fhtom(dnp, v3);
1554 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1555 #ifndef NFS_V2_ONLY
1556 	if (v3) {
1557 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1558 		*tl++ = vtonfsv3_type(vap->va_type);
1559 		nfsm_v3attrbuild(vap, false);
1560 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1561 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1562 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1563 			*tl = txdr_unsigned(minor(vap->va_rdev));
1564 		}
1565 	} else
1566 #endif
1567 	{
1568 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1569 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1570 		sp->sa_uid = nfs_xdrneg1;
1571 		sp->sa_gid = nfs_xdrneg1;
1572 		sp->sa_size = rdev;
1573 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1574 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1575 	}
1576 	nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1577 	if (!error) {
1578 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1579 		if (!gotvp) {
1580 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1581 			    cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1582 			if (!error)
1583 				newvp = NFSTOV(np);
1584 		}
1585 	}
1586 #ifndef NFS_V2_ONLY
1587 	if (v3)
1588 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1589 #endif
1590 	nfsm_reqdone;
1591 	if (error) {
1592 		if (newvp)
1593 			vput(newvp);
1594 	} else {
1595 		if (cnp->cn_flags & MAKEENTRY)
1596 			nfs_cache_enter(dvp, newvp, cnp);
1597 		*vpp = newvp;
1598 	}
1599 	PNBUF_PUT(cnp->cn_pnbuf);
1600 	VTONFS(dvp)->n_flag |= NMODIFIED;
1601 	if (!wccflag)
1602 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1603 	vput(dvp);
1604 	return (error);
1605 }
1606 
1607 /*
1608  * nfs mknod vop
1609  * just call nfs_mknodrpc() to do the work.
1610  */
1611 /* ARGSUSED */
1612 int
1613 nfs_mknod(v)
1614 	void *v;
1615 {
1616 	struct vop_mknod_args /* {
1617 		struct vnode *a_dvp;
1618 		struct vnode **a_vpp;
1619 		struct componentname *a_cnp;
1620 		struct vattr *a_vap;
1621 	} */ *ap = v;
1622 	struct vnode *dvp = ap->a_dvp;
1623 	struct componentname *cnp = ap->a_cnp;
1624 	int error;
1625 
1626 	error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1627 	VN_KNOTE(dvp, NOTE_WRITE);
1628 	if (error == 0 || error == EEXIST)
1629 		cache_purge1(dvp, cnp, 0);
1630 	return (error);
1631 }
1632 
1633 /*
1634  * nfs file create call
1635  */
1636 int
1637 nfs_create(v)
1638 	void *v;
1639 {
1640 	struct vop_create_args /* {
1641 		struct vnode *a_dvp;
1642 		struct vnode **a_vpp;
1643 		struct componentname *a_cnp;
1644 		struct vattr *a_vap;
1645 	} */ *ap = v;
1646 	struct vnode *dvp = ap->a_dvp;
1647 	struct vattr *vap = ap->a_vap;
1648 	struct componentname *cnp = ap->a_cnp;
1649 	struct nfsv2_sattr *sp;
1650 	u_int32_t *tl;
1651 	char *cp;
1652 	int32_t t1, t2;
1653 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1654 	struct vnode *newvp = (struct vnode *)0;
1655 	char *bpos, *dpos, *cp2;
1656 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1657 	struct mbuf *mreq, *mrep, *md, *mb;
1658 	const int v3 = NFS_ISV3(dvp);
1659 	u_int32_t excl_mode = NFSV3CREATE_UNCHECKED;
1660 
1661 	/*
1662 	 * Oops, not for me..
1663 	 */
1664 	if (vap->va_type == VSOCK)
1665 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1666 
1667 	KASSERT(vap->va_type == VREG);
1668 
1669 #ifdef VA_EXCLUSIVE
1670 	if (vap->va_vaflags & VA_EXCLUSIVE) {
1671 		excl_mode = NFSV3CREATE_EXCLUSIVE;
1672 	}
1673 #endif
1674 again:
1675 	error = 0;
1676 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1677 	dnp = VTONFS(dvp);
1678 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1679 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1680 	nfsm_fhtom(dnp, v3);
1681 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1682 #ifndef NFS_V2_ONLY
1683 	if (v3) {
1684 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1685 		if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1686 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1687 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1688 			*tl++ = arc4random();
1689 			*tl = arc4random();
1690 		} else {
1691 			*tl = txdr_unsigned(excl_mode);
1692 			nfsm_v3attrbuild(vap, false);
1693 		}
1694 	} else
1695 #endif
1696 	{
1697 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1698 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1699 		sp->sa_uid = nfs_xdrneg1;
1700 		sp->sa_gid = nfs_xdrneg1;
1701 		sp->sa_size = 0;
1702 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1703 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1704 	}
1705 	nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred);
1706 	if (!error) {
1707 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1708 		if (!gotvp) {
1709 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1710 			    cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1711 			if (!error)
1712 				newvp = NFSTOV(np);
1713 		}
1714 	}
1715 #ifndef NFS_V2_ONLY
1716 	if (v3)
1717 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1718 #endif
1719 	nfsm_reqdone;
1720 	if (error) {
1721 		/*
1722 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1723 		 */
1724 		if (v3 && error == ENOTSUP) {
1725 			if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1726 				excl_mode = NFSV3CREATE_GUARDED;
1727 				goto again;
1728 			} else if (excl_mode == NFSV3CREATE_GUARDED) {
1729 				excl_mode = NFSV3CREATE_UNCHECKED;
1730 				goto again;
1731 			}
1732 		}
1733 	} else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) {
1734 		struct timespec ts;
1735 
1736 		getnanotime(&ts);
1737 
1738 		/*
1739 		 * make sure that we'll update timestamps as
1740 		 * most server implementations use them to store
1741 		 * the create verifier.
1742 		 *
1743 		 * XXX it's better to use TOSERVER always.
1744 		 */
1745 
1746 		if (vap->va_atime.tv_sec == VNOVAL)
1747 			vap->va_atime = ts;
1748 		if (vap->va_mtime.tv_sec == VNOVAL)
1749 			vap->va_mtime = ts;
1750 
1751 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp);
1752 	}
1753 	if (error == 0) {
1754 		if (cnp->cn_flags & MAKEENTRY)
1755 			nfs_cache_enter(dvp, newvp, cnp);
1756 		else
1757 			cache_purge1(dvp, cnp, 0);
1758 		*ap->a_vpp = newvp;
1759 	} else {
1760 		if (newvp)
1761 			vput(newvp);
1762 		if (error == EEXIST)
1763 			cache_purge1(dvp, cnp, 0);
1764 	}
1765 	PNBUF_PUT(cnp->cn_pnbuf);
1766 	VTONFS(dvp)->n_flag |= NMODIFIED;
1767 	if (!wccflag)
1768 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1769 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1770 	vput(dvp);
1771 	return (error);
1772 }
1773 
1774 /*
1775  * nfs file remove call
1776  * To try and make nfs semantics closer to ufs semantics, a file that has
1777  * other processes using the vnode is renamed instead of removed and then
1778  * removed later on the last close.
1779  * - If v_usecount > 1
1780  *	  If a rename is not already in the works
1781  *	     call nfs_sillyrename() to set it up
1782  *     else
1783  *	  do the remove rpc
1784  */
1785 int
1786 nfs_remove(v)
1787 	void *v;
1788 {
1789 	struct vop_remove_args /* {
1790 		struct vnodeop_desc *a_desc;
1791 		struct vnode * a_dvp;
1792 		struct vnode * a_vp;
1793 		struct componentname * a_cnp;
1794 	} */ *ap = v;
1795 	struct vnode *vp = ap->a_vp;
1796 	struct vnode *dvp = ap->a_dvp;
1797 	struct componentname *cnp = ap->a_cnp;
1798 	struct nfsnode *np = VTONFS(vp);
1799 	int error = 0;
1800 	struct vattr vattr;
1801 
1802 #ifndef DIAGNOSTIC
1803 	if ((cnp->cn_flags & HASBUF) == 0)
1804 		panic("nfs_remove: no name");
1805 	if (vp->v_usecount < 1)
1806 		panic("nfs_remove: bad v_usecount");
1807 #endif
1808 	if (vp->v_type == VDIR)
1809 		error = EPERM;
1810 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1811 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1812 	    vattr.va_nlink > 1)) {
1813 		/*
1814 		 * Purge the name cache so that the chance of a lookup for
1815 		 * the name succeeding while the remove is in progress is
1816 		 * minimized. Without node locking it can still happen, such
1817 		 * that an I/O op returns ESTALE, but since you get this if
1818 		 * another host removes the file..
1819 		 */
1820 		cache_purge(vp);
1821 		/*
1822 		 * throw away biocache buffers, mainly to avoid
1823 		 * unnecessary delayed writes later.
1824 		 */
1825 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1);
1826 		/* Do the rpc */
1827 		if (error != EINTR)
1828 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1829 				cnp->cn_namelen, cnp->cn_cred, curlwp);
1830 	} else if (!np->n_sillyrename)
1831 		error = nfs_sillyrename(dvp, vp, cnp, false);
1832 	PNBUF_PUT(cnp->cn_pnbuf);
1833 	if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1834 	    vattr.va_nlink == 1) {
1835 		np->n_flag |= NREMOVED;
1836 	}
1837 	NFS_INVALIDATE_ATTRCACHE(np);
1838 	VN_KNOTE(vp, NOTE_DELETE);
1839 	VN_KNOTE(dvp, NOTE_WRITE);
1840 	if (dvp == vp)
1841 		vrele(vp);
1842 	else
1843 		vput(vp);
1844 	vput(dvp);
1845 	return (error);
1846 }
1847 
1848 /*
1849  * nfs file remove rpc called from nfs_inactive
1850  */
1851 int
1852 nfs_removeit(sp)
1853 	struct sillyrename *sp;
1854 {
1855 
1856 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1857 		(struct lwp *)0));
1858 }
1859 
1860 /*
1861  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1862  */
1863 int
1864 nfs_removerpc(dvp, name, namelen, cred, l)
1865 	struct vnode *dvp;
1866 	const char *name;
1867 	int namelen;
1868 	kauth_cred_t cred;
1869 	struct lwp *l;
1870 {
1871 	u_int32_t *tl;
1872 	char *cp;
1873 #ifndef NFS_V2_ONLY
1874 	int32_t t1;
1875 	char *cp2;
1876 #endif
1877 	int32_t t2;
1878 	char *bpos, *dpos;
1879 	int error = 0, wccflag = NFSV3_WCCRATTR;
1880 	struct mbuf *mreq, *mrep, *md, *mb;
1881 	const int v3 = NFS_ISV3(dvp);
1882 	int rexmit = 0;
1883 	struct nfsnode *dnp = VTONFS(dvp);
1884 
1885 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1886 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1887 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1888 	nfsm_fhtom(dnp, v3);
1889 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1890 	nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1891 #ifndef NFS_V2_ONLY
1892 	if (v3)
1893 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1894 #endif
1895 	nfsm_reqdone;
1896 	VTONFS(dvp)->n_flag |= NMODIFIED;
1897 	if (!wccflag)
1898 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1899 	/*
1900 	 * Kludge City: If the first reply to the remove rpc is lost..
1901 	 *   the reply to the retransmitted request will be ENOENT
1902 	 *   since the file was in fact removed
1903 	 *   Therefore, we cheat and return success.
1904 	 */
1905 	if (rexmit && error == ENOENT)
1906 		error = 0;
1907 	return (error);
1908 }
1909 
1910 /*
1911  * nfs file rename call
1912  */
1913 int
1914 nfs_rename(v)
1915 	void *v;
1916 {
1917 	struct vop_rename_args  /* {
1918 		struct vnode *a_fdvp;
1919 		struct vnode *a_fvp;
1920 		struct componentname *a_fcnp;
1921 		struct vnode *a_tdvp;
1922 		struct vnode *a_tvp;
1923 		struct componentname *a_tcnp;
1924 	} */ *ap = v;
1925 	struct vnode *fvp = ap->a_fvp;
1926 	struct vnode *tvp = ap->a_tvp;
1927 	struct vnode *fdvp = ap->a_fdvp;
1928 	struct vnode *tdvp = ap->a_tdvp;
1929 	struct componentname *tcnp = ap->a_tcnp;
1930 	struct componentname *fcnp = ap->a_fcnp;
1931 	int error;
1932 
1933 #ifndef DIAGNOSTIC
1934 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1935 	    (fcnp->cn_flags & HASBUF) == 0)
1936 		panic("nfs_rename: no name");
1937 #endif
1938 	/* Check for cross-device rename */
1939 	if ((fvp->v_mount != tdvp->v_mount) ||
1940 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1941 		error = EXDEV;
1942 		goto out;
1943 	}
1944 
1945 	/*
1946 	 * If the tvp exists and is in use, sillyrename it before doing the
1947 	 * rename of the new file over it.
1948 	 *
1949 	 * Have sillyrename use link instead of rename if possible,
1950 	 * so that we don't lose the file if the rename fails, and so
1951 	 * that there's no window when the "to" file doesn't exist.
1952 	 */
1953 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1954 	    tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1955 		VN_KNOTE(tvp, NOTE_DELETE);
1956 		vput(tvp);
1957 		tvp = NULL;
1958 	}
1959 
1960 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1961 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1962 		curlwp);
1963 
1964 	VN_KNOTE(fdvp, NOTE_WRITE);
1965 	VN_KNOTE(tdvp, NOTE_WRITE);
1966 	if (error == 0 || error == EEXIST) {
1967 		if (fvp->v_type == VDIR)
1968 			cache_purge(fvp);
1969 		else
1970 			cache_purge1(fdvp, fcnp, 0);
1971 		if (tvp != NULL && tvp->v_type == VDIR)
1972 			cache_purge(tvp);
1973 		else
1974 			cache_purge1(tdvp, tcnp, 0);
1975 	}
1976 out:
1977 	if (tdvp == tvp)
1978 		vrele(tdvp);
1979 	else
1980 		vput(tdvp);
1981 	if (tvp)
1982 		vput(tvp);
1983 	vrele(fdvp);
1984 	vrele(fvp);
1985 	return (error);
1986 }
1987 
1988 /*
1989  * nfs file rename rpc called from nfs_remove() above
1990  */
1991 int
1992 nfs_renameit(sdvp, scnp, sp)
1993 	struct vnode *sdvp;
1994 	struct componentname *scnp;
1995 	struct sillyrename *sp;
1996 {
1997 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1998 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
1999 }
2000 
2001 /*
2002  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2003  */
2004 int
2005 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l)
2006 	struct vnode *fdvp;
2007 	const char *fnameptr;
2008 	int fnamelen;
2009 	struct vnode *tdvp;
2010 	const char *tnameptr;
2011 	int tnamelen;
2012 	kauth_cred_t cred;
2013 	struct lwp *l;
2014 {
2015 	u_int32_t *tl;
2016 	char *cp;
2017 #ifndef NFS_V2_ONLY
2018 	int32_t t1;
2019 	char *cp2;
2020 #endif
2021 	int32_t t2;
2022 	char *bpos, *dpos;
2023 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2024 	struct mbuf *mreq, *mrep, *md, *mb;
2025 	const int v3 = NFS_ISV3(fdvp);
2026 	int rexmit = 0;
2027 	struct nfsnode *fdnp = VTONFS(fdvp);
2028 
2029 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2030 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
2031 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2032 		nfsm_rndup(tnamelen));
2033 	nfsm_fhtom(fdnp, v3);
2034 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2035 	nfsm_fhtom(VTONFS(tdvp), v3);
2036 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2037 	nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
2038 #ifndef NFS_V2_ONLY
2039 	if (v3) {
2040 		nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2041 		nfsm_wcc_data(tdvp, twccflag, 0, !error);
2042 	}
2043 #endif
2044 	nfsm_reqdone;
2045 	VTONFS(fdvp)->n_flag |= NMODIFIED;
2046 	VTONFS(tdvp)->n_flag |= NMODIFIED;
2047 	if (!fwccflag)
2048 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2049 	if (!twccflag)
2050 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2051 	/*
2052 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2053 	 */
2054 	if (rexmit && error == ENOENT)
2055 		error = 0;
2056 	return (error);
2057 }
2058 
2059 /*
2060  * NFS link RPC, called from nfs_link.
2061  * Assumes dvp and vp locked, and leaves them that way.
2062  */
2063 
2064 static int
2065 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2066     size_t namelen, kauth_cred_t cred, struct lwp *l)
2067 {
2068 	u_int32_t *tl;
2069 	char *cp;
2070 #ifndef NFS_V2_ONLY
2071 	int32_t t1;
2072 	char *cp2;
2073 #endif
2074 	int32_t t2;
2075 	char *bpos, *dpos;
2076 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2077 	struct mbuf *mreq, *mrep, *md, *mb;
2078 	const int v3 = NFS_ISV3(dvp);
2079 	int rexmit = 0;
2080 	struct nfsnode *np = VTONFS(vp);
2081 
2082 	nfsstats.rpccnt[NFSPROC_LINK]++;
2083 	nfsm_reqhead(np, NFSPROC_LINK,
2084 	    NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2085 	nfsm_fhtom(np, v3);
2086 	nfsm_fhtom(VTONFS(dvp), v3);
2087 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2088 	nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2089 #ifndef NFS_V2_ONLY
2090 	if (v3) {
2091 		nfsm_postop_attr(vp, attrflag, 0);
2092 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2093 	}
2094 #endif
2095 	nfsm_reqdone;
2096 
2097 	VTONFS(dvp)->n_flag |= NMODIFIED;
2098 	if (!attrflag)
2099 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2100 	if (!wccflag)
2101 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2102 
2103 	/*
2104 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2105 	 */
2106 	if (rexmit && error == EEXIST)
2107 		error = 0;
2108 
2109 	return error;
2110 }
2111 
2112 /*
2113  * nfs hard link create call
2114  */
2115 int
2116 nfs_link(v)
2117 	void *v;
2118 {
2119 	struct vop_link_args /* {
2120 		struct vnode *a_dvp;
2121 		struct vnode *a_vp;
2122 		struct componentname *a_cnp;
2123 	} */ *ap = v;
2124 	struct vnode *vp = ap->a_vp;
2125 	struct vnode *dvp = ap->a_dvp;
2126 	struct componentname *cnp = ap->a_cnp;
2127 	int error = 0;
2128 
2129 	if (dvp->v_mount != vp->v_mount) {
2130 		VOP_ABORTOP(dvp, cnp);
2131 		vput(dvp);
2132 		return (EXDEV);
2133 	}
2134 	if (dvp != vp) {
2135 		error = vn_lock(vp, LK_EXCLUSIVE);
2136 		if (error != 0) {
2137 			VOP_ABORTOP(dvp, cnp);
2138 			vput(dvp);
2139 			return error;
2140 		}
2141 	}
2142 
2143 	/*
2144 	 * Push all writes to the server, so that the attribute cache
2145 	 * doesn't get "out of sync" with the server.
2146 	 * XXX There should be a better way!
2147 	 */
2148 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0);
2149 
2150 	error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2151 	    cnp->cn_cred, curlwp);
2152 
2153 	if (error == 0)
2154 		cache_purge1(dvp, cnp, 0);
2155 	PNBUF_PUT(cnp->cn_pnbuf);
2156 	if (dvp != vp)
2157 		VOP_UNLOCK(vp, 0);
2158 	VN_KNOTE(vp, NOTE_LINK);
2159 	VN_KNOTE(dvp, NOTE_WRITE);
2160 	vput(dvp);
2161 	return (error);
2162 }
2163 
2164 /*
2165  * nfs symbolic link create call
2166  */
2167 int
2168 nfs_symlink(v)
2169 	void *v;
2170 {
2171 	struct vop_symlink_args /* {
2172 		struct vnode *a_dvp;
2173 		struct vnode **a_vpp;
2174 		struct componentname *a_cnp;
2175 		struct vattr *a_vap;
2176 		char *a_target;
2177 	} */ *ap = v;
2178 	struct vnode *dvp = ap->a_dvp;
2179 	struct vattr *vap = ap->a_vap;
2180 	struct componentname *cnp = ap->a_cnp;
2181 	struct nfsv2_sattr *sp;
2182 	u_int32_t *tl;
2183 	char *cp;
2184 	int32_t t1, t2;
2185 	char *bpos, *dpos, *cp2;
2186 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2187 	struct mbuf *mreq, *mrep, *md, *mb;
2188 	struct vnode *newvp = (struct vnode *)0;
2189 	const int v3 = NFS_ISV3(dvp);
2190 	int rexmit = 0;
2191 	struct nfsnode *dnp = VTONFS(dvp);
2192 
2193 	*ap->a_vpp = NULL;
2194 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2195 	slen = strlen(ap->a_target);
2196 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2197 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2198 	nfsm_fhtom(dnp, v3);
2199 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2200 #ifndef NFS_V2_ONlY
2201 	if (v3)
2202 		nfsm_v3attrbuild(vap, false);
2203 #endif
2204 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2205 #ifndef NFS_V2_ONlY
2206 	if (!v3) {
2207 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2208 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2209 		sp->sa_uid = nfs_xdrneg1;
2210 		sp->sa_gid = nfs_xdrneg1;
2211 		sp->sa_size = nfs_xdrneg1;
2212 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2213 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2214 	}
2215 #endif
2216 	nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2217 	    &rexmit);
2218 #ifndef NFS_V2_ONlY
2219 	if (v3) {
2220 		if (!error)
2221 			nfsm_mtofh(dvp, newvp, v3, gotvp);
2222 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2223 	}
2224 #endif
2225 	nfsm_reqdone;
2226 	/*
2227 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2228 	 */
2229 	if (rexmit && error == EEXIST)
2230 		error = 0;
2231 	if (error == 0 || error == EEXIST)
2232 		cache_purge1(dvp, cnp, 0);
2233 	if (error == 0 && newvp == NULL) {
2234 		struct nfsnode *np = NULL;
2235 
2236 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2237 		    cnp->cn_cred, curlwp, &np);
2238 		if (error == 0)
2239 			newvp = NFSTOV(np);
2240 	}
2241 	if (error) {
2242 		if (newvp != NULL)
2243 			vput(newvp);
2244 	} else {
2245 		*ap->a_vpp = newvp;
2246 	}
2247 	PNBUF_PUT(cnp->cn_pnbuf);
2248 	VTONFS(dvp)->n_flag |= NMODIFIED;
2249 	if (!wccflag)
2250 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2251 	VN_KNOTE(dvp, NOTE_WRITE);
2252 	vput(dvp);
2253 	return (error);
2254 }
2255 
2256 /*
2257  * nfs make dir call
2258  */
2259 int
2260 nfs_mkdir(v)
2261 	void *v;
2262 {
2263 	struct vop_mkdir_args /* {
2264 		struct vnode *a_dvp;
2265 		struct vnode **a_vpp;
2266 		struct componentname *a_cnp;
2267 		struct vattr *a_vap;
2268 	} */ *ap = v;
2269 	struct vnode *dvp = ap->a_dvp;
2270 	struct vattr *vap = ap->a_vap;
2271 	struct componentname *cnp = ap->a_cnp;
2272 	struct nfsv2_sattr *sp;
2273 	u_int32_t *tl;
2274 	char *cp;
2275 	int32_t t1, t2;
2276 	int len;
2277 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2278 	struct vnode *newvp = (struct vnode *)0;
2279 	char *bpos, *dpos, *cp2;
2280 	int error = 0, wccflag = NFSV3_WCCRATTR;
2281 	int gotvp = 0;
2282 	int rexmit = 0;
2283 	struct mbuf *mreq, *mrep, *md, *mb;
2284 	const int v3 = NFS_ISV3(dvp);
2285 
2286 	len = cnp->cn_namelen;
2287 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2288 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
2289 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2290 	nfsm_fhtom(dnp, v3);
2291 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2292 #ifndef NFS_V2_ONLY
2293 	if (v3) {
2294 		nfsm_v3attrbuild(vap, false);
2295 	} else
2296 #endif
2297 	{
2298 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2299 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2300 		sp->sa_uid = nfs_xdrneg1;
2301 		sp->sa_gid = nfs_xdrneg1;
2302 		sp->sa_size = nfs_xdrneg1;
2303 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2304 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2305 	}
2306 	nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2307 	if (!error)
2308 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2309 	if (v3)
2310 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2311 	nfsm_reqdone;
2312 	VTONFS(dvp)->n_flag |= NMODIFIED;
2313 	if (!wccflag)
2314 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2315 	/*
2316 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2317 	 * if we can succeed in looking up the directory.
2318 	 */
2319 	if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2320 		if (newvp) {
2321 			vput(newvp);
2322 			newvp = (struct vnode *)0;
2323 		}
2324 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2325 			curlwp, &np);
2326 		if (!error) {
2327 			newvp = NFSTOV(np);
2328 			if (newvp->v_type != VDIR || newvp == dvp)
2329 				error = EEXIST;
2330 		}
2331 	}
2332 	if (error) {
2333 		if (newvp) {
2334 			if (dvp != newvp)
2335 				vput(newvp);
2336 			else
2337 				vrele(newvp);
2338 		}
2339 	} else {
2340 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2341 		if (cnp->cn_flags & MAKEENTRY)
2342 			nfs_cache_enter(dvp, newvp, cnp);
2343 		*ap->a_vpp = newvp;
2344 	}
2345 	PNBUF_PUT(cnp->cn_pnbuf);
2346 	vput(dvp);
2347 	return (error);
2348 }
2349 
2350 /*
2351  * nfs remove directory call
2352  */
2353 int
2354 nfs_rmdir(v)
2355 	void *v;
2356 {
2357 	struct vop_rmdir_args /* {
2358 		struct vnode *a_dvp;
2359 		struct vnode *a_vp;
2360 		struct componentname *a_cnp;
2361 	} */ *ap = v;
2362 	struct vnode *vp = ap->a_vp;
2363 	struct vnode *dvp = ap->a_dvp;
2364 	struct componentname *cnp = ap->a_cnp;
2365 	u_int32_t *tl;
2366 	char *cp;
2367 #ifndef NFS_V2_ONLY
2368 	int32_t t1;
2369 	char *cp2;
2370 #endif
2371 	int32_t t2;
2372 	char *bpos, *dpos;
2373 	int error = 0, wccflag = NFSV3_WCCRATTR;
2374 	int rexmit = 0;
2375 	struct mbuf *mreq, *mrep, *md, *mb;
2376 	const int v3 = NFS_ISV3(dvp);
2377 	struct nfsnode *dnp;
2378 
2379 	if (dvp == vp) {
2380 		vrele(dvp);
2381 		vput(dvp);
2382 		PNBUF_PUT(cnp->cn_pnbuf);
2383 		return (EINVAL);
2384 	}
2385 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2386 	dnp = VTONFS(dvp);
2387 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2388 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2389 	nfsm_fhtom(dnp, v3);
2390 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2391 	nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2392 #ifndef NFS_V2_ONLY
2393 	if (v3)
2394 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2395 #endif
2396 	nfsm_reqdone;
2397 	PNBUF_PUT(cnp->cn_pnbuf);
2398 	VTONFS(dvp)->n_flag |= NMODIFIED;
2399 	if (!wccflag)
2400 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2401 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2402 	VN_KNOTE(vp, NOTE_DELETE);
2403 	cache_purge(vp);
2404 	vput(vp);
2405 	vput(dvp);
2406 	/*
2407 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2408 	 */
2409 	if (rexmit && error == ENOENT)
2410 		error = 0;
2411 	return (error);
2412 }
2413 
2414 /*
2415  * nfs readdir call
2416  */
2417 int
2418 nfs_readdir(v)
2419 	void *v;
2420 {
2421 	struct vop_readdir_args /* {
2422 		struct vnode *a_vp;
2423 		struct uio *a_uio;
2424 		kauth_cred_t a_cred;
2425 		int *a_eofflag;
2426 		off_t **a_cookies;
2427 		int *a_ncookies;
2428 	} */ *ap = v;
2429 	struct vnode *vp = ap->a_vp;
2430 	struct uio *uio = ap->a_uio;
2431 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2432 	char *base = uio->uio_iov->iov_base;
2433 	int tresid, error;
2434 	size_t count, lost;
2435 	struct dirent *dp;
2436 	off_t *cookies = NULL;
2437 	int ncookies = 0, nc;
2438 
2439 	if (vp->v_type != VDIR)
2440 		return (EPERM);
2441 
2442 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2443 	count = uio->uio_resid - lost;
2444 	if (count <= 0)
2445 		return (EINVAL);
2446 
2447 	/*
2448 	 * Call nfs_bioread() to do the real work.
2449 	 */
2450 	tresid = uio->uio_resid = count;
2451 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2452 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2453 
2454 	if (!error && ap->a_cookies) {
2455 		ncookies = count / 16;
2456 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2457 		*ap->a_cookies = cookies;
2458 	}
2459 
2460 	if (!error && uio->uio_resid == tresid) {
2461 		uio->uio_resid += lost;
2462 		nfsstats.direofcache_misses++;
2463 		if (ap->a_cookies)
2464 			*ap->a_ncookies = 0;
2465 		*ap->a_eofflag = 1;
2466 		return (0);
2467 	}
2468 
2469 	if (!error && ap->a_cookies) {
2470 		/*
2471 		 * Only the NFS server and emulations use cookies, and they
2472 		 * load the directory block into system space, so we can
2473 		 * just look at it directly.
2474 		 */
2475 		if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2476 		    uio->uio_iovcnt != 1)
2477 			panic("nfs_readdir: lost in space");
2478 		for (nc = 0; ncookies-- &&
2479 		     base < (char *)uio->uio_iov->iov_base; nc++){
2480 			dp = (struct dirent *) base;
2481 			if (dp->d_reclen == 0)
2482 				break;
2483 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2484 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2485 			else
2486 				*(cookies++) = NFS_GETCOOKIE(dp);
2487 			base += dp->d_reclen;
2488 		}
2489 		uio->uio_resid +=
2490 		    ((char *)uio->uio_iov->iov_base - base);
2491 		uio->uio_iov->iov_len +=
2492 		    ((char *)uio->uio_iov->iov_base - base);
2493 		uio->uio_iov->iov_base = base;
2494 		*ap->a_ncookies = nc;
2495 	}
2496 
2497 	uio->uio_resid += lost;
2498 	*ap->a_eofflag = 0;
2499 	return (error);
2500 }
2501 
2502 /*
2503  * Readdir rpc call.
2504  * Called from below the buffer cache by nfs_doio().
2505  */
2506 int
2507 nfs_readdirrpc(vp, uiop, cred)
2508 	struct vnode *vp;
2509 	struct uio *uiop;
2510 	kauth_cred_t cred;
2511 {
2512 	int len, left;
2513 	struct dirent *dp = NULL;
2514 	u_int32_t *tl;
2515 	char *cp;
2516 	int32_t t1, t2;
2517 	char *bpos, *dpos, *cp2;
2518 	struct mbuf *mreq, *mrep, *md, *mb;
2519 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2520 	struct nfsnode *dnp = VTONFS(vp);
2521 	u_quad_t fileno;
2522 	int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2523 #ifndef NFS_V2_ONLY
2524 	int attrflag;
2525 #endif
2526 	int nrpcs = 0, reclen;
2527 	const int v3 = NFS_ISV3(vp);
2528 
2529 #ifdef DIAGNOSTIC
2530 	/*
2531 	 * Should be called from buffer cache, so only amount of
2532 	 * NFS_DIRBLKSIZ will be requested.
2533 	 */
2534 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2535 		panic("nfs readdirrpc bad uio");
2536 #endif
2537 
2538 	/*
2539 	 * Loop around doing readdir rpc's of size nm_readdirsize
2540 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2541 	 * The stopping criteria is EOF or buffer full.
2542 	 */
2543 	while (more_dirs && bigenough) {
2544 		/*
2545 		 * Heuristic: don't bother to do another RPC to further
2546 		 * fill up this block if there is not much room left. (< 50%
2547 		 * of the readdir RPC size). This wastes some buffer space
2548 		 * but can save up to 50% in RPC calls.
2549 		 */
2550 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2551 			bigenough = 0;
2552 			break;
2553 		}
2554 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2555 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2556 			NFSX_READDIR(v3));
2557 		nfsm_fhtom(dnp, v3);
2558 #ifndef NFS_V2_ONLY
2559 		if (v3) {
2560 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2561 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2562 				txdr_swapcookie3(uiop->uio_offset, tl);
2563 			} else {
2564 				txdr_cookie3(uiop->uio_offset, tl);
2565 			}
2566 			tl += 2;
2567 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2568 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2569 		} else
2570 #endif
2571 		{
2572 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2573 			*tl++ = txdr_unsigned(uiop->uio_offset);
2574 		}
2575 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2576 		nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2577 		nrpcs++;
2578 #ifndef NFS_V2_ONLY
2579 		if (v3) {
2580 			nfsm_postop_attr(vp, attrflag, 0);
2581 			if (!error) {
2582 				nfsm_dissect(tl, u_int32_t *,
2583 				    2 * NFSX_UNSIGNED);
2584 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2585 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2586 			} else {
2587 				m_freem(mrep);
2588 				goto nfsmout;
2589 			}
2590 		}
2591 #endif
2592 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2593 		more_dirs = fxdr_unsigned(int, *tl);
2594 
2595 		/* loop thru the dir entries, doctoring them to 4bsd form */
2596 		while (more_dirs && bigenough) {
2597 #ifndef NFS_V2_ONLY
2598 			if (v3) {
2599 				nfsm_dissect(tl, u_int32_t *,
2600 				    3 * NFSX_UNSIGNED);
2601 				fileno = fxdr_hyper(tl);
2602 				len = fxdr_unsigned(int, *(tl + 2));
2603 			} else
2604 #endif
2605 			{
2606 				nfsm_dissect(tl, u_int32_t *,
2607 				    2 * NFSX_UNSIGNED);
2608 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2609 				len = fxdr_unsigned(int, *tl);
2610 			}
2611 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2612 				error = EBADRPC;
2613 				m_freem(mrep);
2614 				goto nfsmout;
2615 			}
2616 			/* for cookie stashing */
2617 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2618 			left = NFS_DIRFRAGSIZ - blksiz;
2619 			if (reclen > left) {
2620 				memset(uiop->uio_iov->iov_base, 0, left);
2621 				dp->d_reclen += left;
2622 				UIO_ADVANCE(uiop, left);
2623 				blksiz = 0;
2624 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2625 			}
2626 			if (reclen > uiop->uio_resid)
2627 				bigenough = 0;
2628 			if (bigenough) {
2629 				int tlen;
2630 
2631 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2632 				dp->d_fileno = fileno;
2633 				dp->d_namlen = len;
2634 				dp->d_reclen = reclen;
2635 				dp->d_type = DT_UNKNOWN;
2636 				blksiz += reclen;
2637 				if (blksiz == NFS_DIRFRAGSIZ)
2638 					blksiz = 0;
2639 				UIO_ADVANCE(uiop, DIRHDSIZ);
2640 				nfsm_mtouio(uiop, len);
2641 				tlen = reclen - (DIRHDSIZ + len);
2642 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2643 				UIO_ADVANCE(uiop, tlen);
2644 			} else
2645 				nfsm_adv(nfsm_rndup(len));
2646 #ifndef NFS_V2_ONLY
2647 			if (v3) {
2648 				nfsm_dissect(tl, u_int32_t *,
2649 				    3 * NFSX_UNSIGNED);
2650 			} else
2651 #endif
2652 			{
2653 				nfsm_dissect(tl, u_int32_t *,
2654 				    2 * NFSX_UNSIGNED);
2655 			}
2656 			if (bigenough) {
2657 #ifndef NFS_V2_ONLY
2658 				if (v3) {
2659 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2660 						uiop->uio_offset =
2661 						    fxdr_swapcookie3(tl);
2662 					else
2663 						uiop->uio_offset =
2664 						    fxdr_cookie3(tl);
2665 				}
2666 				else
2667 #endif
2668 				{
2669 					uiop->uio_offset =
2670 					    fxdr_unsigned(off_t, *tl);
2671 				}
2672 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2673 			}
2674 			if (v3)
2675 				tl += 2;
2676 			else
2677 				tl++;
2678 			more_dirs = fxdr_unsigned(int, *tl);
2679 		}
2680 		/*
2681 		 * If at end of rpc data, get the eof boolean
2682 		 */
2683 		if (!more_dirs) {
2684 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2685 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2686 
2687 			/*
2688 			 * kludge: if we got no entries, treat it as EOF.
2689 			 * some server sometimes send a reply without any
2690 			 * entries or EOF.
2691 			 * although it might mean the server has very long name,
2692 			 * we can't handle such entries anyway.
2693 			 */
2694 
2695 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2696 				more_dirs = 0;
2697 		}
2698 		m_freem(mrep);
2699 	}
2700 	/*
2701 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2702 	 * by increasing d_reclen for the last record.
2703 	 */
2704 	if (blksiz > 0) {
2705 		left = NFS_DIRFRAGSIZ - blksiz;
2706 		memset(uiop->uio_iov->iov_base, 0, left);
2707 		dp->d_reclen += left;
2708 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2709 		UIO_ADVANCE(uiop, left);
2710 	}
2711 
2712 	/*
2713 	 * We are now either at the end of the directory or have filled the
2714 	 * block.
2715 	 */
2716 	if (bigenough) {
2717 		dnp->n_direofoffset = uiop->uio_offset;
2718 		dnp->n_flag |= NEOFVALID;
2719 	}
2720 nfsmout:
2721 	return (error);
2722 }
2723 
2724 #ifndef NFS_V2_ONLY
2725 /*
2726  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2727  */
2728 int
2729 nfs_readdirplusrpc(vp, uiop, cred)
2730 	struct vnode *vp;
2731 	struct uio *uiop;
2732 	kauth_cred_t cred;
2733 {
2734 	int len, left;
2735 	struct dirent *dp = NULL;
2736 	u_int32_t *tl;
2737 	char *cp;
2738 	int32_t t1, t2;
2739 	struct vnode *newvp;
2740 	char *bpos, *dpos, *cp2;
2741 	struct mbuf *mreq, *mrep, *md, *mb;
2742 	struct nameidata nami, *ndp = &nami;
2743 	struct componentname *cnp = &ndp->ni_cnd;
2744 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2745 	struct nfsnode *dnp = VTONFS(vp), *np;
2746 	nfsfh_t *fhp;
2747 	u_quad_t fileno;
2748 	int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2749 	int attrflag, fhsize, nrpcs = 0, reclen;
2750 	struct nfs_fattr fattr, *fp;
2751 
2752 #ifdef DIAGNOSTIC
2753 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2754 		panic("nfs readdirplusrpc bad uio");
2755 #endif
2756 	ndp->ni_dvp = vp;
2757 	newvp = NULLVP;
2758 
2759 	/*
2760 	 * Loop around doing readdir rpc's of size nm_readdirsize
2761 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2762 	 * The stopping criteria is EOF or buffer full.
2763 	 */
2764 	while (more_dirs && bigenough) {
2765 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2766 			bigenough = 0;
2767 			break;
2768 		}
2769 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2770 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2771 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2772 		nfsm_fhtom(dnp, 1);
2773  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2774 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2775 			txdr_swapcookie3(uiop->uio_offset, tl);
2776 		} else {
2777 			txdr_cookie3(uiop->uio_offset, tl);
2778 		}
2779 		tl += 2;
2780 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2781 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2782 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2783 		*tl = txdr_unsigned(nmp->nm_rsize);
2784 		nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2785 		nfsm_postop_attr(vp, attrflag, 0);
2786 		if (error) {
2787 			m_freem(mrep);
2788 			goto nfsmout;
2789 		}
2790 		nrpcs++;
2791 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2792 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2793 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2794 		more_dirs = fxdr_unsigned(int, *tl);
2795 
2796 		/* loop thru the dir entries, doctoring them to 4bsd form */
2797 		while (more_dirs && bigenough) {
2798 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2799 			fileno = fxdr_hyper(tl);
2800 			len = fxdr_unsigned(int, *(tl + 2));
2801 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2802 				error = EBADRPC;
2803 				m_freem(mrep);
2804 				goto nfsmout;
2805 			}
2806 			/* for cookie stashing */
2807 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2808 			left = NFS_DIRFRAGSIZ - blksiz;
2809 			if (reclen > left) {
2810 				/*
2811 				 * DIRFRAGSIZ is aligned, no need to align
2812 				 * again here.
2813 				 */
2814 				memset(uiop->uio_iov->iov_base, 0, left);
2815 				dp->d_reclen += left;
2816 				UIO_ADVANCE(uiop, left);
2817 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2818 				blksiz = 0;
2819 			}
2820 			if (reclen > uiop->uio_resid)
2821 				bigenough = 0;
2822 			if (bigenough) {
2823 				int tlen;
2824 
2825 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2826 				dp->d_fileno = fileno;
2827 				dp->d_namlen = len;
2828 				dp->d_reclen = reclen;
2829 				dp->d_type = DT_UNKNOWN;
2830 				blksiz += reclen;
2831 				if (blksiz == NFS_DIRFRAGSIZ)
2832 					blksiz = 0;
2833 				UIO_ADVANCE(uiop, DIRHDSIZ);
2834 				nfsm_mtouio(uiop, len);
2835 				tlen = reclen - (DIRHDSIZ + len);
2836 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2837 				UIO_ADVANCE(uiop, tlen);
2838 				cnp->cn_nameptr = dp->d_name;
2839 				cnp->cn_namelen = dp->d_namlen;
2840 			} else
2841 				nfsm_adv(nfsm_rndup(len));
2842 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2843 			if (bigenough) {
2844 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2845 					uiop->uio_offset =
2846 						fxdr_swapcookie3(tl);
2847 				else
2848 					uiop->uio_offset =
2849 						fxdr_cookie3(tl);
2850 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2851 			}
2852 			tl += 2;
2853 
2854 			/*
2855 			 * Since the attributes are before the file handle
2856 			 * (sigh), we must skip over the attributes and then
2857 			 * come back and get them.
2858 			 */
2859 			attrflag = fxdr_unsigned(int, *tl);
2860 			if (attrflag) {
2861 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2862 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2863 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2864 			    doit = fxdr_unsigned(int, *tl);
2865 			    if (doit) {
2866 				nfsm_getfh(fhp, fhsize, 1);
2867 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2868 				    VREF(vp);
2869 				    newvp = vp;
2870 				    np = dnp;
2871 				} else {
2872 				    error = nfs_nget1(vp->v_mount, fhp,
2873 					fhsize, &np, LK_NOWAIT);
2874 				    if (!error)
2875 					newvp = NFSTOV(np);
2876 				}
2877 				if (!error) {
2878 				    const char *xcp;
2879 
2880 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2881 				    if (bigenough) {
2882 					dp->d_type =
2883 					   IFTODT(VTTOIF(np->n_vattr->va_type));
2884 					if (cnp->cn_namelen <= NCHNAMLEN) {
2885 					    ndp->ni_vp = newvp;
2886 					    xcp = cnp->cn_nameptr +
2887 						cnp->cn_namelen;
2888 					    cnp->cn_hash =
2889 					       namei_hash(cnp->cn_nameptr, &xcp);
2890 					    nfs_cache_enter(ndp->ni_dvp,
2891 						ndp->ni_vp, cnp);
2892 					}
2893 				    }
2894 				}
2895 				error = 0;
2896 			   }
2897 			} else {
2898 			    /* Just skip over the file handle */
2899 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2900 			    i = fxdr_unsigned(int, *tl);
2901 			    nfsm_adv(nfsm_rndup(i));
2902 			}
2903 			if (newvp != NULLVP) {
2904 			    if (newvp == vp)
2905 				vrele(newvp);
2906 			    else
2907 				vput(newvp);
2908 			    newvp = NULLVP;
2909 			}
2910 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2911 			more_dirs = fxdr_unsigned(int, *tl);
2912 		}
2913 		/*
2914 		 * If at end of rpc data, get the eof boolean
2915 		 */
2916 		if (!more_dirs) {
2917 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2918 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2919 
2920 			/*
2921 			 * kludge: see a comment in nfs_readdirrpc.
2922 			 */
2923 
2924 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2925 				more_dirs = 0;
2926 		}
2927 		m_freem(mrep);
2928 	}
2929 	/*
2930 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2931 	 * by increasing d_reclen for the last record.
2932 	 */
2933 	if (blksiz > 0) {
2934 		left = NFS_DIRFRAGSIZ - blksiz;
2935 		memset(uiop->uio_iov->iov_base, 0, left);
2936 		dp->d_reclen += left;
2937 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2938 		UIO_ADVANCE(uiop, left);
2939 	}
2940 
2941 	/*
2942 	 * We are now either at the end of the directory or have filled the
2943 	 * block.
2944 	 */
2945 	if (bigenough) {
2946 		dnp->n_direofoffset = uiop->uio_offset;
2947 		dnp->n_flag |= NEOFVALID;
2948 	}
2949 nfsmout:
2950 	if (newvp != NULLVP) {
2951 		if(newvp == vp)
2952 		    vrele(newvp);
2953 		else
2954 		    vput(newvp);
2955 	}
2956 	return (error);
2957 }
2958 #endif
2959 
2960 /*
2961  * Silly rename. To make the NFS filesystem that is stateless look a little
2962  * more like the "ufs" a remove of an active vnode is translated to a rename
2963  * to a funny looking filename that is removed by nfs_inactive on the
2964  * nfsnode. There is the potential for another process on a different client
2965  * to create the same funny name between the nfs_lookitup() fails and the
2966  * nfs_rename() completes, but...
2967  */
2968 int
2969 nfs_sillyrename(dvp, vp, cnp, dolink)
2970 	struct vnode *dvp, *vp;
2971 	struct componentname *cnp;
2972 	bool dolink;
2973 {
2974 	struct sillyrename *sp;
2975 	struct nfsnode *np;
2976 	int error;
2977 	short pid;
2978 
2979 	cache_purge(dvp);
2980 	np = VTONFS(vp);
2981 #ifndef DIAGNOSTIC
2982 	if (vp->v_type == VDIR)
2983 		panic("nfs: sillyrename dir");
2984 #endif
2985 	sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
2986 	sp->s_cred = kauth_cred_dup(cnp->cn_cred);
2987 	sp->s_dvp = dvp;
2988 	VREF(dvp);
2989 
2990 	/* Fudge together a funny name */
2991 	pid = curlwp->l_proc->p_pid;
2992 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2993 	sp->s_namlen = 12;
2994 	sp->s_name[8] = hexdigits[pid & 0xf];
2995 	sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
2996 	sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
2997 	sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
2998 
2999 	/* Try lookitups until we get one that isn't there */
3000 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3001 		curlwp, (struct nfsnode **)0) == 0) {
3002 		sp->s_name[4]++;
3003 		if (sp->s_name[4] > 'z') {
3004 			error = EINVAL;
3005 			goto bad;
3006 		}
3007 	}
3008 	if (dolink) {
3009 		error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
3010 		    sp->s_cred, curlwp);
3011 		/*
3012 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
3013 		 */
3014 		if (error == ENOTSUP) {
3015 			error = nfs_renameit(dvp, cnp, sp);
3016 		}
3017 	} else {
3018 		error = nfs_renameit(dvp, cnp, sp);
3019 	}
3020 	if (error)
3021 		goto bad;
3022 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3023 		curlwp, &np);
3024 	np->n_sillyrename = sp;
3025 	return (0);
3026 bad:
3027 	vrele(sp->s_dvp);
3028 	kauth_cred_free(sp->s_cred);
3029 	kmem_free(sp, sizeof(*sp));
3030 	return (error);
3031 }
3032 
3033 /*
3034  * Look up a file name and optionally either update the file handle or
3035  * allocate an nfsnode, depending on the value of npp.
3036  * npp == NULL	--> just do the lookup
3037  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3038  *			handled too
3039  * *npp != NULL --> update the file handle in the vnode
3040  */
3041 int
3042 nfs_lookitup(dvp, name, len, cred, l, npp)
3043 	struct vnode *dvp;
3044 	const char *name;
3045 	int len;
3046 	kauth_cred_t cred;
3047 	struct lwp *l;
3048 	struct nfsnode **npp;
3049 {
3050 	u_int32_t *tl;
3051 	char *cp;
3052 	int32_t t1, t2;
3053 	struct vnode *newvp = (struct vnode *)0;
3054 	struct nfsnode *np, *dnp = VTONFS(dvp);
3055 	char *bpos, *dpos, *cp2;
3056 	int error = 0, fhlen;
3057 #ifndef NFS_V2_ONLY
3058 	int attrflag;
3059 #endif
3060 	struct mbuf *mreq, *mrep, *md, *mb;
3061 	nfsfh_t *nfhp;
3062 	const int v3 = NFS_ISV3(dvp);
3063 
3064 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3065 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3066 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3067 	nfsm_fhtom(dnp, v3);
3068 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
3069 	nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
3070 	if (npp && !error) {
3071 		nfsm_getfh(nfhp, fhlen, v3);
3072 		if (*npp) {
3073 		    np = *npp;
3074 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3075 			kmem_free(np->n_fhp, np->n_fhsize);
3076 			np->n_fhp = &np->n_fh;
3077 		    }
3078 #if NFS_SMALLFH < NFSX_V3FHMAX
3079 		    else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
3080 			np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
3081 #endif
3082 		    memcpy(np->n_fhp, nfhp, fhlen);
3083 		    np->n_fhsize = fhlen;
3084 		    newvp = NFSTOV(np);
3085 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3086 		    VREF(dvp);
3087 		    newvp = dvp;
3088 		    np = dnp;
3089 		} else {
3090 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3091 		    if (error) {
3092 			m_freem(mrep);
3093 			return (error);
3094 		    }
3095 		    newvp = NFSTOV(np);
3096 		}
3097 #ifndef NFS_V2_ONLY
3098 		if (v3) {
3099 			nfsm_postop_attr(newvp, attrflag, 0);
3100 			if (!attrflag && *npp == NULL) {
3101 				m_freem(mrep);
3102 				vput(newvp);
3103 				return (ENOENT);
3104 			}
3105 		} else
3106 #endif
3107 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
3108 	}
3109 	nfsm_reqdone;
3110 	if (npp && *npp == NULL) {
3111 		if (error) {
3112 			if (newvp)
3113 				vput(newvp);
3114 		} else
3115 			*npp = np;
3116 	}
3117 	return (error);
3118 }
3119 
3120 #ifndef NFS_V2_ONLY
3121 /*
3122  * Nfs Version 3 commit rpc
3123  */
3124 int
3125 nfs_commit(vp, offset, cnt, l)
3126 	struct vnode *vp;
3127 	off_t offset;
3128 	uint32_t cnt;
3129 	struct lwp *l;
3130 {
3131 	char *cp;
3132 	u_int32_t *tl;
3133 	int32_t t1, t2;
3134 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3135 	char *bpos, *dpos, *cp2;
3136 	int error = 0, wccflag = NFSV3_WCCRATTR;
3137 	struct mbuf *mreq, *mrep, *md, *mb;
3138 	struct nfsnode *np;
3139 
3140 	KASSERT(NFS_ISV3(vp));
3141 
3142 #ifdef NFS_DEBUG_COMMIT
3143 	printf("commit %lu - %lu\n", (unsigned long)offset,
3144 	    (unsigned long)(offset + cnt));
3145 #endif
3146 
3147 	mutex_enter(&nmp->nm_lock);
3148 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3149 		mutex_exit(&nmp->nm_lock);
3150 		return (0);
3151 	}
3152 	mutex_exit(&nmp->nm_lock);
3153 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3154 	np = VTONFS(vp);
3155 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3156 	nfsm_fhtom(np, 1);
3157 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3158 	txdr_hyper(offset, tl);
3159 	tl += 2;
3160 	*tl = txdr_unsigned(cnt);
3161 	nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3162 	nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3163 	if (!error) {
3164 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3165 		mutex_enter(&nmp->nm_lock);
3166 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3167 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3168 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3169 			error = NFSERR_STALEWRITEVERF;
3170 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3171 		}
3172 		mutex_exit(&nmp->nm_lock);
3173 	}
3174 	nfsm_reqdone;
3175 	return (error);
3176 }
3177 #endif
3178 
3179 /*
3180  * Kludge City..
3181  * - make nfs_bmap() essentially a no-op that does no translation
3182  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3183  *   (Maybe I could use the process's page mapping, but I was concerned that
3184  *    Kernel Write might not be enabled and also figured copyout() would do
3185  *    a lot more work than memcpy() and also it currently happens in the
3186  *    context of the swapper process (2).
3187  */
3188 int
3189 nfs_bmap(v)
3190 	void *v;
3191 {
3192 	struct vop_bmap_args /* {
3193 		struct vnode *a_vp;
3194 		daddr_t  a_bn;
3195 		struct vnode **a_vpp;
3196 		daddr_t *a_bnp;
3197 		int *a_runp;
3198 	} */ *ap = v;
3199 	struct vnode *vp = ap->a_vp;
3200 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3201 
3202 	if (ap->a_vpp != NULL)
3203 		*ap->a_vpp = vp;
3204 	if (ap->a_bnp != NULL)
3205 		*ap->a_bnp = ap->a_bn << bshift;
3206 	if (ap->a_runp != NULL)
3207 		*ap->a_runp = 1024 * 1024; /* XXX */
3208 	return (0);
3209 }
3210 
3211 /*
3212  * Strategy routine.
3213  * For async requests when nfsiod(s) are running, queue the request by
3214  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3215  * request.
3216  */
3217 int
3218 nfs_strategy(v)
3219 	void *v;
3220 {
3221 	struct vop_strategy_args *ap = v;
3222 	struct buf *bp = ap->a_bp;
3223 	int error = 0;
3224 
3225 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3226 		panic("nfs physio/async");
3227 
3228 	/*
3229 	 * If the op is asynchronous and an i/o daemon is waiting
3230 	 * queue the request, wake it up and wait for completion
3231 	 * otherwise just do it ourselves.
3232 	 */
3233 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3234 		error = nfs_doio(bp);
3235 	return (error);
3236 }
3237 
3238 /*
3239  * fsync vnode op. Just call nfs_flush() with commit == 1.
3240  */
3241 /* ARGSUSED */
3242 int
3243 nfs_fsync(v)
3244 	void *v;
3245 {
3246 	struct vop_fsync_args /* {
3247 		struct vnodeop_desc *a_desc;
3248 		struct vnode * a_vp;
3249 		kauth_cred_t  a_cred;
3250 		int  a_flags;
3251 		off_t offlo;
3252 		off_t offhi;
3253 		struct lwp * a_l;
3254 	} */ *ap = v;
3255 
3256 	struct vnode *vp = ap->a_vp;
3257 
3258 	if (vp->v_type != VREG)
3259 		return 0;
3260 
3261 	return (nfs_flush(vp, ap->a_cred,
3262 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3263 }
3264 
3265 /*
3266  * Flush all the data associated with a vnode.
3267  */
3268 int
3269 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3270     int commit)
3271 {
3272 	struct nfsnode *np = VTONFS(vp);
3273 	int error;
3274 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3275 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3276 
3277 	mutex_enter(&vp->v_interlock);
3278 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3279 	if (np->n_flag & NWRITEERR) {
3280 		error = np->n_error;
3281 		np->n_flag &= ~NWRITEERR;
3282 	}
3283 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3284 	return (error);
3285 }
3286 
3287 /*
3288  * Return POSIX pathconf information applicable to nfs.
3289  *
3290  * N.B. The NFS V2 protocol doesn't support this RPC.
3291  */
3292 /* ARGSUSED */
3293 int
3294 nfs_pathconf(v)
3295 	void *v;
3296 {
3297 	struct vop_pathconf_args /* {
3298 		struct vnode *a_vp;
3299 		int a_name;
3300 		register_t *a_retval;
3301 	} */ *ap = v;
3302 	struct nfsv3_pathconf *pcp;
3303 	struct vnode *vp = ap->a_vp;
3304 	struct mbuf *mreq, *mrep, *md, *mb;
3305 	int32_t t1, t2;
3306 	u_int32_t *tl;
3307 	char *bpos, *dpos, *cp, *cp2;
3308 	int error = 0, attrflag;
3309 #ifndef NFS_V2_ONLY
3310 	struct nfsmount *nmp;
3311 	unsigned int l;
3312 	u_int64_t maxsize;
3313 #endif
3314 	const int v3 = NFS_ISV3(vp);
3315 	struct nfsnode *np = VTONFS(vp);
3316 
3317 	switch (ap->a_name) {
3318 		/* Names that can be resolved locally. */
3319 	case _PC_PIPE_BUF:
3320 		*ap->a_retval = PIPE_BUF;
3321 		break;
3322 	case _PC_SYNC_IO:
3323 		*ap->a_retval = 1;
3324 		break;
3325 	/* Names that cannot be resolved locally; do an RPC, if possible. */
3326 	case _PC_LINK_MAX:
3327 	case _PC_NAME_MAX:
3328 	case _PC_CHOWN_RESTRICTED:
3329 	case _PC_NO_TRUNC:
3330 		if (!v3) {
3331 			error = EINVAL;
3332 			break;
3333 		}
3334 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3335 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3336 		nfsm_fhtom(np, 1);
3337 		nfsm_request(np, NFSPROC_PATHCONF,
3338 		    curlwp, curlwp->l_cred);	/* XXX */
3339 		nfsm_postop_attr(vp, attrflag, 0);
3340 		if (!error) {
3341 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
3342 			    NFSX_V3PATHCONF);
3343 			switch (ap->a_name) {
3344 			case _PC_LINK_MAX:
3345 				*ap->a_retval =
3346 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3347 				break;
3348 			case _PC_NAME_MAX:
3349 				*ap->a_retval =
3350 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3351 				break;
3352 			case _PC_CHOWN_RESTRICTED:
3353 				*ap->a_retval =
3354 				    (pcp->pc_chownrestricted == nfs_true);
3355 				break;
3356 			case _PC_NO_TRUNC:
3357 				*ap->a_retval =
3358 				    (pcp->pc_notrunc == nfs_true);
3359 				break;
3360 			}
3361 		}
3362 		nfsm_reqdone;
3363 		break;
3364 	case _PC_FILESIZEBITS:
3365 #ifndef NFS_V2_ONLY
3366 		if (v3) {
3367 			nmp = VFSTONFS(vp->v_mount);
3368 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3369 				if ((error = nfs_fsinfo(nmp, vp,
3370 				    curlwp->l_cred, curlwp)) != 0) /* XXX */
3371 					break;
3372 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3373 			    (maxsize >> l) > 0; l++)
3374 				;
3375 			*ap->a_retval = l + 1;
3376 		} else
3377 #endif
3378 		{
3379 			*ap->a_retval = 32;	/* NFS V2 limitation */
3380 		}
3381 		break;
3382 	default:
3383 		error = EINVAL;
3384 		break;
3385 	}
3386 
3387 	return (error);
3388 }
3389 
3390 /*
3391  * NFS advisory byte-level locks.
3392  */
3393 int
3394 nfs_advlock(v)
3395 	void *v;
3396 {
3397 	struct vop_advlock_args /* {
3398 		struct vnode *a_vp;
3399 		void *a_id;
3400 		int  a_op;
3401 		struct flock *a_fl;
3402 		int  a_flags;
3403 	} */ *ap = v;
3404 	struct nfsnode *np = VTONFS(ap->a_vp);
3405 
3406 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3407 }
3408 
3409 /*
3410  * Print out the contents of an nfsnode.
3411  */
3412 int
3413 nfs_print(v)
3414 	void *v;
3415 {
3416 	struct vop_print_args /* {
3417 		struct vnode *a_vp;
3418 	} */ *ap = v;
3419 	struct vnode *vp = ap->a_vp;
3420 	struct nfsnode *np = VTONFS(vp);
3421 
3422 	printf("tag VT_NFS, fileid %lld fsid 0x%llx",
3423 	    (unsigned long long)np->n_vattr->va_fileid,
3424 	    (unsigned long long)np->n_vattr->va_fsid);
3425 	if (vp->v_type == VFIFO)
3426 		fifo_printinfo(vp);
3427 	printf("\n");
3428 	return (0);
3429 }
3430 
3431 /*
3432  * nfs unlock wrapper.
3433  */
3434 int
3435 nfs_unlock(void *v)
3436 {
3437 	struct vop_unlock_args /* {
3438 		struct vnode *a_vp;
3439 		int a_flags;
3440 	} */ *ap = v;
3441 	struct vnode *vp = ap->a_vp;
3442 
3443 	/*
3444 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3445 	 * with v_data == 0.
3446 	 */
3447 	if (VTONFS(vp)) {
3448 		nfs_delayedtruncate(vp);
3449 	}
3450 
3451 	return genfs_unlock(v);
3452 }
3453 
3454 /*
3455  * nfs special file access vnode op.
3456  * Essentially just get vattr and then imitate iaccess() since the device is
3457  * local to the client.
3458  */
3459 int
3460 nfsspec_access(v)
3461 	void *v;
3462 {
3463 	struct vop_access_args /* {
3464 		struct vnode *a_vp;
3465 		int  a_mode;
3466 		kauth_cred_t a_cred;
3467 		struct lwp *a_l;
3468 	} */ *ap = v;
3469 	struct vattr va;
3470 	struct vnode *vp = ap->a_vp;
3471 	int error;
3472 
3473 	error = VOP_GETATTR(vp, &va, ap->a_cred);
3474 	if (error)
3475 		return (error);
3476 
3477         /*
3478 	 * Disallow write attempts on filesystems mounted read-only;
3479 	 * unless the file is a socket, fifo, or a block or character
3480 	 * device resident on the filesystem.
3481 	 */
3482 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3483 		switch (vp->v_type) {
3484 		case VREG:
3485 		case VDIR:
3486 		case VLNK:
3487 			return (EROFS);
3488 		default:
3489 			break;
3490 		}
3491 	}
3492 
3493 	return (vaccess(va.va_type, va.va_mode,
3494 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3495 }
3496 
3497 /*
3498  * Read wrapper for special devices.
3499  */
3500 int
3501 nfsspec_read(v)
3502 	void *v;
3503 {
3504 	struct vop_read_args /* {
3505 		struct vnode *a_vp;
3506 		struct uio *a_uio;
3507 		int  a_ioflag;
3508 		kauth_cred_t a_cred;
3509 	} */ *ap = v;
3510 	struct nfsnode *np = VTONFS(ap->a_vp);
3511 
3512 	/*
3513 	 * Set access flag.
3514 	 */
3515 	np->n_flag |= NACC;
3516 	getnanotime(&np->n_atim);
3517 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3518 }
3519 
3520 /*
3521  * Write wrapper for special devices.
3522  */
3523 int
3524 nfsspec_write(v)
3525 	void *v;
3526 {
3527 	struct vop_write_args /* {
3528 		struct vnode *a_vp;
3529 		struct uio *a_uio;
3530 		int  a_ioflag;
3531 		kauth_cred_t a_cred;
3532 	} */ *ap = v;
3533 	struct nfsnode *np = VTONFS(ap->a_vp);
3534 
3535 	/*
3536 	 * Set update flag.
3537 	 */
3538 	np->n_flag |= NUPD;
3539 	getnanotime(&np->n_mtim);
3540 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3541 }
3542 
3543 /*
3544  * Close wrapper for special devices.
3545  *
3546  * Update the times on the nfsnode then do device close.
3547  */
3548 int
3549 nfsspec_close(v)
3550 	void *v;
3551 {
3552 	struct vop_close_args /* {
3553 		struct vnode *a_vp;
3554 		int  a_fflag;
3555 		kauth_cred_t a_cred;
3556 		struct lwp *a_l;
3557 	} */ *ap = v;
3558 	struct vnode *vp = ap->a_vp;
3559 	struct nfsnode *np = VTONFS(vp);
3560 	struct vattr vattr;
3561 
3562 	if (np->n_flag & (NACC | NUPD)) {
3563 		np->n_flag |= NCHG;
3564 		if (vp->v_usecount == 1 &&
3565 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3566 			VATTR_NULL(&vattr);
3567 			if (np->n_flag & NACC)
3568 				vattr.va_atime = np->n_atim;
3569 			if (np->n_flag & NUPD)
3570 				vattr.va_mtime = np->n_mtim;
3571 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3572 		}
3573 	}
3574 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3575 }
3576 
3577 /*
3578  * Read wrapper for fifos.
3579  */
3580 int
3581 nfsfifo_read(v)
3582 	void *v;
3583 {
3584 	struct vop_read_args /* {
3585 		struct vnode *a_vp;
3586 		struct uio *a_uio;
3587 		int  a_ioflag;
3588 		kauth_cred_t a_cred;
3589 	} */ *ap = v;
3590 	struct nfsnode *np = VTONFS(ap->a_vp);
3591 
3592 	/*
3593 	 * Set access flag.
3594 	 */
3595 	np->n_flag |= NACC;
3596 	getnanotime(&np->n_atim);
3597 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3598 }
3599 
3600 /*
3601  * Write wrapper for fifos.
3602  */
3603 int
3604 nfsfifo_write(v)
3605 	void *v;
3606 {
3607 	struct vop_write_args /* {
3608 		struct vnode *a_vp;
3609 		struct uio *a_uio;
3610 		int  a_ioflag;
3611 		kauth_cred_t a_cred;
3612 	} */ *ap = v;
3613 	struct nfsnode *np = VTONFS(ap->a_vp);
3614 
3615 	/*
3616 	 * Set update flag.
3617 	 */
3618 	np->n_flag |= NUPD;
3619 	getnanotime(&np->n_mtim);
3620 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3621 }
3622 
3623 /*
3624  * Close wrapper for fifos.
3625  *
3626  * Update the times on the nfsnode then do fifo close.
3627  */
3628 int
3629 nfsfifo_close(v)
3630 	void *v;
3631 {
3632 	struct vop_close_args /* {
3633 		struct vnode *a_vp;
3634 		int  a_fflag;
3635 		kauth_cred_t a_cred;
3636 		struct lwp *a_l;
3637 	} */ *ap = v;
3638 	struct vnode *vp = ap->a_vp;
3639 	struct nfsnode *np = VTONFS(vp);
3640 	struct vattr vattr;
3641 
3642 	if (np->n_flag & (NACC | NUPD)) {
3643 		struct timespec ts;
3644 
3645 		getnanotime(&ts);
3646 		if (np->n_flag & NACC)
3647 			np->n_atim = ts;
3648 		if (np->n_flag & NUPD)
3649 			np->n_mtim = ts;
3650 		np->n_flag |= NCHG;
3651 		if (vp->v_usecount == 1 &&
3652 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3653 			VATTR_NULL(&vattr);
3654 			if (np->n_flag & NACC)
3655 				vattr.va_atime = np->n_atim;
3656 			if (np->n_flag & NUPD)
3657 				vattr.va_mtime = np->n_mtim;
3658 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3659 		}
3660 	}
3661 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3662 }
3663