xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 8a8f936f250a330d54f8a24ed0e92aadf9743a7b)
1 /*	$NetBSD: nfs_vnops.c,v 1.140 2001/09/22 22:35:19 sommerfeld 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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_vnops.c	8.19 (Berkeley) 7/31/95
39  */
40 
41 /*
42  * vnode op calls for Sun NFS version 2 and 3
43  */
44 
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/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/unistd.h>
65 
66 #include <uvm/uvm_extern.h>
67 #include <uvm/uvm.h>
68 
69 #include <miscfs/fifofs/fifo.h>
70 #include <miscfs/genfs/genfs.h>
71 #include <miscfs/specfs/specdev.h>
72 
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfs/nfs.h>
76 #include <nfs/nfsnode.h>
77 #include <nfs/nfsmount.h>
78 #include <nfs/xdr_subs.h>
79 #include <nfs/nfsm_subs.h>
80 #include <nfs/nqnfs.h>
81 #include <nfs/nfs_var.h>
82 
83 #include <net/if.h>
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86 
87 /* Defs */
88 #define	TRUE	1
89 #define	FALSE	0
90 
91 /*
92  * Global vfs data structures for nfs
93  */
94 int (**nfsv2_vnodeop_p) __P((void *));
95 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
96 	{ &vop_default_desc, vn_default_error },
97 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
98 	{ &vop_create_desc, nfs_create },		/* create */
99 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
100 	{ &vop_open_desc, nfs_open },			/* open */
101 	{ &vop_close_desc, nfs_close },			/* close */
102 	{ &vop_access_desc, nfs_access },		/* access */
103 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
104 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
105 	{ &vop_read_desc, nfs_read },			/* read */
106 	{ &vop_write_desc, nfs_write },			/* write */
107 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
108 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
109 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
110 	{ &vop_poll_desc, nfs_poll },			/* poll */
111 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
112 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
113 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
114 	{ &vop_seek_desc, nfs_seek },			/* seek */
115 	{ &vop_remove_desc, nfs_remove },		/* remove */
116 	{ &vop_link_desc, nfs_link },			/* link */
117 	{ &vop_rename_desc, nfs_rename },		/* rename */
118 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
119 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
120 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
121 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
122 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
123 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
124 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
125 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
126 	{ &vop_lock_desc, nfs_lock },			/* lock */
127 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
128 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
129 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
130 	{ &vop_print_desc, nfs_print },			/* print */
131 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
132 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
133 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
134 	{ &vop_blkatoff_desc, nfs_blkatoff },		/* blkatoff */
135 	{ &vop_valloc_desc, nfs_valloc },		/* valloc */
136 	{ &vop_reallocblks_desc, nfs_reallocblks },	/* reallocblks */
137 	{ &vop_vfree_desc, nfs_vfree },			/* vfree */
138 	{ &vop_truncate_desc, nfs_truncate },		/* truncate */
139 	{ &vop_update_desc, nfs_update },		/* update */
140 	{ &vop_bwrite_desc, nfs_bwrite },		/* bwrite */
141 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
142 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
143 	{ NULL, NULL }
144 };
145 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
146 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
147 
148 /*
149  * Special device vnode ops
150  */
151 int (**spec_nfsv2nodeop_p) __P((void *));
152 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
153 	{ &vop_default_desc, vn_default_error },
154 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
155 	{ &vop_create_desc, spec_create },		/* create */
156 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
157 	{ &vop_open_desc, spec_open },			/* open */
158 	{ &vop_close_desc, nfsspec_close },		/* close */
159 	{ &vop_access_desc, nfsspec_access },		/* access */
160 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
161 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
162 	{ &vop_read_desc, nfsspec_read },		/* read */
163 	{ &vop_write_desc, nfsspec_write },		/* write */
164 	{ &vop_lease_desc, spec_lease_check },		/* lease */
165 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
166 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
167 	{ &vop_poll_desc, spec_poll },			/* poll */
168 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
169 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
170 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
171 	{ &vop_seek_desc, spec_seek },			/* seek */
172 	{ &vop_remove_desc, spec_remove },		/* remove */
173 	{ &vop_link_desc, spec_link },			/* link */
174 	{ &vop_rename_desc, spec_rename },		/* rename */
175 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
176 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
177 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
178 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
179 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
180 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
181 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
182 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
183 	{ &vop_lock_desc, nfs_lock },			/* lock */
184 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
185 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
186 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
187 	{ &vop_print_desc, nfs_print },			/* print */
188 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
189 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
190 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
191 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
192 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
193 	{ &vop_reallocblks_desc, spec_reallocblks },	/* reallocblks */
194 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
195 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
196 	{ &vop_update_desc, nfs_update },		/* update */
197 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
198 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
199 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
200 	{ NULL, NULL }
201 };
202 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
203 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
204 
205 int (**fifo_nfsv2nodeop_p) __P((void *));
206 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
207 	{ &vop_default_desc, vn_default_error },
208 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
209 	{ &vop_create_desc, fifo_create },		/* create */
210 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
211 	{ &vop_open_desc, fifo_open },			/* open */
212 	{ &vop_close_desc, nfsfifo_close },		/* close */
213 	{ &vop_access_desc, nfsspec_access },		/* access */
214 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
215 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
216 	{ &vop_read_desc, nfsfifo_read },		/* read */
217 	{ &vop_write_desc, nfsfifo_write },		/* write */
218 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
219 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
220 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
221 	{ &vop_poll_desc, fifo_poll },			/* poll */
222 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
223 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
224 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
225 	{ &vop_seek_desc, fifo_seek },			/* seek */
226 	{ &vop_remove_desc, fifo_remove },		/* remove */
227 	{ &vop_link_desc, fifo_link },			/* link */
228 	{ &vop_rename_desc, fifo_rename },		/* rename */
229 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
230 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
231 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
232 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
233 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
234 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
235 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
236 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
237 	{ &vop_lock_desc, nfs_lock },			/* lock */
238 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
239 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
240 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
241 	{ &vop_print_desc, nfs_print },			/* print */
242 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
243 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
244 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
245 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
246 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
247 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
248 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
249 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
250 	{ &vop_update_desc, nfs_update },		/* update */
251 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
252 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
253 	{ NULL, NULL }
254 };
255 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
256 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
257 
258 /*
259  * Global variables
260  */
261 extern u_int32_t nfs_true, nfs_false;
262 extern u_int32_t nfs_xdrneg1;
263 extern struct nfsstats nfsstats;
264 extern nfstype nfsv3_type[9];
265 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
266 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
267 int nfs_numasync = 0;
268 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
269 
270 /*
271  * nfs null call from vfs.
272  */
273 int
274 nfs_null(vp, cred, procp)
275 	struct vnode *vp;
276 	struct ucred *cred;
277 	struct proc *procp;
278 {
279 	caddr_t bpos, dpos;
280 	int error = 0;
281 	struct mbuf *mreq, *mrep, *md, *mb;
282 
283 	nfsm_reqhead(vp, NFSPROC_NULL, 0);
284 	nfsm_request(vp, NFSPROC_NULL, procp, cred);
285 	nfsm_reqdone;
286 	return (error);
287 }
288 
289 /*
290  * nfs access vnode op.
291  * For nfs version 2, just return ok. File accesses may fail later.
292  * For nfs version 3, use the access rpc to check accessibility. If file modes
293  * are changed on the server, accesses might still fail later.
294  */
295 int
296 nfs_access(v)
297 	void *v;
298 {
299 	struct vop_access_args /* {
300 		struct vnode *a_vp;
301 		int  a_mode;
302 		struct ucred *a_cred;
303 		struct proc *a_p;
304 	} */ *ap = v;
305 	struct vnode *vp = ap->a_vp;
306 	u_int32_t *tl;
307 	caddr_t cp;
308 	int32_t t1, t2;
309 	caddr_t bpos, dpos, cp2;
310 	int error = 0, attrflag, cachevalid;
311 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
312 	u_int32_t mode, rmode;
313 	const int v3 = NFS_ISV3(vp);
314 	struct nfsnode *np = VTONFS(vp);
315 
316 	cachevalid = (np->n_accstamp != -1 &&
317 	    (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
318 	    np->n_accuid == ap->a_cred->cr_uid);
319 
320 	/*
321 	 * Check access cache first. If this request has been made for this
322 	 * uid shortly before, use the cached result.
323 	 */
324 	if (cachevalid) {
325 		if (!np->n_accerror) {
326 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
327 				return np->n_accerror;
328 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
329 			return np->n_accerror;
330 	}
331 
332 	/*
333 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
334 	 * ufs_access() locally using the vattr. This may not be correct,
335 	 * since the server may apply other access criteria such as
336 	 * client uid-->server uid mapping that we do not know about, but
337 	 * this is better than just returning anything that is lying about
338 	 * in the cache.
339 	 */
340 	if (v3) {
341 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
342 		nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
343 		nfsm_fhtom(vp, v3);
344 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
345 		if (ap->a_mode & VREAD)
346 			mode = NFSV3ACCESS_READ;
347 		else
348 			mode = 0;
349 		if (vp->v_type != VDIR) {
350 			if (ap->a_mode & VWRITE)
351 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
352 			if (ap->a_mode & VEXEC)
353 				mode |= NFSV3ACCESS_EXECUTE;
354 		} else {
355 			if (ap->a_mode & VWRITE)
356 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
357 					 NFSV3ACCESS_DELETE);
358 			if (ap->a_mode & VEXEC)
359 				mode |= NFSV3ACCESS_LOOKUP;
360 		}
361 		*tl = txdr_unsigned(mode);
362 		nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
363 		nfsm_postop_attr(vp, attrflag);
364 		if (!error) {
365 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
366 			rmode = fxdr_unsigned(u_int32_t, *tl);
367 			/*
368 			 * The NFS V3 spec does not clarify whether or not
369 			 * the returned access bits can be a superset of
370 			 * the ones requested, so...
371 			 */
372 			if ((rmode & mode) != mode)
373 				error = EACCES;
374 		}
375 		nfsm_reqdone;
376 	} else
377 		return (nfsspec_access(ap));
378 	/*
379 	 * Disallow write attempts on filesystems mounted read-only;
380 	 * unless the file is a socket, fifo, or a block or character
381 	 * device resident on the filesystem.
382 	 */
383 	if (!error && (ap->a_mode & VWRITE) &&
384 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
385 		switch (vp->v_type) {
386 		case VREG:
387 		case VDIR:
388 		case VLNK:
389 			error = EROFS;
390 		default:
391 			break;
392 		}
393 	}
394 
395 	if (!error || error == EACCES) {
396 		/*
397 		 * If we got the same result as for a previous,
398 		 * different request, OR it in. Don't update
399 		 * the timestamp in that case.
400 		 */
401 		if (cachevalid && error == np->n_accerror) {
402 			if (!error)
403 				np->n_accmode |= ap->a_mode;
404 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
405 				np->n_accmode = ap->a_mode;
406 		} else {
407 			np->n_accstamp = time.tv_sec;
408 			np->n_accuid = ap->a_cred->cr_uid;
409 			np->n_accmode = ap->a_mode;
410 			np->n_accerror = error;
411 		}
412 	}
413 
414 	return (error);
415 }
416 
417 /*
418  * nfs open vnode op
419  * Check to see if the type is ok
420  * and that deletion is not in progress.
421  * For paged in text files, you will need to flush the page cache
422  * if consistency is lost.
423  */
424 /* ARGSUSED */
425 int
426 nfs_open(v)
427 	void *v;
428 {
429 	struct vop_open_args /* {
430 		struct vnode *a_vp;
431 		int  a_mode;
432 		struct ucred *a_cred;
433 		struct proc *a_p;
434 	} */ *ap = v;
435 	struct vnode *vp = ap->a_vp;
436 	struct nfsnode *np = VTONFS(vp);
437 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
438 	struct vattr vattr;
439 	int error;
440 
441 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
442 		return (EACCES);
443 	}
444 
445 	/*
446 	 * Initialize read and write creds here, for swapfiles
447 	 * and other paths that don't set the creds themselves.
448 	 */
449 
450 	if (ap->a_mode & FREAD) {
451 		if (np->n_rcred) {
452 			crfree(np->n_rcred);
453 		}
454 		np->n_rcred = ap->a_cred;
455 		crhold(np->n_rcred);
456 	}
457 	if (ap->a_mode & FWRITE) {
458 		if (np->n_wcred) {
459 			crfree(np->n_wcred);
460 		}
461 		np->n_wcred = ap->a_cred;
462 		crhold(np->n_wcred);
463 	}
464 
465 #ifndef NFS_V2_ONLY
466 	/*
467 	 * Get a valid lease. If cached data is stale, flush it.
468 	 */
469 	if (nmp->nm_flag & NFSMNT_NQNFS) {
470 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
471 		    do {
472 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
473 			    ap->a_p);
474 		    } while (error == NQNFS_EXPIRED);
475 		    if (error)
476 			return (error);
477 		    if (np->n_lrev != np->n_brev ||
478 			(np->n_flag & NQNFSNONCACHE)) {
479 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
480 				ap->a_p, 1)) == EINTR)
481 				return (error);
482 			np->n_brev = np->n_lrev;
483 		    }
484 		}
485 	} else
486 #endif
487 	{
488 		if (np->n_flag & NMODIFIED) {
489 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
490 				ap->a_p, 1)) == EINTR)
491 				return (error);
492 			np->n_attrstamp = 0;
493 			if (vp->v_type == VDIR) {
494 				nfs_invaldircache(vp, 0);
495 				np->n_direofoffset = 0;
496 			}
497 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
498 			if (error)
499 				return (error);
500 			np->n_mtime = vattr.va_mtime.tv_sec;
501 		} else {
502 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
503 			if (error)
504 				return (error);
505 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
506 				if (vp->v_type == VDIR) {
507 					nfs_invaldircache(vp, 0);
508 					np->n_direofoffset = 0;
509 				}
510 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
511 					ap->a_cred, ap->a_p, 1)) == EINTR)
512 					return (error);
513 				np->n_mtime = vattr.va_mtime.tv_sec;
514 			}
515 		}
516 	}
517 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
518 		np->n_attrstamp = 0; /* For Open/Close consistency */
519 	return (0);
520 }
521 
522 /*
523  * nfs close vnode op
524  * What an NFS client should do upon close after writing is a debatable issue.
525  * Most NFS clients push delayed writes to the server upon close, basically for
526  * two reasons:
527  * 1 - So that any write errors may be reported back to the client process
528  *     doing the close system call. By far the two most likely errors are
529  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
530  * 2 - To put a worst case upper bound on cache inconsistency between
531  *     multiple clients for the file.
532  * There is also a consistency problem for Version 2 of the protocol w.r.t.
533  * not being able to tell if other clients are writing a file concurrently,
534  * since there is no way of knowing if the changed modify time in the reply
535  * is only due to the write for this client.
536  * (NFS Version 3 provides weak cache consistency data in the reply that
537  *  should be sufficient to detect and handle this case.)
538  *
539  * The current code does the following:
540  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
541  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
542  *                     or commit them (this satisfies 1 and 2 except for the
543  *                     case where the server crashes after this close but
544  *                     before the commit RPC, which is felt to be "good
545  *                     enough". Changing the last argument to nfs_flush() to
546  *                     a 1 would force a commit operation, if it is felt a
547  *                     commit is necessary now.
548  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
549  *                     1 should be dealt with via an fsync() system call for
550  *                     cases where write errors are important.
551  */
552 /* ARGSUSED */
553 int
554 nfs_close(v)
555 	void *v;
556 {
557 	struct vop_close_args /* {
558 		struct vnodeop_desc *a_desc;
559 		struct vnode *a_vp;
560 		int  a_fflag;
561 		struct ucred *a_cred;
562 		struct proc *a_p;
563 	} */ *ap = v;
564 	struct vnode *vp = ap->a_vp;
565 	struct nfsnode *np = VTONFS(vp);
566 	int error = 0;
567 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
568 
569 	if (vp->v_type == VREG) {
570 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
571 		(np->n_flag & NMODIFIED)) {
572 		if (NFS_ISV3(vp)) {
573 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
574 		    np->n_flag &= ~NMODIFIED;
575 		} else
576 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
577 		np->n_attrstamp = 0;
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, *mb2;
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 	 * First look in the cache.
618 	 */
619 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
620 		return (0);
621 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
622 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
623 	nfsm_fhtom(vp, v3);
624 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
625 	if (!error) {
626 		nfsm_loadattr(vp, ap->a_vap);
627 		if (vp->v_type == VDIR &&
628 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
629 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
630 	}
631 	nfsm_reqdone;
632 	return (error);
633 }
634 
635 /*
636  * nfs setattr call.
637  */
638 int
639 nfs_setattr(v)
640 	void *v;
641 {
642 	struct vop_setattr_args /* {
643 		struct vnodeop_desc *a_desc;
644 		struct vnode *a_vp;
645 		struct vattr *a_vap;
646 		struct ucred *a_cred;
647 		struct proc *a_p;
648 	} */ *ap = v;
649 	struct vnode *vp = ap->a_vp;
650 	struct nfsnode *np = VTONFS(vp);
651 	struct vattr *vap = ap->a_vap;
652 	int error = 0;
653 	u_quad_t tsize = 0;
654 
655 	/*
656 	 * Setting of flags is not supported.
657 	 */
658 	if (vap->va_flags != VNOVAL)
659 		return (EOPNOTSUPP);
660 
661 	/*
662 	 * Disallow write attempts if the filesystem is mounted read-only.
663 	 */
664   	if ((vap->va_uid != (uid_t)VNOVAL ||
665 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
666 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
667 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
668 		return (EROFS);
669 	if (vap->va_size != VNOVAL) {
670  		switch (vp->v_type) {
671  		case VDIR:
672  			return (EISDIR);
673  		case VCHR:
674  		case VBLK:
675  		case VSOCK:
676  		case VFIFO:
677 			if (vap->va_mtime.tv_sec == VNOVAL &&
678 			    vap->va_atime.tv_sec == VNOVAL &&
679 			    vap->va_mode == (mode_t)VNOVAL &&
680 			    vap->va_uid == (uid_t)VNOVAL &&
681 			    vap->va_gid == (gid_t)VNOVAL)
682 				return (0);
683  			vap->va_size = VNOVAL;
684  			break;
685  		default:
686 			/*
687 			 * Disallow write attempts if the filesystem is
688 			 * mounted read-only.
689 			 */
690 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
691 				return (EROFS);
692  			uvm_vnp_setsize(vp, vap->va_size);
693  			if (vap->va_size == 0)
694  				error = nfs_vinvalbuf(vp, 0,
695  				     ap->a_cred, ap->a_p, 1);
696 			else
697 				error = nfs_vinvalbuf(vp, V_SAVE,
698 				     ap->a_cred, ap->a_p, 1);
699 			if (error) {
700 				uvm_vnp_setsize(vp, np->n_size);
701 				return (error);
702 			}
703  			tsize = np->n_size;
704  			np->n_size = np->n_vattr->va_size = vap->va_size;
705   		}
706   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
707 		vap->va_atime.tv_sec != VNOVAL) &&
708 		vp->v_type == VREG &&
709   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
710 		 ap->a_p, 1)) == EINTR)
711 		return (error);
712 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
713 	if (error && vap->va_size != VNOVAL) {
714 		np->n_size = np->n_vattr->va_size = tsize;
715 		uvm_vnp_setsize(vp, np->n_size);
716 	}
717 	return (error);
718 }
719 
720 /*
721  * Do an nfs setattr rpc.
722  */
723 int
724 nfs_setattrrpc(vp, vap, cred, procp)
725 	struct vnode *vp;
726 	struct vattr *vap;
727 	struct ucred *cred;
728 	struct proc *procp;
729 {
730 	struct nfsv2_sattr *sp;
731 	caddr_t cp;
732 	int32_t t1, t2;
733 	caddr_t bpos, dpos, cp2;
734 	u_int32_t *tl;
735 	int error = 0, wccflag = NFSV3_WCCRATTR;
736 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
737 	const int v3 = NFS_ISV3(vp);
738 
739 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
740 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
741 	nfsm_fhtom(vp, v3);
742 	if (v3) {
743 		nfsm_v3attrbuild(vap, TRUE);
744 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
745 		*tl = nfs_false;
746 	} else {
747 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
748 		if (vap->va_mode == (mode_t)VNOVAL)
749 			sp->sa_mode = nfs_xdrneg1;
750 		else
751 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
752 		if (vap->va_uid == (uid_t)VNOVAL)
753 			sp->sa_uid = nfs_xdrneg1;
754 		else
755 			sp->sa_uid = txdr_unsigned(vap->va_uid);
756 		if (vap->va_gid == (gid_t)VNOVAL)
757 			sp->sa_gid = nfs_xdrneg1;
758 		else
759 			sp->sa_gid = txdr_unsigned(vap->va_gid);
760 		sp->sa_size = txdr_unsigned(vap->va_size);
761 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
762 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
763 	}
764 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
765 	if (v3) {
766 		nfsm_wcc_data(vp, wccflag);
767 	} else
768 		nfsm_loadattr(vp, (struct vattr *)0);
769 	nfsm_reqdone;
770 	return (error);
771 }
772 
773 /*
774  * nfs lookup call, one step at a time...
775  * First look in cache
776  * If not found, unlock the directory nfsnode and do the rpc
777  *
778  * This code is full of lock/unlock statements and checks, because
779  * we continue after cache_lookup has finished (we need to check
780  * with the attr cache and do an rpc if it has timed out). This means
781  * that the locking effects of cache_lookup have to be taken into
782  * account.
783  */
784 int
785 nfs_lookup(v)
786 	void *v;
787 {
788 	struct vop_lookup_args /* {
789 		struct vnodeop_desc *a_desc;
790 		struct vnode *a_dvp;
791 		struct vnode **a_vpp;
792 		struct componentname *a_cnp;
793 	} */ *ap = v;
794 	struct componentname *cnp = ap->a_cnp;
795 	struct vnode *dvp = ap->a_dvp;
796 	struct vnode **vpp = ap->a_vpp;
797 	int flags;
798 	struct vnode *newvp;
799 	u_int32_t *tl;
800 	caddr_t cp;
801 	int32_t t1, t2;
802 	struct nfsmount *nmp;
803 	caddr_t bpos, dpos, cp2;
804 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
805 	long len;
806 	nfsfh_t *fhp;
807 	struct nfsnode *np;
808 	int lockparent, wantparent, error = 0, attrflag, fhsize;
809 	const int v3 = NFS_ISV3(dvp);
810 
811 	cnp->cn_flags &= ~PDIRUNLOCK;
812 	flags = cnp->cn_flags;
813 
814 	*vpp = NULLVP;
815 	newvp = NULLVP;
816 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
817 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
818 		return (EROFS);
819 	if (dvp->v_type != VDIR)
820 		return (ENOTDIR);
821 
822 	lockparent = flags & LOCKPARENT;
823 	wantparent = flags & (LOCKPARENT|WANTPARENT);
824 	nmp = VFSTONFS(dvp->v_mount);
825 	np = VTONFS(dvp);
826 
827 	/*
828 	 * Before tediously performing a linear scan of the directory,
829 	 * check the name cache to see if the directory/name pair
830 	 * we are looking for is known already.
831 	 * If the directory/name pair is found in the name cache,
832 	 * we have to ensure the directory has not changed from
833 	 * the time the cache entry has been created. If it has,
834 	 * the cache entry has to be ignored
835 	 */
836 	if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
837 		struct vattr vattr;
838 		int err2;
839 
840 		if (error && error != ENOENT) {
841 			*vpp = NULLVP;
842 			return error;
843 		}
844 
845 		if (cnp->cn_flags & PDIRUNLOCK) {
846 			err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
847 			if (err2 != 0) {
848 				*vpp = NULLVP;
849 				return err2;
850 			}
851 			cnp->cn_flags &= ~PDIRUNLOCK;
852 		}
853 
854 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
855 		if (err2 != 0) {
856 			if (error == 0) {
857 				if (*vpp != dvp)
858 					vput(*vpp);
859 				else
860 					vrele(*vpp);
861 			}
862 			*vpp = NULLVP;
863 			return err2;
864 		}
865 
866 		if (error == ENOENT) {
867 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
868 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
869 			    VTONFS(dvp)->n_nctime)
870 				return ENOENT;
871 			cache_purge(dvp);
872 			np->n_nctime = 0;
873 			goto dorpc;
874 		}
875 
876 		newvp = *vpp;
877 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
878 			&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
879 		{
880 			nfsstats.lookupcache_hits++;
881 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
882 				cnp->cn_flags |= SAVENAME;
883 			if ((!lockparent || !(flags & ISLASTCN)) &&
884 			     newvp != dvp)
885 				VOP_UNLOCK(dvp, 0);
886 			return (0);
887 		}
888 		cache_purge(newvp);
889 		if (newvp != dvp)
890 			vput(newvp);
891 		else
892 			vrele(newvp);
893 		*vpp = NULLVP;
894 	}
895 dorpc:
896 	error = 0;
897 	newvp = NULLVP;
898 	nfsstats.lookupcache_misses++;
899 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
900 	len = cnp->cn_namelen;
901 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
902 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
903 	nfsm_fhtom(dvp, v3);
904 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
905 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
906 	if (error) {
907 		nfsm_postop_attr(dvp, attrflag);
908 		m_freem(mrep);
909 		goto nfsmout;
910 	}
911 	nfsm_getfh(fhp, fhsize, v3);
912 
913 	/*
914 	 * Handle RENAME case...
915 	 */
916 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
917 		if (NFS_CMPFH(np, fhp, fhsize)) {
918 			m_freem(mrep);
919 			return (EISDIR);
920 		}
921 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
922 		if (error) {
923 			m_freem(mrep);
924 			return error;
925 		}
926 		newvp = NFSTOV(np);
927 		if (v3) {
928 			nfsm_postop_attr(newvp, attrflag);
929 			nfsm_postop_attr(dvp, attrflag);
930 		} else
931 			nfsm_loadattr(newvp, (struct vattr *)0);
932 		*vpp = newvp;
933 		m_freem(mrep);
934 		cnp->cn_flags |= SAVENAME;
935 		if (!lockparent) {
936 			VOP_UNLOCK(dvp, 0);
937 			cnp->cn_flags |= PDIRUNLOCK;
938 		}
939 		return (0);
940 	}
941 
942 	/*
943 	 * The postop attr handling is duplicated for each if case,
944 	 * because it should be done while dvp is locked (unlocking
945 	 * dvp is different for each case).
946 	 */
947 
948 	if (NFS_CMPFH(np, fhp, fhsize)) {
949 		/*
950 		 * "." lookup
951 		 */
952 		VREF(dvp);
953 		newvp = dvp;
954 		if (v3) {
955 			nfsm_postop_attr(newvp, attrflag);
956 			nfsm_postop_attr(dvp, attrflag);
957 		} else
958 			nfsm_loadattr(newvp, (struct vattr *)0);
959 	} else if (flags & ISDOTDOT) {
960 		/*
961 		 * ".." lookup
962 		 */
963 		VOP_UNLOCK(dvp, 0);
964 		cnp->cn_flags |= PDIRUNLOCK;
965 
966 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
967 		if (error) {
968 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
969 				cnp->cn_flags &= ~PDIRUNLOCK;
970 			m_freem(mrep);
971 			return error;
972 		}
973 		newvp = NFSTOV(np);
974 
975 		if (v3) {
976 			nfsm_postop_attr(newvp, attrflag);
977 			nfsm_postop_attr(dvp, attrflag);
978 		} else
979 			nfsm_loadattr(newvp, (struct vattr *)0);
980 
981 		if (lockparent && (flags & ISLASTCN)) {
982 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
983 				m_freem(mrep);
984 				vput(newvp);
985 				return error;
986 			}
987 			cnp->cn_flags &= ~PDIRUNLOCK;
988 		}
989 	} else {
990 		/*
991 		 * Other lookups.
992 		 */
993 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
994 		if (error) {
995 			m_freem(mrep);
996 			return error;
997 		}
998 		newvp = NFSTOV(np);
999 		if (v3) {
1000 			nfsm_postop_attr(newvp, attrflag);
1001 			nfsm_postop_attr(dvp, attrflag);
1002 		} else
1003 			nfsm_loadattr(newvp, (struct vattr *)0);
1004 		if (!lockparent || !(flags & ISLASTCN)) {
1005 			VOP_UNLOCK(dvp, 0);
1006 			cnp->cn_flags |= PDIRUNLOCK;
1007 		}
1008 	}
1009 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1010 		cnp->cn_flags |= SAVENAME;
1011 	if ((cnp->cn_flags & MAKEENTRY) &&
1012 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1013 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1014 		cache_enter(dvp, newvp, cnp);
1015 	}
1016 	*vpp = newvp;
1017 	nfsm_reqdone;
1018 	if (error) {
1019 		/*
1020 		 * We get here only because of errors returned by
1021 		 * the RPC. Otherwise we'll have returned above
1022 		 * (the nfsm_* macros will jump to nfsm_reqdone
1023 		 * on error).
1024 		 */
1025 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1026 		    cnp->cn_nameiop != CREATE) {
1027 			if (VTONFS(dvp)->n_nctime == 0)
1028 				VTONFS(dvp)->n_nctime =
1029 				    VTONFS(dvp)->n_vattr->va_mtime.tv_sec;
1030 			cache_enter(dvp, NULL, cnp);
1031 		}
1032 		if (newvp != NULLVP) {
1033 			vrele(newvp);
1034 			if (newvp != dvp)
1035 				VOP_UNLOCK(newvp, 0);
1036 		}
1037 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1038 		    (flags & ISLASTCN) && error == ENOENT) {
1039 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1040 				error = EROFS;
1041 			else
1042 				error = EJUSTRETURN;
1043 		}
1044 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1045 			cnp->cn_flags |= SAVENAME;
1046 		*vpp = NULL;
1047 	}
1048 
1049 	return error;
1050 }
1051 
1052 /*
1053  * nfs read call.
1054  * Just call nfs_bioread() to do the work.
1055  */
1056 int
1057 nfs_read(v)
1058 	void *v;
1059 {
1060 	struct vop_read_args /* {
1061 		struct vnode *a_vp;
1062 		struct uio *a_uio;
1063 		int  a_ioflag;
1064 		struct ucred *a_cred;
1065 	} */ *ap = v;
1066 	struct vnode *vp = ap->a_vp;
1067 
1068 	if (vp->v_type != VREG)
1069 		return (EPERM);
1070 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1071 }
1072 
1073 /*
1074  * nfs readlink call
1075  */
1076 int
1077 nfs_readlink(v)
1078 	void *v;
1079 {
1080 	struct vop_readlink_args /* {
1081 		struct vnode *a_vp;
1082 		struct uio *a_uio;
1083 		struct ucred *a_cred;
1084 	} */ *ap = v;
1085 	struct vnode *vp = ap->a_vp;
1086 
1087 	if (vp->v_type != VLNK)
1088 		return (EPERM);
1089 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1090 }
1091 
1092 /*
1093  * Do a readlink rpc.
1094  * Called by nfs_doio() from below the buffer cache.
1095  */
1096 int
1097 nfs_readlinkrpc(vp, uiop, cred)
1098 	struct vnode *vp;
1099 	struct uio *uiop;
1100 	struct ucred *cred;
1101 {
1102 	u_int32_t *tl;
1103 	caddr_t cp;
1104 	int32_t t1, t2;
1105 	caddr_t bpos, dpos, cp2;
1106 	int error = 0, len, attrflag;
1107 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1108 	const int v3 = NFS_ISV3(vp);
1109 
1110 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1111 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1112 	nfsm_fhtom(vp, v3);
1113 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
1114 	if (v3)
1115 		nfsm_postop_attr(vp, attrflag);
1116 	if (!error) {
1117 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1118 		nfsm_mtouio(uiop, len);
1119 	}
1120 	nfsm_reqdone;
1121 	return (error);
1122 }
1123 
1124 /*
1125  * nfs read rpc call
1126  * Ditto above
1127  */
1128 int
1129 nfs_readrpc(vp, uiop)
1130 	struct vnode *vp;
1131 	struct uio *uiop;
1132 {
1133 	u_int32_t *tl;
1134 	caddr_t cp;
1135 	int32_t t1, t2;
1136 	caddr_t bpos, dpos, cp2;
1137 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1138 	struct nfsmount *nmp;
1139 	int error = 0, len, retlen, tsiz, eof, attrflag;
1140 	const int v3 = NFS_ISV3(vp);
1141 
1142 #ifndef nolint
1143 	eof = 0;
1144 #endif
1145 	nmp = VFSTONFS(vp->v_mount);
1146 	tsiz = uiop->uio_resid;
1147 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1148 		return (EFBIG);
1149 	while (tsiz > 0) {
1150 		nfsstats.rpccnt[NFSPROC_READ]++;
1151 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1152 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1153 		nfsm_fhtom(vp, v3);
1154 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1155 		if (v3) {
1156 			txdr_hyper(uiop->uio_offset, tl);
1157 			*(tl + 2) = txdr_unsigned(len);
1158 		} else {
1159 			*tl++ = txdr_unsigned(uiop->uio_offset);
1160 			*tl++ = txdr_unsigned(len);
1161 			*tl = 0;
1162 		}
1163 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp,
1164 			     VTONFS(vp)->n_rcred);
1165 		if (v3) {
1166 			nfsm_postop_attr(vp, attrflag);
1167 			if (error) {
1168 				m_freem(mrep);
1169 				goto nfsmout;
1170 			}
1171 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1172 			eof = fxdr_unsigned(int, *(tl + 1));
1173 		} else
1174 			nfsm_loadattr(vp, (struct vattr *)0);
1175 		nfsm_strsiz(retlen, nmp->nm_rsize);
1176 		nfsm_mtouio(uiop, retlen);
1177 		m_freem(mrep);
1178 		tsiz -= retlen;
1179 		if (v3) {
1180 			if (eof || retlen == 0)
1181 				tsiz = 0;
1182 		} else if (retlen < len)
1183 			tsiz = 0;
1184 	}
1185 nfsmout:
1186 	return (error);
1187 }
1188 
1189 /*
1190  * nfs write call
1191  */
1192 int
1193 nfs_writerpc(vp, uiop, iomode, must_commit)
1194 	struct vnode *vp;
1195 	struct uio *uiop;
1196 	int *iomode, *must_commit;
1197 {
1198 	u_int32_t *tl;
1199 	caddr_t cp;
1200 	int32_t t1, t2, backup;
1201 	caddr_t bpos, dpos, cp2;
1202 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1203 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1204 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1205 	const int v3 = NFS_ISV3(vp);
1206 	int committed = NFSV3WRITE_FILESYNC;
1207 
1208 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1209 		panic("writerpc readonly vp %p", vp);
1210 	}
1211 
1212 #ifndef DIAGNOSTIC
1213 	if (uiop->uio_iovcnt != 1)
1214 		panic("nfs: writerpc iovcnt > 1");
1215 #endif
1216 	*must_commit = 0;
1217 	tsiz = uiop->uio_resid;
1218 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1219 		return (EFBIG);
1220 	while (tsiz > 0) {
1221 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1222 		len = min(tsiz, nmp->nm_wsize);
1223 		nfsm_reqhead(vp, NFSPROC_WRITE,
1224 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1225 		nfsm_fhtom(vp, v3);
1226 		if (v3) {
1227 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1228 			txdr_hyper(uiop->uio_offset, tl);
1229 			tl += 2;
1230 			*tl++ = txdr_unsigned(len);
1231 			*tl++ = txdr_unsigned(*iomode);
1232 			*tl = txdr_unsigned(len);
1233 		} else {
1234 			u_int32_t x;
1235 
1236 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1237 			/* Set both "begin" and "current" to non-garbage. */
1238 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1239 			*tl++ = x;      /* "begin offset" */
1240 			*tl++ = x;      /* "current offset" */
1241 			x = txdr_unsigned(len);
1242 			*tl++ = x;      /* total to this offset */
1243 			*tl = x;        /* size of this write */
1244 
1245 		}
1246 		nfsm_uiotom(uiop, len);
1247 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp,
1248 			     VTONFS(vp)->n_wcred);
1249 		if (v3) {
1250 			wccflag = NFSV3_WCCCHK;
1251 			nfsm_wcc_data(vp, wccflag);
1252 			if (!error) {
1253 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1254 					+ NFSX_V3WRITEVERF);
1255 				rlen = fxdr_unsigned(int, *tl++);
1256 				if (rlen == 0) {
1257 					error = NFSERR_IO;
1258 					m_freem(mrep);
1259 					break;
1260 				} else if (rlen < len) {
1261 					backup = len - rlen;
1262 					uiop->uio_iov->iov_base =
1263 					    (caddr_t)uiop->uio_iov->iov_base -
1264 					    backup;
1265 					uiop->uio_iov->iov_len += backup;
1266 					uiop->uio_offset -= backup;
1267 					uiop->uio_resid += backup;
1268 					len = rlen;
1269 				}
1270 				commit = fxdr_unsigned(int, *tl++);
1271 
1272 				/*
1273 				 * Return the lowest committment level
1274 				 * obtained by any of the RPCs.
1275 				 */
1276 				if (committed == NFSV3WRITE_FILESYNC)
1277 					committed = commit;
1278 				else if (committed == NFSV3WRITE_DATASYNC &&
1279 					commit == NFSV3WRITE_UNSTABLE)
1280 					committed = commit;
1281 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1282 				    memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
1283 					NFSX_V3WRITEVERF);
1284 				    nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1285 				} else if (memcmp((caddr_t)tl,
1286 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1287 				    *must_commit = 1;
1288 				    memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
1289 					NFSX_V3WRITEVERF);
1290 				}
1291 			}
1292 		} else
1293 		    nfsm_loadattr(vp, (struct vattr *)0);
1294 		if (wccflag)
1295 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1296 		m_freem(mrep);
1297 		if (error)
1298 			break;
1299 		tsiz -= len;
1300 	}
1301 nfsmout:
1302 	*iomode = committed;
1303 	if (error)
1304 		uiop->uio_resid = tsiz;
1305 	return (error);
1306 }
1307 
1308 /*
1309  * nfs mknod rpc
1310  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1311  * mode set to specify the file type and the size field for rdev.
1312  */
1313 int
1314 nfs_mknodrpc(dvp, vpp, cnp, vap)
1315 	struct vnode *dvp;
1316 	struct vnode **vpp;
1317 	struct componentname *cnp;
1318 	struct vattr *vap;
1319 {
1320 	struct nfsv2_sattr *sp;
1321 	u_int32_t *tl;
1322 	caddr_t cp;
1323 	int32_t t1, t2;
1324 	struct vnode *newvp = (struct vnode *)0;
1325 	struct nfsnode *np;
1326 	char *cp2;
1327 	caddr_t bpos, dpos;
1328 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1329 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1330 	u_int32_t rdev;
1331 	const int v3 = NFS_ISV3(dvp);
1332 
1333 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1334 		rdev = txdr_unsigned(vap->va_rdev);
1335 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1336 		rdev = nfs_xdrneg1;
1337 	else {
1338 		VOP_ABORTOP(dvp, cnp);
1339 		vput(dvp);
1340 		return (EOPNOTSUPP);
1341 	}
1342 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1343 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1344 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1345 	nfsm_fhtom(dvp, v3);
1346 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1347 	if (v3) {
1348 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1349 		*tl++ = vtonfsv3_type(vap->va_type);
1350 		nfsm_v3attrbuild(vap, FALSE);
1351 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1352 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1353 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1354 			*tl = txdr_unsigned(minor(vap->va_rdev));
1355 		}
1356 	} else {
1357 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1358 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1359 		sp->sa_uid = nfs_xdrneg1;
1360 		sp->sa_gid = nfs_xdrneg1;
1361 		sp->sa_size = rdev;
1362 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1363 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1364 	}
1365 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1366 	if (!error) {
1367 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1368 		if (!gotvp) {
1369 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1370 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1371 			if (!error)
1372 				newvp = NFSTOV(np);
1373 		}
1374 	}
1375 	if (v3)
1376 		nfsm_wcc_data(dvp, wccflag);
1377 	nfsm_reqdone;
1378 	if (error) {
1379 		if (newvp)
1380 			vput(newvp);
1381 	} else {
1382 		if (cnp->cn_flags & MAKEENTRY)
1383 			cache_enter(dvp, newvp, cnp);
1384 		*vpp = newvp;
1385 	}
1386 	PNBUF_PUT(cnp->cn_pnbuf);
1387 	VTONFS(dvp)->n_flag |= NMODIFIED;
1388 	if (!wccflag)
1389 		VTONFS(dvp)->n_attrstamp = 0;
1390 	vput(dvp);
1391 	return (error);
1392 }
1393 
1394 /*
1395  * nfs mknod vop
1396  * just call nfs_mknodrpc() to do the work.
1397  */
1398 /* ARGSUSED */
1399 int
1400 nfs_mknod(v)
1401 	void *v;
1402 {
1403 	struct vop_mknod_args /* {
1404 		struct vnode *a_dvp;
1405 		struct vnode **a_vpp;
1406 		struct componentname *a_cnp;
1407 		struct vattr *a_vap;
1408 	} */ *ap = v;
1409 	int error;
1410 
1411 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1412 	return (error);
1413 }
1414 
1415 static u_long create_verf;
1416 /*
1417  * nfs file create call
1418  */
1419 int
1420 nfs_create(v)
1421 	void *v;
1422 {
1423 	struct vop_create_args /* {
1424 		struct vnode *a_dvp;
1425 		struct vnode **a_vpp;
1426 		struct componentname *a_cnp;
1427 		struct vattr *a_vap;
1428 	} */ *ap = v;
1429 	struct vnode *dvp = ap->a_dvp;
1430 	struct vattr *vap = ap->a_vap;
1431 	struct componentname *cnp = ap->a_cnp;
1432 	struct nfsv2_sattr *sp;
1433 	u_int32_t *tl;
1434 	caddr_t cp;
1435 	int32_t t1, t2;
1436 	struct nfsnode *np = (struct nfsnode *)0;
1437 	struct vnode *newvp = (struct vnode *)0;
1438 	caddr_t bpos, dpos, cp2;
1439 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1440 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1441 	const int v3 = NFS_ISV3(dvp);
1442 
1443 	/*
1444 	 * Oops, not for me..
1445 	 */
1446 	if (vap->va_type == VSOCK)
1447 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1448 
1449 #ifdef VA_EXCLUSIVE
1450 	if (vap->va_vaflags & VA_EXCLUSIVE)
1451 		fmode |= O_EXCL;
1452 #endif
1453 again:
1454 	error = 0;
1455 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1456 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1457 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1458 	nfsm_fhtom(dvp, v3);
1459 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1460 	if (v3) {
1461 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1462 		if (fmode & O_EXCL) {
1463 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1464 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1465 #ifdef INET
1466 			if (in_ifaddr.tqh_first)
1467 				*tl++ = in_ifaddr.tqh_first->ia_addr.sin_addr.s_addr;
1468 			else
1469 				*tl++ = create_verf;
1470 #else
1471 			*tl++ = create_verf;
1472 #endif
1473 			*tl = ++create_verf;
1474 		} else {
1475 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1476 			nfsm_v3attrbuild(vap, FALSE);
1477 		}
1478 	} else {
1479 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1480 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1481 		sp->sa_uid = nfs_xdrneg1;
1482 		sp->sa_gid = nfs_xdrneg1;
1483 		sp->sa_size = 0;
1484 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1485 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1486 	}
1487 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1488 	if (!error) {
1489 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1490 		if (!gotvp) {
1491 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1492 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1493 			if (!error)
1494 				newvp = NFSTOV(np);
1495 		}
1496 	}
1497 	if (v3)
1498 		nfsm_wcc_data(dvp, wccflag);
1499 	nfsm_reqdone;
1500 	if (error) {
1501 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1502 			fmode &= ~O_EXCL;
1503 			goto again;
1504 		}
1505 		if (newvp)
1506 			vput(newvp);
1507 	} else if (v3 && (fmode & O_EXCL))
1508 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1509 	if (!error) {
1510 		if (cnp->cn_flags & MAKEENTRY)
1511 			cache_enter(dvp, newvp, cnp);
1512 		*ap->a_vpp = newvp;
1513 	}
1514 	PNBUF_PUT(cnp->cn_pnbuf);
1515 	VTONFS(dvp)->n_flag |= NMODIFIED;
1516 	if (!wccflag)
1517 		VTONFS(dvp)->n_attrstamp = 0;
1518 	vput(dvp);
1519 	return (error);
1520 }
1521 
1522 /*
1523  * nfs file remove call
1524  * To try and make nfs semantics closer to ufs semantics, a file that has
1525  * other processes using the vnode is renamed instead of removed and then
1526  * removed later on the last close.
1527  * - If v_usecount > 1
1528  *	  If a rename is not already in the works
1529  *	     call nfs_sillyrename() to set it up
1530  *     else
1531  *	  do the remove rpc
1532  */
1533 int
1534 nfs_remove(v)
1535 	void *v;
1536 {
1537 	struct vop_remove_args /* {
1538 		struct vnodeop_desc *a_desc;
1539 		struct vnode * a_dvp;
1540 		struct vnode * a_vp;
1541 		struct componentname * a_cnp;
1542 	} */ *ap = v;
1543 	struct vnode *vp = ap->a_vp;
1544 	struct vnode *dvp = ap->a_dvp;
1545 	struct componentname *cnp = ap->a_cnp;
1546 	struct nfsnode *np = VTONFS(vp);
1547 	int error = 0;
1548 	struct vattr vattr;
1549 
1550 #ifndef DIAGNOSTIC
1551 	if ((cnp->cn_flags & HASBUF) == 0)
1552 		panic("nfs_remove: no name");
1553 	if (vp->v_usecount < 1)
1554 		panic("nfs_remove: bad v_usecount");
1555 #endif
1556 	if (vp->v_type == VDIR)
1557 		error = EPERM;
1558 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1559 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1560 	    vattr.va_nlink > 1)) {
1561 		/*
1562 		 * Purge the name cache so that the chance of a lookup for
1563 		 * the name succeeding while the remove is in progress is
1564 		 * minimized. Without node locking it can still happen, such
1565 		 * that an I/O op returns ESTALE, but since you get this if
1566 		 * another host removes the file..
1567 		 */
1568 		cache_purge(vp);
1569 		/*
1570 		 * throw away biocache buffers, mainly to avoid
1571 		 * unnecessary delayed writes later.
1572 		 */
1573 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1574 		/* Do the rpc */
1575 		if (error != EINTR)
1576 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1577 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1578 		/*
1579 		 * Kludge City: If the first reply to the remove rpc is lost..
1580 		 *   the reply to the retransmitted request will be ENOENT
1581 		 *   since the file was in fact removed
1582 		 *   Therefore, we cheat and return success.
1583 		 */
1584 		if (error == ENOENT)
1585 			error = 0;
1586 	} else if (!np->n_sillyrename)
1587 		error = nfs_sillyrename(dvp, vp, cnp);
1588 	PNBUF_PUT(cnp->cn_pnbuf);
1589 	np->n_attrstamp = 0;
1590 	vput(dvp);
1591 	vput(vp);
1592 	return (error);
1593 }
1594 
1595 /*
1596  * nfs file remove rpc called from nfs_inactive
1597  */
1598 int
1599 nfs_removeit(sp)
1600 	struct sillyrename *sp;
1601 {
1602 
1603 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1604 		(struct proc *)0));
1605 }
1606 
1607 /*
1608  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1609  */
1610 int
1611 nfs_removerpc(dvp, name, namelen, cred, proc)
1612 	struct vnode *dvp;
1613 	const char *name;
1614 	int namelen;
1615 	struct ucred *cred;
1616 	struct proc *proc;
1617 {
1618 	u_int32_t *tl;
1619 	caddr_t cp;
1620 	int32_t t1, t2;
1621 	caddr_t bpos, dpos, cp2;
1622 	int error = 0, wccflag = NFSV3_WCCRATTR;
1623 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1624 	const int v3 = NFS_ISV3(dvp);
1625 
1626 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1627 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1628 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1629 	nfsm_fhtom(dvp, v3);
1630 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1631 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1632 	if (v3)
1633 		nfsm_wcc_data(dvp, wccflag);
1634 	nfsm_reqdone;
1635 	VTONFS(dvp)->n_flag |= NMODIFIED;
1636 	if (!wccflag)
1637 		VTONFS(dvp)->n_attrstamp = 0;
1638 	return (error);
1639 }
1640 
1641 /*
1642  * nfs file rename call
1643  */
1644 int
1645 nfs_rename(v)
1646 	void *v;
1647 {
1648 	struct vop_rename_args  /* {
1649 		struct vnode *a_fdvp;
1650 		struct vnode *a_fvp;
1651 		struct componentname *a_fcnp;
1652 		struct vnode *a_tdvp;
1653 		struct vnode *a_tvp;
1654 		struct componentname *a_tcnp;
1655 	} */ *ap = v;
1656 	struct vnode *fvp = ap->a_fvp;
1657 	struct vnode *tvp = ap->a_tvp;
1658 	struct vnode *fdvp = ap->a_fdvp;
1659 	struct vnode *tdvp = ap->a_tdvp;
1660 	struct componentname *tcnp = ap->a_tcnp;
1661 	struct componentname *fcnp = ap->a_fcnp;
1662 	int error;
1663 
1664 #ifndef DIAGNOSTIC
1665 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1666 	    (fcnp->cn_flags & HASBUF) == 0)
1667 		panic("nfs_rename: no name");
1668 #endif
1669 	/* Check for cross-device rename */
1670 	if ((fvp->v_mount != tdvp->v_mount) ||
1671 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1672 		error = EXDEV;
1673 		goto out;
1674 	}
1675 
1676 	/*
1677 	 * If the tvp exists and is in use, sillyrename it before doing the
1678 	 * rename of the new file over it.
1679 	 */
1680 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1681 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1682 		vput(tvp);
1683 		tvp = NULL;
1684 	}
1685 
1686 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1687 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1688 		tcnp->cn_proc);
1689 
1690 	if (fvp->v_type == VDIR) {
1691 		if (tvp != NULL && tvp->v_type == VDIR)
1692 			cache_purge(tdvp);
1693 		cache_purge(fdvp);
1694 	}
1695 out:
1696 	if (tdvp == tvp)
1697 		vrele(tdvp);
1698 	else
1699 		vput(tdvp);
1700 	if (tvp)
1701 		vput(tvp);
1702 	vrele(fdvp);
1703 	vrele(fvp);
1704 	/*
1705 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1706 	 */
1707 	if (error == ENOENT)
1708 		error = 0;
1709 	return (error);
1710 }
1711 
1712 /*
1713  * nfs file rename rpc called from nfs_remove() above
1714  */
1715 int
1716 nfs_renameit(sdvp, scnp, sp)
1717 	struct vnode *sdvp;
1718 	struct componentname *scnp;
1719 	struct sillyrename *sp;
1720 {
1721 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1722 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1723 }
1724 
1725 /*
1726  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1727  */
1728 int
1729 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1730 	struct vnode *fdvp;
1731 	const char *fnameptr;
1732 	int fnamelen;
1733 	struct vnode *tdvp;
1734 	const char *tnameptr;
1735 	int tnamelen;
1736 	struct ucred *cred;
1737 	struct proc *proc;
1738 {
1739 	u_int32_t *tl;
1740 	caddr_t cp;
1741 	int32_t t1, t2;
1742 	caddr_t bpos, dpos, cp2;
1743 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1744 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1745 	const int v3 = NFS_ISV3(fdvp);
1746 
1747 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1748 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1749 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1750 		nfsm_rndup(tnamelen));
1751 	nfsm_fhtom(fdvp, v3);
1752 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1753 	nfsm_fhtom(tdvp, v3);
1754 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1755 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1756 	if (v3) {
1757 		nfsm_wcc_data(fdvp, fwccflag);
1758 		nfsm_wcc_data(tdvp, twccflag);
1759 	}
1760 	nfsm_reqdone;
1761 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1762 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1763 	if (!fwccflag)
1764 		VTONFS(fdvp)->n_attrstamp = 0;
1765 	if (!twccflag)
1766 		VTONFS(tdvp)->n_attrstamp = 0;
1767 	return (error);
1768 }
1769 
1770 /*
1771  * nfs hard link create call
1772  */
1773 int
1774 nfs_link(v)
1775 	void *v;
1776 {
1777 	struct vop_link_args /* {
1778 		struct vnode *a_dvp;
1779 		struct vnode *a_vp;
1780 		struct componentname *a_cnp;
1781 	} */ *ap = v;
1782 	struct vnode *vp = ap->a_vp;
1783 	struct vnode *dvp = ap->a_dvp;
1784 	struct componentname *cnp = ap->a_cnp;
1785 	u_int32_t *tl;
1786 	caddr_t cp;
1787 	int32_t t1, t2;
1788 	caddr_t bpos, dpos, cp2;
1789 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1790 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1791 	/* XXX Should be const and initialised? */
1792 	int v3;
1793 
1794 	if (dvp->v_mount != vp->v_mount) {
1795 		VOP_ABORTOP(vp, cnp);
1796 		vput(dvp);
1797 		return (EXDEV);
1798 	}
1799 	if (dvp != vp) {
1800 		error = vn_lock(vp, LK_EXCLUSIVE);
1801 		if (error != 0) {
1802 			VOP_ABORTOP(dvp, cnp);
1803 			vput(dvp);
1804 			return error;
1805 		}
1806 	}
1807 
1808 	/*
1809 	 * Push all writes to the server, so that the attribute cache
1810 	 * doesn't get "out of sync" with the server.
1811 	 * XXX There should be a better way!
1812 	 */
1813 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1814 
1815 	v3 = NFS_ISV3(vp);
1816 	nfsstats.rpccnt[NFSPROC_LINK]++;
1817 	nfsm_reqhead(vp, NFSPROC_LINK,
1818 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1819 	nfsm_fhtom(vp, v3);
1820 	nfsm_fhtom(dvp, v3);
1821 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1822 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1823 	if (v3) {
1824 		nfsm_postop_attr(vp, attrflag);
1825 		nfsm_wcc_data(dvp, wccflag);
1826 	}
1827 	nfsm_reqdone;
1828 	PNBUF_PUT(cnp->cn_pnbuf);
1829 	VTONFS(dvp)->n_flag |= NMODIFIED;
1830 	if (!attrflag)
1831 		VTONFS(vp)->n_attrstamp = 0;
1832 	if (!wccflag)
1833 		VTONFS(dvp)->n_attrstamp = 0;
1834 	if (dvp != vp)
1835 		VOP_UNLOCK(vp, 0);
1836 	vput(dvp);
1837 	/*
1838 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1839 	 */
1840 	if (error == EEXIST)
1841 		error = 0;
1842 	return (error);
1843 }
1844 
1845 /*
1846  * nfs symbolic link create call
1847  */
1848 int
1849 nfs_symlink(v)
1850 	void *v;
1851 {
1852 	struct vop_symlink_args /* {
1853 		struct vnode *a_dvp;
1854 		struct vnode **a_vpp;
1855 		struct componentname *a_cnp;
1856 		struct vattr *a_vap;
1857 		char *a_target;
1858 	} */ *ap = v;
1859 	struct vnode *dvp = ap->a_dvp;
1860 	struct vattr *vap = ap->a_vap;
1861 	struct componentname *cnp = ap->a_cnp;
1862 	struct nfsv2_sattr *sp;
1863 	u_int32_t *tl;
1864 	caddr_t cp;
1865 	int32_t t1, t2;
1866 	caddr_t bpos, dpos, cp2;
1867 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1868 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1869 	struct vnode *newvp = (struct vnode *)0;
1870 	const int v3 = NFS_ISV3(dvp);
1871 
1872 	*ap->a_vpp = NULL;
1873 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1874 	slen = strlen(ap->a_target);
1875 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1876 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1877 	nfsm_fhtom(dvp, v3);
1878 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1879 	if (v3)
1880 		nfsm_v3attrbuild(vap, FALSE);
1881 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1882 	if (!v3) {
1883 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1884 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1885 		sp->sa_uid = nfs_xdrneg1;
1886 		sp->sa_gid = nfs_xdrneg1;
1887 		sp->sa_size = nfs_xdrneg1;
1888 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1889 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1890 	}
1891 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1892 	if (v3) {
1893 		if (!error)
1894 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1895 		nfsm_wcc_data(dvp, wccflag);
1896 	}
1897 	nfsm_reqdone;
1898 	/*
1899 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1900 	 */
1901 	if (error == EEXIST)
1902 		error = 0;
1903 	if (error == 0 && newvp == NULL) {
1904 		struct nfsnode *np = NULL;
1905 
1906 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1907 		    cnp->cn_cred, cnp->cn_proc, &np);
1908 		if (error == 0)
1909 			newvp = NFSTOV(np);
1910 	}
1911 	if (error) {
1912 		if (newvp != NULL)
1913 			vput(newvp);
1914 	} else {
1915 		*ap->a_vpp = newvp;
1916 	}
1917 	PNBUF_PUT(cnp->cn_pnbuf);
1918 	VTONFS(dvp)->n_flag |= NMODIFIED;
1919 	if (!wccflag)
1920 		VTONFS(dvp)->n_attrstamp = 0;
1921 	vput(dvp);
1922 	return (error);
1923 }
1924 
1925 /*
1926  * nfs make dir call
1927  */
1928 int
1929 nfs_mkdir(v)
1930 	void *v;
1931 {
1932 	struct vop_mkdir_args /* {
1933 		struct vnode *a_dvp;
1934 		struct vnode **a_vpp;
1935 		struct componentname *a_cnp;
1936 		struct vattr *a_vap;
1937 	} */ *ap = v;
1938 	struct vnode *dvp = ap->a_dvp;
1939 	struct vattr *vap = ap->a_vap;
1940 	struct componentname *cnp = ap->a_cnp;
1941 	struct nfsv2_sattr *sp;
1942 	u_int32_t *tl;
1943 	caddr_t cp;
1944 	int32_t t1, t2;
1945 	int len;
1946 	struct nfsnode *np = (struct nfsnode *)0;
1947 	struct vnode *newvp = (struct vnode *)0;
1948 	caddr_t bpos, dpos, cp2;
1949 	int error = 0, wccflag = NFSV3_WCCRATTR;
1950 	int gotvp = 0;
1951 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1952 	const int v3 = NFS_ISV3(dvp);
1953 
1954 	len = cnp->cn_namelen;
1955 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1956 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1957 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1958 	nfsm_fhtom(dvp, v3);
1959 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1960 	if (v3) {
1961 		nfsm_v3attrbuild(vap, FALSE);
1962 	} else {
1963 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1964 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1965 		sp->sa_uid = nfs_xdrneg1;
1966 		sp->sa_gid = nfs_xdrneg1;
1967 		sp->sa_size = nfs_xdrneg1;
1968 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1969 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1970 	}
1971 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1972 	if (!error)
1973 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1974 	if (v3)
1975 		nfsm_wcc_data(dvp, wccflag);
1976 	nfsm_reqdone;
1977 	VTONFS(dvp)->n_flag |= NMODIFIED;
1978 	if (!wccflag)
1979 		VTONFS(dvp)->n_attrstamp = 0;
1980 	/*
1981 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1982 	 * if we can succeed in looking up the directory.
1983 	 */
1984 	if (error == EEXIST || (!error && !gotvp)) {
1985 		if (newvp) {
1986 			vput(newvp);
1987 			newvp = (struct vnode *)0;
1988 		}
1989 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1990 			cnp->cn_proc, &np);
1991 		if (!error) {
1992 			newvp = NFSTOV(np);
1993 			if (newvp->v_type != VDIR)
1994 				error = EEXIST;
1995 		}
1996 	}
1997 	if (error) {
1998 		if (newvp)
1999 			vput(newvp);
2000 	} else {
2001 		if (cnp->cn_flags & MAKEENTRY)
2002 			cache_enter(dvp, newvp, cnp);
2003 		*ap->a_vpp = newvp;
2004 	}
2005 	PNBUF_PUT(cnp->cn_pnbuf);
2006 	vput(dvp);
2007 	return (error);
2008 }
2009 
2010 /*
2011  * nfs remove directory call
2012  */
2013 int
2014 nfs_rmdir(v)
2015 	void *v;
2016 {
2017 	struct vop_rmdir_args /* {
2018 		struct vnode *a_dvp;
2019 		struct vnode *a_vp;
2020 		struct componentname *a_cnp;
2021 	} */ *ap = v;
2022 	struct vnode *vp = ap->a_vp;
2023 	struct vnode *dvp = ap->a_dvp;
2024 	struct componentname *cnp = ap->a_cnp;
2025 	u_int32_t *tl;
2026 	caddr_t cp;
2027 	int32_t t1, t2;
2028 	caddr_t bpos, dpos, cp2;
2029 	int error = 0, wccflag = NFSV3_WCCRATTR;
2030 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2031 	const int v3 = NFS_ISV3(dvp);
2032 
2033 	if (dvp == vp) {
2034 		vrele(dvp);
2035 		vput(dvp);
2036 		PNBUF_PUT(cnp->cn_pnbuf);
2037 		return (EINVAL);
2038 	}
2039 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2040 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
2041 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2042 	nfsm_fhtom(dvp, v3);
2043 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2044 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2045 	if (v3)
2046 		nfsm_wcc_data(dvp, wccflag);
2047 	nfsm_reqdone;
2048 	PNBUF_PUT(cnp->cn_pnbuf);
2049 	VTONFS(dvp)->n_flag |= NMODIFIED;
2050 	if (!wccflag)
2051 		VTONFS(dvp)->n_attrstamp = 0;
2052 	cache_purge(dvp);
2053 	cache_purge(vp);
2054 	vput(vp);
2055 	vput(dvp);
2056 	/*
2057 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2058 	 */
2059 	if (error == ENOENT)
2060 		error = 0;
2061 	return (error);
2062 }
2063 
2064 /*
2065  * nfs readdir call
2066  */
2067 int
2068 nfs_readdir(v)
2069 	void *v;
2070 {
2071 	struct vop_readdir_args /* {
2072 		struct vnode *a_vp;
2073 		struct uio *a_uio;
2074 		struct ucred *a_cred;
2075 		int *a_eofflag;
2076 		off_t **a_cookies;
2077 		int *a_ncookies;
2078 	} */ *ap = v;
2079 	struct vnode *vp = ap->a_vp;
2080 	struct uio *uio = ap->a_uio;
2081 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2082 	char *base = uio->uio_iov->iov_base;
2083 	int tresid, error;
2084 	size_t count, lost;
2085 	struct dirent *dp;
2086 	off_t *cookies = NULL;
2087 	int ncookies = 0, nc;
2088 
2089 	if (vp->v_type != VDIR)
2090 		return (EPERM);
2091 
2092 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2093 	count = uio->uio_resid - lost;
2094 	if (count <= 0)
2095 		return (EINVAL);
2096 
2097 	/*
2098 	 * Call nfs_bioread() to do the real work.
2099 	 */
2100 	tresid = uio->uio_resid = count;
2101 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2102 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2103 
2104 	if (!error && ap->a_cookies) {
2105 		ncookies = count / 16;
2106 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2107 		*ap->a_cookies = cookies;
2108 	}
2109 
2110 	if (!error && uio->uio_resid == tresid) {
2111 		uio->uio_resid += lost;
2112 		nfsstats.direofcache_misses++;
2113 		if (ap->a_cookies)
2114 			*ap->a_ncookies = 0;
2115 		*ap->a_eofflag = 1;
2116 		return (0);
2117 	}
2118 
2119 	if (!error && ap->a_cookies) {
2120 		/*
2121 		 * Only the NFS server and emulations use cookies, and they
2122 		 * load the directory block into system space, so we can
2123 		 * just look at it directly.
2124 		 */
2125 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2126 			panic("nfs_readdir: lost in space");
2127 		for (nc = 0; ncookies-- &&
2128 		     base < (char *)uio->uio_iov->iov_base; nc++){
2129 			dp = (struct dirent *) base;
2130 			if (dp->d_reclen == 0)
2131 				break;
2132 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2133 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2134 			else
2135 				*(cookies++) = NFS_GETCOOKIE(dp);
2136 			base += dp->d_reclen;
2137 		}
2138 		uio->uio_resid +=
2139 		    ((caddr_t)uio->uio_iov->iov_base - base);
2140 		uio->uio_iov->iov_len +=
2141 		    ((caddr_t)uio->uio_iov->iov_base - base);
2142 		uio->uio_iov->iov_base = base;
2143 		*ap->a_ncookies = nc;
2144 	}
2145 
2146 	uio->uio_resid += lost;
2147 	*ap->a_eofflag = 0;
2148 	return (error);
2149 }
2150 
2151 /*
2152  * Readdir rpc call.
2153  * Called from below the buffer cache by nfs_doio().
2154  */
2155 int
2156 nfs_readdirrpc(vp, uiop, cred)
2157 	struct vnode *vp;
2158 	struct uio *uiop;
2159 	struct ucred *cred;
2160 {
2161 	int len, left;
2162 	struct dirent *dp = NULL;
2163 	u_int32_t *tl;
2164 	caddr_t cp;
2165 	int32_t t1, t2;
2166 	caddr_t bpos, dpos, cp2;
2167 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2168 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2169 	struct nfsnode *dnp = VTONFS(vp);
2170 	u_quad_t fileno;
2171 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2172 	int attrflag, nrpcs = 0, reclen;
2173 	const int v3 = NFS_ISV3(vp);
2174 	nfsquad_t cookie;
2175 
2176 #ifdef DIAGNOSTIC
2177 	/*
2178 	 * Should be called from buffer cache, so only amount of
2179 	 * NFS_DIRBLKSIZ will be requested.
2180 	 */
2181 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2182 		panic("nfs readdirrpc bad uio");
2183 #endif
2184 
2185 	/*
2186 	 * Loop around doing readdir rpc's of size nm_readdirsize
2187 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2188 	 * The stopping criteria is EOF or buffer full.
2189 	 */
2190 	while (more_dirs && bigenough) {
2191 		/*
2192 		 * Heuristic: don't bother to do another RPC to further
2193 		 * fill up this block if there is not much room left. (< 50%
2194 		 * of the readdir RPC size). This wastes some buffer space
2195 		 * but can save up to 50% in RPC calls.
2196 		 */
2197 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2198 			bigenough = 0;
2199 			break;
2200 		}
2201 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2202 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2203 			NFSX_READDIR(v3));
2204 		nfsm_fhtom(vp, v3);
2205 		if (v3) {
2206 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2207 			cookie.qval = uiop->uio_offset;
2208 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2209 				txdr_swapcookie3(uiop->uio_offset, tl);
2210 			} else {
2211 				txdr_cookie3(uiop->uio_offset, tl);
2212 			}
2213 			tl += 2;
2214 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2215 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2216 		} else {
2217 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2218 			*tl++ = txdr_unsigned(uiop->uio_offset);
2219 		}
2220 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2221 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2222 		nrpcs++;
2223 		if (v3) {
2224 			nfsm_postop_attr(vp, attrflag);
2225 			if (!error) {
2226 				nfsm_dissect(tl, u_int32_t *,
2227 				    2 * NFSX_UNSIGNED);
2228 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2229 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2230 			} else {
2231 				m_freem(mrep);
2232 				goto nfsmout;
2233 			}
2234 		}
2235 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2236 		more_dirs = fxdr_unsigned(int, *tl);
2237 
2238 		/* loop thru the dir entries, doctoring them to 4bsd form */
2239 		while (more_dirs && bigenough) {
2240 			if (v3) {
2241 				nfsm_dissect(tl, u_int32_t *,
2242 				    3 * NFSX_UNSIGNED);
2243 				fileno = fxdr_hyper(tl);
2244 				len = fxdr_unsigned(int, *(tl + 2));
2245 			} else {
2246 				nfsm_dissect(tl, u_int32_t *,
2247 				    2 * NFSX_UNSIGNED);
2248 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2249 				len = fxdr_unsigned(int, *tl);
2250 			}
2251 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2252 				error = EBADRPC;
2253 				m_freem(mrep);
2254 				goto nfsmout;
2255 			}
2256 			tlen = nfsm_rndup(len);
2257 			if (tlen == len)
2258 				tlen += 4;	/* To ensure null termination */
2259 			tlen += sizeof (off_t) + sizeof (int);
2260 			reclen = ALIGN(tlen + DIRHDSIZ);
2261 			tlen = reclen - DIRHDSIZ;
2262 			left = NFS_DIRFRAGSIZ - blksiz;
2263 			if (reclen > left) {
2264 				dp->d_reclen += left;
2265 				uiop->uio_iov->iov_base =
2266 				    (caddr_t)uiop->uio_iov->iov_base + left;
2267 				uiop->uio_iov->iov_len -= left;
2268 				uiop->uio_resid -= left;
2269 				blksiz = 0;
2270 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2271 			}
2272 			if (reclen > uiop->uio_resid)
2273 				bigenough = 0;
2274 			if (bigenough) {
2275 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2276 				dp->d_fileno = (int)fileno;
2277 				dp->d_namlen = len;
2278 				dp->d_reclen = reclen;
2279 				dp->d_type = DT_UNKNOWN;
2280 				blksiz += dp->d_reclen;
2281 				if (blksiz == NFS_DIRFRAGSIZ)
2282 					blksiz = 0;
2283 				uiop->uio_resid -= DIRHDSIZ;
2284 				uiop->uio_iov->iov_base =
2285 				    (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2286 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2287 				nfsm_mtouio(uiop, len);
2288 				cp = uiop->uio_iov->iov_base;
2289 				tlen -= len;
2290 				*cp = '\0';	/* null terminate */
2291 				uiop->uio_iov->iov_base =
2292 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2293 				uiop->uio_iov->iov_len -= tlen;
2294 				uiop->uio_resid -= tlen;
2295 			} else
2296 				nfsm_adv(nfsm_rndup(len));
2297 			if (v3) {
2298 				nfsm_dissect(tl, u_int32_t *,
2299 				    3 * NFSX_UNSIGNED);
2300 			} else {
2301 				nfsm_dissect(tl, u_int32_t *,
2302 				    2 * NFSX_UNSIGNED);
2303 			}
2304 			if (bigenough) {
2305 				if (v3) {
2306 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2307 						uiop->uio_offset =
2308 						    fxdr_swapcookie3(tl);
2309 					else
2310 						uiop->uio_offset =
2311 						    fxdr_cookie3(tl);
2312 				}
2313 				else {
2314 					uiop->uio_offset =
2315 					    fxdr_unsigned(off_t, *tl);
2316 				}
2317 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2318 			}
2319 			if (v3)
2320 				tl += 2;
2321 			else
2322 				tl++;
2323 			more_dirs = fxdr_unsigned(int, *tl);
2324 		}
2325 		/*
2326 		 * If at end of rpc data, get the eof boolean
2327 		 */
2328 		if (!more_dirs) {
2329 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2330 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2331 		}
2332 		m_freem(mrep);
2333 	}
2334 	/*
2335 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2336 	 * by increasing d_reclen for the last record.
2337 	 */
2338 	if (blksiz > 0) {
2339 		left = NFS_DIRFRAGSIZ - blksiz;
2340 		dp->d_reclen += left;
2341 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2342 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2343 		    left;
2344 		uiop->uio_iov->iov_len -= left;
2345 		uiop->uio_resid -= left;
2346 	}
2347 
2348 	/*
2349 	 * We are now either at the end of the directory or have filled the
2350 	 * block.
2351 	 */
2352 	if (bigenough)
2353 		dnp->n_direofoffset = uiop->uio_offset;
2354 nfsmout:
2355 	return (error);
2356 }
2357 
2358 /*
2359  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2360  */
2361 int
2362 nfs_readdirplusrpc(vp, uiop, cred)
2363 	struct vnode *vp;
2364 	struct uio *uiop;
2365 	struct ucred *cred;
2366 {
2367 	int len, left;
2368 	struct dirent *dp = NULL;
2369 	u_int32_t *tl;
2370 	caddr_t cp;
2371 	int32_t t1, t2;
2372 	struct vnode *newvp;
2373 	caddr_t bpos, dpos, cp2;
2374 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2375 	struct nameidata nami, *ndp = &nami;
2376 	struct componentname *cnp = &ndp->ni_cnd;
2377 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2378 	struct nfsnode *dnp = VTONFS(vp), *np;
2379 	const unsigned char *hcp;
2380 	nfsfh_t *fhp;
2381 	u_quad_t fileno;
2382 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2383 	int attrflag, fhsize, nrpcs = 0, reclen;
2384 	struct nfs_fattr fattr, *fp;
2385 
2386 #ifdef DIAGNOSTIC
2387 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2388 		panic("nfs readdirplusrpc bad uio");
2389 #endif
2390 	ndp->ni_dvp = vp;
2391 	newvp = NULLVP;
2392 
2393 	/*
2394 	 * Loop around doing readdir rpc's of size nm_readdirsize
2395 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2396 	 * The stopping criteria is EOF or buffer full.
2397 	 */
2398 	while (more_dirs && bigenough) {
2399 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2400 			bigenough = 0;
2401 			break;
2402 		}
2403 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2404 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2405 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2406 		nfsm_fhtom(vp, 1);
2407  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2408 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2409 			txdr_swapcookie3(uiop->uio_offset, tl);
2410 		} else {
2411 			txdr_cookie3(uiop->uio_offset, tl);
2412 		}
2413 		tl += 2;
2414 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2415 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2416 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2417 		*tl = txdr_unsigned(nmp->nm_rsize);
2418 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2419 		nfsm_postop_attr(vp, attrflag);
2420 		if (error) {
2421 			m_freem(mrep);
2422 			goto nfsmout;
2423 		}
2424 		nrpcs++;
2425 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2426 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2427 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2428 		more_dirs = fxdr_unsigned(int, *tl);
2429 
2430 		/* loop thru the dir entries, doctoring them to 4bsd form */
2431 		while (more_dirs && bigenough) {
2432 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2433 			fileno = fxdr_hyper(tl);
2434 			len = fxdr_unsigned(int, *(tl + 2));
2435 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2436 				error = EBADRPC;
2437 				m_freem(mrep);
2438 				goto nfsmout;
2439 			}
2440 			tlen = nfsm_rndup(len);
2441 			if (tlen == len)
2442 				tlen += 4;	/* To ensure null termination*/
2443 			tlen += sizeof (off_t) + sizeof (int);
2444 			reclen = ALIGN(tlen + DIRHDSIZ);
2445 			tlen = reclen - DIRHDSIZ;
2446 			left = NFS_DIRFRAGSIZ - blksiz;
2447 			if (reclen > left) {
2448 				/*
2449 				 * DIRFRAGSIZ is aligned, no need to align
2450 				 * again here.
2451 				 */
2452 				dp->d_reclen += left;
2453 				uiop->uio_iov->iov_base =
2454 				    (caddr_t)uiop->uio_iov->iov_base + left;
2455 				uiop->uio_iov->iov_len -= left;
2456 				uiop->uio_resid -= left;
2457 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2458 				blksiz = 0;
2459 			}
2460 			if (reclen > uiop->uio_resid)
2461 				bigenough = 0;
2462 			if (bigenough) {
2463 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2464 				dp->d_fileno = (int)fileno;
2465 				dp->d_namlen = len;
2466 				dp->d_reclen = reclen;
2467 				dp->d_type = DT_UNKNOWN;
2468 				blksiz += dp->d_reclen;
2469 				if (blksiz == NFS_DIRFRAGSIZ)
2470 					blksiz = 0;
2471 				uiop->uio_resid -= DIRHDSIZ;
2472 				uiop->uio_iov->iov_base =
2473 				    (caddr_t)uiop->uio_iov->iov_base +
2474 				    DIRHDSIZ;
2475 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2476 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2477 				cnp->cn_namelen = len;
2478 				nfsm_mtouio(uiop, len);
2479 				cp = uiop->uio_iov->iov_base;
2480 				tlen -= len;
2481 				*cp = '\0';
2482 				uiop->uio_iov->iov_base =
2483 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2484 				uiop->uio_iov->iov_len -= tlen;
2485 				uiop->uio_resid -= tlen;
2486 			} else
2487 				nfsm_adv(nfsm_rndup(len));
2488 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2489 			if (bigenough) {
2490 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2491 					uiop->uio_offset =
2492 						fxdr_swapcookie3(tl);
2493 				else
2494 					uiop->uio_offset =
2495 						fxdr_cookie3(tl);
2496 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2497 			}
2498 			tl += 2;
2499 
2500 			/*
2501 			 * Since the attributes are before the file handle
2502 			 * (sigh), we must skip over the attributes and then
2503 			 * come back and get them.
2504 			 */
2505 			attrflag = fxdr_unsigned(int, *tl);
2506 			if (attrflag) {
2507 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2508 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2509 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2510 			    doit = fxdr_unsigned(int, *tl);
2511 			    if (doit) {
2512 				nfsm_getfh(fhp, fhsize, 1);
2513 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2514 				    VREF(vp);
2515 				    newvp = vp;
2516 				    np = dnp;
2517 				} else {
2518 				    error = nfs_nget(vp->v_mount, fhp,
2519 					fhsize, &np);
2520 				    if (!error)
2521 					newvp = NFSTOV(np);
2522 				}
2523 				if (!error) {
2524 				    nfs_loadattrcache(&newvp, &fattr, 0);
2525 				    dp->d_type =
2526 				        IFTODT(VTTOIF(np->n_vattr->va_type));
2527 				    ndp->ni_vp = newvp;
2528 				    cnp->cn_hash = 0;
2529 				    for (hcp = cnp->cn_nameptr, i = 1; i <= len;
2530 				        i++, hcp++)
2531 				        cnp->cn_hash += *hcp * i;
2532 				    if (cnp->cn_namelen <= NCHNAMLEN)
2533 				        cache_enter(ndp->ni_dvp, ndp->ni_vp,
2534 						    cnp);
2535 				}
2536 			   }
2537 			} else {
2538 			    /* Just skip over the file handle */
2539 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2540 			    i = fxdr_unsigned(int, *tl);
2541 			    nfsm_adv(nfsm_rndup(i));
2542 			}
2543 			if (newvp != NULLVP) {
2544 			    vrele(newvp);
2545 			    if (newvp != vp)
2546 				VOP_UNLOCK(vp, 0);
2547 			    newvp = NULLVP;
2548 			}
2549 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2550 			more_dirs = fxdr_unsigned(int, *tl);
2551 		}
2552 		/*
2553 		 * If at end of rpc data, get the eof boolean
2554 		 */
2555 		if (!more_dirs) {
2556 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2557 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2558 		}
2559 		m_freem(mrep);
2560 	}
2561 	/*
2562 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2563 	 * by increasing d_reclen for the last record.
2564 	 */
2565 	if (blksiz > 0) {
2566 		left = NFS_DIRFRAGSIZ - blksiz;
2567 		dp->d_reclen += left;
2568 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2569 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2570 		    left;
2571 		uiop->uio_iov->iov_len -= left;
2572 		uiop->uio_resid -= left;
2573 	}
2574 
2575 	/*
2576 	 * We are now either at the end of the directory or have filled the
2577 	 * block.
2578 	 */
2579 	if (bigenough)
2580 		dnp->n_direofoffset = uiop->uio_offset;
2581 nfsmout:
2582 	if (newvp != NULLVP) {
2583 		vrele(newvp);
2584 		if (newvp != vp)
2585 			VOP_UNLOCK(vp, 0);
2586 	}
2587 	return (error);
2588 }
2589 static char hextoasc[] = "0123456789abcdef";
2590 
2591 /*
2592  * Silly rename. To make the NFS filesystem that is stateless look a little
2593  * more like the "ufs" a remove of an active vnode is translated to a rename
2594  * to a funny looking filename that is removed by nfs_inactive on the
2595  * nfsnode. There is the potential for another process on a different client
2596  * to create the same funny name between the nfs_lookitup() fails and the
2597  * nfs_rename() completes, but...
2598  */
2599 int
2600 nfs_sillyrename(dvp, vp, cnp)
2601 	struct vnode *dvp, *vp;
2602 	struct componentname *cnp;
2603 {
2604 	struct sillyrename *sp;
2605 	struct nfsnode *np;
2606 	int error;
2607 	short pid;
2608 
2609 	cache_purge(dvp);
2610 	np = VTONFS(vp);
2611 #ifndef DIAGNOSTIC
2612 	if (vp->v_type == VDIR)
2613 		panic("nfs: sillyrename dir");
2614 #endif
2615 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2616 		M_NFSREQ, M_WAITOK);
2617 	sp->s_cred = crdup(cnp->cn_cred);
2618 	sp->s_dvp = dvp;
2619 	VREF(dvp);
2620 
2621 	/* Fudge together a funny name */
2622 	pid = cnp->cn_proc->p_pid;
2623 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2624 	sp->s_namlen = 12;
2625 	sp->s_name[8] = hextoasc[pid & 0xf];
2626 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2627 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2628 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2629 
2630 	/* Try lookitups until we get one that isn't there */
2631 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2632 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2633 		sp->s_name[4]++;
2634 		if (sp->s_name[4] > 'z') {
2635 			error = EINVAL;
2636 			goto bad;
2637 		}
2638 	}
2639 	error = nfs_renameit(dvp, cnp, sp);
2640 	if (error)
2641 		goto bad;
2642 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2643 		cnp->cn_proc, &np);
2644 	np->n_sillyrename = sp;
2645 	return (0);
2646 bad:
2647 	vrele(sp->s_dvp);
2648 	crfree(sp->s_cred);
2649 	free((caddr_t)sp, M_NFSREQ);
2650 	return (error);
2651 }
2652 
2653 /*
2654  * Look up a file name and optionally either update the file handle or
2655  * allocate an nfsnode, depending on the value of npp.
2656  * npp == NULL	--> just do the lookup
2657  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2658  *			handled too
2659  * *npp != NULL --> update the file handle in the vnode
2660  */
2661 int
2662 nfs_lookitup(dvp, name, len, cred, procp, npp)
2663 	struct vnode *dvp;
2664 	const char *name;
2665 	int len;
2666 	struct ucred *cred;
2667 	struct proc *procp;
2668 	struct nfsnode **npp;
2669 {
2670 	u_int32_t *tl;
2671 	caddr_t cp;
2672 	int32_t t1, t2;
2673 	struct vnode *newvp = (struct vnode *)0;
2674 	struct nfsnode *np, *dnp = VTONFS(dvp);
2675 	caddr_t bpos, dpos, cp2;
2676 	int error = 0, fhlen, attrflag;
2677 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2678 	nfsfh_t *nfhp;
2679 	const int v3 = NFS_ISV3(dvp);
2680 
2681 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2682 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2683 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2684 	nfsm_fhtom(dvp, v3);
2685 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2686 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2687 	if (npp && !error) {
2688 		nfsm_getfh(nfhp, fhlen, v3);
2689 		if (*npp) {
2690 		    np = *npp;
2691 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2692 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2693 			np->n_fhp = &np->n_fh;
2694 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2695 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2696 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2697 		    np->n_fhsize = fhlen;
2698 		    newvp = NFSTOV(np);
2699 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2700 		    VREF(dvp);
2701 		    newvp = dvp;
2702 		} else {
2703 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2704 		    if (error) {
2705 			m_freem(mrep);
2706 			return (error);
2707 		    }
2708 		    newvp = NFSTOV(np);
2709 		}
2710 		if (v3) {
2711 			nfsm_postop_attr(newvp, attrflag);
2712 			if (!attrflag && *npp == NULL) {
2713 				m_freem(mrep);
2714 				vput(newvp);
2715 				return (ENOENT);
2716 			}
2717 		} else
2718 			nfsm_loadattr(newvp, (struct vattr *)0);
2719 	}
2720 	nfsm_reqdone;
2721 	if (npp && *npp == NULL) {
2722 		if (error) {
2723 			if (newvp)
2724 				vput(newvp);
2725 		} else
2726 			*npp = np;
2727 	}
2728 	return (error);
2729 }
2730 
2731 /*
2732  * Nfs Version 3 commit rpc
2733  */
2734 int
2735 nfs_commit(vp, offset, cnt, procp)
2736 	struct vnode *vp;
2737 	off_t offset;
2738 	uint32_t cnt;
2739 	struct proc *procp;
2740 {
2741 	caddr_t cp;
2742 	u_int32_t *tl;
2743 	int32_t t1, t2;
2744 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2745 	caddr_t bpos, dpos, cp2;
2746 	int error = 0, wccflag = NFSV3_WCCRATTR;
2747 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2748 
2749 #ifdef fvdl_debug
2750 	printf("commit %lu - %lu\n", (unsigned long)offset,
2751 	    (unsigned long)(offset + cnt));
2752 #endif
2753 
2754 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0)
2755 		return (0);
2756 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2757 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2758 	nfsm_fhtom(vp, 1);
2759 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2760 	txdr_hyper(offset, tl);
2761 	tl += 2;
2762 	*tl = txdr_unsigned(cnt);
2763 	nfsm_request(vp, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2764 	nfsm_wcc_data(vp, wccflag);
2765 	if (!error) {
2766 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2767 		if (memcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2768 			NFSX_V3WRITEVERF)) {
2769 			memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
2770 				NFSX_V3WRITEVERF);
2771 			error = NFSERR_STALEWRITEVERF;
2772 		}
2773 	}
2774 	nfsm_reqdone;
2775 	return (error);
2776 }
2777 
2778 /*
2779  * Kludge City..
2780  * - make nfs_bmap() essentially a no-op that does no translation
2781  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2782  *   (Maybe I could use the process's page mapping, but I was concerned that
2783  *    Kernel Write might not be enabled and also figured copyout() would do
2784  *    a lot more work than memcpy() and also it currently happens in the
2785  *    context of the swapper process (2).
2786  */
2787 int
2788 nfs_bmap(v)
2789 	void *v;
2790 {
2791 	struct vop_bmap_args /* {
2792 		struct vnode *a_vp;
2793 		daddr_t  a_bn;
2794 		struct vnode **a_vpp;
2795 		daddr_t *a_bnp;
2796 		int *a_runp;
2797 	} */ *ap = v;
2798 	struct vnode *vp = ap->a_vp;
2799 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2800 
2801 	if (ap->a_vpp != NULL)
2802 		*ap->a_vpp = vp;
2803 	if (ap->a_bnp != NULL)
2804 		*ap->a_bnp = ap->a_bn << bshift;
2805 	if (ap->a_runp != NULL)
2806 		*ap->a_runp = 1024 * 1024; /* XXX */
2807 	return (0);
2808 }
2809 
2810 /*
2811  * Strategy routine.
2812  * For async requests when nfsiod(s) are running, queue the request by
2813  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2814  * request.
2815  */
2816 int
2817 nfs_strategy(v)
2818 	void *v;
2819 {
2820 	struct vop_strategy_args *ap = v;
2821 	struct buf *bp = ap->a_bp;
2822 	struct proc *p;
2823 	int error = 0;
2824 
2825 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2826 		panic("nfs physio/async");
2827 	if (bp->b_flags & B_ASYNC)
2828 		p = NULL;
2829 	else
2830 		p = curproc;	/* XXX */
2831 
2832 	/*
2833 	 * If the op is asynchronous and an i/o daemon is waiting
2834 	 * queue the request, wake it up and wait for completion
2835 	 * otherwise just do it ourselves.
2836 	 */
2837 
2838 	if ((bp->b_flags & B_ASYNC) == 0 ||
2839 	    nfs_asyncio(bp))
2840 		error = nfs_doio(bp, p);
2841 	return (error);
2842 }
2843 
2844 /*
2845  * fsync vnode op. Just call nfs_flush() with commit == 1.
2846  */
2847 /* ARGSUSED */
2848 int
2849 nfs_fsync(v)
2850 	void *v;
2851 {
2852 	struct vop_fsync_args /* {
2853 		struct vnodeop_desc *a_desc;
2854 		struct vnode * a_vp;
2855 		struct ucred * a_cred;
2856 		int  a_flags;
2857 		off_t offlo;
2858 		off_t offhi;
2859 		struct proc * a_p;
2860 	} */ *ap = v;
2861 
2862 	return (nfs_flush(ap->a_vp, ap->a_cred,
2863 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
2864 }
2865 
2866 /*
2867  * Flush all the data associated with a vnode.
2868  */
2869 int
2870 nfs_flush(vp, cred, waitfor, p, commit)
2871 	struct vnode *vp;
2872 	struct ucred *cred;
2873 	int waitfor;
2874 	struct proc *p;
2875 	int commit;
2876 {
2877 	struct uvm_object *uobj = &vp->v_uobj;
2878 	struct nfsnode *np = VTONFS(vp);
2879 	int error;
2880 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
2881 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
2882 
2883 	simple_lock(&uobj->vmobjlock);
2884 	error = (uobj->pgops->pgo_put)(uobj, 0, 0, flushflags);
2885 	if (np->n_flag & NWRITEERR) {
2886 		error = np->n_error;
2887 		np->n_flag &= ~NWRITEERR;
2888 	}
2889 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
2890 	return (error);
2891 }
2892 
2893 /*
2894  * Return POSIX pathconf information applicable to nfs.
2895  *
2896  * N.B. The NFS V2 protocol doesn't support this RPC.
2897  */
2898 /* ARGSUSED */
2899 int
2900 nfs_pathconf(v)
2901 	void *v;
2902 {
2903 	struct vop_pathconf_args /* {
2904 		struct vnode *a_vp;
2905 		int a_name;
2906 		register_t *a_retval;
2907 	} */ *ap = v;
2908 	struct nfsv3_pathconf *pcp;
2909 	struct vnode *vp = ap->a_vp;
2910 	struct nfsmount *nmp;
2911 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2912 	int32_t t1, t2;
2913 	u_int32_t *tl;
2914 	caddr_t bpos, dpos, cp, cp2;
2915 	int error = 0, attrflag;
2916 	unsigned int l;
2917 	u_int64_t maxsize;
2918 	const int v3 = NFS_ISV3(vp);
2919 
2920 	switch (ap->a_name) {
2921 		/* Names that can be resolved locally. */
2922 	case _PC_PIPE_BUF:
2923 		*ap->a_retval = PIPE_BUF;
2924 		break;
2925 	case _PC_SYNC_IO:
2926 		*ap->a_retval = 1;
2927 		break;
2928 	/* Names that cannot be resolved locally; do an RPC, if possible. */
2929 	case _PC_LINK_MAX:
2930 	case _PC_NAME_MAX:
2931 	case _PC_CHOWN_RESTRICTED:
2932 	case _PC_NO_TRUNC:
2933 		if (!v3) {
2934 			error = EINVAL;
2935 			break;
2936 		}
2937 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
2938 		nfsm_reqhead(vp, NFSPROC_PATHCONF, NFSX_FH(1));
2939 		nfsm_fhtom(vp, 1);
2940 		nfsm_request(vp, NFSPROC_PATHCONF,
2941 		    curproc, curproc->p_ucred);	/* XXX */
2942 		nfsm_postop_attr(vp, attrflag);
2943 		if (!error) {
2944 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
2945 			    NFSX_V3PATHCONF);
2946 			switch (ap->a_name) {
2947 			case _PC_LINK_MAX:
2948 				*ap->a_retval =
2949 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
2950 				break;
2951 			case _PC_NAME_MAX:
2952 				*ap->a_retval =
2953 				    fxdr_unsigned(register_t, pcp->pc_namemax);
2954 				break;
2955 			case _PC_CHOWN_RESTRICTED:
2956 				*ap->a_retval =
2957 				    (pcp->pc_chownrestricted == nfs_true);
2958 				break;
2959 			case _PC_NO_TRUNC:
2960 				*ap->a_retval =
2961 				    (pcp->pc_notrunc == nfs_true);
2962 				break;
2963 			}
2964 		}
2965 		nfsm_reqdone;
2966 		break;
2967 	case _PC_FILESIZEBITS:
2968 		if (v3) {
2969 			nmp = VFSTONFS(vp->v_mount);
2970 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
2971 				if ((error = nfs_fsinfo(nmp, vp,
2972 				    curproc->p_ucred, curproc)) != 0) /* XXX */
2973 					break;
2974 			for (l = 0, maxsize = nmp->nm_maxfilesize;
2975 			    (maxsize >> l) > 0; l++)
2976 				;
2977 			*ap->a_retval = l + 1;
2978 		} else {
2979 			*ap->a_retval = 32;	/* NFS V2 limitation */
2980 		}
2981 		break;
2982 	default:
2983 		error = EINVAL;
2984 		break;
2985 	}
2986 
2987 	return (error);
2988 }
2989 
2990 /*
2991  * NFS advisory byte-level locks.
2992  */
2993 int
2994 nfs_advlock(v)
2995 	void *v;
2996 {
2997 	struct vop_advlock_args /* {
2998 		struct vnode *a_vp;
2999 		caddr_t  a_id;
3000 		int  a_op;
3001 		struct flock *a_fl;
3002 		int  a_flags;
3003 	} */ *ap = v;
3004 	struct nfsnode *np = VTONFS(ap->a_vp);
3005 
3006 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3007 }
3008 
3009 /*
3010  * Print out the contents of an nfsnode.
3011  */
3012 int
3013 nfs_print(v)
3014 	void *v;
3015 {
3016 	struct vop_print_args /* {
3017 		struct vnode *a_vp;
3018 	} */ *ap = v;
3019 	struct vnode *vp = ap->a_vp;
3020 	struct nfsnode *np = VTONFS(vp);
3021 
3022 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3023 	    np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3024 	if (vp->v_type == VFIFO)
3025 		fifo_printinfo(vp);
3026 	printf("\n");
3027 	return (0);
3028 }
3029 
3030 /*
3031  * NFS file truncation.
3032  */
3033 int
3034 nfs_truncate(v)
3035 	void *v;
3036 {
3037 #if 0
3038 	struct vop_truncate_args /* {
3039 		struct vnode *a_vp;
3040 		off_t a_length;
3041 		int a_flags;
3042 		struct ucred *a_cred;
3043 		struct proc *a_p;
3044 	} */ *ap = v;
3045 #endif
3046 
3047 	/* Use nfs_setattr */
3048 	return (EOPNOTSUPP);
3049 }
3050 
3051 /*
3052  * NFS update.
3053  */
3054 int
3055 nfs_update(v)
3056 	void *v;
3057 #if 0
3058 	struct vop_update_args /* {
3059 		struct vnode *a_vp;
3060 		struct timespec *a_ta;
3061 		struct timespec *a_tm;
3062 		int a_waitfor;
3063 	} */ *ap = v;
3064 #endif
3065 {
3066 
3067 	/* Use nfs_setattr */
3068 	return (EOPNOTSUPP);
3069 }
3070 
3071 /*
3072  * Just call bwrite().
3073  */
3074 int
3075 nfs_bwrite(v)
3076 	void *v;
3077 {
3078 	struct vop_bwrite_args /* {
3079 		struct vnode *a_bp;
3080 	} */ *ap = v;
3081 
3082 	return (bwrite(ap->a_bp));
3083 }
3084 
3085 /*
3086  * nfs special file access vnode op.
3087  * Essentially just get vattr and then imitate iaccess() since the device is
3088  * local to the client.
3089  */
3090 int
3091 nfsspec_access(v)
3092 	void *v;
3093 {
3094 	struct vop_access_args /* {
3095 		struct vnode *a_vp;
3096 		int  a_mode;
3097 		struct ucred *a_cred;
3098 		struct proc *a_p;
3099 	} */ *ap = v;
3100 	struct vattr va;
3101 	struct vnode *vp = ap->a_vp;
3102 	int error;
3103 
3104 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3105 	if (error)
3106 		return (error);
3107 
3108         /*
3109 	 * Disallow write attempts on filesystems mounted read-only;
3110 	 * unless the file is a socket, fifo, or a block or character
3111 	 * device resident on the filesystem.
3112 	 */
3113 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3114 		switch (vp->v_type) {
3115 		case VREG:
3116 		case VDIR:
3117 		case VLNK:
3118 			return (EROFS);
3119 		default:
3120 			break;
3121 		}
3122 	}
3123 
3124 	return (vaccess(va.va_type, va.va_mode,
3125 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3126 }
3127 
3128 /*
3129  * Read wrapper for special devices.
3130  */
3131 int
3132 nfsspec_read(v)
3133 	void *v;
3134 {
3135 	struct vop_read_args /* {
3136 		struct vnode *a_vp;
3137 		struct uio *a_uio;
3138 		int  a_ioflag;
3139 		struct ucred *a_cred;
3140 	} */ *ap = v;
3141 	struct nfsnode *np = VTONFS(ap->a_vp);
3142 
3143 	/*
3144 	 * Set access flag.
3145 	 */
3146 	np->n_flag |= NACC;
3147 	np->n_atim.tv_sec = time.tv_sec;
3148 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3149 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3150 }
3151 
3152 /*
3153  * Write wrapper for special devices.
3154  */
3155 int
3156 nfsspec_write(v)
3157 	void *v;
3158 {
3159 	struct vop_write_args /* {
3160 		struct vnode *a_vp;
3161 		struct uio *a_uio;
3162 		int  a_ioflag;
3163 		struct ucred *a_cred;
3164 	} */ *ap = v;
3165 	struct nfsnode *np = VTONFS(ap->a_vp);
3166 
3167 	/*
3168 	 * Set update flag.
3169 	 */
3170 	np->n_flag |= NUPD;
3171 	np->n_mtim.tv_sec = time.tv_sec;
3172 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3173 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3174 }
3175 
3176 /*
3177  * Close wrapper for special devices.
3178  *
3179  * Update the times on the nfsnode then do device close.
3180  */
3181 int
3182 nfsspec_close(v)
3183 	void *v;
3184 {
3185 	struct vop_close_args /* {
3186 		struct vnode *a_vp;
3187 		int  a_fflag;
3188 		struct ucred *a_cred;
3189 		struct proc *a_p;
3190 	} */ *ap = v;
3191 	struct vnode *vp = ap->a_vp;
3192 	struct nfsnode *np = VTONFS(vp);
3193 	struct vattr vattr;
3194 
3195 	if (np->n_flag & (NACC | NUPD)) {
3196 		np->n_flag |= NCHG;
3197 		if (vp->v_usecount == 1 &&
3198 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3199 			VATTR_NULL(&vattr);
3200 			if (np->n_flag & NACC)
3201 				vattr.va_atime = np->n_atim;
3202 			if (np->n_flag & NUPD)
3203 				vattr.va_mtime = np->n_mtim;
3204 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3205 		}
3206 	}
3207 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3208 }
3209 
3210 /*
3211  * Read wrapper for fifos.
3212  */
3213 int
3214 nfsfifo_read(v)
3215 	void *v;
3216 {
3217 	struct vop_read_args /* {
3218 		struct vnode *a_vp;
3219 		struct uio *a_uio;
3220 		int  a_ioflag;
3221 		struct ucred *a_cred;
3222 	} */ *ap = v;
3223 	struct nfsnode *np = VTONFS(ap->a_vp);
3224 
3225 	/*
3226 	 * Set access flag.
3227 	 */
3228 	np->n_flag |= NACC;
3229 	np->n_atim.tv_sec = time.tv_sec;
3230 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3231 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3232 }
3233 
3234 /*
3235  * Write wrapper for fifos.
3236  */
3237 int
3238 nfsfifo_write(v)
3239 	void *v;
3240 {
3241 	struct vop_write_args /* {
3242 		struct vnode *a_vp;
3243 		struct uio *a_uio;
3244 		int  a_ioflag;
3245 		struct ucred *a_cred;
3246 	} */ *ap = v;
3247 	struct nfsnode *np = VTONFS(ap->a_vp);
3248 
3249 	/*
3250 	 * Set update flag.
3251 	 */
3252 	np->n_flag |= NUPD;
3253 	np->n_mtim.tv_sec = time.tv_sec;
3254 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3255 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3256 }
3257 
3258 /*
3259  * Close wrapper for fifos.
3260  *
3261  * Update the times on the nfsnode then do fifo close.
3262  */
3263 int
3264 nfsfifo_close(v)
3265 	void *v;
3266 {
3267 	struct vop_close_args /* {
3268 		struct vnode *a_vp;
3269 		int  a_fflag;
3270 		struct ucred *a_cred;
3271 		struct proc *a_p;
3272 	} */ *ap = v;
3273 	struct vnode *vp = ap->a_vp;
3274 	struct nfsnode *np = VTONFS(vp);
3275 	struct vattr vattr;
3276 
3277 	if (np->n_flag & (NACC | NUPD)) {
3278 		if (np->n_flag & NACC) {
3279 			np->n_atim.tv_sec = time.tv_sec;
3280 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3281 		}
3282 		if (np->n_flag & NUPD) {
3283 			np->n_mtim.tv_sec = time.tv_sec;
3284 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3285 		}
3286 		np->n_flag |= NCHG;
3287 		if (vp->v_usecount == 1 &&
3288 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3289 			VATTR_NULL(&vattr);
3290 			if (np->n_flag & NACC)
3291 				vattr.va_atime = np->n_atim;
3292 			if (np->n_flag & NUPD)
3293 				vattr.va_mtime = np->n_mtim;
3294 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3295 		}
3296 	}
3297 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3298 }
3299