xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: nfs_vnops.c,v 1.244 2006/10/14 09:18:57 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_vnops.c	8.19 (Berkeley) 7/31/95
35  */
36 
37 /*
38  * vnode op calls for Sun NFS version 2 and 3
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.244 2006/10/14 09:18:57 yamt Exp $");
43 
44 #include "opt_inet.h"
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 
48 #include <sys/param.h>
49 #include <sys/proc.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/resourcevar.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/disk.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/hash.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/unistd.h>
67 #include <sys/kauth.h>
68 
69 #include <uvm/uvm_extern.h>
70 #include <uvm/uvm.h>
71 
72 #include <miscfs/fifofs/fifo.h>
73 #include <miscfs/genfs/genfs.h>
74 #include <miscfs/genfs/genfs_node.h>
75 #include <miscfs/specfs/specdev.h>
76 
77 #include <nfs/rpcv2.h>
78 #include <nfs/nfsproto.h>
79 #include <nfs/nfs.h>
80 #include <nfs/nfsnode.h>
81 #include <nfs/nfsmount.h>
82 #include <nfs/xdr_subs.h>
83 #include <nfs/nfsm_subs.h>
84 #include <nfs/nqnfs.h>
85 #include <nfs/nfs_var.h>
86 
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 
91 /*
92  * Global vfs data structures for nfs
93  */
94 int (**nfsv2_vnodeop_p) __P((void *));
95 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
96 	{ &vop_default_desc, vn_default_error },
97 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
98 	{ &vop_create_desc, nfs_create },		/* create */
99 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
100 	{ &vop_open_desc, nfs_open },			/* open */
101 	{ &vop_close_desc, nfs_close },			/* close */
102 	{ &vop_access_desc, nfs_access },		/* access */
103 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
104 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
105 	{ &vop_read_desc, nfs_read },			/* read */
106 	{ &vop_write_desc, nfs_write },			/* write */
107 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
108 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
109 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
110 	{ &vop_poll_desc, nfs_poll },			/* poll */
111 	{ &vop_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
112 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
113 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
114 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
115 	{ &vop_seek_desc, nfs_seek },			/* seek */
116 	{ &vop_remove_desc, nfs_remove },		/* remove */
117 	{ &vop_link_desc, nfs_link },			/* link */
118 	{ &vop_rename_desc, nfs_rename },		/* rename */
119 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
120 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
121 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
122 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
123 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
124 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
125 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
126 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
127 	{ &vop_lock_desc, nfs_lock },			/* lock */
128 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
129 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
130 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
131 	{ &vop_print_desc, nfs_print },			/* print */
132 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
133 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
134 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
135 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
136 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
137 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
138 	{ NULL, NULL }
139 };
140 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
141 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
142 
143 /*
144  * Special device vnode ops
145  */
146 int (**spec_nfsv2nodeop_p) __P((void *));
147 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
148 	{ &vop_default_desc, vn_default_error },
149 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
150 	{ &vop_create_desc, spec_create },		/* create */
151 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
152 	{ &vop_open_desc, spec_open },			/* open */
153 	{ &vop_close_desc, nfsspec_close },		/* close */
154 	{ &vop_access_desc, nfsspec_access },		/* access */
155 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
156 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
157 	{ &vop_read_desc, nfsspec_read },		/* read */
158 	{ &vop_write_desc, nfsspec_write },		/* write */
159 	{ &vop_lease_desc, spec_lease_check },		/* lease */
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_lease_desc, fifo_lease_check },		/* lease */
209 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
210 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
211 	{ &vop_poll_desc, fifo_poll },			/* poll */
212 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
213 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
214 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
215 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
216 	{ &vop_seek_desc, fifo_seek },			/* seek */
217 	{ &vop_remove_desc, fifo_remove },		/* remove */
218 	{ &vop_link_desc, fifo_link },			/* link */
219 	{ &vop_rename_desc, fifo_rename },		/* rename */
220 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
221 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
222 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
223 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
224 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
225 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
226 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
227 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
228 	{ &vop_lock_desc, nfs_lock },			/* lock */
229 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
230 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
231 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
232 	{ &vop_print_desc, nfs_print },			/* print */
233 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
234 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
235 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
236 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
237 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
238 	{ NULL, NULL }
239 };
240 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
241 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
242 
243 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
244     size_t, kauth_cred_t, struct lwp *);
245 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
246 
247 /*
248  * Global variables
249  */
250 extern u_int32_t nfs_true, nfs_false;
251 extern u_int32_t nfs_xdrneg1;
252 extern const nfstype nfsv3_type[9];
253 
254 int nfs_numasync = 0;
255 #define	DIRHDSIZ	_DIRENT_NAMEOFF(dp)
256 #define UIO_ADVANCE(uio, siz) \
257     (void)((uio)->uio_resid -= (siz), \
258     (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
259     (uio)->uio_iov->iov_len -= (siz))
260 
261 static void nfs_cache_enter(struct vnode *, struct vnode *,
262     struct componentname *);
263 
264 static void
265 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
266     struct componentname *cnp)
267 {
268 	struct nfsnode *dnp = VTONFS(dvp);
269 
270 	if (vp != NULL) {
271 		struct nfsnode *np = VTONFS(vp);
272 
273 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
274 	}
275 
276 	if (!timespecisset(&dnp->n_nctime))
277 		dnp->n_nctime = dnp->n_vattr->va_mtime;
278 
279 	cache_enter(dvp, vp, cnp);
280 }
281 
282 /*
283  * nfs null call from vfs.
284  */
285 int
286 nfs_null(vp, cred, l)
287 	struct vnode *vp;
288 	kauth_cred_t cred;
289 	struct lwp *l;
290 {
291 	caddr_t bpos, dpos;
292 	int error = 0;
293 	struct mbuf *mreq, *mrep, *md, *mb;
294 	struct nfsnode *np = VTONFS(vp);
295 
296 	nfsm_reqhead(np, NFSPROC_NULL, 0);
297 	nfsm_request(np, NFSPROC_NULL, l, cred);
298 	nfsm_reqdone;
299 	return (error);
300 }
301 
302 /*
303  * nfs access vnode op.
304  * For nfs version 2, just return ok. File accesses may fail later.
305  * For nfs version 3, use the access rpc to check accessibility. If file modes
306  * are changed on the server, accesses might still fail later.
307  */
308 int
309 nfs_access(v)
310 	void *v;
311 {
312 	struct vop_access_args /* {
313 		struct vnode *a_vp;
314 		int  a_mode;
315 		kauth_cred_t a_cred;
316 		struct lwp *a_l;
317 	} */ *ap = v;
318 	struct vnode *vp = ap->a_vp;
319 #ifndef NFS_V2_ONLY
320 	u_int32_t *tl;
321 	caddr_t cp;
322 	int32_t t1, t2;
323 	caddr_t bpos, dpos, cp2;
324 	int error = 0, attrflag;
325 	struct mbuf *mreq, *mrep, *md, *mb;
326 	u_int32_t mode, rmode;
327 	const int v3 = NFS_ISV3(vp);
328 #endif
329 	int cachevalid;
330 	struct nfsnode *np = VTONFS(vp);
331 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
332 
333 	cachevalid = (np->n_accstamp != -1 &&
334 	    (time_uptime - np->n_accstamp) < NFS_ATTRTIMEO(nmp, np) &&
335 	    np->n_accuid == kauth_cred_geteuid(ap->a_cred));
336 
337 	/*
338 	 * Check access cache first. If this request has been made for this
339 	 * uid shortly before, use the cached result.
340 	 */
341 	if (cachevalid) {
342 		if (!np->n_accerror) {
343 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
344 				return np->n_accerror;
345 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
346 			return np->n_accerror;
347 	}
348 
349 #ifndef NFS_V2_ONLY
350 	/*
351 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
352 	 * ufs_access() locally using the vattr. This may not be correct,
353 	 * since the server may apply other access criteria such as
354 	 * client uid-->server uid mapping that we do not know about, but
355 	 * this is better than just returning anything that is lying about
356 	 * in the cache.
357 	 */
358 	if (v3) {
359 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
360 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
361 		nfsm_fhtom(np, v3);
362 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
363 		if (ap->a_mode & VREAD)
364 			mode = NFSV3ACCESS_READ;
365 		else
366 			mode = 0;
367 		if (vp->v_type != VDIR) {
368 			if (ap->a_mode & VWRITE)
369 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
370 			if (ap->a_mode & VEXEC)
371 				mode |= NFSV3ACCESS_EXECUTE;
372 		} else {
373 			if (ap->a_mode & VWRITE)
374 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
375 					 NFSV3ACCESS_DELETE);
376 			if (ap->a_mode & VEXEC)
377 				mode |= NFSV3ACCESS_LOOKUP;
378 		}
379 		*tl = txdr_unsigned(mode);
380 		nfsm_request(np, NFSPROC_ACCESS, ap->a_l, ap->a_cred);
381 		nfsm_postop_attr(vp, attrflag, 0);
382 		if (!error) {
383 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
384 			rmode = fxdr_unsigned(u_int32_t, *tl);
385 			/*
386 			 * The NFS V3 spec does not clarify whether or not
387 			 * the returned access bits can be a superset of
388 			 * the ones requested, so...
389 			 */
390 			if ((rmode & mode) != mode)
391 				error = EACCES;
392 		}
393 		nfsm_reqdone;
394 	} else
395 #endif
396 		return (nfsspec_access(ap));
397 #ifndef NFS_V2_ONLY
398 	/*
399 	 * Disallow write attempts on filesystems mounted read-only;
400 	 * unless the file is a socket, fifo, or a block or character
401 	 * device resident on the filesystem.
402 	 */
403 	if (!error && (ap->a_mode & VWRITE) &&
404 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
405 		switch (vp->v_type) {
406 		case VREG:
407 		case VDIR:
408 		case VLNK:
409 			error = EROFS;
410 		default:
411 			break;
412 		}
413 	}
414 
415 	if (!error || error == EACCES) {
416 		/*
417 		 * If we got the same result as for a previous,
418 		 * different request, OR it in. Don't update
419 		 * the timestamp in that case.
420 		 */
421 		if (cachevalid && np->n_accstamp != -1 &&
422 		    error == np->n_accerror) {
423 			if (!error)
424 				np->n_accmode |= ap->a_mode;
425 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
426 				np->n_accmode = ap->a_mode;
427 		} else {
428 			np->n_accstamp = time_uptime;
429 			np->n_accuid = kauth_cred_geteuid(ap->a_cred);
430 			np->n_accmode = ap->a_mode;
431 			np->n_accerror = error;
432 		}
433 	}
434 
435 	return (error);
436 #endif
437 }
438 
439 /*
440  * nfs open vnode op
441  * Check to see if the type is ok
442  * and that deletion is not in progress.
443  * For paged in text files, you will need to flush the page cache
444  * if consistency is lost.
445  */
446 /* ARGSUSED */
447 int
448 nfs_open(v)
449 	void *v;
450 {
451 	struct vop_open_args /* {
452 		struct vnode *a_vp;
453 		int  a_mode;
454 		kauth_cred_t a_cred;
455 		struct lwp *a_l;
456 	} */ *ap = v;
457 	struct vnode *vp = ap->a_vp;
458 	struct nfsnode *np = VTONFS(vp);
459 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
460 	int error;
461 
462 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
463 		return (EACCES);
464 	}
465 
466 	if (ap->a_mode & FREAD) {
467 		if (np->n_rcred != NULL)
468 			kauth_cred_free(np->n_rcred);
469 		np->n_rcred = ap->a_cred;
470 		kauth_cred_hold(np->n_rcred);
471 	}
472 	if (ap->a_mode & FWRITE) {
473 		if (np->n_wcred != NULL)
474 			kauth_cred_free(np->n_wcred);
475 		np->n_wcred = ap->a_cred;
476 		kauth_cred_hold(np->n_wcred);
477 	}
478 
479 #ifndef NFS_V2_ONLY
480 	/*
481 	 * Get a valid lease. If cached data is stale, flush it.
482 	 */
483 	if (nmp->nm_flag & NFSMNT_NQNFS) {
484 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
485 		    do {
486 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
487 			    ap->a_l);
488 		    } while (error == NQNFS_EXPIRED);
489 		    if (error)
490 			return (error);
491 		    if (np->n_lrev != np->n_brev ||
492 			(np->n_flag & NQNFSNONCACHE)) {
493 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
494 				ap->a_l, 1)) == EINTR)
495 				return (error);
496 			np->n_brev = np->n_lrev;
497 		    }
498 		}
499 	} else
500 #endif
501 	{
502 		error = nfs_flushstalebuf(vp, ap->a_cred, ap->a_l, 0);
503 		if (error)
504 			return error;
505 	}
506 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
507 		NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
508 	return (0);
509 }
510 
511 /*
512  * nfs close vnode op
513  * What an NFS client should do upon close after writing is a debatable issue.
514  * Most NFS clients push delayed writes to the server upon close, basically for
515  * two reasons:
516  * 1 - So that any write errors may be reported back to the client process
517  *     doing the close system call. By far the two most likely errors are
518  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
519  * 2 - To put a worst case upper bound on cache inconsistency between
520  *     multiple clients for the file.
521  * There is also a consistency problem for Version 2 of the protocol w.r.t.
522  * not being able to tell if other clients are writing a file concurrently,
523  * since there is no way of knowing if the changed modify time in the reply
524  * is only due to the write for this client.
525  * (NFS Version 3 provides weak cache consistency data in the reply that
526  *  should be sufficient to detect and handle this case.)
527  *
528  * The current code does the following:
529  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
530  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
531  *                     or commit them (this satisfies 1 and 2 except for the
532  *                     case where the server crashes after this close but
533  *                     before the commit RPC, which is felt to be "good
534  *                     enough". Changing the last argument to nfs_flush() to
535  *                     a 1 would force a commit operation, if it is felt a
536  *                     commit is necessary now.
537  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
538  *                     1 should be dealt with via an fsync() system call for
539  *                     cases where write errors are important.
540  */
541 /* ARGSUSED */
542 int
543 nfs_close(v)
544 	void *v;
545 {
546 	struct vop_close_args /* {
547 		struct vnodeop_desc *a_desc;
548 		struct vnode *a_vp;
549 		int  a_fflag;
550 		kauth_cred_t a_cred;
551 		struct lwp *a_l;
552 	} */ *ap = v;
553 	struct vnode *vp = ap->a_vp;
554 	struct nfsnode *np = VTONFS(vp);
555 	int error = 0;
556 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
557 
558 	if (vp->v_type == VREG) {
559 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
560 		(np->n_flag & NMODIFIED)) {
561 #ifndef NFS_V2_ONLY
562 		if (NFS_ISV3(vp)) {
563 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_l, 0);
564 		    np->n_flag &= ~NMODIFIED;
565 		} else
566 #endif
567 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 1);
568 		NFS_INVALIDATE_ATTRCACHE(np);
569 	    }
570 	    if (np->n_flag & NWRITEERR) {
571 		np->n_flag &= ~NWRITEERR;
572 		error = np->n_error;
573 	    }
574 	}
575 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
576 	return (error);
577 }
578 
579 /*
580  * nfs getattr call from vfs.
581  */
582 int
583 nfs_getattr(v)
584 	void *v;
585 {
586 	struct vop_getattr_args /* {
587 		struct vnode *a_vp;
588 		struct vattr *a_vap;
589 		kauth_cred_t a_cred;
590 		struct lwp *a_l;
591 	} */ *ap = v;
592 	struct vnode *vp = ap->a_vp;
593 	struct nfsnode *np = VTONFS(vp);
594 	caddr_t cp;
595 	u_int32_t *tl;
596 	int32_t t1, t2;
597 	caddr_t bpos, dpos;
598 	int error = 0;
599 	struct mbuf *mreq, *mrep, *md, *mb;
600 	const int v3 = NFS_ISV3(vp);
601 
602 	/*
603 	 * Update local times for special files.
604 	 */
605 	if (np->n_flag & (NACC | NUPD))
606 		np->n_flag |= NCHG;
607 
608 	/*
609 	 * if we have delayed truncation, do it now.
610 	 */
611 	nfs_delayedtruncate(vp);
612 
613 	/*
614 	 * First look in the cache.
615 	 */
616 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
617 		return (0);
618 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
619 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
620 	nfsm_fhtom(np, v3);
621 	nfsm_request(np, NFSPROC_GETATTR, ap->a_l, ap->a_cred);
622 	if (!error) {
623 		nfsm_loadattr(vp, ap->a_vap, 0);
624 		if (vp->v_type == VDIR &&
625 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
626 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
627 	}
628 	nfsm_reqdone;
629 	return (error);
630 }
631 
632 /*
633  * nfs setattr call.
634  */
635 int
636 nfs_setattr(v)
637 	void *v;
638 {
639 	struct vop_setattr_args /* {
640 		struct vnodeop_desc *a_desc;
641 		struct vnode *a_vp;
642 		struct vattr *a_vap;
643 		kauth_cred_t a_cred;
644 		struct lwp *a_l;
645 	} */ *ap = v;
646 	struct vnode *vp = ap->a_vp;
647 	struct nfsnode *np = VTONFS(vp);
648 	struct vattr *vap = ap->a_vap;
649 	int error = 0;
650 	u_quad_t tsize = 0;
651 
652 	/*
653 	 * Setting of flags is not supported.
654 	 */
655 	if (vap->va_flags != VNOVAL)
656 		return (EOPNOTSUPP);
657 
658 	/*
659 	 * Disallow write attempts if the filesystem is mounted read-only.
660 	 */
661   	if ((vap->va_uid != (uid_t)VNOVAL ||
662 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
663 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
664 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
665 		return (EROFS);
666 	if (vap->va_size != VNOVAL) {
667  		switch (vp->v_type) {
668  		case VDIR:
669  			return (EISDIR);
670  		case VCHR:
671  		case VBLK:
672  		case VSOCK:
673  		case VFIFO:
674 			if (vap->va_mtime.tv_sec == VNOVAL &&
675 			    vap->va_atime.tv_sec == VNOVAL &&
676 			    vap->va_mode == (mode_t)VNOVAL &&
677 			    vap->va_uid == (uid_t)VNOVAL &&
678 			    vap->va_gid == (gid_t)VNOVAL)
679 				return (0);
680  			vap->va_size = VNOVAL;
681  			break;
682  		default:
683 			/*
684 			 * Disallow write attempts if the filesystem is
685 			 * mounted read-only.
686 			 */
687 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
688 				return (EROFS);
689 			genfs_node_wrlock(vp);
690  			uvm_vnp_setsize(vp, vap->va_size);
691  			tsize = np->n_size;
692 			np->n_size = vap->va_size;
693  			if (vap->va_size == 0)
694  				error = nfs_vinvalbuf(vp, 0,
695  				     ap->a_cred, ap->a_l, 1);
696 			else
697 				error = nfs_vinvalbuf(vp, V_SAVE,
698 				     ap->a_cred, ap->a_l, 1);
699 			if (error) {
700 				uvm_vnp_setsize(vp, tsize);
701 				genfs_node_unlock(vp);
702 				return (error);
703 			}
704  			np->n_vattr->va_size = vap->va_size;
705   		}
706   	} else {
707 		/*
708 		 * flush files before setattr because a later write of
709 		 * cached data might change timestamps or reset sugid bits
710 		 */
711 		if ((vap->va_mtime.tv_sec != VNOVAL ||
712 		     vap->va_atime.tv_sec != VNOVAL ||
713 		     vap->va_mode != VNOVAL) &&
714 		    vp->v_type == VREG &&
715   		    (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
716 		 			   ap->a_l, 1)) == EINTR)
717 			return (error);
718 	}
719 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_l);
720 	if (vap->va_size != VNOVAL) {
721 		if (error) {
722 			np->n_size = np->n_vattr->va_size = tsize;
723 			uvm_vnp_setsize(vp, np->n_size);
724 		}
725 		genfs_node_unlock(vp);
726 	}
727 	VN_KNOTE(vp, NOTE_ATTRIB);
728 	return (error);
729 }
730 
731 /*
732  * Do an nfs setattr rpc.
733  */
734 int
735 nfs_setattrrpc(vp, vap, cred, l)
736 	struct vnode *vp;
737 	struct vattr *vap;
738 	kauth_cred_t cred;
739 	struct lwp *l;
740 {
741 	struct nfsv2_sattr *sp;
742 	caddr_t cp;
743 	int32_t t1, t2;
744 	caddr_t bpos, dpos;
745 	u_int32_t *tl;
746 	int error = 0;
747 	struct mbuf *mreq, *mrep, *md, *mb;
748 	const int v3 = NFS_ISV3(vp);
749 	struct nfsnode *np = VTONFS(vp);
750 #ifndef NFS_V2_ONLY
751 	int wccflag = NFSV3_WCCRATTR;
752 	caddr_t cp2;
753 #endif
754 
755 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
756 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
757 	nfsm_fhtom(np, v3);
758 #ifndef NFS_V2_ONLY
759 	if (v3) {
760 		nfsm_v3attrbuild(vap, TRUE);
761 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
762 		*tl = nfs_false;
763 	} else {
764 #endif
765 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
766 		if (vap->va_mode == (mode_t)VNOVAL)
767 			sp->sa_mode = nfs_xdrneg1;
768 		else
769 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
770 		if (vap->va_uid == (uid_t)VNOVAL)
771 			sp->sa_uid = nfs_xdrneg1;
772 		else
773 			sp->sa_uid = txdr_unsigned(vap->va_uid);
774 		if (vap->va_gid == (gid_t)VNOVAL)
775 			sp->sa_gid = nfs_xdrneg1;
776 		else
777 			sp->sa_gid = txdr_unsigned(vap->va_gid);
778 		sp->sa_size = txdr_unsigned(vap->va_size);
779 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
780 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
781 #ifndef NFS_V2_ONLY
782 	}
783 #endif
784 	nfsm_request(np, NFSPROC_SETATTR, l, cred);
785 #ifndef NFS_V2_ONLY
786 	if (v3) {
787 		nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, FALSE);
788 	} else
789 #endif
790 		nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
791 	nfsm_reqdone;
792 	return (error);
793 }
794 
795 /*
796  * nfs lookup call, one step at a time...
797  * First look in cache
798  * If not found, unlock the directory nfsnode and do the rpc
799  *
800  * This code is full of lock/unlock statements and checks, because
801  * we continue after cache_lookup has finished (we need to check
802  * with the attr cache and do an rpc if it has timed out). This means
803  * that the locking effects of cache_lookup have to be taken into
804  * account.
805  */
806 int
807 nfs_lookup(v)
808 	void *v;
809 {
810 	struct vop_lookup_args /* {
811 		struct vnodeop_desc *a_desc;
812 		struct vnode *a_dvp;
813 		struct vnode **a_vpp;
814 		struct componentname *a_cnp;
815 	} */ *ap = v;
816 	struct componentname *cnp = ap->a_cnp;
817 	struct vnode *dvp = ap->a_dvp;
818 	struct vnode **vpp = ap->a_vpp;
819 	int flags;
820 	struct vnode *newvp;
821 	u_int32_t *tl;
822 	caddr_t cp;
823 	int32_t t1, t2;
824 	caddr_t bpos, dpos, cp2;
825 	struct mbuf *mreq, *mrep, *md, *mb;
826 	long len;
827 	nfsfh_t *fhp;
828 	struct nfsnode *np;
829 	int lockparent, wantparent, error = 0, attrflag, fhsize;
830 	const int v3 = NFS_ISV3(dvp);
831 
832 	cnp->cn_flags &= ~PDIRUNLOCK;
833 	flags = cnp->cn_flags;
834 
835 	*vpp = NULLVP;
836 	newvp = NULLVP;
837 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
838 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
839 		return (EROFS);
840 	if (dvp->v_type != VDIR)
841 		return (ENOTDIR);
842 
843 	/*
844 	 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
845 	 */
846 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
847 		error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp);
848 		if (error)
849 			return error;
850 		if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
851 			return EISDIR;
852 		VREF(dvp);
853 		*vpp = dvp;
854 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
855 				cnp->cn_flags |= SAVENAME;
856 		return 0;
857 	}
858 
859 	lockparent = flags & LOCKPARENT;
860 	wantparent = flags & (LOCKPARENT|WANTPARENT);
861 	np = VTONFS(dvp);
862 
863 	/*
864 	 * Before tediously performing a linear scan of the directory,
865 	 * check the name cache to see if the directory/name pair
866 	 * we are looking for is known already.
867 	 * If the directory/name pair is found in the name cache,
868 	 * we have to ensure the directory has not changed from
869 	 * the time the cache entry has been created. If it has,
870 	 * the cache entry has to be ignored.
871 	 */
872 	error = cache_lookup_raw(dvp, vpp, cnp);
873 	KASSERT(dvp != *vpp);
874 	if (error >= 0) {
875 		struct vattr vattr;
876 		int err2;
877 
878 		if (error && error != ENOENT) {
879 			*vpp = NULLVP;
880 			return error;
881 		}
882 
883 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp);
884 		if (err2 != 0) {
885 			if (error == 0)
886 				vrele(*vpp);
887 			*vpp = NULLVP;
888 			return err2;
889 		}
890 
891 		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
892 		    cnp->cn_lwp) || timespeccmp(&vattr.va_mtime,
893 		    &VTONFS(dvp)->n_nctime, !=)) {
894 			if (error == 0) {
895 				vrele(*vpp);
896 				*vpp = NULLVP;
897 			}
898 			cache_purge1(dvp, NULL, PURGE_CHILDREN);
899 			timespecclear(&np->n_nctime);
900 			goto dorpc;
901 		}
902 
903 		if (error == ENOENT) {
904 			goto noentry;
905 		}
906 
907 		newvp = *vpp;
908 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp)
909 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
910 			nfsstats.lookupcache_hits++;
911 			if ((flags & ISDOTDOT) != 0 ||
912 			    (~flags & (LOCKPARENT|ISLASTCN)) != 0) {
913 				VOP_UNLOCK(dvp, 0);
914 				cnp->cn_flags |= PDIRUNLOCK;
915 			}
916 			error = vn_lock(newvp, LK_EXCLUSIVE);
917 			if (error) {
918 				/* newvp has been revoked. */
919 				vrele(newvp);
920 				*vpp = NULL;
921 				return error;
922 			}
923 			if ((~flags & (LOCKPARENT|ISLASTCN)) == 0
924 			    && (cnp->cn_flags & PDIRUNLOCK)) {
925 				KASSERT(flags & ISDOTDOT);
926 				error = vn_lock(dvp, LK_EXCLUSIVE);
927 				if (error) {
928 					vput(newvp);
929 					*vpp = NULL;
930 					return error;
931 				}
932 				cnp->cn_flags &= ~PDIRUNLOCK;
933 			}
934 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
935 				cnp->cn_flags |= SAVENAME;
936 			KASSERT(newvp->v_type != VNON);
937 			return (0);
938 		}
939 		cache_purge1(newvp, NULL, PURGE_PARENTS);
940 		vrele(newvp);
941 		*vpp = NULLVP;
942 	}
943 dorpc:
944 #if 0
945 	/*
946 	 * because nfsv3 has the same CREATE semantics as ours,
947 	 * we don't have to perform LOOKUPs beforehand.
948 	 *
949 	 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
950 	 * XXX although we have no way to know if O_EXCL is requested or not.
951 	 */
952 
953 	if (v3 && cnp->cn_nameiop == CREATE &&
954 	    (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
955 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
956 		KASSERT(lockparent);
957 		cnp->cn_flags |= SAVENAME;
958 		return (EJUSTRETURN);
959 	}
960 #endif /* 0 */
961 
962 	error = 0;
963 	newvp = NULLVP;
964 	nfsstats.lookupcache_misses++;
965 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
966 	len = cnp->cn_namelen;
967 	nfsm_reqhead(np, NFSPROC_LOOKUP,
968 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
969 	nfsm_fhtom(np, v3);
970 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
971 	nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_lwp, cnp->cn_cred);
972 	if (error) {
973 		nfsm_postop_attr(dvp, attrflag, 0);
974 		m_freem(mrep);
975 		goto nfsmout;
976 	}
977 	nfsm_getfh(fhp, fhsize, v3);
978 
979 	/*
980 	 * Handle RENAME case...
981 	 */
982 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
983 		if (NFS_CMPFH(np, fhp, fhsize)) {
984 			m_freem(mrep);
985 			return (EISDIR);
986 		}
987 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
988 		if (error) {
989 			m_freem(mrep);
990 			return error;
991 		}
992 		newvp = NFSTOV(np);
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 		*vpp = newvp;
1001 		m_freem(mrep);
1002 		cnp->cn_flags |= SAVENAME;
1003 		if (!lockparent) {
1004 			VOP_UNLOCK(dvp, 0);
1005 			cnp->cn_flags |= PDIRUNLOCK;
1006 		}
1007 		goto validate;
1008 	}
1009 
1010 	/*
1011 	 * The postop attr handling is duplicated for each if case,
1012 	 * because it should be done while dvp is locked (unlocking
1013 	 * dvp is different for each case).
1014 	 */
1015 
1016 	if (NFS_CMPFH(np, fhp, fhsize)) {
1017 		/*
1018 		 * "." lookup
1019 		 */
1020 		VREF(dvp);
1021 		newvp = dvp;
1022 #ifndef NFS_V2_ONLY
1023 		if (v3) {
1024 			nfsm_postop_attr(newvp, attrflag, 0);
1025 			nfsm_postop_attr(dvp, attrflag, 0);
1026 		} else
1027 #endif
1028 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1029 	} else if (flags & ISDOTDOT) {
1030 		/*
1031 		 * ".." lookup
1032 		 */
1033 		VOP_UNLOCK(dvp, 0);
1034 		cnp->cn_flags |= PDIRUNLOCK;
1035 
1036 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1037 		if (error) {
1038 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
1039 				cnp->cn_flags &= ~PDIRUNLOCK;
1040 			m_freem(mrep);
1041 			return error;
1042 		}
1043 		newvp = NFSTOV(np);
1044 
1045 #ifndef NFS_V2_ONLY
1046 		if (v3) {
1047 			nfsm_postop_attr(newvp, attrflag, 0);
1048 			nfsm_postop_attr(dvp, attrflag, 0);
1049 		} else
1050 #endif
1051 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1052 
1053 		if (lockparent && (flags & ISLASTCN)) {
1054 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
1055 				m_freem(mrep);
1056 				vput(newvp);
1057 				return error;
1058 			}
1059 			cnp->cn_flags &= ~PDIRUNLOCK;
1060 		}
1061 	} else {
1062 		/*
1063 		 * Other lookups.
1064 		 */
1065 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1066 		if (error) {
1067 			m_freem(mrep);
1068 			return error;
1069 		}
1070 		newvp = NFSTOV(np);
1071 #ifndef NFS_V2_ONLY
1072 		if (v3) {
1073 			nfsm_postop_attr(newvp, attrflag, 0);
1074 			nfsm_postop_attr(dvp, attrflag, 0);
1075 		} else
1076 #endif
1077 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1078 		if (!lockparent || !(flags & ISLASTCN)) {
1079 			VOP_UNLOCK(dvp, 0);
1080 			cnp->cn_flags |= PDIRUNLOCK;
1081 		}
1082 	}
1083 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1084 		cnp->cn_flags |= SAVENAME;
1085 	if ((cnp->cn_flags & MAKEENTRY) &&
1086 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1087 		nfs_cache_enter(dvp, newvp, cnp);
1088 	}
1089 	*vpp = newvp;
1090 	nfsm_reqdone;
1091 	if (error) {
1092 		/*
1093 		 * We get here only because of errors returned by
1094 		 * the RPC. Otherwise we'll have returned above
1095 		 * (the nfsm_* macros will jump to nfsm_reqdone
1096 		 * on error).
1097 		 */
1098 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1099 		    cnp->cn_nameiop != CREATE) {
1100 			nfs_cache_enter(dvp, NULL, cnp);
1101 		}
1102 		if (newvp != NULLVP) {
1103 			vrele(newvp);
1104 			if (newvp != dvp)
1105 				VOP_UNLOCK(newvp, 0);
1106 		}
1107 noentry:
1108 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1109 		    (flags & ISLASTCN) && error == ENOENT) {
1110 			if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1111 				error = EROFS;
1112 			} else {
1113 				error = EJUSTRETURN;
1114 				cnp->cn_flags |= SAVENAME;
1115 			}
1116 		}
1117 		*vpp = NULL;
1118 		return error;
1119 	}
1120 
1121 validate:
1122 	/*
1123 	 * make sure we have valid type and size.
1124 	 */
1125 
1126 	newvp = *vpp;
1127 	if (newvp->v_type == VNON) {
1128 		struct vattr vattr; /* dummy */
1129 
1130 		KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1131 		error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp);
1132 		if (error) {
1133 			vput(newvp);
1134 			*vpp = NULL;
1135 		}
1136 	}
1137 
1138 	return error;
1139 }
1140 
1141 /*
1142  * nfs read call.
1143  * Just call nfs_bioread() to do the work.
1144  */
1145 int
1146 nfs_read(v)
1147 	void *v;
1148 {
1149 	struct vop_read_args /* {
1150 		struct vnode *a_vp;
1151 		struct uio *a_uio;
1152 		int  a_ioflag;
1153 		kauth_cred_t a_cred;
1154 	} */ *ap = v;
1155 	struct vnode *vp = ap->a_vp;
1156 
1157 	if (vp->v_type != VREG)
1158 		return EISDIR;
1159 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1160 }
1161 
1162 /*
1163  * nfs readlink call
1164  */
1165 int
1166 nfs_readlink(v)
1167 	void *v;
1168 {
1169 	struct vop_readlink_args /* {
1170 		struct vnode *a_vp;
1171 		struct uio *a_uio;
1172 		kauth_cred_t a_cred;
1173 	} */ *ap = v;
1174 	struct vnode *vp = ap->a_vp;
1175 	struct nfsnode *np = VTONFS(vp);
1176 
1177 	if (vp->v_type != VLNK)
1178 		return (EPERM);
1179 
1180 	if (np->n_rcred != NULL) {
1181 		kauth_cred_free(np->n_rcred);
1182 	}
1183 	np->n_rcred = ap->a_cred;
1184 	kauth_cred_hold(np->n_rcred);
1185 
1186 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1187 }
1188 
1189 /*
1190  * Do a readlink rpc.
1191  * Called by nfs_doio() from below the buffer cache.
1192  */
1193 int
1194 nfs_readlinkrpc(vp, uiop, cred)
1195 	struct vnode *vp;
1196 	struct uio *uiop;
1197 	kauth_cred_t cred;
1198 {
1199 	u_int32_t *tl;
1200 	caddr_t cp;
1201 	int32_t t1, t2;
1202 	caddr_t bpos, dpos, cp2;
1203 	int error = 0;
1204 	uint32_t len;
1205 	struct mbuf *mreq, *mrep, *md, *mb;
1206 	const int v3 = NFS_ISV3(vp);
1207 	struct nfsnode *np = VTONFS(vp);
1208 #ifndef NFS_V2_ONLY
1209 	int attrflag;
1210 #endif
1211 
1212 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1213 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1214 	nfsm_fhtom(np, v3);
1215 	nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1216 #ifndef NFS_V2_ONLY
1217 	if (v3)
1218 		nfsm_postop_attr(vp, attrflag, 0);
1219 #endif
1220 	if (!error) {
1221 #ifndef NFS_V2_ONLY
1222 		if (v3) {
1223 			nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1224 			len = fxdr_unsigned(uint32_t, *tl);
1225 			if (len > MAXPATHLEN) {
1226 				/*
1227 				 * this pathname is too long for us.
1228 				 */
1229 				m_freem(mrep);
1230 				/* Solaris returns EINVAL. should we follow? */
1231 				error = ENAMETOOLONG;
1232 				goto nfsmout;
1233 			}
1234 		} else
1235 #endif
1236 		{
1237 			nfsm_strsiz(len, NFS_MAXPATHLEN);
1238 		}
1239 		nfsm_mtouio(uiop, len);
1240 	}
1241 	nfsm_reqdone;
1242 	return (error);
1243 }
1244 
1245 /*
1246  * nfs read rpc call
1247  * Ditto above
1248  */
1249 int
1250 nfs_readrpc(vp, uiop)
1251 	struct vnode *vp;
1252 	struct uio *uiop;
1253 {
1254 	u_int32_t *tl;
1255 	caddr_t cp;
1256 	int32_t t1, t2;
1257 	caddr_t bpos, dpos, cp2;
1258 	struct mbuf *mreq, *mrep, *md, *mb;
1259 	struct nfsmount *nmp;
1260 	int error = 0, len, retlen, tsiz, eof, byte_count;
1261 	const int v3 = NFS_ISV3(vp);
1262 	struct nfsnode *np = VTONFS(vp);
1263 #ifndef NFS_V2_ONLY
1264 	int attrflag;
1265 #endif
1266 
1267 #ifndef nolint
1268 	eof = 0;
1269 #endif
1270 	nmp = VFSTONFS(vp->v_mount);
1271 	tsiz = uiop->uio_resid;
1272 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1273 		return (EFBIG);
1274 	iostat_busy(nmp->nm_stats);
1275 	byte_count = 0; /* count bytes actually transferred */
1276 	while (tsiz > 0) {
1277 		nfsstats.rpccnt[NFSPROC_READ]++;
1278 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1279 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1280 		nfsm_fhtom(np, v3);
1281 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1282 #ifndef NFS_V2_ONLY
1283 		if (v3) {
1284 			txdr_hyper(uiop->uio_offset, tl);
1285 			*(tl + 2) = txdr_unsigned(len);
1286 		} else
1287 #endif
1288 		{
1289 			*tl++ = txdr_unsigned(uiop->uio_offset);
1290 			*tl++ = txdr_unsigned(len);
1291 			*tl = 0;
1292 		}
1293 		nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1294 #ifndef NFS_V2_ONLY
1295 		if (v3) {
1296 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1297 			if (error) {
1298 				m_freem(mrep);
1299 				goto nfsmout;
1300 			}
1301 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1302 			eof = fxdr_unsigned(int, *(tl + 1));
1303 		} else
1304 #endif
1305 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1306 		nfsm_strsiz(retlen, nmp->nm_rsize);
1307 		nfsm_mtouio(uiop, retlen);
1308 		m_freem(mrep);
1309 		tsiz -= retlen;
1310 		byte_count += retlen;
1311 #ifndef NFS_V2_ONLY
1312 		if (v3) {
1313 			if (eof || retlen == 0)
1314 				tsiz = 0;
1315 		} else
1316 #endif
1317 		if (retlen < len)
1318 			tsiz = 0;
1319 	}
1320 nfsmout:
1321 	iostat_unbusy(nmp->nm_stats, byte_count, 1);
1322 	return (error);
1323 }
1324 
1325 struct nfs_writerpc_context {
1326 	struct simplelock nwc_slock;
1327 	volatile int nwc_mbufcount;
1328 };
1329 
1330 /*
1331  * free mbuf used to refer protected pages while write rpc call.
1332  * called at splvm.
1333  */
1334 static void
1335 nfs_writerpc_extfree(struct mbuf *m, caddr_t tbuf __unused,
1336     size_t size __unused, void *arg)
1337 {
1338 	struct nfs_writerpc_context *ctx = arg;
1339 
1340 	KASSERT(m != NULL);
1341 	KASSERT(ctx != NULL);
1342 	pool_cache_put(&mbpool_cache, m);
1343 	simple_lock(&ctx->nwc_slock);
1344 	if (--ctx->nwc_mbufcount == 0) {
1345 		wakeup(ctx);
1346 	}
1347 	simple_unlock(&ctx->nwc_slock);
1348 }
1349 
1350 /*
1351  * nfs write call
1352  */
1353 int
1354 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1355 	struct vnode *vp;
1356 	struct uio *uiop;
1357 	int *iomode;
1358 	boolean_t pageprotected;
1359 	boolean_t *stalewriteverfp;
1360 {
1361 	u_int32_t *tl;
1362 	caddr_t cp;
1363 	int32_t t1, t2;
1364 	caddr_t bpos, dpos;
1365 	struct mbuf *mreq, *mrep, *md, *mb;
1366 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1367 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1368 	const int v3 = NFS_ISV3(vp);
1369 	int committed = NFSV3WRITE_FILESYNC;
1370 	struct nfsnode *np = VTONFS(vp);
1371 	struct nfs_writerpc_context ctx;
1372 	int s, byte_count;
1373 	struct lwp *l = NULL;
1374 	size_t origresid;
1375 #ifndef NFS_V2_ONLY
1376 	caddr_t cp2;
1377 	int rlen, commit;
1378 #endif
1379 
1380 	simple_lock_init(&ctx.nwc_slock);
1381 	ctx.nwc_mbufcount = 1;
1382 
1383 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1384 		panic("writerpc readonly vp %p", vp);
1385 	}
1386 
1387 #ifdef DIAGNOSTIC
1388 	if (uiop->uio_iovcnt != 1)
1389 		panic("nfs: writerpc iovcnt > 1");
1390 #endif
1391 	tsiz = uiop->uio_resid;
1392 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1393 		return (EFBIG);
1394 	if (pageprotected) {
1395 		l = curlwp;
1396 		PHOLD(l);
1397 	}
1398 retry:
1399 	origresid = uiop->uio_resid;
1400 	KASSERT(origresid == uiop->uio_iov->iov_len);
1401 	iostat_busy(nmp->nm_stats);
1402 	byte_count = 0; /* count of bytes actually written */
1403 	while (tsiz > 0) {
1404 		uint32_t datalen; /* data bytes need to be allocated in mbuf */
1405 		uint32_t backup;
1406 		boolean_t stalewriteverf = FALSE;
1407 
1408 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1409 		len = min(tsiz, nmp->nm_wsize);
1410 		datalen = pageprotected ? 0 : nfsm_rndup(len);
1411 		nfsm_reqhead(np, NFSPROC_WRITE,
1412 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1413 		nfsm_fhtom(np, v3);
1414 #ifndef NFS_V2_ONLY
1415 		if (v3) {
1416 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1417 			txdr_hyper(uiop->uio_offset, tl);
1418 			tl += 2;
1419 			*tl++ = txdr_unsigned(len);
1420 			*tl++ = txdr_unsigned(*iomode);
1421 			*tl = txdr_unsigned(len);
1422 		} else
1423 #endif
1424 		{
1425 			u_int32_t x;
1426 
1427 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1428 			/* Set both "begin" and "current" to non-garbage. */
1429 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1430 			*tl++ = x;      /* "begin offset" */
1431 			*tl++ = x;      /* "current offset" */
1432 			x = txdr_unsigned(len);
1433 			*tl++ = x;      /* total to this offset */
1434 			*tl = x;        /* size of this write */
1435 
1436 		}
1437 		if (pageprotected) {
1438 			/*
1439 			 * since we know pages can't be modified during i/o,
1440 			 * no need to copy them for us.
1441 			 */
1442 			struct mbuf *m;
1443 			struct iovec *iovp = uiop->uio_iov;
1444 
1445 			m = m_get(M_WAIT, MT_DATA);
1446 			MCLAIM(m, &nfs_mowner);
1447 			MEXTADD(m, iovp->iov_base, len, M_MBUF,
1448 			    nfs_writerpc_extfree, &ctx);
1449 			m->m_flags |= M_EXT_ROMAP;
1450 			m->m_len = len;
1451 			mb->m_next = m;
1452 			/*
1453 			 * no need to maintain mb and bpos here
1454 			 * because no one care them later.
1455 			 */
1456 #if 0
1457 			mb = m;
1458 			bpos = mtod(caddr_t, mb) + mb->m_len;
1459 #endif
1460 			UIO_ADVANCE(uiop, len);
1461 			uiop->uio_offset += len;
1462 			s = splvm();
1463 			simple_lock(&ctx.nwc_slock);
1464 			ctx.nwc_mbufcount++;
1465 			simple_unlock(&ctx.nwc_slock);
1466 			splx(s);
1467 			nfs_zeropad(mb, 0, nfsm_padlen(len));
1468 		} else {
1469 			nfsm_uiotom(uiop, len);
1470 		}
1471 		nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1472 #ifndef NFS_V2_ONLY
1473 		if (v3) {
1474 			wccflag = NFSV3_WCCCHK;
1475 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1476 			if (!error) {
1477 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1478 					+ NFSX_V3WRITEVERF);
1479 				rlen = fxdr_unsigned(int, *tl++);
1480 				if (rlen == 0) {
1481 					error = NFSERR_IO;
1482 					m_freem(mrep);
1483 					break;
1484 				} else if (rlen < len) {
1485 					backup = len - rlen;
1486 					UIO_ADVANCE(uiop, -backup);
1487 					uiop->uio_offset -= backup;
1488 					len = rlen;
1489 				}
1490 				commit = fxdr_unsigned(int, *tl++);
1491 
1492 				/*
1493 				 * Return the lowest committment level
1494 				 * obtained by any of the RPCs.
1495 				 */
1496 				if (committed == NFSV3WRITE_FILESYNC)
1497 					committed = commit;
1498 				else if (committed == NFSV3WRITE_DATASYNC &&
1499 					commit == NFSV3WRITE_UNSTABLE)
1500 					committed = commit;
1501 				simple_lock(&nmp->nm_slock);
1502 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1503 					memcpy(nmp->nm_writeverf, tl,
1504 					    NFSX_V3WRITEVERF);
1505 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1506 				} else if ((nmp->nm_iflag &
1507 				    NFSMNT_STALEWRITEVERF) ||
1508 				    memcmp(tl, nmp->nm_writeverf,
1509 				    NFSX_V3WRITEVERF)) {
1510 					memcpy(nmp->nm_writeverf, tl,
1511 					    NFSX_V3WRITEVERF);
1512 					/*
1513 					 * note NFSMNT_STALEWRITEVERF
1514 					 * if we're the first thread to
1515 					 * notice it.
1516 					 */
1517 					if ((nmp->nm_iflag &
1518 					    NFSMNT_STALEWRITEVERF) == 0) {
1519 						stalewriteverf = TRUE;
1520 						nmp->nm_iflag |=
1521 						    NFSMNT_STALEWRITEVERF;
1522 					}
1523 				}
1524 				simple_unlock(&nmp->nm_slock);
1525 			}
1526 		} else
1527 #endif
1528 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1529 		if (wccflag)
1530 			VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1531 		m_freem(mrep);
1532 		if (error)
1533 			break;
1534 		tsiz -= len;
1535 		byte_count += len;
1536 		if (stalewriteverf) {
1537 			*stalewriteverfp = TRUE;
1538 			stalewriteverf = FALSE;
1539 			if (committed == NFSV3WRITE_UNSTABLE &&
1540 			    len != origresid) {
1541 				/*
1542 				 * if our write requests weren't atomic but
1543 				 * unstable, datas in previous iterations
1544 				 * might have already been lost now.
1545 				 * then, we should resend them to nfsd.
1546 				 */
1547 				backup = origresid - tsiz;
1548 				UIO_ADVANCE(uiop, -backup);
1549 				uiop->uio_offset -= backup;
1550 				tsiz = origresid;
1551 				goto retry;
1552 			}
1553 		}
1554 	}
1555 nfsmout:
1556 	iostat_unbusy(nmp->nm_stats, byte_count, 0);
1557 	if (pageprotected) {
1558 		/*
1559 		 * wait until mbufs go away.
1560 		 * retransmitted mbufs can survive longer than rpc requests
1561 		 * themselves.
1562 		 */
1563 		s = splvm();
1564 		simple_lock(&ctx.nwc_slock);
1565 		ctx.nwc_mbufcount--;
1566 		while (ctx.nwc_mbufcount > 0) {
1567 			ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1568 		}
1569 		simple_unlock(&ctx.nwc_slock);
1570 		splx(s);
1571 		PRELE(l);
1572 	}
1573 	*iomode = committed;
1574 	if (error)
1575 		uiop->uio_resid = tsiz;
1576 	return (error);
1577 }
1578 
1579 /*
1580  * nfs mknod rpc
1581  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1582  * mode set to specify the file type and the size field for rdev.
1583  */
1584 int
1585 nfs_mknodrpc(dvp, vpp, cnp, vap)
1586 	struct vnode *dvp;
1587 	struct vnode **vpp;
1588 	struct componentname *cnp;
1589 	struct vattr *vap;
1590 {
1591 	struct nfsv2_sattr *sp;
1592 	u_int32_t *tl;
1593 	caddr_t cp;
1594 	int32_t t1, t2;
1595 	struct vnode *newvp = (struct vnode *)0;
1596 	struct nfsnode *dnp, *np;
1597 	char *cp2;
1598 	caddr_t bpos, dpos;
1599 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1600 	struct mbuf *mreq, *mrep, *md, *mb;
1601 	u_int32_t rdev;
1602 	const int v3 = NFS_ISV3(dvp);
1603 
1604 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1605 		rdev = txdr_unsigned(vap->va_rdev);
1606 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1607 		rdev = nfs_xdrneg1;
1608 	else {
1609 		VOP_ABORTOP(dvp, cnp);
1610 		vput(dvp);
1611 		return (EOPNOTSUPP);
1612 	}
1613 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1614 	dnp = VTONFS(dvp);
1615 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1616 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1617 	nfsm_fhtom(dnp, v3);
1618 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1619 #ifndef NFS_V2_ONLY
1620 	if (v3) {
1621 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1622 		*tl++ = vtonfsv3_type(vap->va_type);
1623 		nfsm_v3attrbuild(vap, FALSE);
1624 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1625 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1626 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1627 			*tl = txdr_unsigned(minor(vap->va_rdev));
1628 		}
1629 	} else
1630 #endif
1631 	{
1632 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1633 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1634 		sp->sa_uid = nfs_xdrneg1;
1635 		sp->sa_gid = nfs_xdrneg1;
1636 		sp->sa_size = rdev;
1637 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1638 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1639 	}
1640 	nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_lwp, cnp->cn_cred);
1641 	if (!error) {
1642 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1643 		if (!gotvp) {
1644 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1645 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np);
1646 			if (!error)
1647 				newvp = NFSTOV(np);
1648 		}
1649 	}
1650 #ifndef NFS_V2_ONLY
1651 	if (v3)
1652 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1653 #endif
1654 	nfsm_reqdone;
1655 	if (error) {
1656 		if (newvp)
1657 			vput(newvp);
1658 	} else {
1659 		if (cnp->cn_flags & MAKEENTRY)
1660 			nfs_cache_enter(dvp, newvp, cnp);
1661 		*vpp = newvp;
1662 	}
1663 	PNBUF_PUT(cnp->cn_pnbuf);
1664 	VTONFS(dvp)->n_flag |= NMODIFIED;
1665 	if (!wccflag)
1666 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1667 	vput(dvp);
1668 	return (error);
1669 }
1670 
1671 /*
1672  * nfs mknod vop
1673  * just call nfs_mknodrpc() to do the work.
1674  */
1675 /* ARGSUSED */
1676 int
1677 nfs_mknod(v)
1678 	void *v;
1679 {
1680 	struct vop_mknod_args /* {
1681 		struct vnode *a_dvp;
1682 		struct vnode **a_vpp;
1683 		struct componentname *a_cnp;
1684 		struct vattr *a_vap;
1685 	} */ *ap = v;
1686 	struct vnode *dvp = ap->a_dvp;
1687 	struct componentname *cnp = ap->a_cnp;
1688 	int error;
1689 
1690 	error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1691 	VN_KNOTE(dvp, NOTE_WRITE);
1692 	if (error == 0 || error == EEXIST)
1693 		cache_purge1(dvp, cnp, 0);
1694 	return (error);
1695 }
1696 
1697 #ifndef NFS_V2_ONLY
1698 static u_long create_verf;
1699 #endif
1700 /*
1701  * nfs file create call
1702  */
1703 int
1704 nfs_create(v)
1705 	void *v;
1706 {
1707 	struct vop_create_args /* {
1708 		struct vnode *a_dvp;
1709 		struct vnode **a_vpp;
1710 		struct componentname *a_cnp;
1711 		struct vattr *a_vap;
1712 	} */ *ap = v;
1713 	struct vnode *dvp = ap->a_dvp;
1714 	struct vattr *vap = ap->a_vap;
1715 	struct componentname *cnp = ap->a_cnp;
1716 	struct nfsv2_sattr *sp;
1717 	u_int32_t *tl;
1718 	caddr_t cp;
1719 	int32_t t1, t2;
1720 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1721 	struct vnode *newvp = (struct vnode *)0;
1722 	caddr_t bpos, dpos, cp2;
1723 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1724 	struct mbuf *mreq, *mrep, *md, *mb;
1725 	const int v3 = NFS_ISV3(dvp);
1726 
1727 	/*
1728 	 * Oops, not for me..
1729 	 */
1730 	if (vap->va_type == VSOCK)
1731 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1732 
1733 	KASSERT(vap->va_type == VREG);
1734 
1735 #ifdef VA_EXCLUSIVE
1736 	if (vap->va_vaflags & VA_EXCLUSIVE)
1737 		fmode |= O_EXCL;
1738 #endif
1739 again:
1740 	error = 0;
1741 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1742 	dnp = VTONFS(dvp);
1743 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1744 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1745 	nfsm_fhtom(dnp, v3);
1746 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1747 #ifndef NFS_V2_ONLY
1748 	if (v3) {
1749 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1750 		if (fmode & O_EXCL) {
1751 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1752 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1753 #ifdef INET
1754 			if (TAILQ_FIRST(&in_ifaddrhead))
1755 				*tl++ = TAILQ_FIRST(&in_ifaddrhead)->
1756 				    ia_addr.sin_addr.s_addr;
1757 			else
1758 				*tl++ = create_verf;
1759 #else
1760 			*tl++ = create_verf;
1761 #endif
1762 			*tl = ++create_verf;
1763 		} else {
1764 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1765 			nfsm_v3attrbuild(vap, FALSE);
1766 		}
1767 	} else
1768 #endif
1769 	{
1770 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1771 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1772 		sp->sa_uid = nfs_xdrneg1;
1773 		sp->sa_gid = nfs_xdrneg1;
1774 		sp->sa_size = 0;
1775 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1776 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1777 	}
1778 	nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_lwp, cnp->cn_cred);
1779 	if (!error) {
1780 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1781 		if (!gotvp) {
1782 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1783 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np);
1784 			if (!error)
1785 				newvp = NFSTOV(np);
1786 		}
1787 	}
1788 #ifndef NFS_V2_ONLY
1789 	if (v3)
1790 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1791 #endif
1792 	nfsm_reqdone;
1793 	if (error) {
1794 		/*
1795 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1796 		 */
1797 		if (v3 && (fmode & O_EXCL) && error == ENOTSUP) {
1798 			fmode &= ~O_EXCL;
1799 			goto again;
1800 		}
1801 	} else if (v3 && (fmode & O_EXCL)) {
1802 		struct timespec ts;
1803 
1804 		getnanotime(&ts);
1805 
1806 		/*
1807 		 * make sure that we'll update timestamps as
1808 		 * most server implementations use them to store
1809 		 * the create verifier.
1810 		 *
1811 		 * XXX it's better to use TOSERVER always.
1812 		 */
1813 
1814 		if (vap->va_atime.tv_sec == VNOVAL)
1815 			vap->va_atime = ts;
1816 		if (vap->va_mtime.tv_sec == VNOVAL)
1817 			vap->va_mtime = ts;
1818 
1819 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_lwp);
1820 	}
1821 	if (error == 0) {
1822 		if (cnp->cn_flags & MAKEENTRY)
1823 			nfs_cache_enter(dvp, newvp, cnp);
1824 		else
1825 			cache_purge1(dvp, cnp, 0);
1826 		*ap->a_vpp = newvp;
1827 	} else {
1828 		if (newvp)
1829 			vput(newvp);
1830 		if (error == EEXIST)
1831 			cache_purge1(dvp, cnp, 0);
1832 	}
1833 	PNBUF_PUT(cnp->cn_pnbuf);
1834 	VTONFS(dvp)->n_flag |= NMODIFIED;
1835 	if (!wccflag)
1836 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1837 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1838 	vput(dvp);
1839 	return (error);
1840 }
1841 
1842 /*
1843  * nfs file remove call
1844  * To try and make nfs semantics closer to ufs semantics, a file that has
1845  * other processes using the vnode is renamed instead of removed and then
1846  * removed later on the last close.
1847  * - If v_usecount > 1
1848  *	  If a rename is not already in the works
1849  *	     call nfs_sillyrename() to set it up
1850  *     else
1851  *	  do the remove rpc
1852  */
1853 int
1854 nfs_remove(v)
1855 	void *v;
1856 {
1857 	struct vop_remove_args /* {
1858 		struct vnodeop_desc *a_desc;
1859 		struct vnode * a_dvp;
1860 		struct vnode * a_vp;
1861 		struct componentname * a_cnp;
1862 	} */ *ap = v;
1863 	struct vnode *vp = ap->a_vp;
1864 	struct vnode *dvp = ap->a_dvp;
1865 	struct componentname *cnp = ap->a_cnp;
1866 	struct nfsnode *np = VTONFS(vp);
1867 	int error = 0;
1868 	struct vattr vattr;
1869 
1870 #ifndef DIAGNOSTIC
1871 	if ((cnp->cn_flags & HASBUF) == 0)
1872 		panic("nfs_remove: no name");
1873 	if (vp->v_usecount < 1)
1874 		panic("nfs_remove: bad v_usecount");
1875 #endif
1876 	if (vp->v_type == VDIR)
1877 		error = EPERM;
1878 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1879 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_lwp) == 0 &&
1880 	    vattr.va_nlink > 1)) {
1881 		/*
1882 		 * Purge the name cache so that the chance of a lookup for
1883 		 * the name succeeding while the remove is in progress is
1884 		 * minimized. Without node locking it can still happen, such
1885 		 * that an I/O op returns ESTALE, but since you get this if
1886 		 * another host removes the file..
1887 		 */
1888 		cache_purge(vp);
1889 		/*
1890 		 * throw away biocache buffers, mainly to avoid
1891 		 * unnecessary delayed writes later.
1892 		 */
1893 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_lwp, 1);
1894 		/* Do the rpc */
1895 		if (error != EINTR)
1896 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1897 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp);
1898 	} else if (!np->n_sillyrename)
1899 		error = nfs_sillyrename(dvp, vp, cnp, FALSE);
1900 	PNBUF_PUT(cnp->cn_pnbuf);
1901 	if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1902 	    vattr.va_nlink == 1) {
1903 		np->n_flag |= NREMOVED;
1904 	}
1905 	NFS_INVALIDATE_ATTRCACHE(np);
1906 	VN_KNOTE(vp, NOTE_DELETE);
1907 	VN_KNOTE(dvp, NOTE_WRITE);
1908 	if (dvp == vp)
1909 		vrele(vp);
1910 	else
1911 		vput(vp);
1912 	vput(dvp);
1913 	return (error);
1914 }
1915 
1916 /*
1917  * nfs file remove rpc called from nfs_inactive
1918  */
1919 int
1920 nfs_removeit(sp)
1921 	struct sillyrename *sp;
1922 {
1923 
1924 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1925 		(struct lwp *)0));
1926 }
1927 
1928 /*
1929  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1930  */
1931 int
1932 nfs_removerpc(dvp, name, namelen, cred, l)
1933 	struct vnode *dvp;
1934 	const char *name;
1935 	int namelen;
1936 	kauth_cred_t cred;
1937 	struct lwp *l;
1938 {
1939 	u_int32_t *tl;
1940 	caddr_t cp;
1941 #ifndef NFS_V2_ONLY
1942 	int32_t t1;
1943 	caddr_t cp2;
1944 #endif
1945 	int32_t t2;
1946 	caddr_t bpos, dpos;
1947 	int error = 0, wccflag = NFSV3_WCCRATTR;
1948 	struct mbuf *mreq, *mrep, *md, *mb;
1949 	const int v3 = NFS_ISV3(dvp);
1950 	int rexmit = 0;
1951 	struct nfsnode *dnp = VTONFS(dvp);
1952 
1953 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1954 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1955 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1956 	nfsm_fhtom(dnp, v3);
1957 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1958 	nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1959 #ifndef NFS_V2_ONLY
1960 	if (v3)
1961 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1962 #endif
1963 	nfsm_reqdone;
1964 	VTONFS(dvp)->n_flag |= NMODIFIED;
1965 	if (!wccflag)
1966 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1967 	/*
1968 	 * Kludge City: If the first reply to the remove rpc is lost..
1969 	 *   the reply to the retransmitted request will be ENOENT
1970 	 *   since the file was in fact removed
1971 	 *   Therefore, we cheat and return success.
1972 	 */
1973 	if (rexmit && error == ENOENT)
1974 		error = 0;
1975 	return (error);
1976 }
1977 
1978 /*
1979  * nfs file rename call
1980  */
1981 int
1982 nfs_rename(v)
1983 	void *v;
1984 {
1985 	struct vop_rename_args  /* {
1986 		struct vnode *a_fdvp;
1987 		struct vnode *a_fvp;
1988 		struct componentname *a_fcnp;
1989 		struct vnode *a_tdvp;
1990 		struct vnode *a_tvp;
1991 		struct componentname *a_tcnp;
1992 	} */ *ap = v;
1993 	struct vnode *fvp = ap->a_fvp;
1994 	struct vnode *tvp = ap->a_tvp;
1995 	struct vnode *fdvp = ap->a_fdvp;
1996 	struct vnode *tdvp = ap->a_tdvp;
1997 	struct componentname *tcnp = ap->a_tcnp;
1998 	struct componentname *fcnp = ap->a_fcnp;
1999 	int error;
2000 
2001 #ifndef DIAGNOSTIC
2002 	if ((tcnp->cn_flags & HASBUF) == 0 ||
2003 	    (fcnp->cn_flags & HASBUF) == 0)
2004 		panic("nfs_rename: no name");
2005 #endif
2006 	/* Check for cross-device rename */
2007 	if ((fvp->v_mount != tdvp->v_mount) ||
2008 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
2009 		error = EXDEV;
2010 		goto out;
2011 	}
2012 
2013 	/*
2014 	 * If the tvp exists and is in use, sillyrename it before doing the
2015 	 * rename of the new file over it.
2016 	 *
2017 	 * Have sillyrename use link instead of rename if possible,
2018 	 * so that we don't lose the file if the rename fails, and so
2019 	 * that there's no window when the "to" file doesn't exist.
2020 	 */
2021 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
2022 	    tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, TRUE)) {
2023 		VN_KNOTE(tvp, NOTE_DELETE);
2024 		vput(tvp);
2025 		tvp = NULL;
2026 	}
2027 
2028 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2029 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2030 		tcnp->cn_lwp);
2031 
2032 	VN_KNOTE(fdvp, NOTE_WRITE);
2033 	VN_KNOTE(tdvp, NOTE_WRITE);
2034 	if (error == 0 || error == EEXIST) {
2035 		if (fvp->v_type == VDIR)
2036 			cache_purge(fvp);
2037 		else
2038 			cache_purge1(fdvp, fcnp, 0);
2039 		if (tvp != NULL && tvp->v_type == VDIR)
2040 			cache_purge(tvp);
2041 		else
2042 			cache_purge1(tdvp, tcnp, 0);
2043 	}
2044 out:
2045 	if (tdvp == tvp)
2046 		vrele(tdvp);
2047 	else
2048 		vput(tdvp);
2049 	if (tvp)
2050 		vput(tvp);
2051 	vrele(fdvp);
2052 	vrele(fvp);
2053 	return (error);
2054 }
2055 
2056 /*
2057  * nfs file rename rpc called from nfs_remove() above
2058  */
2059 int
2060 nfs_renameit(sdvp, scnp, sp)
2061 	struct vnode *sdvp;
2062 	struct componentname *scnp;
2063 	struct sillyrename *sp;
2064 {
2065 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2066 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_lwp));
2067 }
2068 
2069 /*
2070  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2071  */
2072 int
2073 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l)
2074 	struct vnode *fdvp;
2075 	const char *fnameptr;
2076 	int fnamelen;
2077 	struct vnode *tdvp;
2078 	const char *tnameptr;
2079 	int tnamelen;
2080 	kauth_cred_t cred;
2081 	struct lwp *l;
2082 {
2083 	u_int32_t *tl;
2084 	caddr_t cp;
2085 #ifndef NFS_V2_ONLY
2086 	int32_t t1;
2087 	caddr_t cp2;
2088 #endif
2089 	int32_t t2;
2090 	caddr_t bpos, dpos;
2091 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2092 	struct mbuf *mreq, *mrep, *md, *mb;
2093 	const int v3 = NFS_ISV3(fdvp);
2094 	int rexmit = 0;
2095 	struct nfsnode *fdnp = VTONFS(fdvp);
2096 
2097 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2098 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
2099 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2100 		nfsm_rndup(tnamelen));
2101 	nfsm_fhtom(fdnp, v3);
2102 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2103 	nfsm_fhtom(VTONFS(tdvp), v3);
2104 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2105 	nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
2106 #ifndef NFS_V2_ONLY
2107 	if (v3) {
2108 		nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2109 		nfsm_wcc_data(tdvp, twccflag, 0, !error);
2110 	}
2111 #endif
2112 	nfsm_reqdone;
2113 	VTONFS(fdvp)->n_flag |= NMODIFIED;
2114 	VTONFS(tdvp)->n_flag |= NMODIFIED;
2115 	if (!fwccflag)
2116 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2117 	if (!twccflag)
2118 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2119 	/*
2120 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2121 	 */
2122 	if (rexmit && error == ENOENT)
2123 		error = 0;
2124 	return (error);
2125 }
2126 
2127 /*
2128  * NFS link RPC, called from nfs_link.
2129  * Assumes dvp and vp locked, and leaves them that way.
2130  */
2131 
2132 static int
2133 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2134     size_t namelen, kauth_cred_t cred, struct lwp *l)
2135 {
2136 	u_int32_t *tl;
2137 	caddr_t cp;
2138 #ifndef NFS_V2_ONLY
2139 	int32_t t1;
2140 	caddr_t cp2;
2141 #endif
2142 	int32_t t2;
2143 	caddr_t bpos, dpos;
2144 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2145 	struct mbuf *mreq, *mrep, *md, *mb;
2146 	const int v3 = NFS_ISV3(dvp);
2147 	int rexmit = 0;
2148 	struct nfsnode *np = VTONFS(vp);
2149 
2150 	nfsstats.rpccnt[NFSPROC_LINK]++;
2151 	nfsm_reqhead(np, NFSPROC_LINK,
2152 	    NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2153 	nfsm_fhtom(np, v3);
2154 	nfsm_fhtom(VTONFS(dvp), v3);
2155 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2156 	nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2157 #ifndef NFS_V2_ONLY
2158 	if (v3) {
2159 		nfsm_postop_attr(vp, attrflag, 0);
2160 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2161 	}
2162 #endif
2163 	nfsm_reqdone;
2164 
2165 	VTONFS(dvp)->n_flag |= NMODIFIED;
2166 	if (!attrflag)
2167 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2168 	if (!wccflag)
2169 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2170 
2171 	/*
2172 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2173 	 */
2174 	if (rexmit && error == EEXIST)
2175 		error = 0;
2176 
2177 	return error;
2178 }
2179 
2180 /*
2181  * nfs hard link create call
2182  */
2183 int
2184 nfs_link(v)
2185 	void *v;
2186 {
2187 	struct vop_link_args /* {
2188 		struct vnode *a_dvp;
2189 		struct vnode *a_vp;
2190 		struct componentname *a_cnp;
2191 	} */ *ap = v;
2192 	struct vnode *vp = ap->a_vp;
2193 	struct vnode *dvp = ap->a_dvp;
2194 	struct componentname *cnp = ap->a_cnp;
2195 	int error = 0;
2196 
2197 	if (dvp->v_mount != vp->v_mount) {
2198 		VOP_ABORTOP(dvp, cnp);
2199 		vput(dvp);
2200 		return (EXDEV);
2201 	}
2202 	if (dvp != vp) {
2203 		error = vn_lock(vp, LK_EXCLUSIVE);
2204 		if (error != 0) {
2205 			VOP_ABORTOP(dvp, cnp);
2206 			vput(dvp);
2207 			return error;
2208 		}
2209 	}
2210 
2211 	/*
2212 	 * Push all writes to the server, so that the attribute cache
2213 	 * doesn't get "out of sync" with the server.
2214 	 * XXX There should be a better way!
2215 	 */
2216 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_lwp);
2217 
2218 	error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2219 	    cnp->cn_cred, cnp->cn_lwp);
2220 
2221 	if (error == 0)
2222 		cache_purge1(dvp, cnp, 0);
2223 	PNBUF_PUT(cnp->cn_pnbuf);
2224 	if (dvp != vp)
2225 		VOP_UNLOCK(vp, 0);
2226 	VN_KNOTE(vp, NOTE_LINK);
2227 	VN_KNOTE(dvp, NOTE_WRITE);
2228 	vput(dvp);
2229 	return (error);
2230 }
2231 
2232 /*
2233  * nfs symbolic link create call
2234  */
2235 int
2236 nfs_symlink(v)
2237 	void *v;
2238 {
2239 	struct vop_symlink_args /* {
2240 		struct vnode *a_dvp;
2241 		struct vnode **a_vpp;
2242 		struct componentname *a_cnp;
2243 		struct vattr *a_vap;
2244 		char *a_target;
2245 	} */ *ap = v;
2246 	struct vnode *dvp = ap->a_dvp;
2247 	struct vattr *vap = ap->a_vap;
2248 	struct componentname *cnp = ap->a_cnp;
2249 	struct nfsv2_sattr *sp;
2250 	u_int32_t *tl;
2251 	caddr_t cp;
2252 	int32_t t1, t2;
2253 	caddr_t bpos, dpos, cp2;
2254 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2255 	struct mbuf *mreq, *mrep, *md, *mb;
2256 	struct vnode *newvp = (struct vnode *)0;
2257 	const int v3 = NFS_ISV3(dvp);
2258 	int rexmit = 0;
2259 	struct nfsnode *dnp = VTONFS(dvp);
2260 
2261 	*ap->a_vpp = NULL;
2262 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2263 	slen = strlen(ap->a_target);
2264 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2265 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2266 	nfsm_fhtom(dnp, v3);
2267 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2268 #ifndef NFS_V2_ONlY
2269 	if (v3)
2270 		nfsm_v3attrbuild(vap, FALSE);
2271 #endif
2272 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2273 #ifndef NFS_V2_ONlY
2274 	if (!v3) {
2275 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2276 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2277 		sp->sa_uid = nfs_xdrneg1;
2278 		sp->sa_gid = nfs_xdrneg1;
2279 		sp->sa_size = nfs_xdrneg1;
2280 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2281 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2282 	}
2283 #endif
2284 	nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_lwp, cnp->cn_cred,
2285 	    &rexmit);
2286 #ifndef NFS_V2_ONlY
2287 	if (v3) {
2288 		if (!error)
2289 			nfsm_mtofh(dvp, newvp, v3, gotvp);
2290 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2291 	}
2292 #endif
2293 	nfsm_reqdone;
2294 	/*
2295 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2296 	 */
2297 	if (rexmit && error == EEXIST)
2298 		error = 0;
2299 	if (error == 0 || error == EEXIST)
2300 		cache_purge1(dvp, cnp, 0);
2301 	if (error == 0 && newvp == NULL) {
2302 		struct nfsnode *np = NULL;
2303 
2304 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2305 		    cnp->cn_cred, cnp->cn_lwp, &np);
2306 		if (error == 0)
2307 			newvp = NFSTOV(np);
2308 	}
2309 	if (error) {
2310 		if (newvp != NULL)
2311 			vput(newvp);
2312 	} else {
2313 		*ap->a_vpp = newvp;
2314 	}
2315 	PNBUF_PUT(cnp->cn_pnbuf);
2316 	VTONFS(dvp)->n_flag |= NMODIFIED;
2317 	if (!wccflag)
2318 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2319 	VN_KNOTE(dvp, NOTE_WRITE);
2320 	vput(dvp);
2321 	return (error);
2322 }
2323 
2324 /*
2325  * nfs make dir call
2326  */
2327 int
2328 nfs_mkdir(v)
2329 	void *v;
2330 {
2331 	struct vop_mkdir_args /* {
2332 		struct vnode *a_dvp;
2333 		struct vnode **a_vpp;
2334 		struct componentname *a_cnp;
2335 		struct vattr *a_vap;
2336 	} */ *ap = v;
2337 	struct vnode *dvp = ap->a_dvp;
2338 	struct vattr *vap = ap->a_vap;
2339 	struct componentname *cnp = ap->a_cnp;
2340 	struct nfsv2_sattr *sp;
2341 	u_int32_t *tl;
2342 	caddr_t cp;
2343 	int32_t t1, t2;
2344 	int len;
2345 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2346 	struct vnode *newvp = (struct vnode *)0;
2347 	caddr_t bpos, dpos, cp2;
2348 	int error = 0, wccflag = NFSV3_WCCRATTR;
2349 	int gotvp = 0;
2350 	int rexmit = 0;
2351 	struct mbuf *mreq, *mrep, *md, *mb;
2352 	const int v3 = NFS_ISV3(dvp);
2353 
2354 	len = cnp->cn_namelen;
2355 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2356 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
2357 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2358 	nfsm_fhtom(dnp, v3);
2359 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2360 #ifndef NFS_V2_ONLY
2361 	if (v3) {
2362 		nfsm_v3attrbuild(vap, FALSE);
2363 	} else
2364 #endif
2365 	{
2366 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2367 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2368 		sp->sa_uid = nfs_xdrneg1;
2369 		sp->sa_gid = nfs_xdrneg1;
2370 		sp->sa_size = nfs_xdrneg1;
2371 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2372 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2373 	}
2374 	nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit);
2375 	if (!error)
2376 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2377 	if (v3)
2378 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2379 	nfsm_reqdone;
2380 	VTONFS(dvp)->n_flag |= NMODIFIED;
2381 	if (!wccflag)
2382 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2383 	/*
2384 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2385 	 * if we can succeed in looking up the directory.
2386 	 */
2387 	if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2388 		if (newvp) {
2389 			vput(newvp);
2390 			newvp = (struct vnode *)0;
2391 		}
2392 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2393 			cnp->cn_lwp, &np);
2394 		if (!error) {
2395 			newvp = NFSTOV(np);
2396 			if (newvp->v_type != VDIR || newvp == dvp)
2397 				error = EEXIST;
2398 		}
2399 	}
2400 	if (error) {
2401 		if (newvp) {
2402 			if (dvp != newvp)
2403 				vput(newvp);
2404 			else
2405 				vrele(newvp);
2406 		}
2407 	} else {
2408 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2409 		if (cnp->cn_flags & MAKEENTRY)
2410 			nfs_cache_enter(dvp, newvp, cnp);
2411 		*ap->a_vpp = newvp;
2412 	}
2413 	PNBUF_PUT(cnp->cn_pnbuf);
2414 	vput(dvp);
2415 	return (error);
2416 }
2417 
2418 /*
2419  * nfs remove directory call
2420  */
2421 int
2422 nfs_rmdir(v)
2423 	void *v;
2424 {
2425 	struct vop_rmdir_args /* {
2426 		struct vnode *a_dvp;
2427 		struct vnode *a_vp;
2428 		struct componentname *a_cnp;
2429 	} */ *ap = v;
2430 	struct vnode *vp = ap->a_vp;
2431 	struct vnode *dvp = ap->a_dvp;
2432 	struct componentname *cnp = ap->a_cnp;
2433 	u_int32_t *tl;
2434 	caddr_t cp;
2435 #ifndef NFS_V2_ONLY
2436 	int32_t t1;
2437 	caddr_t cp2;
2438 #endif
2439 	int32_t t2;
2440 	caddr_t bpos, dpos;
2441 	int error = 0, wccflag = NFSV3_WCCRATTR;
2442 	int rexmit = 0;
2443 	struct mbuf *mreq, *mrep, *md, *mb;
2444 	const int v3 = NFS_ISV3(dvp);
2445 	struct nfsnode *dnp;
2446 
2447 	if (dvp == vp) {
2448 		vrele(dvp);
2449 		vput(dvp);
2450 		PNBUF_PUT(cnp->cn_pnbuf);
2451 		return (EINVAL);
2452 	}
2453 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2454 	dnp = VTONFS(dvp);
2455 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2456 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2457 	nfsm_fhtom(dnp, v3);
2458 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2459 	nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit);
2460 #ifndef NFS_V2_ONLY
2461 	if (v3)
2462 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2463 #endif
2464 	nfsm_reqdone;
2465 	PNBUF_PUT(cnp->cn_pnbuf);
2466 	VTONFS(dvp)->n_flag |= NMODIFIED;
2467 	if (!wccflag)
2468 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2469 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2470 	VN_KNOTE(vp, NOTE_DELETE);
2471 	cache_purge(vp);
2472 	vput(vp);
2473 	vput(dvp);
2474 	/*
2475 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2476 	 */
2477 	if (rexmit && error == ENOENT)
2478 		error = 0;
2479 	return (error);
2480 }
2481 
2482 /*
2483  * nfs readdir call
2484  */
2485 int
2486 nfs_readdir(v)
2487 	void *v;
2488 {
2489 	struct vop_readdir_args /* {
2490 		struct vnode *a_vp;
2491 		struct uio *a_uio;
2492 		kauth_cred_t a_cred;
2493 		int *a_eofflag;
2494 		off_t **a_cookies;
2495 		int *a_ncookies;
2496 	} */ *ap = v;
2497 	struct vnode *vp = ap->a_vp;
2498 	struct uio *uio = ap->a_uio;
2499 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2500 	char *base = uio->uio_iov->iov_base;
2501 	int tresid, error;
2502 	size_t count, lost;
2503 	struct dirent *dp;
2504 	off_t *cookies = NULL;
2505 	int ncookies = 0, nc;
2506 
2507 	if (vp->v_type != VDIR)
2508 		return (EPERM);
2509 
2510 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2511 	count = uio->uio_resid - lost;
2512 	if (count <= 0)
2513 		return (EINVAL);
2514 
2515 	/*
2516 	 * Call nfs_bioread() to do the real work.
2517 	 */
2518 	tresid = uio->uio_resid = count;
2519 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2520 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2521 
2522 	if (!error && ap->a_cookies) {
2523 		ncookies = count / 16;
2524 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2525 		*ap->a_cookies = cookies;
2526 	}
2527 
2528 	if (!error && uio->uio_resid == tresid) {
2529 		uio->uio_resid += lost;
2530 		nfsstats.direofcache_misses++;
2531 		if (ap->a_cookies)
2532 			*ap->a_ncookies = 0;
2533 		*ap->a_eofflag = 1;
2534 		return (0);
2535 	}
2536 
2537 	if (!error && ap->a_cookies) {
2538 		/*
2539 		 * Only the NFS server and emulations use cookies, and they
2540 		 * load the directory block into system space, so we can
2541 		 * just look at it directly.
2542 		 */
2543 		if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2544 		    uio->uio_iovcnt != 1)
2545 			panic("nfs_readdir: lost in space");
2546 		for (nc = 0; ncookies-- &&
2547 		     base < (char *)uio->uio_iov->iov_base; nc++){
2548 			dp = (struct dirent *) base;
2549 			if (dp->d_reclen == 0)
2550 				break;
2551 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2552 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2553 			else
2554 				*(cookies++) = NFS_GETCOOKIE(dp);
2555 			base += dp->d_reclen;
2556 		}
2557 		uio->uio_resid +=
2558 		    ((caddr_t)uio->uio_iov->iov_base - base);
2559 		uio->uio_iov->iov_len +=
2560 		    ((caddr_t)uio->uio_iov->iov_base - base);
2561 		uio->uio_iov->iov_base = base;
2562 		*ap->a_ncookies = nc;
2563 	}
2564 
2565 	uio->uio_resid += lost;
2566 	*ap->a_eofflag = 0;
2567 	return (error);
2568 }
2569 
2570 /*
2571  * Readdir rpc call.
2572  * Called from below the buffer cache by nfs_doio().
2573  */
2574 int
2575 nfs_readdirrpc(vp, uiop, cred)
2576 	struct vnode *vp;
2577 	struct uio *uiop;
2578 	kauth_cred_t cred;
2579 {
2580 	int len, left;
2581 	struct dirent *dp = NULL;
2582 	u_int32_t *tl;
2583 	caddr_t cp;
2584 	int32_t t1, t2;
2585 	caddr_t bpos, dpos, cp2;
2586 	struct mbuf *mreq, *mrep, *md, *mb;
2587 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2588 	struct nfsnode *dnp = VTONFS(vp);
2589 	u_quad_t fileno;
2590 	int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2591 #ifndef NFS_V2_ONLY
2592 	int attrflag;
2593 #endif
2594 	int nrpcs = 0, reclen;
2595 	const int v3 = NFS_ISV3(vp);
2596 
2597 #ifdef DIAGNOSTIC
2598 	/*
2599 	 * Should be called from buffer cache, so only amount of
2600 	 * NFS_DIRBLKSIZ will be requested.
2601 	 */
2602 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2603 		panic("nfs readdirrpc bad uio");
2604 #endif
2605 
2606 	/*
2607 	 * Loop around doing readdir rpc's of size nm_readdirsize
2608 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2609 	 * The stopping criteria is EOF or buffer full.
2610 	 */
2611 	while (more_dirs && bigenough) {
2612 		/*
2613 		 * Heuristic: don't bother to do another RPC to further
2614 		 * fill up this block if there is not much room left. (< 50%
2615 		 * of the readdir RPC size). This wastes some buffer space
2616 		 * but can save up to 50% in RPC calls.
2617 		 */
2618 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2619 			bigenough = 0;
2620 			break;
2621 		}
2622 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2623 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2624 			NFSX_READDIR(v3));
2625 		nfsm_fhtom(dnp, v3);
2626 #ifndef NFS_V2_ONLY
2627 		if (v3) {
2628 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2629 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2630 				txdr_swapcookie3(uiop->uio_offset, tl);
2631 			} else {
2632 				txdr_cookie3(uiop->uio_offset, tl);
2633 			}
2634 			tl += 2;
2635 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2636 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2637 		} else
2638 #endif
2639 		{
2640 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2641 			*tl++ = txdr_unsigned(uiop->uio_offset);
2642 		}
2643 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2644 		nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2645 		nrpcs++;
2646 #ifndef NFS_V2_ONLY
2647 		if (v3) {
2648 			nfsm_postop_attr(vp, attrflag, 0);
2649 			if (!error) {
2650 				nfsm_dissect(tl, u_int32_t *,
2651 				    2 * NFSX_UNSIGNED);
2652 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2653 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2654 			} else {
2655 				m_freem(mrep);
2656 				goto nfsmout;
2657 			}
2658 		}
2659 #endif
2660 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2661 		more_dirs = fxdr_unsigned(int, *tl);
2662 
2663 		/* loop thru the dir entries, doctoring them to 4bsd form */
2664 		while (more_dirs && bigenough) {
2665 #ifndef NFS_V2_ONLY
2666 			if (v3) {
2667 				nfsm_dissect(tl, u_int32_t *,
2668 				    3 * NFSX_UNSIGNED);
2669 				fileno = fxdr_hyper(tl);
2670 				len = fxdr_unsigned(int, *(tl + 2));
2671 			} else
2672 #endif
2673 			{
2674 				nfsm_dissect(tl, u_int32_t *,
2675 				    2 * NFSX_UNSIGNED);
2676 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2677 				len = fxdr_unsigned(int, *tl);
2678 			}
2679 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2680 				error = EBADRPC;
2681 				m_freem(mrep);
2682 				goto nfsmout;
2683 			}
2684 			/* for cookie stashing */
2685 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2686 			left = NFS_DIRFRAGSIZ - blksiz;
2687 			if (reclen > left) {
2688 				memset(uiop->uio_iov->iov_base, 0, left);
2689 				dp->d_reclen += left;
2690 				UIO_ADVANCE(uiop, left);
2691 				blksiz = 0;
2692 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2693 			}
2694 			if (reclen > uiop->uio_resid)
2695 				bigenough = 0;
2696 			if (bigenough) {
2697 				int tlen;
2698 
2699 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2700 				dp->d_fileno = fileno;
2701 				dp->d_namlen = len;
2702 				dp->d_reclen = reclen;
2703 				dp->d_type = DT_UNKNOWN;
2704 				blksiz += reclen;
2705 				if (blksiz == NFS_DIRFRAGSIZ)
2706 					blksiz = 0;
2707 				UIO_ADVANCE(uiop, DIRHDSIZ);
2708 				nfsm_mtouio(uiop, len);
2709 				tlen = reclen - (DIRHDSIZ + len);
2710 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2711 				UIO_ADVANCE(uiop, tlen);
2712 			} else
2713 				nfsm_adv(nfsm_rndup(len));
2714 #ifndef NFS_V2_ONLY
2715 			if (v3) {
2716 				nfsm_dissect(tl, u_int32_t *,
2717 				    3 * NFSX_UNSIGNED);
2718 			} else
2719 #endif
2720 			{
2721 				nfsm_dissect(tl, u_int32_t *,
2722 				    2 * NFSX_UNSIGNED);
2723 			}
2724 			if (bigenough) {
2725 #ifndef NFS_V2_ONLY
2726 				if (v3) {
2727 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2728 						uiop->uio_offset =
2729 						    fxdr_swapcookie3(tl);
2730 					else
2731 						uiop->uio_offset =
2732 						    fxdr_cookie3(tl);
2733 				}
2734 				else
2735 #endif
2736 				{
2737 					uiop->uio_offset =
2738 					    fxdr_unsigned(off_t, *tl);
2739 				}
2740 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2741 			}
2742 			if (v3)
2743 				tl += 2;
2744 			else
2745 				tl++;
2746 			more_dirs = fxdr_unsigned(int, *tl);
2747 		}
2748 		/*
2749 		 * If at end of rpc data, get the eof boolean
2750 		 */
2751 		if (!more_dirs) {
2752 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2753 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2754 
2755 			/*
2756 			 * kludge: if we got no entries, treat it as EOF.
2757 			 * some server sometimes send a reply without any
2758 			 * entries or EOF.
2759 			 * although it might mean the server has very long name,
2760 			 * we can't handle such entries anyway.
2761 			 */
2762 
2763 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2764 				more_dirs = 0;
2765 		}
2766 		m_freem(mrep);
2767 	}
2768 	/*
2769 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2770 	 * by increasing d_reclen for the last record.
2771 	 */
2772 	if (blksiz > 0) {
2773 		left = NFS_DIRFRAGSIZ - blksiz;
2774 		memset(uiop->uio_iov->iov_base, 0, left);
2775 		dp->d_reclen += left;
2776 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2777 		UIO_ADVANCE(uiop, left);
2778 	}
2779 
2780 	/*
2781 	 * We are now either at the end of the directory or have filled the
2782 	 * block.
2783 	 */
2784 	if (bigenough) {
2785 		dnp->n_direofoffset = uiop->uio_offset;
2786 		dnp->n_flag |= NEOFVALID;
2787 	}
2788 nfsmout:
2789 	return (error);
2790 }
2791 
2792 #ifndef NFS_V2_ONLY
2793 /*
2794  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2795  */
2796 int
2797 nfs_readdirplusrpc(vp, uiop, cred)
2798 	struct vnode *vp;
2799 	struct uio *uiop;
2800 	kauth_cred_t cred;
2801 {
2802 	int len, left;
2803 	struct dirent *dp = NULL;
2804 	u_int32_t *tl;
2805 	caddr_t cp;
2806 	int32_t t1, t2;
2807 	struct vnode *newvp;
2808 	caddr_t bpos, dpos, cp2;
2809 	struct mbuf *mreq, *mrep, *md, *mb;
2810 	struct nameidata nami, *ndp = &nami;
2811 	struct componentname *cnp = &ndp->ni_cnd;
2812 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2813 	struct nfsnode *dnp = VTONFS(vp), *np;
2814 	nfsfh_t *fhp;
2815 	u_quad_t fileno;
2816 	int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2817 	int attrflag, fhsize, nrpcs = 0, reclen;
2818 	struct nfs_fattr fattr, *fp;
2819 
2820 #ifdef DIAGNOSTIC
2821 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2822 		panic("nfs readdirplusrpc bad uio");
2823 #endif
2824 	ndp->ni_dvp = vp;
2825 	newvp = NULLVP;
2826 
2827 	/*
2828 	 * Loop around doing readdir rpc's of size nm_readdirsize
2829 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2830 	 * The stopping criteria is EOF or buffer full.
2831 	 */
2832 	while (more_dirs && bigenough) {
2833 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2834 			bigenough = 0;
2835 			break;
2836 		}
2837 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2838 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2839 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2840 		nfsm_fhtom(dnp, 1);
2841  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2842 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2843 			txdr_swapcookie3(uiop->uio_offset, tl);
2844 		} else {
2845 			txdr_cookie3(uiop->uio_offset, tl);
2846 		}
2847 		tl += 2;
2848 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2849 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2850 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2851 		*tl = txdr_unsigned(nmp->nm_rsize);
2852 		nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2853 		nfsm_postop_attr(vp, attrflag, 0);
2854 		if (error) {
2855 			m_freem(mrep);
2856 			goto nfsmout;
2857 		}
2858 		nrpcs++;
2859 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2860 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2861 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2862 		more_dirs = fxdr_unsigned(int, *tl);
2863 
2864 		/* loop thru the dir entries, doctoring them to 4bsd form */
2865 		while (more_dirs && bigenough) {
2866 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2867 			fileno = fxdr_hyper(tl);
2868 			len = fxdr_unsigned(int, *(tl + 2));
2869 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2870 				error = EBADRPC;
2871 				m_freem(mrep);
2872 				goto nfsmout;
2873 			}
2874 			/* for cookie stashing */
2875 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2876 			left = NFS_DIRFRAGSIZ - blksiz;
2877 			if (reclen > left) {
2878 				/*
2879 				 * DIRFRAGSIZ is aligned, no need to align
2880 				 * again here.
2881 				 */
2882 				memset(uiop->uio_iov->iov_base, 0, left);
2883 				dp->d_reclen += left;
2884 				UIO_ADVANCE(uiop, left);
2885 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2886 				blksiz = 0;
2887 			}
2888 			if (reclen > uiop->uio_resid)
2889 				bigenough = 0;
2890 			if (bigenough) {
2891 				int tlen;
2892 
2893 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2894 				dp->d_fileno = fileno;
2895 				dp->d_namlen = len;
2896 				dp->d_reclen = reclen;
2897 				dp->d_type = DT_UNKNOWN;
2898 				blksiz += reclen;
2899 				if (blksiz == NFS_DIRFRAGSIZ)
2900 					blksiz = 0;
2901 				UIO_ADVANCE(uiop, DIRHDSIZ);
2902 				nfsm_mtouio(uiop, len);
2903 				tlen = reclen - (DIRHDSIZ + len);
2904 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2905 				UIO_ADVANCE(uiop, tlen);
2906 				cnp->cn_nameptr = dp->d_name;
2907 				cnp->cn_namelen = dp->d_namlen;
2908 			} else
2909 				nfsm_adv(nfsm_rndup(len));
2910 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2911 			if (bigenough) {
2912 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2913 					uiop->uio_offset =
2914 						fxdr_swapcookie3(tl);
2915 				else
2916 					uiop->uio_offset =
2917 						fxdr_cookie3(tl);
2918 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2919 			}
2920 			tl += 2;
2921 
2922 			/*
2923 			 * Since the attributes are before the file handle
2924 			 * (sigh), we must skip over the attributes and then
2925 			 * come back and get them.
2926 			 */
2927 			attrflag = fxdr_unsigned(int, *tl);
2928 			if (attrflag) {
2929 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2930 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2931 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2932 			    doit = fxdr_unsigned(int, *tl);
2933 			    if (doit) {
2934 				nfsm_getfh(fhp, fhsize, 1);
2935 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2936 				    VREF(vp);
2937 				    newvp = vp;
2938 				    np = dnp;
2939 				} else {
2940 				    error = nfs_nget1(vp->v_mount, fhp,
2941 					fhsize, &np, LK_NOWAIT);
2942 				    if (!error)
2943 					newvp = NFSTOV(np);
2944 				}
2945 				if (!error) {
2946 				    const char *xcp;
2947 
2948 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2949 				    if (bigenough) {
2950 					dp->d_type =
2951 					   IFTODT(VTTOIF(np->n_vattr->va_type));
2952 					if (cnp->cn_namelen <= NCHNAMLEN) {
2953 					    ndp->ni_vp = newvp;
2954 					    xcp = cnp->cn_nameptr +
2955 						cnp->cn_namelen;
2956 					    cnp->cn_hash =
2957 					       namei_hash(cnp->cn_nameptr, &xcp);
2958 					    nfs_cache_enter(ndp->ni_dvp,
2959 						ndp->ni_vp, cnp);
2960 					}
2961 				    }
2962 				}
2963 				error = 0;
2964 			   }
2965 			} else {
2966 			    /* Just skip over the file handle */
2967 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2968 			    i = fxdr_unsigned(int, *tl);
2969 			    nfsm_adv(nfsm_rndup(i));
2970 			}
2971 			if (newvp != NULLVP) {
2972 			    if (newvp == vp)
2973 				vrele(newvp);
2974 			    else
2975 				vput(newvp);
2976 			    newvp = NULLVP;
2977 			}
2978 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2979 			more_dirs = fxdr_unsigned(int, *tl);
2980 		}
2981 		/*
2982 		 * If at end of rpc data, get the eof boolean
2983 		 */
2984 		if (!more_dirs) {
2985 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2986 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2987 
2988 			/*
2989 			 * kludge: see a comment in nfs_readdirrpc.
2990 			 */
2991 
2992 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2993 				more_dirs = 0;
2994 		}
2995 		m_freem(mrep);
2996 	}
2997 	/*
2998 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2999 	 * by increasing d_reclen for the last record.
3000 	 */
3001 	if (blksiz > 0) {
3002 		left = NFS_DIRFRAGSIZ - blksiz;
3003 		memset(uiop->uio_iov->iov_base, 0, left);
3004 		dp->d_reclen += left;
3005 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
3006 		UIO_ADVANCE(uiop, left);
3007 	}
3008 
3009 	/*
3010 	 * We are now either at the end of the directory or have filled the
3011 	 * block.
3012 	 */
3013 	if (bigenough) {
3014 		dnp->n_direofoffset = uiop->uio_offset;
3015 		dnp->n_flag |= NEOFVALID;
3016 	}
3017 nfsmout:
3018 	if (newvp != NULLVP) {
3019 		if(newvp == vp)
3020 		    vrele(newvp);
3021 		else
3022 		    vput(newvp);
3023 	}
3024 	return (error);
3025 }
3026 #endif
3027 
3028 /*
3029  * Silly rename. To make the NFS filesystem that is stateless look a little
3030  * more like the "ufs" a remove of an active vnode is translated to a rename
3031  * to a funny looking filename that is removed by nfs_inactive on the
3032  * nfsnode. There is the potential for another process on a different client
3033  * to create the same funny name between the nfs_lookitup() fails and the
3034  * nfs_rename() completes, but...
3035  */
3036 int
3037 nfs_sillyrename(dvp, vp, cnp, dolink)
3038 	struct vnode *dvp, *vp;
3039 	struct componentname *cnp;
3040 	boolean_t dolink;
3041 {
3042 	struct sillyrename *sp;
3043 	struct nfsnode *np;
3044 	int error;
3045 	short pid;
3046 
3047 	cache_purge(dvp);
3048 	np = VTONFS(vp);
3049 #ifndef DIAGNOSTIC
3050 	if (vp->v_type == VDIR)
3051 		panic("nfs: sillyrename dir");
3052 #endif
3053 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
3054 		M_NFSREQ, M_WAITOK);
3055 	sp->s_cred = kauth_cred_dup(cnp->cn_cred);
3056 	sp->s_dvp = dvp;
3057 	VREF(dvp);
3058 
3059 	/* Fudge together a funny name */
3060 	pid = cnp->cn_lwp->l_proc->p_pid;
3061 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
3062 	sp->s_namlen = 12;
3063 	sp->s_name[8] = hexdigits[pid & 0xf];
3064 	sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
3065 	sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
3066 	sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
3067 
3068 	/* Try lookitups until we get one that isn't there */
3069 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3070 		cnp->cn_lwp, (struct nfsnode **)0) == 0) {
3071 		sp->s_name[4]++;
3072 		if (sp->s_name[4] > 'z') {
3073 			error = EINVAL;
3074 			goto bad;
3075 		}
3076 	}
3077 	if (dolink) {
3078 		error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
3079 		    sp->s_cred, cnp->cn_lwp);
3080 		/*
3081 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
3082 		 */
3083 		if (error == ENOTSUP) {
3084 			error = nfs_renameit(dvp, cnp, sp);
3085 		}
3086 	} else {
3087 		error = nfs_renameit(dvp, cnp, sp);
3088 	}
3089 	if (error)
3090 		goto bad;
3091 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3092 		cnp->cn_lwp, &np);
3093 	np->n_sillyrename = sp;
3094 	return (0);
3095 bad:
3096 	vrele(sp->s_dvp);
3097 	kauth_cred_free(sp->s_cred);
3098 	free((caddr_t)sp, M_NFSREQ);
3099 	return (error);
3100 }
3101 
3102 /*
3103  * Look up a file name and optionally either update the file handle or
3104  * allocate an nfsnode, depending on the value of npp.
3105  * npp == NULL	--> just do the lookup
3106  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3107  *			handled too
3108  * *npp != NULL --> update the file handle in the vnode
3109  */
3110 int
3111 nfs_lookitup(dvp, name, len, cred, l, npp)
3112 	struct vnode *dvp;
3113 	const char *name;
3114 	int len;
3115 	kauth_cred_t cred;
3116 	struct lwp *l;
3117 	struct nfsnode **npp;
3118 {
3119 	u_int32_t *tl;
3120 	caddr_t cp;
3121 	int32_t t1, t2;
3122 	struct vnode *newvp = (struct vnode *)0;
3123 	struct nfsnode *np, *dnp = VTONFS(dvp);
3124 	caddr_t bpos, dpos, cp2;
3125 	int error = 0, fhlen;
3126 #ifndef NFS_V2_ONLY
3127 	int attrflag;
3128 #endif
3129 	struct mbuf *mreq, *mrep, *md, *mb;
3130 	nfsfh_t *nfhp;
3131 	const int v3 = NFS_ISV3(dvp);
3132 
3133 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3134 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3135 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3136 	nfsm_fhtom(dnp, v3);
3137 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
3138 	nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
3139 	if (npp && !error) {
3140 		nfsm_getfh(nfhp, fhlen, v3);
3141 		if (*npp) {
3142 		    np = *npp;
3143 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3144 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
3145 			np->n_fhp = &np->n_fh;
3146 		    }
3147 #if NFS_SMALLFH < NFSX_V3FHMAX
3148 		    else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
3149 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
3150 #endif
3151 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
3152 		    np->n_fhsize = fhlen;
3153 		    newvp = NFSTOV(np);
3154 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3155 		    VREF(dvp);
3156 		    newvp = dvp;
3157 		    np = dnp;
3158 		} else {
3159 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3160 		    if (error) {
3161 			m_freem(mrep);
3162 			return (error);
3163 		    }
3164 		    newvp = NFSTOV(np);
3165 		}
3166 #ifndef NFS_V2_ONLY
3167 		if (v3) {
3168 			nfsm_postop_attr(newvp, attrflag, 0);
3169 			if (!attrflag && *npp == NULL) {
3170 				m_freem(mrep);
3171 				vput(newvp);
3172 				return (ENOENT);
3173 			}
3174 		} else
3175 #endif
3176 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
3177 	}
3178 	nfsm_reqdone;
3179 	if (npp && *npp == NULL) {
3180 		if (error) {
3181 			if (newvp)
3182 				vput(newvp);
3183 		} else
3184 			*npp = np;
3185 	}
3186 	return (error);
3187 }
3188 
3189 #ifndef NFS_V2_ONLY
3190 /*
3191  * Nfs Version 3 commit rpc
3192  */
3193 int
3194 nfs_commit(vp, offset, cnt, l)
3195 	struct vnode *vp;
3196 	off_t offset;
3197 	uint32_t cnt;
3198 	struct lwp *l;
3199 {
3200 	caddr_t cp;
3201 	u_int32_t *tl;
3202 	int32_t t1, t2;
3203 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3204 	caddr_t bpos, dpos, cp2;
3205 	int error = 0, wccflag = NFSV3_WCCRATTR;
3206 	struct mbuf *mreq, *mrep, *md, *mb;
3207 	struct nfsnode *np;
3208 
3209 	KASSERT(NFS_ISV3(vp));
3210 
3211 #ifdef NFS_DEBUG_COMMIT
3212 	printf("commit %lu - %lu\n", (unsigned long)offset,
3213 	    (unsigned long)(offset + cnt));
3214 #endif
3215 
3216 	simple_lock(&nmp->nm_slock);
3217 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3218 		simple_unlock(&nmp->nm_slock);
3219 		return (0);
3220 	}
3221 	simple_unlock(&nmp->nm_slock);
3222 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3223 	np = VTONFS(vp);
3224 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3225 	nfsm_fhtom(np, 1);
3226 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3227 	txdr_hyper(offset, tl);
3228 	tl += 2;
3229 	*tl = txdr_unsigned(cnt);
3230 	nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3231 	nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, FALSE);
3232 	if (!error) {
3233 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3234 		simple_lock(&nmp->nm_slock);
3235 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3236 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3237 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3238 			error = NFSERR_STALEWRITEVERF;
3239 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3240 		}
3241 		simple_unlock(&nmp->nm_slock);
3242 	}
3243 	nfsm_reqdone;
3244 	return (error);
3245 }
3246 #endif
3247 
3248 /*
3249  * Kludge City..
3250  * - make nfs_bmap() essentially a no-op that does no translation
3251  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3252  *   (Maybe I could use the process's page mapping, but I was concerned that
3253  *    Kernel Write might not be enabled and also figured copyout() would do
3254  *    a lot more work than memcpy() and also it currently happens in the
3255  *    context of the swapper process (2).
3256  */
3257 int
3258 nfs_bmap(v)
3259 	void *v;
3260 {
3261 	struct vop_bmap_args /* {
3262 		struct vnode *a_vp;
3263 		daddr_t  a_bn;
3264 		struct vnode **a_vpp;
3265 		daddr_t *a_bnp;
3266 		int *a_runp;
3267 	} */ *ap = v;
3268 	struct vnode *vp = ap->a_vp;
3269 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3270 
3271 	if (ap->a_vpp != NULL)
3272 		*ap->a_vpp = vp;
3273 	if (ap->a_bnp != NULL)
3274 		*ap->a_bnp = ap->a_bn << bshift;
3275 	if (ap->a_runp != NULL)
3276 		*ap->a_runp = 1024 * 1024; /* XXX */
3277 	return (0);
3278 }
3279 
3280 /*
3281  * Strategy routine.
3282  * For async requests when nfsiod(s) are running, queue the request by
3283  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3284  * request.
3285  */
3286 int
3287 nfs_strategy(v)
3288 	void *v;
3289 {
3290 	struct vop_strategy_args *ap = v;
3291 	struct buf *bp = ap->a_bp;
3292 	int error = 0;
3293 
3294 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3295 		panic("nfs physio/async");
3296 
3297 	/*
3298 	 * If the op is asynchronous and an i/o daemon is waiting
3299 	 * queue the request, wake it up and wait for completion
3300 	 * otherwise just do it ourselves.
3301 	 */
3302 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3303 		error = nfs_doio(bp);
3304 	return (error);
3305 }
3306 
3307 /*
3308  * fsync vnode op. Just call nfs_flush() with commit == 1.
3309  */
3310 /* ARGSUSED */
3311 int
3312 nfs_fsync(v)
3313 	void *v;
3314 {
3315 	struct vop_fsync_args /* {
3316 		struct vnodeop_desc *a_desc;
3317 		struct vnode * a_vp;
3318 		kauth_cred_t  a_cred;
3319 		int  a_flags;
3320 		off_t offlo;
3321 		off_t offhi;
3322 		struct lwp * a_l;
3323 	} */ *ap = v;
3324 
3325 	struct vnode *vp = ap->a_vp;
3326 
3327 	if (vp->v_type != VREG)
3328 		return 0;
3329 
3330 	return (nfs_flush(vp, ap->a_cred,
3331 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_l, 1));
3332 }
3333 
3334 /*
3335  * Flush all the data associated with a vnode.
3336  */
3337 int
3338 nfs_flush(
3339     struct vnode *vp,
3340     kauth_cred_t cred __unused,
3341     int waitfor __unused,
3342     struct lwp *l __unused,
3343     int commit __unused
3344 )
3345 {
3346 	struct nfsnode *np = VTONFS(vp);
3347 	int error;
3348 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3349 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3350 
3351 	simple_lock(&vp->v_interlock);
3352 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3353 	if (np->n_flag & NWRITEERR) {
3354 		error = np->n_error;
3355 		np->n_flag &= ~NWRITEERR;
3356 	}
3357 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3358 	return (error);
3359 }
3360 
3361 /*
3362  * Return POSIX pathconf information applicable to nfs.
3363  *
3364  * N.B. The NFS V2 protocol doesn't support this RPC.
3365  */
3366 /* ARGSUSED */
3367 int
3368 nfs_pathconf(v)
3369 	void *v;
3370 {
3371 	struct vop_pathconf_args /* {
3372 		struct vnode *a_vp;
3373 		int a_name;
3374 		register_t *a_retval;
3375 	} */ *ap = v;
3376 	struct nfsv3_pathconf *pcp;
3377 	struct vnode *vp = ap->a_vp;
3378 	struct mbuf *mreq, *mrep, *md, *mb;
3379 	int32_t t1, t2;
3380 	u_int32_t *tl;
3381 	caddr_t bpos, dpos, cp, cp2;
3382 	int error = 0, attrflag;
3383 #ifndef NFS_V2_ONLY
3384 	struct nfsmount *nmp;
3385 	unsigned int l;
3386 	u_int64_t maxsize;
3387 #endif
3388 	const int v3 = NFS_ISV3(vp);
3389 	struct nfsnode *np = VTONFS(vp);
3390 
3391 	switch (ap->a_name) {
3392 		/* Names that can be resolved locally. */
3393 	case _PC_PIPE_BUF:
3394 		*ap->a_retval = PIPE_BUF;
3395 		break;
3396 	case _PC_SYNC_IO:
3397 		*ap->a_retval = 1;
3398 		break;
3399 	/* Names that cannot be resolved locally; do an RPC, if possible. */
3400 	case _PC_LINK_MAX:
3401 	case _PC_NAME_MAX:
3402 	case _PC_CHOWN_RESTRICTED:
3403 	case _PC_NO_TRUNC:
3404 		if (!v3) {
3405 			error = EINVAL;
3406 			break;
3407 		}
3408 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3409 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3410 		nfsm_fhtom(np, 1);
3411 		nfsm_request(np, NFSPROC_PATHCONF,
3412 		    curlwp, curlwp->l_cred);	/* XXX */
3413 		nfsm_postop_attr(vp, attrflag, 0);
3414 		if (!error) {
3415 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
3416 			    NFSX_V3PATHCONF);
3417 			switch (ap->a_name) {
3418 			case _PC_LINK_MAX:
3419 				*ap->a_retval =
3420 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3421 				break;
3422 			case _PC_NAME_MAX:
3423 				*ap->a_retval =
3424 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3425 				break;
3426 			case _PC_CHOWN_RESTRICTED:
3427 				*ap->a_retval =
3428 				    (pcp->pc_chownrestricted == nfs_true);
3429 				break;
3430 			case _PC_NO_TRUNC:
3431 				*ap->a_retval =
3432 				    (pcp->pc_notrunc == nfs_true);
3433 				break;
3434 			}
3435 		}
3436 		nfsm_reqdone;
3437 		break;
3438 	case _PC_FILESIZEBITS:
3439 #ifndef NFS_V2_ONLY
3440 		if (v3) {
3441 			nmp = VFSTONFS(vp->v_mount);
3442 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3443 				if ((error = nfs_fsinfo(nmp, vp,
3444 				    curlwp->l_cred, curlwp)) != 0) /* XXX */
3445 					break;
3446 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3447 			    (maxsize >> l) > 0; l++)
3448 				;
3449 			*ap->a_retval = l + 1;
3450 		} else
3451 #endif
3452 		{
3453 			*ap->a_retval = 32;	/* NFS V2 limitation */
3454 		}
3455 		break;
3456 	default:
3457 		error = EINVAL;
3458 		break;
3459 	}
3460 
3461 	return (error);
3462 }
3463 
3464 /*
3465  * NFS advisory byte-level locks.
3466  */
3467 int
3468 nfs_advlock(v)
3469 	void *v;
3470 {
3471 	struct vop_advlock_args /* {
3472 		struct vnode *a_vp;
3473 		caddr_t  a_id;
3474 		int  a_op;
3475 		struct flock *a_fl;
3476 		int  a_flags;
3477 	} */ *ap = v;
3478 	struct nfsnode *np = VTONFS(ap->a_vp);
3479 
3480 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3481 }
3482 
3483 /*
3484  * Print out the contents of an nfsnode.
3485  */
3486 int
3487 nfs_print(v)
3488 	void *v;
3489 {
3490 	struct vop_print_args /* {
3491 		struct vnode *a_vp;
3492 	} */ *ap = v;
3493 	struct vnode *vp = ap->a_vp;
3494 	struct nfsnode *np = VTONFS(vp);
3495 
3496 	printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3497 	    (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3498 	if (vp->v_type == VFIFO)
3499 		fifo_printinfo(vp);
3500 	printf("\n");
3501 	return (0);
3502 }
3503 
3504 /*
3505  * nfs unlock wrapper.
3506  */
3507 int
3508 nfs_unlock(void *v)
3509 {
3510 	struct vop_unlock_args /* {
3511 		struct vnode *a_vp;
3512 		int a_flags;
3513 	} */ *ap = v;
3514 	struct vnode *vp = ap->a_vp;
3515 
3516 	/*
3517 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3518 	 * with v_data == 0.
3519 	 */
3520 	if (VTONFS(vp)) {
3521 		nfs_delayedtruncate(vp);
3522 	}
3523 
3524 	return genfs_unlock(v);
3525 }
3526 
3527 /*
3528  * nfs special file access vnode op.
3529  * Essentially just get vattr and then imitate iaccess() since the device is
3530  * local to the client.
3531  */
3532 int
3533 nfsspec_access(v)
3534 	void *v;
3535 {
3536 	struct vop_access_args /* {
3537 		struct vnode *a_vp;
3538 		int  a_mode;
3539 		kauth_cred_t a_cred;
3540 		struct lwp *a_l;
3541 	} */ *ap = v;
3542 	struct vattr va;
3543 	struct vnode *vp = ap->a_vp;
3544 	int error;
3545 
3546 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_l);
3547 	if (error)
3548 		return (error);
3549 
3550         /*
3551 	 * Disallow write attempts on filesystems mounted read-only;
3552 	 * unless the file is a socket, fifo, or a block or character
3553 	 * device resident on the filesystem.
3554 	 */
3555 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3556 		switch (vp->v_type) {
3557 		case VREG:
3558 		case VDIR:
3559 		case VLNK:
3560 			return (EROFS);
3561 		default:
3562 			break;
3563 		}
3564 	}
3565 
3566 	return (vaccess(va.va_type, va.va_mode,
3567 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3568 }
3569 
3570 /*
3571  * Read wrapper for special devices.
3572  */
3573 int
3574 nfsspec_read(v)
3575 	void *v;
3576 {
3577 	struct vop_read_args /* {
3578 		struct vnode *a_vp;
3579 		struct uio *a_uio;
3580 		int  a_ioflag;
3581 		kauth_cred_t a_cred;
3582 	} */ *ap = v;
3583 	struct nfsnode *np = VTONFS(ap->a_vp);
3584 
3585 	/*
3586 	 * Set access flag.
3587 	 */
3588 	np->n_flag |= NACC;
3589 	getnanotime(&np->n_atim);
3590 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3591 }
3592 
3593 /*
3594  * Write wrapper for special devices.
3595  */
3596 int
3597 nfsspec_write(v)
3598 	void *v;
3599 {
3600 	struct vop_write_args /* {
3601 		struct vnode *a_vp;
3602 		struct uio *a_uio;
3603 		int  a_ioflag;
3604 		kauth_cred_t a_cred;
3605 	} */ *ap = v;
3606 	struct nfsnode *np = VTONFS(ap->a_vp);
3607 
3608 	/*
3609 	 * Set update flag.
3610 	 */
3611 	np->n_flag |= NUPD;
3612 	getnanotime(&np->n_mtim);
3613 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3614 }
3615 
3616 /*
3617  * Close wrapper for special devices.
3618  *
3619  * Update the times on the nfsnode then do device close.
3620  */
3621 int
3622 nfsspec_close(v)
3623 	void *v;
3624 {
3625 	struct vop_close_args /* {
3626 		struct vnode *a_vp;
3627 		int  a_fflag;
3628 		kauth_cred_t a_cred;
3629 		struct lwp *a_l;
3630 	} */ *ap = v;
3631 	struct vnode *vp = ap->a_vp;
3632 	struct nfsnode *np = VTONFS(vp);
3633 	struct vattr vattr;
3634 
3635 	if (np->n_flag & (NACC | NUPD)) {
3636 		np->n_flag |= NCHG;
3637 		if (vp->v_usecount == 1 &&
3638 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3639 			VATTR_NULL(&vattr);
3640 			if (np->n_flag & NACC)
3641 				vattr.va_atime = np->n_atim;
3642 			if (np->n_flag & NUPD)
3643 				vattr.va_mtime = np->n_mtim;
3644 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l);
3645 		}
3646 	}
3647 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3648 }
3649 
3650 /*
3651  * Read wrapper for fifos.
3652  */
3653 int
3654 nfsfifo_read(v)
3655 	void *v;
3656 {
3657 	struct vop_read_args /* {
3658 		struct vnode *a_vp;
3659 		struct uio *a_uio;
3660 		int  a_ioflag;
3661 		kauth_cred_t a_cred;
3662 	} */ *ap = v;
3663 	struct nfsnode *np = VTONFS(ap->a_vp);
3664 
3665 	/*
3666 	 * Set access flag.
3667 	 */
3668 	np->n_flag |= NACC;
3669 	getnanotime(&np->n_atim);
3670 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3671 }
3672 
3673 /*
3674  * Write wrapper for fifos.
3675  */
3676 int
3677 nfsfifo_write(v)
3678 	void *v;
3679 {
3680 	struct vop_write_args /* {
3681 		struct vnode *a_vp;
3682 		struct uio *a_uio;
3683 		int  a_ioflag;
3684 		kauth_cred_t a_cred;
3685 	} */ *ap = v;
3686 	struct nfsnode *np = VTONFS(ap->a_vp);
3687 
3688 	/*
3689 	 * Set update flag.
3690 	 */
3691 	np->n_flag |= NUPD;
3692 	getnanotime(&np->n_mtim);
3693 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3694 }
3695 
3696 /*
3697  * Close wrapper for fifos.
3698  *
3699  * Update the times on the nfsnode then do fifo close.
3700  */
3701 int
3702 nfsfifo_close(v)
3703 	void *v;
3704 {
3705 	struct vop_close_args /* {
3706 		struct vnode *a_vp;
3707 		int  a_fflag;
3708 		kauth_cred_t a_cred;
3709 		struct lwp *a_l;
3710 	} */ *ap = v;
3711 	struct vnode *vp = ap->a_vp;
3712 	struct nfsnode *np = VTONFS(vp);
3713 	struct vattr vattr;
3714 
3715 	if (np->n_flag & (NACC | NUPD)) {
3716 		struct timespec ts;
3717 
3718 		getnanotime(&ts);
3719 		if (np->n_flag & NACC)
3720 			np->n_atim = ts;
3721 		if (np->n_flag & NUPD)
3722 			np->n_mtim = ts;
3723 		np->n_flag |= NCHG;
3724 		if (vp->v_usecount == 1 &&
3725 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3726 			VATTR_NULL(&vattr);
3727 			if (np->n_flag & NACC)
3728 				vattr.va_atime = np->n_atim;
3729 			if (np->n_flag & NUPD)
3730 				vattr.va_mtime = np->n_mtim;
3731 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l);
3732 		}
3733 	}
3734 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3735 }
3736