xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 5b84b3983f71fd20a534cfa5d1556623a8aaa717)
1 /*	$NetBSD: nfs_vnops.c,v 1.226 2005/08/19 10:08:48 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 1.226 2005/08/19 10:08:48 yamt Exp $");
43 
44 #include "opt_inet.h"
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 
48 #include <sys/param.h>
49 #include <sys/proc.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/resourcevar.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/namei.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/hash.h>
63 #include <sys/lockf.h>
64 #include <sys/stat.h>
65 #include <sys/unistd.h>
66 
67 #include <uvm/uvm_extern.h>
68 #include <uvm/uvm.h>
69 
70 #include <miscfs/fifofs/fifo.h>
71 #include <miscfs/genfs/genfs.h>
72 #include <miscfs/specfs/specdev.h>
73 
74 #include <nfs/rpcv2.h>
75 #include <nfs/nfsproto.h>
76 #include <nfs/nfs.h>
77 #include <nfs/nfsnode.h>
78 #include <nfs/nfsmount.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfs/nfsm_subs.h>
81 #include <nfs/nqnfs.h>
82 #include <nfs/nfs_var.h>
83 
84 #include <net/if.h>
85 #include <netinet/in.h>
86 #include <netinet/in_var.h>
87 
88 /*
89  * Global vfs data structures for nfs
90  */
91 int (**nfsv2_vnodeop_p) __P((void *));
92 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
93 	{ &vop_default_desc, vn_default_error },
94 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
95 	{ &vop_create_desc, nfs_create },		/* create */
96 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
97 	{ &vop_open_desc, nfs_open },			/* open */
98 	{ &vop_close_desc, nfs_close },			/* close */
99 	{ &vop_access_desc, nfs_access },		/* access */
100 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
101 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
102 	{ &vop_read_desc, nfs_read },			/* read */
103 	{ &vop_write_desc, nfs_write },			/* write */
104 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
105 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
106 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
107 	{ &vop_poll_desc, nfs_poll },			/* poll */
108 	{ &vop_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
109 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
110 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
111 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
112 	{ &vop_seek_desc, nfs_seek },			/* seek */
113 	{ &vop_remove_desc, nfs_remove },		/* remove */
114 	{ &vop_link_desc, nfs_link },			/* link */
115 	{ &vop_rename_desc, nfs_rename },		/* rename */
116 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
117 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
118 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
119 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
120 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
121 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
122 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
123 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
124 	{ &vop_lock_desc, nfs_lock },			/* lock */
125 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
126 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
127 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
128 	{ &vop_print_desc, nfs_print },			/* print */
129 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
130 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
131 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
132 	{ &vop_blkatoff_desc, genfs_badop },		/* blkatoff */
133 	{ &vop_valloc_desc, genfs_badop },		/* valloc */
134 	{ &vop_reallocblks_desc, genfs_badop },		/* reallocblks */
135 	{ &vop_vfree_desc, genfs_badop },		/* vfree */
136 	{ &vop_truncate_desc, genfs_badop },		/* truncate */
137 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
138 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
139 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
140 	{ NULL, NULL }
141 };
142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
143 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
144 
145 /*
146  * Special device vnode ops
147  */
148 int (**spec_nfsv2nodeop_p) __P((void *));
149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
150 	{ &vop_default_desc, vn_default_error },
151 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
152 	{ &vop_create_desc, spec_create },		/* create */
153 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
154 	{ &vop_open_desc, spec_open },			/* open */
155 	{ &vop_close_desc, nfsspec_close },		/* close */
156 	{ &vop_access_desc, nfsspec_access },		/* access */
157 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
158 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
159 	{ &vop_read_desc, nfsspec_read },		/* read */
160 	{ &vop_write_desc, nfsspec_write },		/* write */
161 	{ &vop_lease_desc, spec_lease_check },		/* lease */
162 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
163 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
164 	{ &vop_poll_desc, spec_poll },			/* poll */
165 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
166 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
167 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
168 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
169 	{ &vop_seek_desc, spec_seek },			/* seek */
170 	{ &vop_remove_desc, spec_remove },		/* remove */
171 	{ &vop_link_desc, spec_link },			/* link */
172 	{ &vop_rename_desc, spec_rename },		/* rename */
173 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
174 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
175 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
176 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
177 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
178 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
179 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
180 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
181 	{ &vop_lock_desc, nfs_lock },			/* lock */
182 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
183 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
184 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
185 	{ &vop_print_desc, nfs_print },			/* print */
186 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
187 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
188 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
189 	{ &vop_blkatoff_desc, genfs_badop },		/* blkatoff */
190 	{ &vop_valloc_desc, genfs_badop },		/* valloc */
191 	{ &vop_reallocblks_desc, genfs_badop },		/* reallocblks */
192 	{ &vop_vfree_desc, genfs_badop },		/* vfree */
193 	{ &vop_truncate_desc, genfs_badop },		/* truncate */
194 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
195 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
196 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
197 	{ NULL, NULL }
198 };
199 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
200 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
201 
202 int (**fifo_nfsv2nodeop_p) __P((void *));
203 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
204 	{ &vop_default_desc, vn_default_error },
205 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
206 	{ &vop_create_desc, fifo_create },		/* create */
207 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
208 	{ &vop_open_desc, fifo_open },			/* open */
209 	{ &vop_close_desc, nfsfifo_close },		/* close */
210 	{ &vop_access_desc, nfsspec_access },		/* access */
211 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
212 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
213 	{ &vop_read_desc, nfsfifo_read },		/* read */
214 	{ &vop_write_desc, nfsfifo_write },		/* write */
215 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
216 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
217 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
218 	{ &vop_poll_desc, fifo_poll },			/* poll */
219 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
220 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
221 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
222 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
223 	{ &vop_seek_desc, fifo_seek },			/* seek */
224 	{ &vop_remove_desc, fifo_remove },		/* remove */
225 	{ &vop_link_desc, fifo_link },			/* link */
226 	{ &vop_rename_desc, fifo_rename },		/* rename */
227 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
228 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
229 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
230 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
231 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
232 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
233 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
234 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
235 	{ &vop_lock_desc, nfs_lock },			/* lock */
236 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
237 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
238 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
239 	{ &vop_print_desc, nfs_print },			/* print */
240 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
241 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
242 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
243 	{ &vop_blkatoff_desc, genfs_badop },		/* blkatoff */
244 	{ &vop_valloc_desc, genfs_badop },		/* valloc */
245 	{ &vop_reallocblks_desc, genfs_badop },	/* reallocblks */
246 	{ &vop_vfree_desc, genfs_badop },		/* vfree */
247 	{ &vop_truncate_desc, genfs_badop },		/* truncate */
248 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
249 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
250 	{ NULL, NULL }
251 };
252 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
253 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
254 
255 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
256 
257 /*
258  * Global variables
259  */
260 extern u_int32_t nfs_true, nfs_false;
261 extern u_int32_t nfs_xdrneg1;
262 extern const nfstype nfsv3_type[9];
263 
264 int nfs_numasync = 0;
265 #define	DIRHDSIZ	_DIRENT_NAMEOFF(dp)
266 #define UIO_ADVANCE(uio, siz) \
267     (void)((uio)->uio_resid -= (siz), \
268     (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
269     (uio)->uio_iov->iov_len -= (siz))
270 
271 static void nfs_cache_enter(struct vnode *, struct vnode *,
272     struct componentname *);
273 
274 static void
275 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
276     struct componentname *cnp)
277 {
278 	struct nfsnode *dnp = VTONFS(dvp);
279 
280 	if (vp != NULL) {
281 		struct nfsnode *np = VTONFS(vp);
282 
283 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
284 	}
285 
286 	if (!timespecisset(&dnp->n_nctime))
287 		dnp->n_nctime = dnp->n_vattr->va_mtime;
288 
289 	cache_enter(dvp, vp, cnp);
290 }
291 
292 /*
293  * nfs null call from vfs.
294  */
295 int
296 nfs_null(vp, cred, procp)
297 	struct vnode *vp;
298 	struct ucred *cred;
299 	struct proc *procp;
300 {
301 	caddr_t bpos, dpos;
302 	int error = 0;
303 	struct mbuf *mreq, *mrep, *md, *mb;
304 	struct nfsnode *np = VTONFS(vp);
305 
306 	nfsm_reqhead(np, NFSPROC_NULL, 0);
307 	nfsm_request(np, NFSPROC_NULL, procp, cred);
308 	nfsm_reqdone;
309 	return (error);
310 }
311 
312 /*
313  * nfs access vnode op.
314  * For nfs version 2, just return ok. File accesses may fail later.
315  * For nfs version 3, use the access rpc to check accessibility. If file modes
316  * are changed on the server, accesses might still fail later.
317  */
318 int
319 nfs_access(v)
320 	void *v;
321 {
322 	struct vop_access_args /* {
323 		struct vnode *a_vp;
324 		int  a_mode;
325 		struct ucred *a_cred;
326 		struct proc *a_p;
327 	} */ *ap = v;
328 	struct vnode *vp = ap->a_vp;
329 #ifndef NFS_V2_ONLY
330 	u_int32_t *tl;
331 	caddr_t cp;
332 	int32_t t1, t2;
333 	caddr_t bpos, dpos, cp2;
334 	int error = 0, attrflag;
335 	struct mbuf *mreq, *mrep, *md, *mb;
336 	u_int32_t mode, rmode;
337 	const int v3 = NFS_ISV3(vp);
338 #endif
339 	int cachevalid;
340 	struct nfsnode *np = VTONFS(vp);
341 
342 	cachevalid = (np->n_accstamp != -1 &&
343 	    (mono_time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
344 	    np->n_accuid == ap->a_cred->cr_uid);
345 
346 	/*
347 	 * Check access cache first. If this request has been made for this
348 	 * uid shortly before, use the cached result.
349 	 */
350 	if (cachevalid) {
351 		if (!np->n_accerror) {
352 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
353 				return np->n_accerror;
354 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
355 			return np->n_accerror;
356 	}
357 
358 #ifndef NFS_V2_ONLY
359 	/*
360 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
361 	 * ufs_access() locally using the vattr. This may not be correct,
362 	 * since the server may apply other access criteria such as
363 	 * client uid-->server uid mapping that we do not know about, but
364 	 * this is better than just returning anything that is lying about
365 	 * in the cache.
366 	 */
367 	if (v3) {
368 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
369 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
370 		nfsm_fhtom(np, v3);
371 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
372 		if (ap->a_mode & VREAD)
373 			mode = NFSV3ACCESS_READ;
374 		else
375 			mode = 0;
376 		if (vp->v_type != VDIR) {
377 			if (ap->a_mode & VWRITE)
378 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
379 			if (ap->a_mode & VEXEC)
380 				mode |= NFSV3ACCESS_EXECUTE;
381 		} else {
382 			if (ap->a_mode & VWRITE)
383 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
384 					 NFSV3ACCESS_DELETE);
385 			if (ap->a_mode & VEXEC)
386 				mode |= NFSV3ACCESS_LOOKUP;
387 		}
388 		*tl = txdr_unsigned(mode);
389 		nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
390 		nfsm_postop_attr(vp, attrflag, 0);
391 		if (!error) {
392 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
393 			rmode = fxdr_unsigned(u_int32_t, *tl);
394 			/*
395 			 * The NFS V3 spec does not clarify whether or not
396 			 * the returned access bits can be a superset of
397 			 * the ones requested, so...
398 			 */
399 			if ((rmode & mode) != mode)
400 				error = EACCES;
401 		}
402 		nfsm_reqdone;
403 	} else
404 #endif
405 		return (nfsspec_access(ap));
406 #ifndef NFS_V2_ONLY
407 	/*
408 	 * Disallow write attempts on filesystems mounted read-only;
409 	 * unless the file is a socket, fifo, or a block or character
410 	 * device resident on the filesystem.
411 	 */
412 	if (!error && (ap->a_mode & VWRITE) &&
413 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
414 		switch (vp->v_type) {
415 		case VREG:
416 		case VDIR:
417 		case VLNK:
418 			error = EROFS;
419 		default:
420 			break;
421 		}
422 	}
423 
424 	if (!error || error == EACCES) {
425 		/*
426 		 * If we got the same result as for a previous,
427 		 * different request, OR it in. Don't update
428 		 * the timestamp in that case.
429 		 */
430 		if (cachevalid && np->n_accstamp != -1 &&
431 		    error == np->n_accerror) {
432 			if (!error)
433 				np->n_accmode |= ap->a_mode;
434 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
435 				np->n_accmode = ap->a_mode;
436 		} else {
437 			np->n_accstamp = mono_time.tv_sec;
438 			np->n_accuid = ap->a_cred->cr_uid;
439 			np->n_accmode = ap->a_mode;
440 			np->n_accerror = error;
441 		}
442 	}
443 
444 	return (error);
445 #endif
446 }
447 
448 /*
449  * nfs open vnode op
450  * Check to see if the type is ok
451  * and that deletion is not in progress.
452  * For paged in text files, you will need to flush the page cache
453  * if consistency is lost.
454  */
455 /* ARGSUSED */
456 int
457 nfs_open(v)
458 	void *v;
459 {
460 	struct vop_open_args /* {
461 		struct vnode *a_vp;
462 		int  a_mode;
463 		struct ucred *a_cred;
464 		struct proc *a_p;
465 	} */ *ap = v;
466 	struct vnode *vp = ap->a_vp;
467 	struct nfsnode *np = VTONFS(vp);
468 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
469 	int error;
470 
471 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
472 		return (EACCES);
473 	}
474 
475 	if (ap->a_mode & FREAD) {
476 		if (np->n_rcred != NULL)
477 			crfree(np->n_rcred);
478 		np->n_rcred = ap->a_cred;
479 		crhold(np->n_rcred);
480 	}
481 	if (ap->a_mode & FWRITE) {
482 		if (np->n_wcred != NULL)
483 			crfree(np->n_wcred);
484 		np->n_wcred = ap->a_cred;
485 		crhold(np->n_wcred);
486 	}
487 
488 #ifndef NFS_V2_ONLY
489 	/*
490 	 * Get a valid lease. If cached data is stale, flush it.
491 	 */
492 	if (nmp->nm_flag & NFSMNT_NQNFS) {
493 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
494 		    do {
495 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
496 			    ap->a_p);
497 		    } while (error == NQNFS_EXPIRED);
498 		    if (error)
499 			return (error);
500 		    if (np->n_lrev != np->n_brev ||
501 			(np->n_flag & NQNFSNONCACHE)) {
502 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
503 				ap->a_p, 1)) == EINTR)
504 				return (error);
505 			np->n_brev = np->n_lrev;
506 		    }
507 		}
508 	} else
509 #endif
510 	{
511 		error = nfs_flushstalebuf(vp, ap->a_cred, ap->a_p, 0);
512 		if (error)
513 			return error;
514 	}
515 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
516 		NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
517 	return (0);
518 }
519 
520 /*
521  * nfs close vnode op
522  * What an NFS client should do upon close after writing is a debatable issue.
523  * Most NFS clients push delayed writes to the server upon close, basically for
524  * two reasons:
525  * 1 - So that any write errors may be reported back to the client process
526  *     doing the close system call. By far the two most likely errors are
527  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
528  * 2 - To put a worst case upper bound on cache inconsistency between
529  *     multiple clients for the file.
530  * There is also a consistency problem for Version 2 of the protocol w.r.t.
531  * not being able to tell if other clients are writing a file concurrently,
532  * since there is no way of knowing if the changed modify time in the reply
533  * is only due to the write for this client.
534  * (NFS Version 3 provides weak cache consistency data in the reply that
535  *  should be sufficient to detect and handle this case.)
536  *
537  * The current code does the following:
538  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
539  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
540  *                     or commit them (this satisfies 1 and 2 except for the
541  *                     case where the server crashes after this close but
542  *                     before the commit RPC, which is felt to be "good
543  *                     enough". Changing the last argument to nfs_flush() to
544  *                     a 1 would force a commit operation, if it is felt a
545  *                     commit is necessary now.
546  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
547  *                     1 should be dealt with via an fsync() system call for
548  *                     cases where write errors are important.
549  */
550 /* ARGSUSED */
551 int
552 nfs_close(v)
553 	void *v;
554 {
555 	struct vop_close_args /* {
556 		struct vnodeop_desc *a_desc;
557 		struct vnode *a_vp;
558 		int  a_fflag;
559 		struct ucred *a_cred;
560 		struct proc *a_p;
561 	} */ *ap = v;
562 	struct vnode *vp = ap->a_vp;
563 	struct nfsnode *np = VTONFS(vp);
564 	int error = 0;
565 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
566 
567 	if (vp->v_type == VREG) {
568 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
569 		(np->n_flag & NMODIFIED)) {
570 #ifndef NFS_V2_ONLY
571 		if (NFS_ISV3(vp)) {
572 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
573 		    np->n_flag &= ~NMODIFIED;
574 		} else
575 #endif
576 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
577 		NFS_INVALIDATE_ATTRCACHE(np);
578 	    }
579 	    if (np->n_flag & NWRITEERR) {
580 		np->n_flag &= ~NWRITEERR;
581 		error = np->n_error;
582 	    }
583 	}
584 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
585 	return (error);
586 }
587 
588 /*
589  * nfs getattr call from vfs.
590  */
591 int
592 nfs_getattr(v)
593 	void *v;
594 {
595 	struct vop_getattr_args /* {
596 		struct vnode *a_vp;
597 		struct vattr *a_vap;
598 		struct ucred *a_cred;
599 		struct proc *a_p;
600 	} */ *ap = v;
601 	struct vnode *vp = ap->a_vp;
602 	struct nfsnode *np = VTONFS(vp);
603 	caddr_t cp;
604 	u_int32_t *tl;
605 	int32_t t1, t2;
606 	caddr_t bpos, dpos;
607 	int error = 0;
608 	struct mbuf *mreq, *mrep, *md, *mb;
609 	const int v3 = NFS_ISV3(vp);
610 
611 	/*
612 	 * Update local times for special files.
613 	 */
614 	if (np->n_flag & (NACC | NUPD))
615 		np->n_flag |= NCHG;
616 
617 	/*
618 	 * if we have delayed truncation, do it now.
619 	 */
620 	nfs_delayedtruncate(vp);
621 
622 	/*
623 	 * First look in the cache.
624 	 */
625 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
626 		return (0);
627 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
628 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
629 	nfsm_fhtom(np, v3);
630 	nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
631 	if (!error) {
632 		nfsm_loadattr(vp, ap->a_vap, 0);
633 		if (vp->v_type == VDIR &&
634 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
635 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
636 	}
637 	nfsm_reqdone;
638 	return (error);
639 }
640 
641 /*
642  * nfs setattr call.
643  */
644 int
645 nfs_setattr(v)
646 	void *v;
647 {
648 	struct vop_setattr_args /* {
649 		struct vnodeop_desc *a_desc;
650 		struct vnode *a_vp;
651 		struct vattr *a_vap;
652 		struct ucred *a_cred;
653 		struct proc *a_p;
654 	} */ *ap = v;
655 	struct vnode *vp = ap->a_vp;
656 	struct nfsnode *np = VTONFS(vp);
657 	struct vattr *vap = ap->a_vap;
658 	int error = 0;
659 	u_quad_t tsize = 0;
660 
661 	/*
662 	 * Setting of flags is not supported.
663 	 */
664 	if (vap->va_flags != VNOVAL)
665 		return (EOPNOTSUPP);
666 
667 	/*
668 	 * Disallow write attempts if the filesystem is mounted read-only.
669 	 */
670   	if ((vap->va_uid != (uid_t)VNOVAL ||
671 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
672 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
673 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
674 		return (EROFS);
675 	if (vap->va_size != VNOVAL) {
676  		switch (vp->v_type) {
677  		case VDIR:
678  			return (EISDIR);
679  		case VCHR:
680  		case VBLK:
681  		case VSOCK:
682  		case VFIFO:
683 			if (vap->va_mtime.tv_sec == VNOVAL &&
684 			    vap->va_atime.tv_sec == VNOVAL &&
685 			    vap->va_mode == (mode_t)VNOVAL &&
686 			    vap->va_uid == (uid_t)VNOVAL &&
687 			    vap->va_gid == (gid_t)VNOVAL)
688 				return (0);
689  			vap->va_size = VNOVAL;
690  			break;
691  		default:
692 			/*
693 			 * Disallow write attempts if the filesystem is
694 			 * mounted read-only.
695 			 */
696 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
697 				return (EROFS);
698  			uvm_vnp_setsize(vp, vap->va_size);
699  			tsize = np->n_size;
700 			np->n_size = vap->va_size;
701  			if (vap->va_size == 0)
702  				error = nfs_vinvalbuf(vp, 0,
703  				     ap->a_cred, ap->a_p, 1);
704 			else
705 				error = nfs_vinvalbuf(vp, V_SAVE,
706 				     ap->a_cred, ap->a_p, 1);
707 			if (error) {
708 				uvm_vnp_setsize(vp, tsize);
709 				return (error);
710 			}
711  			np->n_vattr->va_size = vap->va_size;
712   		}
713   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
714 		vap->va_atime.tv_sec != VNOVAL) &&
715 		vp->v_type == VREG &&
716   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
717 		 ap->a_p, 1)) == EINTR)
718 		return (error);
719 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
720 	if (error && vap->va_size != VNOVAL) {
721 		np->n_size = np->n_vattr->va_size = tsize;
722 		uvm_vnp_setsize(vp, np->n_size);
723 	}
724 	VN_KNOTE(vp, NOTE_ATTRIB);
725 	return (error);
726 }
727 
728 /*
729  * Do an nfs setattr rpc.
730  */
731 int
732 nfs_setattrrpc(vp, vap, cred, procp)
733 	struct vnode *vp;
734 	struct vattr *vap;
735 	struct ucred *cred;
736 	struct proc *procp;
737 {
738 	struct nfsv2_sattr *sp;
739 	caddr_t cp;
740 	int32_t t1, t2;
741 	caddr_t bpos, dpos;
742 	u_int32_t *tl;
743 	int error = 0;
744 	struct mbuf *mreq, *mrep, *md, *mb;
745 	const int v3 = NFS_ISV3(vp);
746 	struct nfsnode *np = VTONFS(vp);
747 #ifndef NFS_V2_ONLY
748 	int wccflag = NFSV3_WCCRATTR;
749 	caddr_t cp2;
750 #endif
751 
752 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
753 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
754 	nfsm_fhtom(np, v3);
755 #ifndef NFS_V2_ONLY
756 	if (v3) {
757 		nfsm_v3attrbuild(vap, TRUE);
758 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
759 		*tl = nfs_false;
760 	} else {
761 #endif
762 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
763 		if (vap->va_mode == (mode_t)VNOVAL)
764 			sp->sa_mode = nfs_xdrneg1;
765 		else
766 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
767 		if (vap->va_uid == (uid_t)VNOVAL)
768 			sp->sa_uid = nfs_xdrneg1;
769 		else
770 			sp->sa_uid = txdr_unsigned(vap->va_uid);
771 		if (vap->va_gid == (gid_t)VNOVAL)
772 			sp->sa_gid = nfs_xdrneg1;
773 		else
774 			sp->sa_gid = txdr_unsigned(vap->va_gid);
775 		sp->sa_size = txdr_unsigned(vap->va_size);
776 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
777 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
778 #ifndef NFS_V2_ONLY
779 	}
780 #endif
781 	nfsm_request(np, NFSPROC_SETATTR, procp, cred);
782 #ifndef NFS_V2_ONLY
783 	if (v3) {
784 		nfsm_wcc_data(vp, wccflag, 0, FALSE);
785 	} else
786 #endif
787 		nfsm_loadattr(vp, (struct vattr *)0, 0);
788 	nfsm_reqdone;
789 	return (error);
790 }
791 
792 /*
793  * nfs lookup call, one step at a time...
794  * First look in cache
795  * If not found, unlock the directory nfsnode and do the rpc
796  *
797  * This code is full of lock/unlock statements and checks, because
798  * we continue after cache_lookup has finished (we need to check
799  * with the attr cache and do an rpc if it has timed out). This means
800  * that the locking effects of cache_lookup have to be taken into
801  * account.
802  */
803 int
804 nfs_lookup(v)
805 	void *v;
806 {
807 	struct vop_lookup_args /* {
808 		struct vnodeop_desc *a_desc;
809 		struct vnode *a_dvp;
810 		struct vnode **a_vpp;
811 		struct componentname *a_cnp;
812 	} */ *ap = v;
813 	struct componentname *cnp = ap->a_cnp;
814 	struct vnode *dvp = ap->a_dvp;
815 	struct vnode **vpp = ap->a_vpp;
816 	int flags;
817 	struct vnode *newvp;
818 	u_int32_t *tl;
819 	caddr_t cp;
820 	int32_t t1, t2;
821 	caddr_t bpos, dpos, cp2;
822 	struct mbuf *mreq, *mrep, *md, *mb;
823 	long len;
824 	nfsfh_t *fhp;
825 	struct nfsnode *np;
826 	int lockparent, wantparent, error = 0, attrflag, fhsize;
827 	const int v3 = NFS_ISV3(dvp);
828 
829 	cnp->cn_flags &= ~PDIRUNLOCK;
830 	flags = cnp->cn_flags;
831 
832 	*vpp = NULLVP;
833 	newvp = NULLVP;
834 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
835 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
836 		return (EROFS);
837 	if (dvp->v_type != VDIR)
838 		return (ENOTDIR);
839 
840 	/*
841 	 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
842 	 */
843 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
844 		error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
845 		if (error)
846 			return error;
847 		if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
848 			return EISDIR;
849 		VREF(dvp);
850 		*vpp = dvp;
851 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
852 				cnp->cn_flags |= SAVENAME;
853 		return 0;
854 	}
855 
856 	lockparent = flags & LOCKPARENT;
857 	wantparent = flags & (LOCKPARENT|WANTPARENT);
858 	np = VTONFS(dvp);
859 
860 	/*
861 	 * Before tediously performing a linear scan of the directory,
862 	 * check the name cache to see if the directory/name pair
863 	 * we are looking for is known already.
864 	 * If the directory/name pair is found in the name cache,
865 	 * we have to ensure the directory has not changed from
866 	 * the time the cache entry has been created. If it has,
867 	 * the cache entry has to be ignored.
868 	 */
869 	error = cache_lookup_raw(dvp, vpp, cnp);
870 	KASSERT(dvp != *vpp);
871 	if (error >= 0) {
872 		struct vattr vattr;
873 		int err2;
874 
875 		if (error && error != ENOENT) {
876 			*vpp = NULLVP;
877 			return error;
878 		}
879 
880 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
881 		if (err2 != 0) {
882 			if (error == 0)
883 				vrele(*vpp);
884 			*vpp = NULLVP;
885 			return err2;
886 		}
887 
888 		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
889 		    cnp->cn_proc) || timespeccmp(&vattr.va_mtime,
890 		    &VTONFS(dvp)->n_nctime, !=)) {
891 			if (error == 0) {
892 				vrele(*vpp);
893 				*vpp = NULLVP;
894 			}
895 			cache_purge1(dvp, NULL, PURGE_CHILDREN);
896 			timespecclear(&np->n_nctime);
897 			goto dorpc;
898 		}
899 
900 		if (error == ENOENT) {
901 			goto noentry;
902 		}
903 
904 		newvp = *vpp;
905 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
906 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
907 			nfsstats.lookupcache_hits++;
908 			if ((flags & ISDOTDOT) != 0 ||
909 			    (~flags & (LOCKPARENT|ISLASTCN)) != 0) {
910 				VOP_UNLOCK(dvp, 0);
911 				cnp->cn_flags |= PDIRUNLOCK;
912 			}
913 			error = vn_lock(newvp, LK_EXCLUSIVE);
914 			if (error) {
915 				/* newvp has been revoked. */
916 				vrele(newvp);
917 				*vpp = NULL;
918 				return error;
919 			}
920 			if ((~flags & (LOCKPARENT|ISLASTCN)) == 0
921 			    && (cnp->cn_flags & PDIRUNLOCK)) {
922 				KASSERT(flags & ISDOTDOT);
923 				error = vn_lock(dvp, LK_EXCLUSIVE);
924 				if (error) {
925 					vput(newvp);
926 					*vpp = NULL;
927 					return error;
928 				}
929 				cnp->cn_flags &= ~PDIRUNLOCK;
930 			}
931 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
932 				cnp->cn_flags |= SAVENAME;
933 			KASSERT(newvp->v_type != VNON);
934 			return (0);
935 		}
936 		cache_purge1(newvp, NULL, PURGE_PARENTS);
937 		vrele(newvp);
938 		*vpp = NULLVP;
939 	}
940 dorpc:
941 #if 0
942 	/*
943 	 * because nfsv3 has the same CREATE semantics as ours,
944 	 * we don't have to perform LOOKUPs beforehand.
945 	 *
946 	 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
947 	 * XXX although we have no way to know if O_EXCL is requested or not.
948 	 */
949 
950 	if (v3 && cnp->cn_nameiop == CREATE &&
951 	    (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
952 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
953 		KASSERT(lockparent);
954 		cnp->cn_flags |= SAVENAME;
955 		return (EJUSTRETURN);
956 	}
957 #endif /* 0 */
958 
959 	error = 0;
960 	newvp = NULLVP;
961 	nfsstats.lookupcache_misses++;
962 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
963 	len = cnp->cn_namelen;
964 	nfsm_reqhead(np, NFSPROC_LOOKUP,
965 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
966 	nfsm_fhtom(np, v3);
967 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
968 	nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
969 	if (error) {
970 		nfsm_postop_attr(dvp, attrflag, 0);
971 		m_freem(mrep);
972 		goto nfsmout;
973 	}
974 	nfsm_getfh(fhp, fhsize, v3);
975 
976 	/*
977 	 * Handle RENAME case...
978 	 */
979 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
980 		if (NFS_CMPFH(np, fhp, fhsize)) {
981 			m_freem(mrep);
982 			return (EISDIR);
983 		}
984 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
985 		if (error) {
986 			m_freem(mrep);
987 			return error;
988 		}
989 		newvp = NFSTOV(np);
990 #ifndef NFS_V2_ONLY
991 		if (v3) {
992 			nfsm_postop_attr(newvp, attrflag, 0);
993 			nfsm_postop_attr(dvp, attrflag, 0);
994 		} else
995 #endif
996 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
997 		*vpp = newvp;
998 		m_freem(mrep);
999 		cnp->cn_flags |= SAVENAME;
1000 		if (!lockparent) {
1001 			VOP_UNLOCK(dvp, 0);
1002 			cnp->cn_flags |= PDIRUNLOCK;
1003 		}
1004 		goto validate;
1005 	}
1006 
1007 	/*
1008 	 * The postop attr handling is duplicated for each if case,
1009 	 * because it should be done while dvp is locked (unlocking
1010 	 * dvp is different for each case).
1011 	 */
1012 
1013 	if (NFS_CMPFH(np, fhp, fhsize)) {
1014 		/*
1015 		 * "." lookup
1016 		 */
1017 		VREF(dvp);
1018 		newvp = dvp;
1019 #ifndef NFS_V2_ONLY
1020 		if (v3) {
1021 			nfsm_postop_attr(newvp, attrflag, 0);
1022 			nfsm_postop_attr(dvp, attrflag, 0);
1023 		} else
1024 #endif
1025 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1026 	} else if (flags & ISDOTDOT) {
1027 		/*
1028 		 * ".." lookup
1029 		 */
1030 		VOP_UNLOCK(dvp, 0);
1031 		cnp->cn_flags |= PDIRUNLOCK;
1032 
1033 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1034 		if (error) {
1035 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
1036 				cnp->cn_flags &= ~PDIRUNLOCK;
1037 			m_freem(mrep);
1038 			return error;
1039 		}
1040 		newvp = NFSTOV(np);
1041 
1042 #ifndef NFS_V2_ONLY
1043 		if (v3) {
1044 			nfsm_postop_attr(newvp, attrflag, 0);
1045 			nfsm_postop_attr(dvp, attrflag, 0);
1046 		} else
1047 #endif
1048 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1049 
1050 		if (lockparent && (flags & ISLASTCN)) {
1051 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
1052 				m_freem(mrep);
1053 				vput(newvp);
1054 				return error;
1055 			}
1056 			cnp->cn_flags &= ~PDIRUNLOCK;
1057 		}
1058 	} else {
1059 		/*
1060 		 * Other lookups.
1061 		 */
1062 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1063 		if (error) {
1064 			m_freem(mrep);
1065 			return error;
1066 		}
1067 		newvp = NFSTOV(np);
1068 #ifndef NFS_V2_ONLY
1069 		if (v3) {
1070 			nfsm_postop_attr(newvp, attrflag, 0);
1071 			nfsm_postop_attr(dvp, attrflag, 0);
1072 		} else
1073 #endif
1074 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1075 		if (!lockparent || !(flags & ISLASTCN)) {
1076 			VOP_UNLOCK(dvp, 0);
1077 			cnp->cn_flags |= PDIRUNLOCK;
1078 		}
1079 	}
1080 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1081 		cnp->cn_flags |= SAVENAME;
1082 	if ((cnp->cn_flags & MAKEENTRY) &&
1083 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1084 		nfs_cache_enter(dvp, newvp, cnp);
1085 	}
1086 	*vpp = newvp;
1087 	nfsm_reqdone;
1088 	if (error) {
1089 		/*
1090 		 * We get here only because of errors returned by
1091 		 * the RPC. Otherwise we'll have returned above
1092 		 * (the nfsm_* macros will jump to nfsm_reqdone
1093 		 * on error).
1094 		 */
1095 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1096 		    cnp->cn_nameiop != CREATE) {
1097 			nfs_cache_enter(dvp, NULL, cnp);
1098 		}
1099 		if (newvp != NULLVP) {
1100 			vrele(newvp);
1101 			if (newvp != dvp)
1102 				VOP_UNLOCK(newvp, 0);
1103 		}
1104 noentry:
1105 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1106 		    (flags & ISLASTCN) && error == ENOENT) {
1107 			if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1108 				error = EROFS;
1109 			} else {
1110 				error = EJUSTRETURN;
1111 				cnp->cn_flags |= SAVENAME;
1112 			}
1113 		}
1114 		*vpp = NULL;
1115 		return error;
1116 	}
1117 
1118 validate:
1119 	/*
1120 	 * make sure we have valid type and size.
1121 	 */
1122 
1123 	newvp = *vpp;
1124 	if (newvp->v_type == VNON) {
1125 		struct vattr vattr; /* dummy */
1126 
1127 		KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1128 		error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc);
1129 		if (error) {
1130 			vput(newvp);
1131 			*vpp = NULL;
1132 		}
1133 	}
1134 
1135 	return error;
1136 }
1137 
1138 /*
1139  * nfs read call.
1140  * Just call nfs_bioread() to do the work.
1141  */
1142 int
1143 nfs_read(v)
1144 	void *v;
1145 {
1146 	struct vop_read_args /* {
1147 		struct vnode *a_vp;
1148 		struct uio *a_uio;
1149 		int  a_ioflag;
1150 		struct ucred *a_cred;
1151 	} */ *ap = v;
1152 	struct vnode *vp = ap->a_vp;
1153 
1154 	if (vp->v_type != VREG)
1155 		return (EPERM);
1156 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1157 }
1158 
1159 /*
1160  * nfs readlink call
1161  */
1162 int
1163 nfs_readlink(v)
1164 	void *v;
1165 {
1166 	struct vop_readlink_args /* {
1167 		struct vnode *a_vp;
1168 		struct uio *a_uio;
1169 		struct ucred *a_cred;
1170 	} */ *ap = v;
1171 	struct vnode *vp = ap->a_vp;
1172 	struct nfsnode *np = VTONFS(vp);
1173 
1174 	if (vp->v_type != VLNK)
1175 		return (EPERM);
1176 
1177 	if (np->n_rcred != NULL) {
1178 		crfree(np->n_rcred);
1179 	}
1180 	np->n_rcred = ap->a_cred;
1181 	crhold(np->n_rcred);
1182 
1183 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1184 }
1185 
1186 /*
1187  * Do a readlink rpc.
1188  * Called by nfs_doio() from below the buffer cache.
1189  */
1190 int
1191 nfs_readlinkrpc(vp, uiop, cred)
1192 	struct vnode *vp;
1193 	struct uio *uiop;
1194 	struct ucred *cred;
1195 {
1196 	u_int32_t *tl;
1197 	caddr_t cp;
1198 	int32_t t1, t2;
1199 	caddr_t bpos, dpos, cp2;
1200 	int error = 0;
1201 	uint32_t len;
1202 	struct mbuf *mreq, *mrep, *md, *mb;
1203 	const int v3 = NFS_ISV3(vp);
1204 	struct nfsnode *np = VTONFS(vp);
1205 #ifndef NFS_V2_ONLY
1206 	int attrflag;
1207 #endif
1208 
1209 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1210 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1211 	nfsm_fhtom(np, v3);
1212 	nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1213 #ifndef NFS_V2_ONLY
1214 	if (v3)
1215 		nfsm_postop_attr(vp, attrflag, 0);
1216 #endif
1217 	if (!error) {
1218 #ifndef NFS_V2_ONLY
1219 		if (v3) {
1220 			nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1221 			len = fxdr_unsigned(uint32_t, *tl);
1222 			if (len > MAXPATHLEN) {
1223 				/*
1224 				 * this pathname is too long for us.
1225 				 */
1226 				m_freem(mrep);
1227 				/* Solaris returns EINVAL. should we follow? */
1228 				error = ENAMETOOLONG;
1229 				goto nfsmout;
1230 			}
1231 		} else
1232 #endif
1233 		{
1234 			nfsm_strsiz(len, NFS_MAXPATHLEN);
1235 		}
1236 		nfsm_mtouio(uiop, len);
1237 	}
1238 	nfsm_reqdone;
1239 	return (error);
1240 }
1241 
1242 /*
1243  * nfs read rpc call
1244  * Ditto above
1245  */
1246 int
1247 nfs_readrpc(vp, uiop)
1248 	struct vnode *vp;
1249 	struct uio *uiop;
1250 {
1251 	u_int32_t *tl;
1252 	caddr_t cp;
1253 	int32_t t1, t2;
1254 	caddr_t bpos, dpos, cp2;
1255 	struct mbuf *mreq, *mrep, *md, *mb;
1256 	struct nfsmount *nmp;
1257 	int error = 0, len, retlen, tsiz, eof;
1258 	const int v3 = NFS_ISV3(vp);
1259 	struct nfsnode *np = VTONFS(vp);
1260 #ifndef NFS_V2_ONLY
1261 	int attrflag;
1262 #endif
1263 
1264 #ifndef nolint
1265 	eof = 0;
1266 #endif
1267 	nmp = VFSTONFS(vp->v_mount);
1268 	tsiz = uiop->uio_resid;
1269 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1270 		return (EFBIG);
1271 	while (tsiz > 0) {
1272 		nfsstats.rpccnt[NFSPROC_READ]++;
1273 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1274 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1275 		nfsm_fhtom(np, v3);
1276 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1277 #ifndef NFS_V2_ONLY
1278 		if (v3) {
1279 			txdr_hyper(uiop->uio_offset, tl);
1280 			*(tl + 2) = txdr_unsigned(len);
1281 		} else
1282 #endif
1283 		{
1284 			*tl++ = txdr_unsigned(uiop->uio_offset);
1285 			*tl++ = txdr_unsigned(len);
1286 			*tl = 0;
1287 		}
1288 		nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred);
1289 #ifndef NFS_V2_ONLY
1290 		if (v3) {
1291 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1292 			if (error) {
1293 				m_freem(mrep);
1294 				goto nfsmout;
1295 			}
1296 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1297 			eof = fxdr_unsigned(int, *(tl + 1));
1298 		} else
1299 #endif
1300 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1301 		nfsm_strsiz(retlen, nmp->nm_rsize);
1302 		nfsm_mtouio(uiop, retlen);
1303 		m_freem(mrep);
1304 		tsiz -= retlen;
1305 #ifndef NFS_V2_ONLY
1306 		if (v3) {
1307 			if (eof || retlen == 0)
1308 				tsiz = 0;
1309 		} else
1310 #endif
1311 		if (retlen < len)
1312 			tsiz = 0;
1313 	}
1314 nfsmout:
1315 	return (error);
1316 }
1317 
1318 struct nfs_writerpc_context {
1319 	struct simplelock nwc_slock;
1320 	volatile int nwc_mbufcount;
1321 };
1322 
1323 /*
1324  * free mbuf used to refer protected pages while write rpc call.
1325  * called at splvm.
1326  */
1327 static void
1328 nfs_writerpc_extfree(struct mbuf *m, caddr_t tbuf, size_t size, void *arg)
1329 {
1330 	struct nfs_writerpc_context *ctx = arg;
1331 
1332 	KASSERT(m != NULL);
1333 	KASSERT(ctx != NULL);
1334 	pool_cache_put(&mbpool_cache, m);
1335 	simple_lock(&ctx->nwc_slock);
1336 	if (--ctx->nwc_mbufcount == 0) {
1337 		wakeup(ctx);
1338 	}
1339 	simple_unlock(&ctx->nwc_slock);
1340 }
1341 
1342 /*
1343  * nfs write call
1344  */
1345 int
1346 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1347 	struct vnode *vp;
1348 	struct uio *uiop;
1349 	int *iomode;
1350 	boolean_t pageprotected;
1351 	boolean_t *stalewriteverfp;
1352 {
1353 	u_int32_t *tl;
1354 	caddr_t cp;
1355 	int32_t t1, t2;
1356 	caddr_t bpos, dpos;
1357 	struct mbuf *mreq, *mrep, *md, *mb;
1358 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1359 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1360 	const int v3 = NFS_ISV3(vp);
1361 	int committed = NFSV3WRITE_FILESYNC;
1362 	struct nfsnode *np = VTONFS(vp);
1363 	struct nfs_writerpc_context ctx;
1364 	int s;
1365 	struct lwp *l = NULL;
1366 	size_t origresid;
1367 #ifndef NFS_V2_ONLY
1368 	caddr_t cp2;
1369 	int rlen, commit;
1370 #endif
1371 
1372 	simple_lock_init(&ctx.nwc_slock);
1373 	ctx.nwc_mbufcount = 1;
1374 
1375 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1376 		panic("writerpc readonly vp %p", vp);
1377 	}
1378 
1379 #ifdef DIAGNOSTIC
1380 	if (uiop->uio_iovcnt != 1)
1381 		panic("nfs: writerpc iovcnt > 1");
1382 #endif
1383 	tsiz = uiop->uio_resid;
1384 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1385 		return (EFBIG);
1386 	if (pageprotected) {
1387 		l = curlwp;
1388 		PHOLD(l);
1389 	}
1390 retry:
1391 	origresid = uiop->uio_resid;
1392 	KASSERT(origresid == uiop->uio_iov->iov_len);
1393 	while (tsiz > 0) {
1394 		uint32_t datalen; /* data bytes need to be allocated in mbuf */
1395 		uint32_t backup;
1396 		boolean_t stalewriteverf = FALSE;
1397 
1398 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1399 		len = min(tsiz, nmp->nm_wsize);
1400 		datalen = pageprotected ? 0 : nfsm_rndup(len);
1401 		nfsm_reqhead(np, NFSPROC_WRITE,
1402 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1403 		nfsm_fhtom(np, v3);
1404 #ifndef NFS_V2_ONLY
1405 		if (v3) {
1406 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1407 			txdr_hyper(uiop->uio_offset, tl);
1408 			tl += 2;
1409 			*tl++ = txdr_unsigned(len);
1410 			*tl++ = txdr_unsigned(*iomode);
1411 			*tl = txdr_unsigned(len);
1412 		} else
1413 #endif
1414 		{
1415 			u_int32_t x;
1416 
1417 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1418 			/* Set both "begin" and "current" to non-garbage. */
1419 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1420 			*tl++ = x;      /* "begin offset" */
1421 			*tl++ = x;      /* "current offset" */
1422 			x = txdr_unsigned(len);
1423 			*tl++ = x;      /* total to this offset */
1424 			*tl = x;        /* size of this write */
1425 
1426 		}
1427 		if (pageprotected) {
1428 			/*
1429 			 * since we know pages can't be modified during i/o,
1430 			 * no need to copy them for us.
1431 			 */
1432 			struct mbuf *m;
1433 			struct iovec *iovp = uiop->uio_iov;
1434 
1435 			m = m_get(M_WAIT, MT_DATA);
1436 			MCLAIM(m, &nfs_mowner);
1437 			MEXTADD(m, iovp->iov_base, len, M_MBUF,
1438 			    nfs_writerpc_extfree, &ctx);
1439 			m->m_flags |= M_EXT_ROMAP;
1440 			m->m_len = len;
1441 			mb->m_next = m;
1442 			/*
1443 			 * no need to maintain mb and bpos here
1444 			 * because no one care them later.
1445 			 */
1446 #if 0
1447 			mb = m;
1448 			bpos = mtod(caddr_t, mb) + mb->m_len;
1449 #endif
1450 			UIO_ADVANCE(uiop, len);
1451 			uiop->uio_offset += len;
1452 			s = splvm();
1453 			simple_lock(&ctx.nwc_slock);
1454 			ctx.nwc_mbufcount++;
1455 			simple_unlock(&ctx.nwc_slock);
1456 			splx(s);
1457 			nfs_zeropad(mb, 0, nfsm_padlen(len));
1458 		} else {
1459 			nfsm_uiotom(uiop, len);
1460 		}
1461 		nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred);
1462 #ifndef NFS_V2_ONLY
1463 		if (v3) {
1464 			wccflag = NFSV3_WCCCHK;
1465 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1466 			if (!error) {
1467 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1468 					+ NFSX_V3WRITEVERF);
1469 				rlen = fxdr_unsigned(int, *tl++);
1470 				if (rlen == 0) {
1471 					error = NFSERR_IO;
1472 					m_freem(mrep);
1473 					break;
1474 				} else if (rlen < len) {
1475 					backup = len - rlen;
1476 					UIO_ADVANCE(uiop, -backup);
1477 					uiop->uio_offset -= backup;
1478 					len = rlen;
1479 				}
1480 				commit = fxdr_unsigned(int, *tl++);
1481 
1482 				/*
1483 				 * Return the lowest committment level
1484 				 * obtained by any of the RPCs.
1485 				 */
1486 				if (committed == NFSV3WRITE_FILESYNC)
1487 					committed = commit;
1488 				else if (committed == NFSV3WRITE_DATASYNC &&
1489 					commit == NFSV3WRITE_UNSTABLE)
1490 					committed = commit;
1491 				simple_lock(&nmp->nm_slock);
1492 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1493 					memcpy(nmp->nm_writeverf, tl,
1494 					    NFSX_V3WRITEVERF);
1495 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1496 				} else if ((nmp->nm_iflag &
1497 				    NFSMNT_STALEWRITEVERF) ||
1498 				    memcmp(tl, nmp->nm_writeverf,
1499 				    NFSX_V3WRITEVERF)) {
1500 					memcpy(nmp->nm_writeverf, tl,
1501 					    NFSX_V3WRITEVERF);
1502 					/*
1503 					 * note NFSMNT_STALEWRITEVERF
1504 					 * if we're the first thread to
1505 					 * notice it.
1506 					 */
1507 					if ((nmp->nm_iflag &
1508 					    NFSMNT_STALEWRITEVERF) == 0) {
1509 						stalewriteverf = TRUE;
1510 						nmp->nm_iflag |=
1511 						    NFSMNT_STALEWRITEVERF;
1512 					}
1513 				}
1514 				simple_unlock(&nmp->nm_slock);
1515 			}
1516 		} else
1517 #endif
1518 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1519 		if (wccflag)
1520 			VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1521 		m_freem(mrep);
1522 		if (error)
1523 			break;
1524 		tsiz -= len;
1525 		if (stalewriteverf) {
1526 			*stalewriteverfp = TRUE;
1527 			stalewriteverf = FALSE;
1528 			if (committed == NFSV3WRITE_UNSTABLE &&
1529 			    len != origresid) {
1530 				/*
1531 				 * if our write requests weren't atomic but
1532 				 * unstable, datas in previous iterations
1533 				 * might have already been lost now.
1534 				 * then, we should resend them to nfsd.
1535 				 */
1536 				backup = origresid - tsiz;
1537 				UIO_ADVANCE(uiop, -backup);
1538 				uiop->uio_offset -= backup;
1539 				tsiz = origresid;
1540 				goto retry;
1541 			}
1542 		}
1543 	}
1544 nfsmout:
1545 	if (pageprotected) {
1546 		/*
1547 		 * wait until mbufs go away.
1548 		 * retransmitted mbufs can survive longer than rpc requests
1549 		 * themselves.
1550 		 */
1551 		s = splvm();
1552 		simple_lock(&ctx.nwc_slock);
1553 		ctx.nwc_mbufcount--;
1554 		while (ctx.nwc_mbufcount > 0) {
1555 			ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1556 		}
1557 		simple_unlock(&ctx.nwc_slock);
1558 		splx(s);
1559 		PRELE(l);
1560 	}
1561 	*iomode = committed;
1562 	if (error)
1563 		uiop->uio_resid = tsiz;
1564 	return (error);
1565 }
1566 
1567 /*
1568  * nfs mknod rpc
1569  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1570  * mode set to specify the file type and the size field for rdev.
1571  */
1572 int
1573 nfs_mknodrpc(dvp, vpp, cnp, vap)
1574 	struct vnode *dvp;
1575 	struct vnode **vpp;
1576 	struct componentname *cnp;
1577 	struct vattr *vap;
1578 {
1579 	struct nfsv2_sattr *sp;
1580 	u_int32_t *tl;
1581 	caddr_t cp;
1582 	int32_t t1, t2;
1583 	struct vnode *newvp = (struct vnode *)0;
1584 	struct nfsnode *dnp, *np;
1585 	char *cp2;
1586 	caddr_t bpos, dpos;
1587 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1588 	struct mbuf *mreq, *mrep, *md, *mb;
1589 	u_int32_t rdev;
1590 	const int v3 = NFS_ISV3(dvp);
1591 
1592 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1593 		rdev = txdr_unsigned(vap->va_rdev);
1594 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1595 		rdev = nfs_xdrneg1;
1596 	else {
1597 		VOP_ABORTOP(dvp, cnp);
1598 		vput(dvp);
1599 		return (EOPNOTSUPP);
1600 	}
1601 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1602 	dnp = VTONFS(dvp);
1603 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1604 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1605 	nfsm_fhtom(dnp, v3);
1606 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1607 #ifndef NFS_V2_ONLY
1608 	if (v3) {
1609 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1610 		*tl++ = vtonfsv3_type(vap->va_type);
1611 		nfsm_v3attrbuild(vap, FALSE);
1612 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1613 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1614 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1615 			*tl = txdr_unsigned(minor(vap->va_rdev));
1616 		}
1617 	} else
1618 #endif
1619 	{
1620 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1621 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1622 		sp->sa_uid = nfs_xdrneg1;
1623 		sp->sa_gid = nfs_xdrneg1;
1624 		sp->sa_size = rdev;
1625 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1626 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1627 	}
1628 	nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1629 	if (!error) {
1630 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1631 		if (!gotvp) {
1632 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1633 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1634 			if (!error)
1635 				newvp = NFSTOV(np);
1636 		}
1637 	}
1638 #ifndef NFS_V2_ONLY
1639 	if (v3)
1640 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1641 #endif
1642 	nfsm_reqdone;
1643 	if (error) {
1644 		if (newvp)
1645 			vput(newvp);
1646 	} else {
1647 		if (cnp->cn_flags & MAKEENTRY)
1648 			nfs_cache_enter(dvp, newvp, cnp);
1649 		*vpp = newvp;
1650 	}
1651 	PNBUF_PUT(cnp->cn_pnbuf);
1652 	VTONFS(dvp)->n_flag |= NMODIFIED;
1653 	if (!wccflag)
1654 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1655 	vput(dvp);
1656 	return (error);
1657 }
1658 
1659 /*
1660  * nfs mknod vop
1661  * just call nfs_mknodrpc() to do the work.
1662  */
1663 /* ARGSUSED */
1664 int
1665 nfs_mknod(v)
1666 	void *v;
1667 {
1668 	struct vop_mknod_args /* {
1669 		struct vnode *a_dvp;
1670 		struct vnode **a_vpp;
1671 		struct componentname *a_cnp;
1672 		struct vattr *a_vap;
1673 	} */ *ap = v;
1674 	struct vnode *dvp = ap->a_dvp;
1675 	struct componentname *cnp = ap->a_cnp;
1676 	int error;
1677 
1678 	error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1679 	VN_KNOTE(dvp, NOTE_WRITE);
1680 	if (error == 0 || error == EEXIST)
1681 		cache_purge1(dvp, cnp, 0);
1682 	return (error);
1683 }
1684 
1685 #ifndef NFS_V2_ONLY
1686 static u_long create_verf;
1687 #endif
1688 /*
1689  * nfs file create call
1690  */
1691 int
1692 nfs_create(v)
1693 	void *v;
1694 {
1695 	struct vop_create_args /* {
1696 		struct vnode *a_dvp;
1697 		struct vnode **a_vpp;
1698 		struct componentname *a_cnp;
1699 		struct vattr *a_vap;
1700 	} */ *ap = v;
1701 	struct vnode *dvp = ap->a_dvp;
1702 	struct vattr *vap = ap->a_vap;
1703 	struct componentname *cnp = ap->a_cnp;
1704 	struct nfsv2_sattr *sp;
1705 	u_int32_t *tl;
1706 	caddr_t cp;
1707 	int32_t t1, t2;
1708 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1709 	struct vnode *newvp = (struct vnode *)0;
1710 	caddr_t bpos, dpos, cp2;
1711 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1712 	struct mbuf *mreq, *mrep, *md, *mb;
1713 	const int v3 = NFS_ISV3(dvp);
1714 
1715 	/*
1716 	 * Oops, not for me..
1717 	 */
1718 	if (vap->va_type == VSOCK)
1719 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1720 
1721 	KASSERT(vap->va_type == VREG);
1722 
1723 #ifdef VA_EXCLUSIVE
1724 	if (vap->va_vaflags & VA_EXCLUSIVE)
1725 		fmode |= O_EXCL;
1726 #endif
1727 again:
1728 	error = 0;
1729 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1730 	dnp = VTONFS(dvp);
1731 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1732 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1733 	nfsm_fhtom(dnp, v3);
1734 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1735 #ifndef NFS_V2_ONLY
1736 	if (v3) {
1737 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1738 		if (fmode & O_EXCL) {
1739 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1740 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1741 #ifdef INET
1742 			if (TAILQ_FIRST(&in_ifaddrhead))
1743 				*tl++ = TAILQ_FIRST(&in_ifaddrhead)->
1744 				    ia_addr.sin_addr.s_addr;
1745 			else
1746 				*tl++ = create_verf;
1747 #else
1748 			*tl++ = create_verf;
1749 #endif
1750 			*tl = ++create_verf;
1751 		} else {
1752 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1753 			nfsm_v3attrbuild(vap, FALSE);
1754 		}
1755 	} else
1756 #endif
1757 	{
1758 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1759 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1760 		sp->sa_uid = nfs_xdrneg1;
1761 		sp->sa_gid = nfs_xdrneg1;
1762 		sp->sa_size = 0;
1763 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1764 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1765 	}
1766 	nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1767 	if (!error) {
1768 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1769 		if (!gotvp) {
1770 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1771 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1772 			if (!error)
1773 				newvp = NFSTOV(np);
1774 		}
1775 	}
1776 #ifndef NFS_V2_ONLY
1777 	if (v3)
1778 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1779 #endif
1780 	nfsm_reqdone;
1781 	if (error) {
1782 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1783 			fmode &= ~O_EXCL;
1784 			goto again;
1785 		}
1786 	} else if (v3 && (fmode & O_EXCL)) {
1787 		struct timeval tm = time;
1788 
1789 		/*
1790 		 * make sure that we'll update timestamps as
1791 		 * most server implementations use them to store
1792 		 * the create verifier.
1793 		 *
1794 		 * XXX it's better to use TOSERVER always.
1795 		 */
1796 
1797 		if (vap->va_atime.tv_sec == VNOVAL) {
1798 			vap->va_atime.tv_sec = tm.tv_sec;
1799 			vap->va_atime.tv_nsec = tm.tv_usec * 1000;
1800 		}
1801 		if (vap->va_mtime.tv_sec == VNOVAL) {
1802 			vap->va_mtime.tv_sec = tm.tv_sec;
1803 			vap->va_mtime.tv_nsec = tm.tv_usec * 1000;
1804 		}
1805 
1806 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1807 	}
1808 	if (error == 0) {
1809 		if (cnp->cn_flags & MAKEENTRY)
1810 			nfs_cache_enter(dvp, newvp, cnp);
1811 		else
1812 			cache_purge1(dvp, cnp, 0);
1813 		*ap->a_vpp = newvp;
1814 	} else {
1815 		if (newvp)
1816 			vput(newvp);
1817 		if (error == EEXIST)
1818 			cache_purge1(dvp, cnp, 0);
1819 	}
1820 	PNBUF_PUT(cnp->cn_pnbuf);
1821 	VTONFS(dvp)->n_flag |= NMODIFIED;
1822 	if (!wccflag)
1823 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1824 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1825 	vput(dvp);
1826 	return (error);
1827 }
1828 
1829 /*
1830  * nfs file remove call
1831  * To try and make nfs semantics closer to ufs semantics, a file that has
1832  * other processes using the vnode is renamed instead of removed and then
1833  * removed later on the last close.
1834  * - If v_usecount > 1
1835  *	  If a rename is not already in the works
1836  *	     call nfs_sillyrename() to set it up
1837  *     else
1838  *	  do the remove rpc
1839  */
1840 int
1841 nfs_remove(v)
1842 	void *v;
1843 {
1844 	struct vop_remove_args /* {
1845 		struct vnodeop_desc *a_desc;
1846 		struct vnode * a_dvp;
1847 		struct vnode * a_vp;
1848 		struct componentname * a_cnp;
1849 	} */ *ap = v;
1850 	struct vnode *vp = ap->a_vp;
1851 	struct vnode *dvp = ap->a_dvp;
1852 	struct componentname *cnp = ap->a_cnp;
1853 	struct nfsnode *np = VTONFS(vp);
1854 	int error = 0;
1855 	struct vattr vattr;
1856 
1857 #ifndef DIAGNOSTIC
1858 	if ((cnp->cn_flags & HASBUF) == 0)
1859 		panic("nfs_remove: no name");
1860 	if (vp->v_usecount < 1)
1861 		panic("nfs_remove: bad v_usecount");
1862 #endif
1863 	if (vp->v_type == VDIR)
1864 		error = EPERM;
1865 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1866 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1867 	    vattr.va_nlink > 1)) {
1868 		/*
1869 		 * Purge the name cache so that the chance of a lookup for
1870 		 * the name succeeding while the remove is in progress is
1871 		 * minimized. Without node locking it can still happen, such
1872 		 * that an I/O op returns ESTALE, but since you get this if
1873 		 * another host removes the file..
1874 		 */
1875 		cache_purge(vp);
1876 		/*
1877 		 * throw away biocache buffers, mainly to avoid
1878 		 * unnecessary delayed writes later.
1879 		 */
1880 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1881 		/* Do the rpc */
1882 		if (error != EINTR)
1883 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1884 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1885 	} else if (!np->n_sillyrename)
1886 		error = nfs_sillyrename(dvp, vp, cnp);
1887 	PNBUF_PUT(cnp->cn_pnbuf);
1888 	if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1889 	    vattr.va_nlink == 1) {
1890 		np->n_flag |= NREMOVED;
1891 	}
1892 	NFS_INVALIDATE_ATTRCACHE(np);
1893 	VN_KNOTE(vp, NOTE_DELETE);
1894 	VN_KNOTE(dvp, NOTE_WRITE);
1895 	if (dvp == vp)
1896 		vrele(vp);
1897 	else
1898 		vput(vp);
1899 	vput(dvp);
1900 	return (error);
1901 }
1902 
1903 /*
1904  * nfs file remove rpc called from nfs_inactive
1905  */
1906 int
1907 nfs_removeit(sp)
1908 	struct sillyrename *sp;
1909 {
1910 
1911 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1912 		(struct proc *)0));
1913 }
1914 
1915 /*
1916  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1917  */
1918 int
1919 nfs_removerpc(dvp, name, namelen, cred, proc)
1920 	struct vnode *dvp;
1921 	const char *name;
1922 	int namelen;
1923 	struct ucred *cred;
1924 	struct proc *proc;
1925 {
1926 	u_int32_t *tl;
1927 	caddr_t cp;
1928 #ifndef NFS_V2_ONLY
1929 	int32_t t1;
1930 	caddr_t cp2;
1931 #endif
1932 	int32_t t2;
1933 	caddr_t bpos, dpos;
1934 	int error = 0, wccflag = NFSV3_WCCRATTR;
1935 	struct mbuf *mreq, *mrep, *md, *mb;
1936 	const int v3 = NFS_ISV3(dvp);
1937 	int rexmit;
1938 	struct nfsnode *dnp = VTONFS(dvp);
1939 
1940 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1941 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1942 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1943 	nfsm_fhtom(dnp, v3);
1944 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1945 	nfsm_request1(dnp, NFSPROC_REMOVE, proc, cred, &rexmit);
1946 #ifndef NFS_V2_ONLY
1947 	if (v3)
1948 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1949 #endif
1950 	nfsm_reqdone;
1951 	VTONFS(dvp)->n_flag |= NMODIFIED;
1952 	if (!wccflag)
1953 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1954 	/*
1955 	 * Kludge City: If the first reply to the remove rpc is lost..
1956 	 *   the reply to the retransmitted request will be ENOENT
1957 	 *   since the file was in fact removed
1958 	 *   Therefore, we cheat and return success.
1959 	 */
1960 	if (rexmit && error == ENOENT)
1961 		error = 0;
1962 	return (error);
1963 }
1964 
1965 /*
1966  * nfs file rename call
1967  */
1968 int
1969 nfs_rename(v)
1970 	void *v;
1971 {
1972 	struct vop_rename_args  /* {
1973 		struct vnode *a_fdvp;
1974 		struct vnode *a_fvp;
1975 		struct componentname *a_fcnp;
1976 		struct vnode *a_tdvp;
1977 		struct vnode *a_tvp;
1978 		struct componentname *a_tcnp;
1979 	} */ *ap = v;
1980 	struct vnode *fvp = ap->a_fvp;
1981 	struct vnode *tvp = ap->a_tvp;
1982 	struct vnode *fdvp = ap->a_fdvp;
1983 	struct vnode *tdvp = ap->a_tdvp;
1984 	struct componentname *tcnp = ap->a_tcnp;
1985 	struct componentname *fcnp = ap->a_fcnp;
1986 	int error;
1987 
1988 #ifndef DIAGNOSTIC
1989 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1990 	    (fcnp->cn_flags & HASBUF) == 0)
1991 		panic("nfs_rename: no name");
1992 #endif
1993 	/* Check for cross-device rename */
1994 	if ((fvp->v_mount != tdvp->v_mount) ||
1995 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1996 		error = EXDEV;
1997 		goto out;
1998 	}
1999 
2000 	/*
2001 	 * If the tvp exists and is in use, sillyrename it before doing the
2002 	 * rename of the new file over it.
2003 	 */
2004 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
2005 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2006 		VN_KNOTE(tvp, NOTE_DELETE);
2007 		vput(tvp);
2008 		tvp = NULL;
2009 	}
2010 
2011 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2012 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2013 		tcnp->cn_proc);
2014 
2015 	VN_KNOTE(fdvp, NOTE_WRITE);
2016 	VN_KNOTE(tdvp, NOTE_WRITE);
2017 	if (error == 0 || error == EEXIST) {
2018 		if (fvp->v_type == VDIR)
2019 			cache_purge(fvp);
2020 		else
2021 			cache_purge1(fdvp, fcnp, 0);
2022 		if (tvp != NULL && tvp->v_type == VDIR)
2023 			cache_purge(tvp);
2024 		else
2025 			cache_purge1(tdvp, tcnp, 0);
2026 	}
2027 out:
2028 	if (tdvp == tvp)
2029 		vrele(tdvp);
2030 	else
2031 		vput(tdvp);
2032 	if (tvp)
2033 		vput(tvp);
2034 	vrele(fdvp);
2035 	vrele(fvp);
2036 	return (error);
2037 }
2038 
2039 /*
2040  * nfs file rename rpc called from nfs_remove() above
2041  */
2042 int
2043 nfs_renameit(sdvp, scnp, sp)
2044 	struct vnode *sdvp;
2045 	struct componentname *scnp;
2046 	struct sillyrename *sp;
2047 {
2048 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2049 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
2050 }
2051 
2052 /*
2053  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2054  */
2055 int
2056 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
2057 	struct vnode *fdvp;
2058 	const char *fnameptr;
2059 	int fnamelen;
2060 	struct vnode *tdvp;
2061 	const char *tnameptr;
2062 	int tnamelen;
2063 	struct ucred *cred;
2064 	struct proc *proc;
2065 {
2066 	u_int32_t *tl;
2067 	caddr_t cp;
2068 #ifndef NFS_V2_ONLY
2069 	int32_t t1;
2070 	caddr_t cp2;
2071 #endif
2072 	int32_t t2;
2073 	caddr_t bpos, dpos;
2074 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2075 	struct mbuf *mreq, *mrep, *md, *mb;
2076 	const int v3 = NFS_ISV3(fdvp);
2077 	int rexmit;
2078 	struct nfsnode *fdnp = VTONFS(fdvp);
2079 
2080 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2081 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
2082 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2083 		nfsm_rndup(tnamelen));
2084 	nfsm_fhtom(fdnp, v3);
2085 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2086 	nfsm_fhtom(VTONFS(tdvp), v3);
2087 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2088 	nfsm_request1(fdnp, NFSPROC_RENAME, proc, cred, &rexmit);
2089 #ifndef NFS_V2_ONLY
2090 	if (v3) {
2091 		nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2092 		nfsm_wcc_data(tdvp, twccflag, 0, !error);
2093 	}
2094 #endif
2095 	nfsm_reqdone;
2096 	VTONFS(fdvp)->n_flag |= NMODIFIED;
2097 	VTONFS(tdvp)->n_flag |= NMODIFIED;
2098 	if (!fwccflag)
2099 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2100 	if (!twccflag)
2101 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2102 	/*
2103 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2104 	 */
2105 	if (rexmit && error == ENOENT)
2106 		error = 0;
2107 	return (error);
2108 }
2109 
2110 /*
2111  * nfs hard link create call
2112  */
2113 int
2114 nfs_link(v)
2115 	void *v;
2116 {
2117 	struct vop_link_args /* {
2118 		struct vnode *a_dvp;
2119 		struct vnode *a_vp;
2120 		struct componentname *a_cnp;
2121 	} */ *ap = v;
2122 	struct vnode *vp = ap->a_vp;
2123 	struct vnode *dvp = ap->a_dvp;
2124 	struct componentname *cnp = ap->a_cnp;
2125 	u_int32_t *tl;
2126 	caddr_t cp;
2127 #ifndef NFS_V2_ONLY
2128 	int32_t t1;
2129 	caddr_t cp2;
2130 #endif
2131 	int32_t t2;
2132 	caddr_t bpos, dpos;
2133 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2134 	struct mbuf *mreq, *mrep, *md, *mb;
2135 	/* XXX Should be const and initialised? */
2136 	int v3;
2137 	int rexmit;
2138 	struct nfsnode *np;
2139 
2140 	if (dvp->v_mount != vp->v_mount) {
2141 		VOP_ABORTOP(dvp, cnp);
2142 		vput(dvp);
2143 		return (EXDEV);
2144 	}
2145 	if (dvp != vp) {
2146 		error = vn_lock(vp, LK_EXCLUSIVE);
2147 		if (error != 0) {
2148 			VOP_ABORTOP(dvp, cnp);
2149 			vput(dvp);
2150 			return error;
2151 		}
2152 	}
2153 
2154 	/*
2155 	 * Push all writes to the server, so that the attribute cache
2156 	 * doesn't get "out of sync" with the server.
2157 	 * XXX There should be a better way!
2158 	 */
2159 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
2160 
2161 	v3 = NFS_ISV3(vp);
2162 	nfsstats.rpccnt[NFSPROC_LINK]++;
2163 	np = VTONFS(vp);
2164 	nfsm_reqhead(np, NFSPROC_LINK,
2165 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2166 	nfsm_fhtom(np, v3);
2167 	nfsm_fhtom(VTONFS(dvp), v3);
2168 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2169 	nfsm_request1(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred, &rexmit);
2170 #ifndef NFS_V2_ONLY
2171 	if (v3) {
2172 		nfsm_postop_attr(vp, attrflag, 0);
2173 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2174 	}
2175 #endif
2176 	nfsm_reqdone;
2177 	if (error == 0 || error == EEXIST)
2178 		cache_purge1(dvp, cnp, 0);
2179 	PNBUF_PUT(cnp->cn_pnbuf);
2180 	VTONFS(dvp)->n_flag |= NMODIFIED;
2181 	if (!attrflag)
2182 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2183 	if (!wccflag)
2184 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2185 	if (dvp != vp)
2186 		VOP_UNLOCK(vp, 0);
2187 	VN_KNOTE(vp, NOTE_LINK);
2188 	VN_KNOTE(dvp, NOTE_WRITE);
2189 	vput(dvp);
2190 	/*
2191 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2192 	 */
2193 	if (rexmit && error == EEXIST)
2194 		error = 0;
2195 	return (error);
2196 }
2197 
2198 /*
2199  * nfs symbolic link create call
2200  */
2201 int
2202 nfs_symlink(v)
2203 	void *v;
2204 {
2205 	struct vop_symlink_args /* {
2206 		struct vnode *a_dvp;
2207 		struct vnode **a_vpp;
2208 		struct componentname *a_cnp;
2209 		struct vattr *a_vap;
2210 		char *a_target;
2211 	} */ *ap = v;
2212 	struct vnode *dvp = ap->a_dvp;
2213 	struct vattr *vap = ap->a_vap;
2214 	struct componentname *cnp = ap->a_cnp;
2215 	struct nfsv2_sattr *sp;
2216 	u_int32_t *tl;
2217 	caddr_t cp;
2218 	int32_t t1, t2;
2219 	caddr_t bpos, dpos, cp2;
2220 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2221 	struct mbuf *mreq, *mrep, *md, *mb;
2222 	struct vnode *newvp = (struct vnode *)0;
2223 	const int v3 = NFS_ISV3(dvp);
2224 	int rexmit;
2225 	struct nfsnode *dnp = VTONFS(dvp);
2226 
2227 	*ap->a_vpp = NULL;
2228 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2229 	slen = strlen(ap->a_target);
2230 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2231 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2232 	nfsm_fhtom(dnp, v3);
2233 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2234 #ifndef NFS_V2_ONlY
2235 	if (v3)
2236 		nfsm_v3attrbuild(vap, FALSE);
2237 #endif
2238 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2239 #ifndef NFS_V2_ONlY
2240 	if (!v3) {
2241 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2242 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2243 		sp->sa_uid = nfs_xdrneg1;
2244 		sp->sa_gid = nfs_xdrneg1;
2245 		sp->sa_size = nfs_xdrneg1;
2246 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2247 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2248 	}
2249 #endif
2250 	nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred,
2251 	    &rexmit);
2252 #ifndef NFS_V2_ONlY
2253 	if (v3) {
2254 		if (!error)
2255 			nfsm_mtofh(dvp, newvp, v3, gotvp);
2256 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2257 	}
2258 #endif
2259 	nfsm_reqdone;
2260 	/*
2261 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2262 	 */
2263 	if (rexmit && error == EEXIST)
2264 		error = 0;
2265 	if (error == 0 || error == EEXIST)
2266 		cache_purge1(dvp, cnp, 0);
2267 	if (error == 0 && newvp == NULL) {
2268 		struct nfsnode *np = NULL;
2269 
2270 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2271 		    cnp->cn_cred, cnp->cn_proc, &np);
2272 		if (error == 0)
2273 			newvp = NFSTOV(np);
2274 	}
2275 	if (error) {
2276 		if (newvp != NULL)
2277 			vput(newvp);
2278 	} else {
2279 		*ap->a_vpp = newvp;
2280 	}
2281 	PNBUF_PUT(cnp->cn_pnbuf);
2282 	VTONFS(dvp)->n_flag |= NMODIFIED;
2283 	if (!wccflag)
2284 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2285 	VN_KNOTE(dvp, NOTE_WRITE);
2286 	vput(dvp);
2287 	return (error);
2288 }
2289 
2290 /*
2291  * nfs make dir call
2292  */
2293 int
2294 nfs_mkdir(v)
2295 	void *v;
2296 {
2297 	struct vop_mkdir_args /* {
2298 		struct vnode *a_dvp;
2299 		struct vnode **a_vpp;
2300 		struct componentname *a_cnp;
2301 		struct vattr *a_vap;
2302 	} */ *ap = v;
2303 	struct vnode *dvp = ap->a_dvp;
2304 	struct vattr *vap = ap->a_vap;
2305 	struct componentname *cnp = ap->a_cnp;
2306 	struct nfsv2_sattr *sp;
2307 	u_int32_t *tl;
2308 	caddr_t cp;
2309 	int32_t t1, t2;
2310 	int len;
2311 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2312 	struct vnode *newvp = (struct vnode *)0;
2313 	caddr_t bpos, dpos, cp2;
2314 	int error = 0, wccflag = NFSV3_WCCRATTR;
2315 	int gotvp = 0;
2316 	int rexmit;
2317 	struct mbuf *mreq, *mrep, *md, *mb;
2318 	const int v3 = NFS_ISV3(dvp);
2319 
2320 	len = cnp->cn_namelen;
2321 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2322 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
2323 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2324 	nfsm_fhtom(dnp, v3);
2325 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2326 #ifndef NFS_V2_ONLY
2327 	if (v3) {
2328 		nfsm_v3attrbuild(vap, FALSE);
2329 	} else
2330 #endif
2331 	{
2332 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2333 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2334 		sp->sa_uid = nfs_xdrneg1;
2335 		sp->sa_gid = nfs_xdrneg1;
2336 		sp->sa_size = nfs_xdrneg1;
2337 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2338 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2339 	}
2340 	nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred, &rexmit);
2341 	if (!error)
2342 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2343 	if (v3)
2344 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2345 	nfsm_reqdone;
2346 	VTONFS(dvp)->n_flag |= NMODIFIED;
2347 	if (!wccflag)
2348 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2349 	/*
2350 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2351 	 * if we can succeed in looking up the directory.
2352 	 */
2353 	if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2354 		if (newvp) {
2355 			vput(newvp);
2356 			newvp = (struct vnode *)0;
2357 		}
2358 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2359 			cnp->cn_proc, &np);
2360 		if (!error) {
2361 			newvp = NFSTOV(np);
2362 			if (newvp->v_type != VDIR || newvp == dvp)
2363 				error = EEXIST;
2364 		}
2365 	}
2366 	if (error) {
2367 		if (newvp) {
2368 			if (dvp != newvp)
2369 				vput(newvp);
2370 			else
2371 				vrele(newvp);
2372 		}
2373 	} else {
2374 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2375 		if (cnp->cn_flags & MAKEENTRY)
2376 			nfs_cache_enter(dvp, newvp, cnp);
2377 		*ap->a_vpp = newvp;
2378 	}
2379 	PNBUF_PUT(cnp->cn_pnbuf);
2380 	vput(dvp);
2381 	return (error);
2382 }
2383 
2384 /*
2385  * nfs remove directory call
2386  */
2387 int
2388 nfs_rmdir(v)
2389 	void *v;
2390 {
2391 	struct vop_rmdir_args /* {
2392 		struct vnode *a_dvp;
2393 		struct vnode *a_vp;
2394 		struct componentname *a_cnp;
2395 	} */ *ap = v;
2396 	struct vnode *vp = ap->a_vp;
2397 	struct vnode *dvp = ap->a_dvp;
2398 	struct componentname *cnp = ap->a_cnp;
2399 	u_int32_t *tl;
2400 	caddr_t cp;
2401 #ifndef NFS_V2_ONLY
2402 	int32_t t1;
2403 	caddr_t cp2;
2404 #endif
2405 	int32_t t2;
2406 	caddr_t bpos, dpos;
2407 	int error = 0, wccflag = NFSV3_WCCRATTR;
2408 	int rexmit;
2409 	struct mbuf *mreq, *mrep, *md, *mb;
2410 	const int v3 = NFS_ISV3(dvp);
2411 	struct nfsnode *dnp;
2412 
2413 	if (dvp == vp) {
2414 		vrele(dvp);
2415 		vput(dvp);
2416 		PNBUF_PUT(cnp->cn_pnbuf);
2417 		return (EINVAL);
2418 	}
2419 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2420 	dnp = VTONFS(dvp);
2421 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2422 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2423 	nfsm_fhtom(dnp, v3);
2424 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2425 	nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred, &rexmit);
2426 #ifndef NFS_V2_ONLY
2427 	if (v3)
2428 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2429 #endif
2430 	nfsm_reqdone;
2431 	PNBUF_PUT(cnp->cn_pnbuf);
2432 	VTONFS(dvp)->n_flag |= NMODIFIED;
2433 	if (!wccflag)
2434 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2435 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2436 	VN_KNOTE(vp, NOTE_DELETE);
2437 	cache_purge(vp);
2438 	vput(vp);
2439 	vput(dvp);
2440 	/*
2441 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2442 	 */
2443 	if (rexmit && error == ENOENT)
2444 		error = 0;
2445 	return (error);
2446 }
2447 
2448 /*
2449  * nfs readdir call
2450  */
2451 int
2452 nfs_readdir(v)
2453 	void *v;
2454 {
2455 	struct vop_readdir_args /* {
2456 		struct vnode *a_vp;
2457 		struct uio *a_uio;
2458 		struct ucred *a_cred;
2459 		int *a_eofflag;
2460 		off_t **a_cookies;
2461 		int *a_ncookies;
2462 	} */ *ap = v;
2463 	struct vnode *vp = ap->a_vp;
2464 	struct uio *uio = ap->a_uio;
2465 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2466 	char *base = uio->uio_iov->iov_base;
2467 	int tresid, error;
2468 	size_t count, lost;
2469 	struct dirent *dp;
2470 	off_t *cookies = NULL;
2471 	int ncookies = 0, nc;
2472 
2473 	if (vp->v_type != VDIR)
2474 		return (EPERM);
2475 
2476 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2477 	count = uio->uio_resid - lost;
2478 	if (count <= 0)
2479 		return (EINVAL);
2480 
2481 	/*
2482 	 * Call nfs_bioread() to do the real work.
2483 	 */
2484 	tresid = uio->uio_resid = count;
2485 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2486 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2487 
2488 	if (!error && ap->a_cookies) {
2489 		ncookies = count / 16;
2490 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2491 		*ap->a_cookies = cookies;
2492 	}
2493 
2494 	if (!error && uio->uio_resid == tresid) {
2495 		uio->uio_resid += lost;
2496 		nfsstats.direofcache_misses++;
2497 		if (ap->a_cookies)
2498 			*ap->a_ncookies = 0;
2499 		*ap->a_eofflag = 1;
2500 		return (0);
2501 	}
2502 
2503 	if (!error && ap->a_cookies) {
2504 		/*
2505 		 * Only the NFS server and emulations use cookies, and they
2506 		 * load the directory block into system space, so we can
2507 		 * just look at it directly.
2508 		 */
2509 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2510 			panic("nfs_readdir: lost in space");
2511 		for (nc = 0; ncookies-- &&
2512 		     base < (char *)uio->uio_iov->iov_base; nc++){
2513 			dp = (struct dirent *) base;
2514 			if (dp->d_reclen == 0)
2515 				break;
2516 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2517 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2518 			else
2519 				*(cookies++) = NFS_GETCOOKIE(dp);
2520 			base += dp->d_reclen;
2521 		}
2522 		uio->uio_resid +=
2523 		    ((caddr_t)uio->uio_iov->iov_base - base);
2524 		uio->uio_iov->iov_len +=
2525 		    ((caddr_t)uio->uio_iov->iov_base - base);
2526 		uio->uio_iov->iov_base = base;
2527 		*ap->a_ncookies = nc;
2528 	}
2529 
2530 	uio->uio_resid += lost;
2531 	*ap->a_eofflag = 0;
2532 	return (error);
2533 }
2534 
2535 /*
2536  * Readdir rpc call.
2537  * Called from below the buffer cache by nfs_doio().
2538  */
2539 int
2540 nfs_readdirrpc(vp, uiop, cred)
2541 	struct vnode *vp;
2542 	struct uio *uiop;
2543 	struct ucred *cred;
2544 {
2545 	int len, left;
2546 	struct dirent *dp = NULL;
2547 	u_int32_t *tl;
2548 	caddr_t cp;
2549 	int32_t t1, t2;
2550 	caddr_t bpos, dpos, cp2;
2551 	struct mbuf *mreq, *mrep, *md, *mb;
2552 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2553 	struct nfsnode *dnp = VTONFS(vp);
2554 	u_quad_t fileno;
2555 	int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2556 #ifndef NFS_V2_ONLY
2557 	int attrflag;
2558 #endif
2559 	int nrpcs = 0, reclen;
2560 	const int v3 = NFS_ISV3(vp);
2561 
2562 #ifdef DIAGNOSTIC
2563 	/*
2564 	 * Should be called from buffer cache, so only amount of
2565 	 * NFS_DIRBLKSIZ will be requested.
2566 	 */
2567 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2568 		panic("nfs readdirrpc bad uio");
2569 #endif
2570 
2571 	/*
2572 	 * Loop around doing readdir rpc's of size nm_readdirsize
2573 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2574 	 * The stopping criteria is EOF or buffer full.
2575 	 */
2576 	while (more_dirs && bigenough) {
2577 		/*
2578 		 * Heuristic: don't bother to do another RPC to further
2579 		 * fill up this block if there is not much room left. (< 50%
2580 		 * of the readdir RPC size). This wastes some buffer space
2581 		 * but can save up to 50% in RPC calls.
2582 		 */
2583 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2584 			bigenough = 0;
2585 			break;
2586 		}
2587 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2588 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2589 			NFSX_READDIR(v3));
2590 		nfsm_fhtom(dnp, v3);
2591 #ifndef NFS_V2_ONLY
2592 		if (v3) {
2593 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2594 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2595 				txdr_swapcookie3(uiop->uio_offset, tl);
2596 			} else {
2597 				txdr_cookie3(uiop->uio_offset, tl);
2598 			}
2599 			tl += 2;
2600 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2601 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2602 		} else
2603 #endif
2604 		{
2605 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2606 			*tl++ = txdr_unsigned(uiop->uio_offset);
2607 		}
2608 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2609 		nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2610 		nrpcs++;
2611 #ifndef NFS_V2_ONLY
2612 		if (v3) {
2613 			nfsm_postop_attr(vp, attrflag, 0);
2614 			if (!error) {
2615 				nfsm_dissect(tl, u_int32_t *,
2616 				    2 * NFSX_UNSIGNED);
2617 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2618 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2619 			} else {
2620 				m_freem(mrep);
2621 				goto nfsmout;
2622 			}
2623 		}
2624 #endif
2625 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2626 		more_dirs = fxdr_unsigned(int, *tl);
2627 
2628 		/* loop thru the dir entries, doctoring them to 4bsd form */
2629 		while (more_dirs && bigenough) {
2630 #ifndef NFS_V2_ONLY
2631 			if (v3) {
2632 				nfsm_dissect(tl, u_int32_t *,
2633 				    3 * NFSX_UNSIGNED);
2634 				fileno = fxdr_hyper(tl);
2635 				len = fxdr_unsigned(int, *(tl + 2));
2636 			} else
2637 #endif
2638 			{
2639 				nfsm_dissect(tl, u_int32_t *,
2640 				    2 * NFSX_UNSIGNED);
2641 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2642 				len = fxdr_unsigned(int, *tl);
2643 			}
2644 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2645 				error = EBADRPC;
2646 				m_freem(mrep);
2647 				goto nfsmout;
2648 			}
2649 			/* for cookie stashing */
2650 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2651 			left = NFS_DIRFRAGSIZ - blksiz;
2652 			if (reclen > left) {
2653 				memset(uiop->uio_iov->iov_base, 0, left);
2654 				dp->d_reclen += left;
2655 				UIO_ADVANCE(uiop, left);
2656 				blksiz = 0;
2657 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2658 			}
2659 			if (reclen > uiop->uio_resid)
2660 				bigenough = 0;
2661 			if (bigenough) {
2662 				int tlen;
2663 
2664 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2665 				dp->d_fileno = fileno;
2666 				dp->d_namlen = len;
2667 				dp->d_reclen = reclen;
2668 				dp->d_type = DT_UNKNOWN;
2669 				blksiz += reclen;
2670 				if (blksiz == NFS_DIRFRAGSIZ)
2671 					blksiz = 0;
2672 				UIO_ADVANCE(uiop, DIRHDSIZ);
2673 				nfsm_mtouio(uiop, len);
2674 				tlen = reclen - (DIRHDSIZ + len);
2675 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2676 				UIO_ADVANCE(uiop, tlen);
2677 			} else
2678 				nfsm_adv(nfsm_rndup(len));
2679 #ifndef NFS_V2_ONLY
2680 			if (v3) {
2681 				nfsm_dissect(tl, u_int32_t *,
2682 				    3 * NFSX_UNSIGNED);
2683 			} else
2684 #endif
2685 			{
2686 				nfsm_dissect(tl, u_int32_t *,
2687 				    2 * NFSX_UNSIGNED);
2688 			}
2689 			if (bigenough) {
2690 #ifndef NFS_V2_ONLY
2691 				if (v3) {
2692 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2693 						uiop->uio_offset =
2694 						    fxdr_swapcookie3(tl);
2695 					else
2696 						uiop->uio_offset =
2697 						    fxdr_cookie3(tl);
2698 				}
2699 				else
2700 #endif
2701 				{
2702 					uiop->uio_offset =
2703 					    fxdr_unsigned(off_t, *tl);
2704 				}
2705 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2706 			}
2707 			if (v3)
2708 				tl += 2;
2709 			else
2710 				tl++;
2711 			more_dirs = fxdr_unsigned(int, *tl);
2712 		}
2713 		/*
2714 		 * If at end of rpc data, get the eof boolean
2715 		 */
2716 		if (!more_dirs) {
2717 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2718 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2719 
2720 			/*
2721 			 * kludge: if we got no entries, treat it as EOF.
2722 			 * some server sometimes send a reply without any
2723 			 * entries or EOF.
2724 			 * although it might mean the server has very long name,
2725 			 * we can't handle such entries anyway.
2726 			 */
2727 
2728 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2729 				more_dirs = 0;
2730 		}
2731 		m_freem(mrep);
2732 	}
2733 	/*
2734 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2735 	 * by increasing d_reclen for the last record.
2736 	 */
2737 	if (blksiz > 0) {
2738 		left = NFS_DIRFRAGSIZ - blksiz;
2739 		memset(uiop->uio_iov->iov_base, 0, left);
2740 		dp->d_reclen += left;
2741 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2742 		UIO_ADVANCE(uiop, left);
2743 	}
2744 
2745 	/*
2746 	 * We are now either at the end of the directory or have filled the
2747 	 * block.
2748 	 */
2749 	if (bigenough) {
2750 		dnp->n_direofoffset = uiop->uio_offset;
2751 		dnp->n_flag |= NEOFVALID;
2752 	}
2753 nfsmout:
2754 	return (error);
2755 }
2756 
2757 #ifndef NFS_V2_ONLY
2758 /*
2759  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2760  */
2761 int
2762 nfs_readdirplusrpc(vp, uiop, cred)
2763 	struct vnode *vp;
2764 	struct uio *uiop;
2765 	struct ucred *cred;
2766 {
2767 	int len, left;
2768 	struct dirent *dp = NULL;
2769 	u_int32_t *tl;
2770 	caddr_t cp;
2771 	int32_t t1, t2;
2772 	struct vnode *newvp;
2773 	caddr_t bpos, dpos, cp2;
2774 	struct mbuf *mreq, *mrep, *md, *mb;
2775 	struct nameidata nami, *ndp = &nami;
2776 	struct componentname *cnp = &ndp->ni_cnd;
2777 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2778 	struct nfsnode *dnp = VTONFS(vp), *np;
2779 	nfsfh_t *fhp;
2780 	u_quad_t fileno;
2781 	int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2782 	int attrflag, fhsize, nrpcs = 0, reclen;
2783 	struct nfs_fattr fattr, *fp;
2784 
2785 #ifdef DIAGNOSTIC
2786 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2787 		panic("nfs readdirplusrpc bad uio");
2788 #endif
2789 	ndp->ni_dvp = vp;
2790 	newvp = NULLVP;
2791 
2792 	/*
2793 	 * Loop around doing readdir rpc's of size nm_readdirsize
2794 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2795 	 * The stopping criteria is EOF or buffer full.
2796 	 */
2797 	while (more_dirs && bigenough) {
2798 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2799 			bigenough = 0;
2800 			break;
2801 		}
2802 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2803 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2804 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2805 		nfsm_fhtom(dnp, 1);
2806  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2807 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2808 			txdr_swapcookie3(uiop->uio_offset, tl);
2809 		} else {
2810 			txdr_cookie3(uiop->uio_offset, tl);
2811 		}
2812 		tl += 2;
2813 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2814 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2815 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2816 		*tl = txdr_unsigned(nmp->nm_rsize);
2817 		nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2818 		nfsm_postop_attr(vp, attrflag, 0);
2819 		if (error) {
2820 			m_freem(mrep);
2821 			goto nfsmout;
2822 		}
2823 		nrpcs++;
2824 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2825 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2826 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2827 		more_dirs = fxdr_unsigned(int, *tl);
2828 
2829 		/* loop thru the dir entries, doctoring them to 4bsd form */
2830 		while (more_dirs && bigenough) {
2831 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2832 			fileno = fxdr_hyper(tl);
2833 			len = fxdr_unsigned(int, *(tl + 2));
2834 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2835 				error = EBADRPC;
2836 				m_freem(mrep);
2837 				goto nfsmout;
2838 			}
2839 			/* for cookie stashing */
2840 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2841 			left = NFS_DIRFRAGSIZ - blksiz;
2842 			if (reclen > left) {
2843 				/*
2844 				 * DIRFRAGSIZ is aligned, no need to align
2845 				 * again here.
2846 				 */
2847 				memset(uiop->uio_iov->iov_base, 0, left);
2848 				dp->d_reclen += left;
2849 				UIO_ADVANCE(uiop, left);
2850 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2851 				blksiz = 0;
2852 			}
2853 			if (reclen > uiop->uio_resid)
2854 				bigenough = 0;
2855 			if (bigenough) {
2856 				int tlen;
2857 
2858 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2859 				dp->d_fileno = fileno;
2860 				dp->d_namlen = len;
2861 				dp->d_reclen = reclen;
2862 				dp->d_type = DT_UNKNOWN;
2863 				blksiz += reclen;
2864 				if (blksiz == NFS_DIRFRAGSIZ)
2865 					blksiz = 0;
2866 				UIO_ADVANCE(uiop, DIRHDSIZ);
2867 				nfsm_mtouio(uiop, len);
2868 				tlen = reclen - (DIRHDSIZ + len);
2869 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2870 				UIO_ADVANCE(uiop, tlen);
2871 				cnp->cn_nameptr = dp->d_name;
2872 				cnp->cn_namelen = dp->d_namlen;
2873 			} else
2874 				nfsm_adv(nfsm_rndup(len));
2875 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2876 			if (bigenough) {
2877 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2878 					uiop->uio_offset =
2879 						fxdr_swapcookie3(tl);
2880 				else
2881 					uiop->uio_offset =
2882 						fxdr_cookie3(tl);
2883 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2884 			}
2885 			tl += 2;
2886 
2887 			/*
2888 			 * Since the attributes are before the file handle
2889 			 * (sigh), we must skip over the attributes and then
2890 			 * come back and get them.
2891 			 */
2892 			attrflag = fxdr_unsigned(int, *tl);
2893 			if (attrflag) {
2894 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2895 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2896 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2897 			    doit = fxdr_unsigned(int, *tl);
2898 			    if (doit) {
2899 				nfsm_getfh(fhp, fhsize, 1);
2900 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2901 				    VREF(vp);
2902 				    newvp = vp;
2903 				    np = dnp;
2904 				} else {
2905 				    error = nfs_nget1(vp->v_mount, fhp,
2906 					fhsize, &np, LK_NOWAIT);
2907 				    if (!error)
2908 					newvp = NFSTOV(np);
2909 				}
2910 				if (!error) {
2911 				    const char *xcp;
2912 
2913 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2914 				    if (bigenough) {
2915 					dp->d_type =
2916 					   IFTODT(VTTOIF(np->n_vattr->va_type));
2917 					if (cnp->cn_namelen <= NCHNAMLEN) {
2918 					    ndp->ni_vp = newvp;
2919 					    xcp = cnp->cn_nameptr +
2920 						cnp->cn_namelen;
2921 					    cnp->cn_hash =
2922 					       namei_hash(cnp->cn_nameptr, &xcp);
2923 					    nfs_cache_enter(ndp->ni_dvp,
2924 						ndp->ni_vp, cnp);
2925 					}
2926 				    }
2927 				}
2928 				error = 0;
2929 			   }
2930 			} else {
2931 			    /* Just skip over the file handle */
2932 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2933 			    i = fxdr_unsigned(int, *tl);
2934 			    nfsm_adv(nfsm_rndup(i));
2935 			}
2936 			if (newvp != NULLVP) {
2937 			    if (newvp == vp)
2938 				vrele(newvp);
2939 			    else
2940 				vput(newvp);
2941 			    newvp = NULLVP;
2942 			}
2943 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2944 			more_dirs = fxdr_unsigned(int, *tl);
2945 		}
2946 		/*
2947 		 * If at end of rpc data, get the eof boolean
2948 		 */
2949 		if (!more_dirs) {
2950 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2951 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2952 
2953 			/*
2954 			 * kludge: see a comment in nfs_readdirrpc.
2955 			 */
2956 
2957 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2958 				more_dirs = 0;
2959 		}
2960 		m_freem(mrep);
2961 	}
2962 	/*
2963 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2964 	 * by increasing d_reclen for the last record.
2965 	 */
2966 	if (blksiz > 0) {
2967 		left = NFS_DIRFRAGSIZ - blksiz;
2968 		memset(uiop->uio_iov->iov_base, 0, left);
2969 		dp->d_reclen += left;
2970 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2971 		UIO_ADVANCE(uiop, left);
2972 	}
2973 
2974 	/*
2975 	 * We are now either at the end of the directory or have filled the
2976 	 * block.
2977 	 */
2978 	if (bigenough) {
2979 		dnp->n_direofoffset = uiop->uio_offset;
2980 		dnp->n_flag |= NEOFVALID;
2981 	}
2982 nfsmout:
2983 	if (newvp != NULLVP) {
2984 		if(newvp == vp)
2985 		    vrele(newvp);
2986 		else
2987 		    vput(newvp);
2988 	}
2989 	return (error);
2990 }
2991 #endif
2992 
2993 /*
2994  * Silly rename. To make the NFS filesystem that is stateless look a little
2995  * more like the "ufs" a remove of an active vnode is translated to a rename
2996  * to a funny looking filename that is removed by nfs_inactive on the
2997  * nfsnode. There is the potential for another process on a different client
2998  * to create the same funny name between the nfs_lookitup() fails and the
2999  * nfs_rename() completes, but...
3000  */
3001 int
3002 nfs_sillyrename(dvp, vp, cnp)
3003 	struct vnode *dvp, *vp;
3004 	struct componentname *cnp;
3005 {
3006 	struct sillyrename *sp;
3007 	struct nfsnode *np;
3008 	int error;
3009 	short pid;
3010 
3011 	cache_purge(dvp);
3012 	np = VTONFS(vp);
3013 #ifndef DIAGNOSTIC
3014 	if (vp->v_type == VDIR)
3015 		panic("nfs: sillyrename dir");
3016 #endif
3017 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
3018 		M_NFSREQ, M_WAITOK);
3019 	sp->s_cred = crdup(cnp->cn_cred);
3020 	sp->s_dvp = dvp;
3021 	VREF(dvp);
3022 
3023 	/* Fudge together a funny name */
3024 	pid = cnp->cn_proc->p_pid;
3025 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
3026 	sp->s_namlen = 12;
3027 	sp->s_name[8] = hexdigits[pid & 0xf];
3028 	sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
3029 	sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
3030 	sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
3031 
3032 	/* Try lookitups until we get one that isn't there */
3033 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3034 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
3035 		sp->s_name[4]++;
3036 		if (sp->s_name[4] > 'z') {
3037 			error = EINVAL;
3038 			goto bad;
3039 		}
3040 	}
3041 	error = nfs_renameit(dvp, cnp, sp);
3042 	if (error)
3043 		goto bad;
3044 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3045 		cnp->cn_proc, &np);
3046 	np->n_sillyrename = sp;
3047 	return (0);
3048 bad:
3049 	vrele(sp->s_dvp);
3050 	crfree(sp->s_cred);
3051 	free((caddr_t)sp, M_NFSREQ);
3052 	return (error);
3053 }
3054 
3055 /*
3056  * Look up a file name and optionally either update the file handle or
3057  * allocate an nfsnode, depending on the value of npp.
3058  * npp == NULL	--> just do the lookup
3059  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3060  *			handled too
3061  * *npp != NULL --> update the file handle in the vnode
3062  */
3063 int
3064 nfs_lookitup(dvp, name, len, cred, procp, npp)
3065 	struct vnode *dvp;
3066 	const char *name;
3067 	int len;
3068 	struct ucred *cred;
3069 	struct proc *procp;
3070 	struct nfsnode **npp;
3071 {
3072 	u_int32_t *tl;
3073 	caddr_t cp;
3074 	int32_t t1, t2;
3075 	struct vnode *newvp = (struct vnode *)0;
3076 	struct nfsnode *np, *dnp = VTONFS(dvp);
3077 	caddr_t bpos, dpos, cp2;
3078 	int error = 0, fhlen;
3079 #ifndef NFS_V2_ONLY
3080 	int attrflag;
3081 #endif
3082 	struct mbuf *mreq, *mrep, *md, *mb;
3083 	nfsfh_t *nfhp;
3084 	const int v3 = NFS_ISV3(dvp);
3085 
3086 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3087 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3088 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3089 	nfsm_fhtom(dnp, v3);
3090 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
3091 	nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
3092 	if (npp && !error) {
3093 		nfsm_getfh(nfhp, fhlen, v3);
3094 		if (*npp) {
3095 		    np = *npp;
3096 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3097 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
3098 			np->n_fhp = &np->n_fh;
3099 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
3100 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
3101 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
3102 		    np->n_fhsize = fhlen;
3103 		    newvp = NFSTOV(np);
3104 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3105 		    VREF(dvp);
3106 		    newvp = dvp;
3107 		    np = dnp;
3108 		} else {
3109 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3110 		    if (error) {
3111 			m_freem(mrep);
3112 			return (error);
3113 		    }
3114 		    newvp = NFSTOV(np);
3115 		}
3116 #ifndef NFS_V2_ONLY
3117 		if (v3) {
3118 			nfsm_postop_attr(newvp, attrflag, 0);
3119 			if (!attrflag && *npp == NULL) {
3120 				m_freem(mrep);
3121 				vput(newvp);
3122 				return (ENOENT);
3123 			}
3124 		} else
3125 #endif
3126 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
3127 	}
3128 	nfsm_reqdone;
3129 	if (npp && *npp == NULL) {
3130 		if (error) {
3131 			if (newvp)
3132 				vput(newvp);
3133 		} else
3134 			*npp = np;
3135 	}
3136 	return (error);
3137 }
3138 
3139 #ifndef NFS_V2_ONLY
3140 /*
3141  * Nfs Version 3 commit rpc
3142  */
3143 int
3144 nfs_commit(vp, offset, cnt, procp)
3145 	struct vnode *vp;
3146 	off_t offset;
3147 	uint32_t cnt;
3148 	struct proc *procp;
3149 {
3150 	caddr_t cp;
3151 	u_int32_t *tl;
3152 	int32_t t1, t2;
3153 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3154 	caddr_t bpos, dpos, cp2;
3155 	int error = 0, wccflag = NFSV3_WCCRATTR;
3156 	struct mbuf *mreq, *mrep, *md, *mb;
3157 	struct nfsnode *np;
3158 
3159 	KASSERT(NFS_ISV3(vp));
3160 
3161 #ifdef NFS_DEBUG_COMMIT
3162 	printf("commit %lu - %lu\n", (unsigned long)offset,
3163 	    (unsigned long)(offset + cnt));
3164 #endif
3165 
3166 	simple_lock(&nmp->nm_slock);
3167 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3168 		simple_unlock(&nmp->nm_slock);
3169 		return (0);
3170 	}
3171 	simple_unlock(&nmp->nm_slock);
3172 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3173 	np = VTONFS(vp);
3174 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3175 	nfsm_fhtom(np, 1);
3176 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3177 	txdr_hyper(offset, tl);
3178 	tl += 2;
3179 	*tl = txdr_unsigned(cnt);
3180 	nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred);
3181 	nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, FALSE);
3182 	if (!error) {
3183 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3184 		simple_lock(&nmp->nm_slock);
3185 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3186 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3187 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3188 			error = NFSERR_STALEWRITEVERF;
3189 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3190 		}
3191 		simple_unlock(&nmp->nm_slock);
3192 	}
3193 	nfsm_reqdone;
3194 	return (error);
3195 }
3196 #endif
3197 
3198 /*
3199  * Kludge City..
3200  * - make nfs_bmap() essentially a no-op that does no translation
3201  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3202  *   (Maybe I could use the process's page mapping, but I was concerned that
3203  *    Kernel Write might not be enabled and also figured copyout() would do
3204  *    a lot more work than memcpy() and also it currently happens in the
3205  *    context of the swapper process (2).
3206  */
3207 int
3208 nfs_bmap(v)
3209 	void *v;
3210 {
3211 	struct vop_bmap_args /* {
3212 		struct vnode *a_vp;
3213 		daddr_t  a_bn;
3214 		struct vnode **a_vpp;
3215 		daddr_t *a_bnp;
3216 		int *a_runp;
3217 	} */ *ap = v;
3218 	struct vnode *vp = ap->a_vp;
3219 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3220 
3221 	if (ap->a_vpp != NULL)
3222 		*ap->a_vpp = vp;
3223 	if (ap->a_bnp != NULL)
3224 		*ap->a_bnp = ap->a_bn << bshift;
3225 	if (ap->a_runp != NULL)
3226 		*ap->a_runp = 1024 * 1024; /* XXX */
3227 	return (0);
3228 }
3229 
3230 /*
3231  * Strategy routine.
3232  * For async requests when nfsiod(s) are running, queue the request by
3233  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3234  * request.
3235  */
3236 int
3237 nfs_strategy(v)
3238 	void *v;
3239 {
3240 	struct vop_strategy_args *ap = v;
3241 	struct buf *bp = ap->a_bp;
3242 	int error = 0;
3243 
3244 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3245 		panic("nfs physio/async");
3246 
3247 	/*
3248 	 * If the op is asynchronous and an i/o daemon is waiting
3249 	 * queue the request, wake it up and wait for completion
3250 	 * otherwise just do it ourselves.
3251 	 */
3252 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3253 		error = nfs_doio(bp);
3254 	return (error);
3255 }
3256 
3257 /*
3258  * fsync vnode op. Just call nfs_flush() with commit == 1.
3259  */
3260 /* ARGSUSED */
3261 int
3262 nfs_fsync(v)
3263 	void *v;
3264 {
3265 	struct vop_fsync_args /* {
3266 		struct vnodeop_desc *a_desc;
3267 		struct vnode * a_vp;
3268 		struct ucred * a_cred;
3269 		int  a_flags;
3270 		off_t offlo;
3271 		off_t offhi;
3272 		struct proc * a_p;
3273 	} */ *ap = v;
3274 
3275 	struct vnode *vp = ap->a_vp;
3276 
3277 	if (vp->v_type != VREG)
3278 		return 0;
3279 
3280 	return (nfs_flush(vp, ap->a_cred,
3281 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3282 }
3283 
3284 /*
3285  * Flush all the data associated with a vnode.
3286  */
3287 int
3288 nfs_flush(vp, cred, waitfor, p, commit)
3289 	struct vnode *vp;
3290 	struct ucred *cred;
3291 	int waitfor;
3292 	struct proc *p;
3293 	int commit;
3294 {
3295 	struct nfsnode *np = VTONFS(vp);
3296 	int error;
3297 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3298 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3299 
3300 	simple_lock(&vp->v_interlock);
3301 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3302 	if (np->n_flag & NWRITEERR) {
3303 		error = np->n_error;
3304 		np->n_flag &= ~NWRITEERR;
3305 	}
3306 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3307 	return (error);
3308 }
3309 
3310 /*
3311  * Return POSIX pathconf information applicable to nfs.
3312  *
3313  * N.B. The NFS V2 protocol doesn't support this RPC.
3314  */
3315 /* ARGSUSED */
3316 int
3317 nfs_pathconf(v)
3318 	void *v;
3319 {
3320 	struct vop_pathconf_args /* {
3321 		struct vnode *a_vp;
3322 		int a_name;
3323 		register_t *a_retval;
3324 	} */ *ap = v;
3325 	struct nfsv3_pathconf *pcp;
3326 	struct vnode *vp = ap->a_vp;
3327 	struct mbuf *mreq, *mrep, *md, *mb;
3328 	int32_t t1, t2;
3329 	u_int32_t *tl;
3330 	caddr_t bpos, dpos, cp, cp2;
3331 	int error = 0, attrflag;
3332 #ifndef NFS_V2_ONLY
3333 	struct nfsmount *nmp;
3334 	unsigned int l;
3335 	u_int64_t maxsize;
3336 #endif
3337 	const int v3 = NFS_ISV3(vp);
3338 	struct nfsnode *np = VTONFS(vp);
3339 
3340 	switch (ap->a_name) {
3341 		/* Names that can be resolved locally. */
3342 	case _PC_PIPE_BUF:
3343 		*ap->a_retval = PIPE_BUF;
3344 		break;
3345 	case _PC_SYNC_IO:
3346 		*ap->a_retval = 1;
3347 		break;
3348 	/* Names that cannot be resolved locally; do an RPC, if possible. */
3349 	case _PC_LINK_MAX:
3350 	case _PC_NAME_MAX:
3351 	case _PC_CHOWN_RESTRICTED:
3352 	case _PC_NO_TRUNC:
3353 		if (!v3) {
3354 			error = EINVAL;
3355 			break;
3356 		}
3357 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3358 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3359 		nfsm_fhtom(np, 1);
3360 		nfsm_request(np, NFSPROC_PATHCONF,
3361 		    curproc, curproc->p_ucred);	/* XXX */
3362 		nfsm_postop_attr(vp, attrflag, 0);
3363 		if (!error) {
3364 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
3365 			    NFSX_V3PATHCONF);
3366 			switch (ap->a_name) {
3367 			case _PC_LINK_MAX:
3368 				*ap->a_retval =
3369 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3370 				break;
3371 			case _PC_NAME_MAX:
3372 				*ap->a_retval =
3373 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3374 				break;
3375 			case _PC_CHOWN_RESTRICTED:
3376 				*ap->a_retval =
3377 				    (pcp->pc_chownrestricted == nfs_true);
3378 				break;
3379 			case _PC_NO_TRUNC:
3380 				*ap->a_retval =
3381 				    (pcp->pc_notrunc == nfs_true);
3382 				break;
3383 			}
3384 		}
3385 		nfsm_reqdone;
3386 		break;
3387 	case _PC_FILESIZEBITS:
3388 #ifndef NFS_V2_ONLY
3389 		if (v3) {
3390 			nmp = VFSTONFS(vp->v_mount);
3391 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3392 				if ((error = nfs_fsinfo(nmp, vp,
3393 				    curproc->p_ucred, curproc)) != 0) /* XXX */
3394 					break;
3395 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3396 			    (maxsize >> l) > 0; l++)
3397 				;
3398 			*ap->a_retval = l + 1;
3399 		} else
3400 #endif
3401 		{
3402 			*ap->a_retval = 32;	/* NFS V2 limitation */
3403 		}
3404 		break;
3405 	default:
3406 		error = EINVAL;
3407 		break;
3408 	}
3409 
3410 	return (error);
3411 }
3412 
3413 /*
3414  * NFS advisory byte-level locks.
3415  */
3416 int
3417 nfs_advlock(v)
3418 	void *v;
3419 {
3420 	struct vop_advlock_args /* {
3421 		struct vnode *a_vp;
3422 		caddr_t  a_id;
3423 		int  a_op;
3424 		struct flock *a_fl;
3425 		int  a_flags;
3426 	} */ *ap = v;
3427 	struct nfsnode *np = VTONFS(ap->a_vp);
3428 
3429 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3430 }
3431 
3432 /*
3433  * Print out the contents of an nfsnode.
3434  */
3435 int
3436 nfs_print(v)
3437 	void *v;
3438 {
3439 	struct vop_print_args /* {
3440 		struct vnode *a_vp;
3441 	} */ *ap = v;
3442 	struct vnode *vp = ap->a_vp;
3443 	struct nfsnode *np = VTONFS(vp);
3444 
3445 	printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3446 	    (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3447 	if (vp->v_type == VFIFO)
3448 		fifo_printinfo(vp);
3449 	printf("\n");
3450 	return (0);
3451 }
3452 
3453 /*
3454  * nfs unlock wrapper.
3455  */
3456 int
3457 nfs_unlock(void *v)
3458 {
3459 	struct vop_unlock_args /* {
3460 		struct vnode *a_vp;
3461 		int a_flags;
3462 	} */ *ap = v;
3463 	struct vnode *vp = ap->a_vp;
3464 
3465 	/*
3466 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3467 	 * with v_data == 0.
3468 	 */
3469 	if (VTONFS(vp)) {
3470 		nfs_delayedtruncate(vp);
3471 	}
3472 
3473 	return genfs_unlock(v);
3474 }
3475 
3476 /*
3477  * nfs special file access vnode op.
3478  * Essentially just get vattr and then imitate iaccess() since the device is
3479  * local to the client.
3480  */
3481 int
3482 nfsspec_access(v)
3483 	void *v;
3484 {
3485 	struct vop_access_args /* {
3486 		struct vnode *a_vp;
3487 		int  a_mode;
3488 		struct ucred *a_cred;
3489 		struct proc *a_p;
3490 	} */ *ap = v;
3491 	struct vattr va;
3492 	struct vnode *vp = ap->a_vp;
3493 	int error;
3494 
3495 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3496 	if (error)
3497 		return (error);
3498 
3499         /*
3500 	 * Disallow write attempts on filesystems mounted read-only;
3501 	 * unless the file is a socket, fifo, or a block or character
3502 	 * device resident on the filesystem.
3503 	 */
3504 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3505 		switch (vp->v_type) {
3506 		case VREG:
3507 		case VDIR:
3508 		case VLNK:
3509 			return (EROFS);
3510 		default:
3511 			break;
3512 		}
3513 	}
3514 
3515 	return (vaccess(va.va_type, va.va_mode,
3516 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3517 }
3518 
3519 /*
3520  * Read wrapper for special devices.
3521  */
3522 int
3523 nfsspec_read(v)
3524 	void *v;
3525 {
3526 	struct vop_read_args /* {
3527 		struct vnode *a_vp;
3528 		struct uio *a_uio;
3529 		int  a_ioflag;
3530 		struct ucred *a_cred;
3531 	} */ *ap = v;
3532 	struct nfsnode *np = VTONFS(ap->a_vp);
3533 
3534 	/*
3535 	 * Set access flag.
3536 	 */
3537 	np->n_flag |= NACC;
3538 	np->n_atim.tv_sec = time.tv_sec;
3539 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3540 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3541 }
3542 
3543 /*
3544  * Write wrapper for special devices.
3545  */
3546 int
3547 nfsspec_write(v)
3548 	void *v;
3549 {
3550 	struct vop_write_args /* {
3551 		struct vnode *a_vp;
3552 		struct uio *a_uio;
3553 		int  a_ioflag;
3554 		struct ucred *a_cred;
3555 	} */ *ap = v;
3556 	struct nfsnode *np = VTONFS(ap->a_vp);
3557 
3558 	/*
3559 	 * Set update flag.
3560 	 */
3561 	np->n_flag |= NUPD;
3562 	np->n_mtim.tv_sec = time.tv_sec;
3563 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3564 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3565 }
3566 
3567 /*
3568  * Close wrapper for special devices.
3569  *
3570  * Update the times on the nfsnode then do device close.
3571  */
3572 int
3573 nfsspec_close(v)
3574 	void *v;
3575 {
3576 	struct vop_close_args /* {
3577 		struct vnode *a_vp;
3578 		int  a_fflag;
3579 		struct ucred *a_cred;
3580 		struct proc *a_p;
3581 	} */ *ap = v;
3582 	struct vnode *vp = ap->a_vp;
3583 	struct nfsnode *np = VTONFS(vp);
3584 	struct vattr vattr;
3585 
3586 	if (np->n_flag & (NACC | NUPD)) {
3587 		np->n_flag |= NCHG;
3588 		if (vp->v_usecount == 1 &&
3589 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3590 			VATTR_NULL(&vattr);
3591 			if (np->n_flag & NACC)
3592 				vattr.va_atime = np->n_atim;
3593 			if (np->n_flag & NUPD)
3594 				vattr.va_mtime = np->n_mtim;
3595 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3596 		}
3597 	}
3598 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3599 }
3600 
3601 /*
3602  * Read wrapper for fifos.
3603  */
3604 int
3605 nfsfifo_read(v)
3606 	void *v;
3607 {
3608 	struct vop_read_args /* {
3609 		struct vnode *a_vp;
3610 		struct uio *a_uio;
3611 		int  a_ioflag;
3612 		struct ucred *a_cred;
3613 	} */ *ap = v;
3614 	struct nfsnode *np = VTONFS(ap->a_vp);
3615 
3616 	/*
3617 	 * Set access flag.
3618 	 */
3619 	np->n_flag |= NACC;
3620 	np->n_atim.tv_sec = time.tv_sec;
3621 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3622 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3623 }
3624 
3625 /*
3626  * Write wrapper for fifos.
3627  */
3628 int
3629 nfsfifo_write(v)
3630 	void *v;
3631 {
3632 	struct vop_write_args /* {
3633 		struct vnode *a_vp;
3634 		struct uio *a_uio;
3635 		int  a_ioflag;
3636 		struct ucred *a_cred;
3637 	} */ *ap = v;
3638 	struct nfsnode *np = VTONFS(ap->a_vp);
3639 
3640 	/*
3641 	 * Set update flag.
3642 	 */
3643 	np->n_flag |= NUPD;
3644 	np->n_mtim.tv_sec = time.tv_sec;
3645 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3646 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3647 }
3648 
3649 /*
3650  * Close wrapper for fifos.
3651  *
3652  * Update the times on the nfsnode then do fifo close.
3653  */
3654 int
3655 nfsfifo_close(v)
3656 	void *v;
3657 {
3658 	struct vop_close_args /* {
3659 		struct vnode *a_vp;
3660 		int  a_fflag;
3661 		struct ucred *a_cred;
3662 		struct proc *a_p;
3663 	} */ *ap = v;
3664 	struct vnode *vp = ap->a_vp;
3665 	struct nfsnode *np = VTONFS(vp);
3666 	struct vattr vattr;
3667 
3668 	if (np->n_flag & (NACC | NUPD)) {
3669 		if (np->n_flag & NACC) {
3670 			np->n_atim.tv_sec = time.tv_sec;
3671 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3672 		}
3673 		if (np->n_flag & NUPD) {
3674 			np->n_mtim.tv_sec = time.tv_sec;
3675 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3676 		}
3677 		np->n_flag |= NCHG;
3678 		if (vp->v_usecount == 1 &&
3679 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3680 			VATTR_NULL(&vattr);
3681 			if (np->n_flag & NACC)
3682 				vattr.va_atime = np->n_atim;
3683 			if (np->n_flag & NUPD)
3684 				vattr.va_mtime = np->n_mtim;
3685 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3686 		}
3687 	}
3688 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3689 }
3690