xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision e55cffd8e520e9b03f18a1bd98bb04223e79f69f)
1 /*	$NetBSD: nfs_vnops.c,v 1.131 2001/04/20 11:22:02 fvdl 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, nfs_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 	{ NULL, NULL }
199 };
200 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
201 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
202 
203 int (**fifo_nfsv2nodeop_p) __P((void *));
204 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
205 	{ &vop_default_desc, vn_default_error },
206 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
207 	{ &vop_create_desc, fifo_create },		/* create */
208 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
209 	{ &vop_open_desc, fifo_open },			/* open */
210 	{ &vop_close_desc, nfsfifo_close },		/* close */
211 	{ &vop_access_desc, nfsspec_access },		/* access */
212 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
213 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
214 	{ &vop_read_desc, nfsfifo_read },		/* read */
215 	{ &vop_write_desc, nfsfifo_write },		/* write */
216 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
217 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
218 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
219 	{ &vop_poll_desc, fifo_poll },			/* poll */
220 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
221 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
222 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
223 	{ &vop_seek_desc, fifo_seek },			/* seek */
224 	{ &vop_remove_desc, fifo_remove },		/* remove */
225 	{ &vop_link_desc, fifo_link },			/* link */
226 	{ &vop_rename_desc, fifo_rename },		/* rename */
227 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
228 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
229 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
230 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
231 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
232 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
233 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
234 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
235 	{ &vop_lock_desc, nfs_lock },			/* lock */
236 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
237 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
238 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
239 	{ &vop_print_desc, nfs_print },			/* print */
240 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
241 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
242 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
243 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
244 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
245 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
246 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
247 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
248 	{ &vop_update_desc, nfs_update },		/* update */
249 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
250 	{ NULL, NULL }
251 };
252 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
253 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
254 
255 /*
256  * Global variables
257  */
258 extern u_int32_t nfs_true, nfs_false;
259 extern u_int32_t nfs_xdrneg1;
260 extern struct nfsstats nfsstats;
261 extern nfstype nfsv3_type[9];
262 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
263 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
264 int nfs_numasync = 0;
265 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
266 
267 /*
268  * nfs null call from vfs.
269  */
270 int
271 nfs_null(vp, cred, procp)
272 	struct vnode *vp;
273 	struct ucred *cred;
274 	struct proc *procp;
275 {
276 	caddr_t bpos, dpos;
277 	int error = 0;
278 	struct mbuf *mreq, *mrep, *md, *mb;
279 
280 	nfsm_reqhead(vp, NFSPROC_NULL, 0);
281 	nfsm_request(vp, NFSPROC_NULL, procp, cred);
282 	nfsm_reqdone;
283 	return (error);
284 }
285 
286 /*
287  * nfs access vnode op.
288  * For nfs version 2, just return ok. File accesses may fail later.
289  * For nfs version 3, use the access rpc to check accessibility. If file modes
290  * are changed on the server, accesses might still fail later.
291  */
292 int
293 nfs_access(v)
294 	void *v;
295 {
296 	struct vop_access_args /* {
297 		struct vnode *a_vp;
298 		int  a_mode;
299 		struct ucred *a_cred;
300 		struct proc *a_p;
301 	} */ *ap = v;
302 	struct vnode *vp = ap->a_vp;
303 	u_int32_t *tl;
304 	caddr_t cp;
305 	int32_t t1, t2;
306 	caddr_t bpos, dpos, cp2;
307 	int error = 0, attrflag, cachevalid;
308 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
309 	u_int32_t mode, rmode;
310 	const int v3 = NFS_ISV3(vp);
311 	struct nfsnode *np = VTONFS(vp);
312 
313 	cachevalid = (np->n_accstamp != -1 &&
314 	    (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
315 	    np->n_accuid == ap->a_cred->cr_uid);
316 
317 	/*
318 	 * Check access cache first. If this request has been made for this
319 	 * uid shortly before, use the cached result.
320 	 */
321 	if (cachevalid) {
322 		if (!np->n_accerror) {
323 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
324 				return np->n_accerror;
325 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
326 			return np->n_accerror;
327 	}
328 
329 	/*
330 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
331 	 * ufs_access() locally using the vattr. This may not be correct,
332 	 * since the server may apply other access criteria such as
333 	 * client uid-->server uid mapping that we do not know about, but
334 	 * this is better than just returning anything that is lying about
335 	 * in the cache.
336 	 */
337 	if (v3) {
338 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
339 		nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
340 		nfsm_fhtom(vp, v3);
341 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
342 		if (ap->a_mode & VREAD)
343 			mode = NFSV3ACCESS_READ;
344 		else
345 			mode = 0;
346 		if (vp->v_type != VDIR) {
347 			if (ap->a_mode & VWRITE)
348 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
349 			if (ap->a_mode & VEXEC)
350 				mode |= NFSV3ACCESS_EXECUTE;
351 		} else {
352 			if (ap->a_mode & VWRITE)
353 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
354 					 NFSV3ACCESS_DELETE);
355 			if (ap->a_mode & VEXEC)
356 				mode |= NFSV3ACCESS_LOOKUP;
357 		}
358 		*tl = txdr_unsigned(mode);
359 		nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
360 		nfsm_postop_attr(vp, attrflag);
361 		if (!error) {
362 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
363 			rmode = fxdr_unsigned(u_int32_t, *tl);
364 			/*
365 			 * The NFS V3 spec does not clarify whether or not
366 			 * the returned access bits can be a superset of
367 			 * the ones requested, so...
368 			 */
369 			if ((rmode & mode) != mode)
370 				error = EACCES;
371 		}
372 		nfsm_reqdone;
373 	} else
374 		return (nfsspec_access(ap));
375 	/*
376 	 * Disallow write attempts on filesystems mounted read-only;
377 	 * unless the file is a socket, fifo, or a block or character
378 	 * device resident on the filesystem.
379 	 */
380 	if (!error && (ap->a_mode & VWRITE) &&
381 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
382 		switch (vp->v_type) {
383 		case VREG:
384 		case VDIR:
385 		case VLNK:
386 			error = EROFS;
387 		default:
388 			break;
389 		}
390 	}
391 
392 	if (!error || error == EACCES) {
393 		/*
394 		 * If we got the same result as for a previous,
395 		 * different request, OR it in. Don't update
396 		 * the timestamp in that case.
397 		 */
398 		if (cachevalid && error == np->n_accerror) {
399 			if (!error)
400 				np->n_accmode |= ap->a_mode;
401 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
402 				np->n_accmode = ap->a_mode;
403 		} else {
404 			np->n_accstamp = time.tv_sec;
405 			np->n_accuid = ap->a_cred->cr_uid;
406 			np->n_accmode = ap->a_mode;
407 			np->n_accerror = error;
408 		}
409 	}
410 
411 	return (error);
412 }
413 
414 /*
415  * nfs open vnode op
416  * Check to see if the type is ok
417  * and that deletion is not in progress.
418  * For paged in text files, you will need to flush the page cache
419  * if consistency is lost.
420  */
421 /* ARGSUSED */
422 int
423 nfs_open(v)
424 	void *v;
425 {
426 	struct vop_open_args /* {
427 		struct vnode *a_vp;
428 		int  a_mode;
429 		struct ucred *a_cred;
430 		struct proc *a_p;
431 	} */ *ap = v;
432 	struct vnode *vp = ap->a_vp;
433 	struct nfsnode *np = VTONFS(vp);
434 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
435 	struct vattr vattr;
436 	int error;
437 
438 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
439 		return (EACCES);
440 	}
441 
442 	/*
443 	 * Initialize read and write creds here, for swapfiles
444 	 * and other paths that don't set the creds themselves.
445 	 */
446 
447 	if (ap->a_mode & FREAD) {
448 		if (np->n_rcred) {
449 			crfree(np->n_rcred);
450 		}
451 		np->n_rcred = ap->a_cred;
452 		crhold(np->n_rcred);
453 	}
454 	if (ap->a_mode & FWRITE) {
455 		if (np->n_wcred) {
456 			crfree(np->n_wcred);
457 		}
458 		np->n_wcred = ap->a_cred;
459 		crhold(np->n_wcred);
460 	}
461 
462 #ifndef NFS_V2_ONLY
463 	/*
464 	 * Get a valid lease. If cached data is stale, flush it.
465 	 */
466 	if (nmp->nm_flag & NFSMNT_NQNFS) {
467 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
468 		    do {
469 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
470 			    ap->a_p);
471 		    } while (error == NQNFS_EXPIRED);
472 		    if (error)
473 			return (error);
474 		    if (np->n_lrev != np->n_brev ||
475 			(np->n_flag & NQNFSNONCACHE)) {
476 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
477 				ap->a_p, 1)) == EINTR)
478 				return (error);
479 			np->n_brev = np->n_lrev;
480 		    }
481 		}
482 	} else
483 #endif
484 	{
485 		if (np->n_flag & NMODIFIED) {
486 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
487 				ap->a_p, 1)) == EINTR)
488 				return (error);
489 			np->n_attrstamp = 0;
490 			if (vp->v_type == VDIR) {
491 				nfs_invaldircache(vp, 0);
492 				np->n_direofoffset = 0;
493 			}
494 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
495 			if (error)
496 				return (error);
497 			np->n_mtime = vattr.va_mtime.tv_sec;
498 		} else {
499 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
500 			if (error)
501 				return (error);
502 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
503 				if (vp->v_type == VDIR) {
504 					nfs_invaldircache(vp, 0);
505 					np->n_direofoffset = 0;
506 				}
507 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
508 					ap->a_cred, ap->a_p, 1)) == EINTR)
509 					return (error);
510 				np->n_mtime = vattr.va_mtime.tv_sec;
511 			}
512 		}
513 	}
514 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
515 		np->n_attrstamp = 0; /* For Open/Close consistency */
516 	return (0);
517 }
518 
519 /*
520  * nfs close vnode op
521  * What an NFS client should do upon close after writing is a debatable issue.
522  * Most NFS clients push delayed writes to the server upon close, basically for
523  * two reasons:
524  * 1 - So that any write errors may be reported back to the client process
525  *     doing the close system call. By far the two most likely errors are
526  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
527  * 2 - To put a worst case upper bound on cache inconsistency between
528  *     multiple clients for the file.
529  * There is also a consistency problem for Version 2 of the protocol w.r.t.
530  * not being able to tell if other clients are writing a file concurrently,
531  * since there is no way of knowing if the changed modify time in the reply
532  * is only due to the write for this client.
533  * (NFS Version 3 provides weak cache consistency data in the reply that
534  *  should be sufficient to detect and handle this case.)
535  *
536  * The current code does the following:
537  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
538  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
539  *                     or commit them (this satisfies 1 and 2 except for the
540  *                     case where the server crashes after this close but
541  *                     before the commit RPC, which is felt to be "good
542  *                     enough". Changing the last argument to nfs_flush() to
543  *                     a 1 would force a commit operation, if it is felt a
544  *                     commit is necessary now.
545  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
546  *                     1 should be dealt with via an fsync() system call for
547  *                     cases where write errors are important.
548  */
549 /* ARGSUSED */
550 int
551 nfs_close(v)
552 	void *v;
553 {
554 	struct vop_close_args /* {
555 		struct vnodeop_desc *a_desc;
556 		struct vnode *a_vp;
557 		int  a_fflag;
558 		struct ucred *a_cred;
559 		struct proc *a_p;
560 	} */ *ap = v;
561 	struct vnode *vp = ap->a_vp;
562 	struct nfsnode *np = VTONFS(vp);
563 	int error = 0;
564 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
565 
566 	if (vp->v_type == VREG) {
567 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
568 		(np->n_flag & NMODIFIED)) {
569 		if (NFS_ISV3(vp)) {
570 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
571 		    np->n_flag &= ~NMODIFIED;
572 		} else
573 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
574 		np->n_attrstamp = 0;
575 	    }
576 	    if (np->n_flag & NWRITEERR) {
577 		np->n_flag &= ~NWRITEERR;
578 		error = np->n_error;
579 	    }
580 	}
581 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
582 	return (error);
583 }
584 
585 /*
586  * nfs getattr call from vfs.
587  */
588 int
589 nfs_getattr(v)
590 	void *v;
591 {
592 	struct vop_getattr_args /* {
593 		struct vnode *a_vp;
594 		struct vattr *a_vap;
595 		struct ucred *a_cred;
596 		struct proc *a_p;
597 	} */ *ap = v;
598 	struct vnode *vp = ap->a_vp;
599 	struct nfsnode *np = VTONFS(vp);
600 	caddr_t cp;
601 	u_int32_t *tl;
602 	int32_t t1, t2;
603 	caddr_t bpos, dpos;
604 	int error = 0;
605 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
606 	const int v3 = NFS_ISV3(vp);
607 
608 	/*
609 	 * Update local times for special files.
610 	 */
611 	if (np->n_flag & (NACC | NUPD))
612 		np->n_flag |= NCHG;
613 	/*
614 	 * First look in the cache.
615 	 */
616 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
617 		return (0);
618 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
619 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
620 	nfsm_fhtom(vp, v3);
621 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
622 	if (!error) {
623 		nfsm_loadattr(vp, ap->a_vap);
624 		if (vp->v_type == VDIR &&
625 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
626 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
627 	}
628 	nfsm_reqdone;
629 	return (error);
630 }
631 
632 /*
633  * nfs setattr call.
634  */
635 int
636 nfs_setattr(v)
637 	void *v;
638 {
639 	struct vop_setattr_args /* {
640 		struct vnodeop_desc *a_desc;
641 		struct vnode *a_vp;
642 		struct vattr *a_vap;
643 		struct ucred *a_cred;
644 		struct proc *a_p;
645 	} */ *ap = v;
646 	struct vnode *vp = ap->a_vp;
647 	struct nfsnode *np = VTONFS(vp);
648 	struct vattr *vap = ap->a_vap;
649 	int error = 0;
650 	u_quad_t tsize = 0;
651 
652 	/*
653 	 * Setting of flags is not supported.
654 	 */
655 	if (vap->va_flags != VNOVAL)
656 		return (EOPNOTSUPP);
657 
658 	/*
659 	 * Disallow write attempts if the filesystem is mounted read-only.
660 	 */
661   	if ((vap->va_uid != (uid_t)VNOVAL ||
662 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
663 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
664 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
665 		return (EROFS);
666 	if (vap->va_size != VNOVAL) {
667  		switch (vp->v_type) {
668  		case VDIR:
669  			return (EISDIR);
670  		case VCHR:
671  		case VBLK:
672  		case VSOCK:
673  		case VFIFO:
674 			if (vap->va_mtime.tv_sec == VNOVAL &&
675 			    vap->va_atime.tv_sec == VNOVAL &&
676 			    vap->va_mode == (mode_t)VNOVAL &&
677 			    vap->va_uid == (uid_t)VNOVAL &&
678 			    vap->va_gid == (gid_t)VNOVAL)
679 				return (0);
680  			vap->va_size = VNOVAL;
681  			break;
682  		default:
683 			/*
684 			 * Disallow write attempts if the filesystem is
685 			 * mounted read-only.
686 			 */
687 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
688 				return (EROFS);
689  			uvm_vnp_setsize(vp, vap->va_size);
690  			if (vap->va_size == 0)
691  				error = nfs_vinvalbuf(vp, 0,
692  				     ap->a_cred, ap->a_p, 1);
693 			else
694 				error = nfs_vinvalbuf(vp, V_SAVE,
695 				     ap->a_cred, ap->a_p, 1);
696 			if (error) {
697 				uvm_vnp_setsize(vp, np->n_size);
698 				return (error);
699 			}
700  			tsize = np->n_size;
701  			np->n_size = np->n_vattr->va_size = vap->va_size;
702   		}
703   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
704 		vap->va_atime.tv_sec != VNOVAL) &&
705 		vp->v_type == VREG &&
706   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
707 		 ap->a_p, 1)) == EINTR)
708 		return (error);
709 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
710 	if (error && vap->va_size != VNOVAL) {
711 		np->n_size = np->n_vattr->va_size = tsize;
712 		uvm_vnp_setsize(vp, np->n_size);
713 	}
714 	return (error);
715 }
716 
717 /*
718  * Do an nfs setattr rpc.
719  */
720 int
721 nfs_setattrrpc(vp, vap, cred, procp)
722 	struct vnode *vp;
723 	struct vattr *vap;
724 	struct ucred *cred;
725 	struct proc *procp;
726 {
727 	struct nfsv2_sattr *sp;
728 	caddr_t cp;
729 	int32_t t1, t2;
730 	caddr_t bpos, dpos, cp2;
731 	u_int32_t *tl;
732 	int error = 0, wccflag = NFSV3_WCCRATTR;
733 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
734 	const int v3 = NFS_ISV3(vp);
735 
736 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
737 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
738 	nfsm_fhtom(vp, v3);
739 	if (v3) {
740 		nfsm_v3attrbuild(vap, TRUE);
741 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
742 		*tl = nfs_false;
743 	} else {
744 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
745 		if (vap->va_mode == (mode_t)VNOVAL)
746 			sp->sa_mode = nfs_xdrneg1;
747 		else
748 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
749 		if (vap->va_uid == (uid_t)VNOVAL)
750 			sp->sa_uid = nfs_xdrneg1;
751 		else
752 			sp->sa_uid = txdr_unsigned(vap->va_uid);
753 		if (vap->va_gid == (gid_t)VNOVAL)
754 			sp->sa_gid = nfs_xdrneg1;
755 		else
756 			sp->sa_gid = txdr_unsigned(vap->va_gid);
757 		sp->sa_size = txdr_unsigned(vap->va_size);
758 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
759 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
760 	}
761 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
762 	if (v3) {
763 		nfsm_wcc_data(vp, wccflag);
764 	} else
765 		nfsm_loadattr(vp, (struct vattr *)0);
766 	nfsm_reqdone;
767 	return (error);
768 }
769 
770 /*
771  * nfs lookup call, one step at a time...
772  * First look in cache
773  * If not found, unlock the directory nfsnode and do the rpc
774  *
775  * This code is full of lock/unlock statements and checks, because
776  * we continue after cache_lookup has finished (we need to check
777  * with the attr cache and do an rpc if it has timed out). This means
778  * that the locking effects of cache_lookup have to be taken into
779  * account.
780  */
781 int
782 nfs_lookup(v)
783 	void *v;
784 {
785 	struct vop_lookup_args /* {
786 		struct vnodeop_desc *a_desc;
787 		struct vnode *a_dvp;
788 		struct vnode **a_vpp;
789 		struct componentname *a_cnp;
790 	} */ *ap = v;
791 	struct componentname *cnp = ap->a_cnp;
792 	struct vnode *dvp = ap->a_dvp;
793 	struct vnode **vpp = ap->a_vpp;
794 	int flags;
795 	struct vnode *newvp;
796 	u_int32_t *tl;
797 	caddr_t cp;
798 	int32_t t1, t2;
799 	struct nfsmount *nmp;
800 	caddr_t bpos, dpos, cp2;
801 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
802 	long len;
803 	nfsfh_t *fhp;
804 	struct nfsnode *np;
805 	int lockparent, wantparent, error = 0, attrflag, fhsize;
806 	const int v3 = NFS_ISV3(dvp);
807 
808 	cnp->cn_flags &= ~PDIRUNLOCK;
809 	flags = cnp->cn_flags;
810 
811 	*vpp = NULLVP;
812 	newvp = NULLVP;
813 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
814 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
815 		return (EROFS);
816 	if (dvp->v_type != VDIR)
817 		return (ENOTDIR);
818 
819 	lockparent = flags & LOCKPARENT;
820 	wantparent = flags & (LOCKPARENT|WANTPARENT);
821 	nmp = VFSTONFS(dvp->v_mount);
822 	np = VTONFS(dvp);
823 
824 	/*
825 	 * Before tediously performing a linear scan of the directory,
826 	 * check the name cache to see if the directory/name pair
827 	 * we are looking for is known already.
828 	 * If the directory/name pair is found in the name cache,
829 	 * we have to ensure the directory has not changed from
830 	 * the time the cache entry has been created. If it has,
831 	 * the cache entry has to be ignored
832 	 */
833 	if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
834 		struct vattr vattr;
835 		int err2;
836 
837 		if (error && error != ENOENT) {
838 			*vpp = NULLVP;
839 			return error;
840 		}
841 
842 		if (cnp->cn_flags & PDIRUNLOCK) {
843 			err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
844 			if (err2 != 0) {
845 				*vpp = NULLVP;
846 				return err2;
847 			}
848 			cnp->cn_flags &= ~PDIRUNLOCK;
849 		}
850 
851 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
852 		if (err2 != 0) {
853 			if (error == 0) {
854 				if (*vpp != dvp)
855 					vput(*vpp);
856 				else
857 					vrele(*vpp);
858 			}
859 			*vpp = NULLVP;
860 			return err2;
861 		}
862 
863 		if (error == ENOENT) {
864 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
865 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
866 			    VTONFS(dvp)->n_nctime)
867 				return ENOENT;
868 			cache_purge(dvp);
869 			np->n_nctime = 0;
870 			goto dorpc;
871 		}
872 
873 		newvp = *vpp;
874 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
875 			&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
876 		{
877 			nfsstats.lookupcache_hits++;
878 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
879 				cnp->cn_flags |= SAVENAME;
880 			if ((!lockparent || !(flags & ISLASTCN)) &&
881 			     newvp != dvp)
882 				VOP_UNLOCK(dvp, 0);
883 			return (0);
884 		}
885 		cache_purge(newvp);
886 		if (newvp != dvp)
887 			vput(newvp);
888 		else
889 			vrele(newvp);
890 		*vpp = NULLVP;
891 	}
892 dorpc:
893 	error = 0;
894 	newvp = NULLVP;
895 	nfsstats.lookupcache_misses++;
896 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
897 	len = cnp->cn_namelen;
898 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
899 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
900 	nfsm_fhtom(dvp, v3);
901 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
902 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
903 	if (error) {
904 		nfsm_postop_attr(dvp, attrflag);
905 		m_freem(mrep);
906 		goto nfsmout;
907 	}
908 	nfsm_getfh(fhp, fhsize, v3);
909 
910 	/*
911 	 * Handle RENAME case...
912 	 */
913 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
914 		if (NFS_CMPFH(np, fhp, fhsize)) {
915 			m_freem(mrep);
916 			return (EISDIR);
917 		}
918 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
919 		if (error) {
920 			m_freem(mrep);
921 			return error;
922 		}
923 		newvp = NFSTOV(np);
924 		if (v3) {
925 			nfsm_postop_attr(newvp, attrflag);
926 			nfsm_postop_attr(dvp, attrflag);
927 		} else
928 			nfsm_loadattr(newvp, (struct vattr *)0);
929 		*vpp = newvp;
930 		m_freem(mrep);
931 		cnp->cn_flags |= SAVENAME;
932 		if (!lockparent) {
933 			VOP_UNLOCK(dvp, 0);
934 			cnp->cn_flags |= PDIRUNLOCK;
935 		}
936 		return (0);
937 	}
938 
939 	/*
940 	 * The postop attr handling is duplicated for each if case,
941 	 * because it should be done while dvp is locked (unlocking
942 	 * dvp is different for each case).
943 	 */
944 
945 	if (NFS_CMPFH(np, fhp, fhsize)) {
946 		/*
947 		 * "." lookup
948 		 */
949 		VREF(dvp);
950 		newvp = dvp;
951 		if (v3) {
952 			nfsm_postop_attr(newvp, attrflag);
953 			nfsm_postop_attr(dvp, attrflag);
954 		} else
955 			nfsm_loadattr(newvp, (struct vattr *)0);
956 	} else if (flags & ISDOTDOT) {
957 		/*
958 		 * ".." lookup
959 		 */
960 		VOP_UNLOCK(dvp, 0);
961 		cnp->cn_flags |= PDIRUNLOCK;
962 
963 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
964 		if (error) {
965 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
966 				cnp->cn_flags &= ~PDIRUNLOCK;
967 			m_freem(mrep);
968 			return error;
969 		}
970 		newvp = NFSTOV(np);
971 
972 		if (v3) {
973 			nfsm_postop_attr(newvp, attrflag);
974 			nfsm_postop_attr(dvp, attrflag);
975 		} else
976 			nfsm_loadattr(newvp, (struct vattr *)0);
977 
978 		if (lockparent && (flags & ISLASTCN)) {
979 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
980 				m_freem(mrep);
981 				vput(newvp);
982 				return error;
983 			}
984 			cnp->cn_flags &= ~PDIRUNLOCK;
985 		}
986 	} else {
987 		/*
988 		 * Other lookups.
989 		 */
990 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
991 		if (error) {
992 			m_freem(mrep);
993 			return error;
994 		}
995 		newvp = NFSTOV(np);
996 		if (v3) {
997 			nfsm_postop_attr(newvp, attrflag);
998 			nfsm_postop_attr(dvp, attrflag);
999 		} else
1000 			nfsm_loadattr(newvp, (struct vattr *)0);
1001 		if (!lockparent || !(flags & ISLASTCN)) {
1002 			VOP_UNLOCK(dvp, 0);
1003 			cnp->cn_flags |= PDIRUNLOCK;
1004 		}
1005 	}
1006 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1007 		cnp->cn_flags |= SAVENAME;
1008 	if ((cnp->cn_flags & MAKEENTRY) &&
1009 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1010 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1011 		cache_enter(dvp, newvp, cnp);
1012 	}
1013 	*vpp = newvp;
1014 	nfsm_reqdone;
1015 	if (error) {
1016 		/*
1017 		 * We get here only because of errors returned by
1018 		 * the RPC. Otherwise we'll have returned above
1019 		 * (the nfsm_* macros will jump to nfsm_reqdone
1020 		 * on error).
1021 		 */
1022 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1023 		    cnp->cn_nameiop != CREATE) {
1024 			if (VTONFS(dvp)->n_nctime == 0)
1025 				VTONFS(dvp)->n_nctime =
1026 				    VTONFS(dvp)->n_vattr->va_mtime.tv_sec;
1027 			cache_enter(dvp, NULL, cnp);
1028 		}
1029 		if (newvp != NULLVP) {
1030 			vrele(newvp);
1031 			if (newvp != dvp)
1032 				VOP_UNLOCK(newvp, 0);
1033 		}
1034 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1035 		    (flags & ISLASTCN) && error == ENOENT) {
1036 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1037 				error = EROFS;
1038 			else
1039 				error = EJUSTRETURN;
1040 		}
1041 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1042 			cnp->cn_flags |= SAVENAME;
1043 		*vpp = NULL;
1044 	}
1045 
1046 	return error;
1047 }
1048 
1049 /*
1050  * nfs read call.
1051  * Just call nfs_bioread() to do the work.
1052  */
1053 int
1054 nfs_read(v)
1055 	void *v;
1056 {
1057 	struct vop_read_args /* {
1058 		struct vnode *a_vp;
1059 		struct uio *a_uio;
1060 		int  a_ioflag;
1061 		struct ucred *a_cred;
1062 	} */ *ap = v;
1063 	struct vnode *vp = ap->a_vp;
1064 
1065 	if (vp->v_type != VREG)
1066 		return (EPERM);
1067 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1068 }
1069 
1070 /*
1071  * nfs readlink call
1072  */
1073 int
1074 nfs_readlink(v)
1075 	void *v;
1076 {
1077 	struct vop_readlink_args /* {
1078 		struct vnode *a_vp;
1079 		struct uio *a_uio;
1080 		struct ucred *a_cred;
1081 	} */ *ap = v;
1082 	struct vnode *vp = ap->a_vp;
1083 
1084 	if (vp->v_type != VLNK)
1085 		return (EPERM);
1086 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1087 }
1088 
1089 /*
1090  * Do a readlink rpc.
1091  * Called by nfs_doio() from below the buffer cache.
1092  */
1093 int
1094 nfs_readlinkrpc(vp, uiop, cred)
1095 	struct vnode *vp;
1096 	struct uio *uiop;
1097 	struct ucred *cred;
1098 {
1099 	u_int32_t *tl;
1100 	caddr_t cp;
1101 	int32_t t1, t2;
1102 	caddr_t bpos, dpos, cp2;
1103 	int error = 0, len, attrflag;
1104 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1105 	const int v3 = NFS_ISV3(vp);
1106 
1107 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1108 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1109 	nfsm_fhtom(vp, v3);
1110 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
1111 	if (v3)
1112 		nfsm_postop_attr(vp, attrflag);
1113 	if (!error) {
1114 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1115 		nfsm_mtouio(uiop, len);
1116 	}
1117 	nfsm_reqdone;
1118 	return (error);
1119 }
1120 
1121 /*
1122  * nfs read rpc call
1123  * Ditto above
1124  */
1125 int
1126 nfs_readrpc(vp, uiop)
1127 	struct vnode *vp;
1128 	struct uio *uiop;
1129 {
1130 	u_int32_t *tl;
1131 	caddr_t cp;
1132 	int32_t t1, t2;
1133 	caddr_t bpos, dpos, cp2;
1134 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1135 	struct nfsmount *nmp;
1136 	int error = 0, len, retlen, tsiz, eof, attrflag;
1137 	const int v3 = NFS_ISV3(vp);
1138 
1139 #ifndef nolint
1140 	eof = 0;
1141 #endif
1142 	nmp = VFSTONFS(vp->v_mount);
1143 	tsiz = uiop->uio_resid;
1144 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1145 		return (EFBIG);
1146 	while (tsiz > 0) {
1147 		nfsstats.rpccnt[NFSPROC_READ]++;
1148 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1149 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1150 		nfsm_fhtom(vp, v3);
1151 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1152 		if (v3) {
1153 			txdr_hyper(uiop->uio_offset, tl);
1154 			*(tl + 2) = txdr_unsigned(len);
1155 		} else {
1156 			*tl++ = txdr_unsigned(uiop->uio_offset);
1157 			*tl++ = txdr_unsigned(len);
1158 			*tl = 0;
1159 		}
1160 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp,
1161 			     VTONFS(vp)->n_rcred);
1162 		if (v3) {
1163 			nfsm_postop_attr(vp, attrflag);
1164 			if (error) {
1165 				m_freem(mrep);
1166 				goto nfsmout;
1167 			}
1168 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1169 			eof = fxdr_unsigned(int, *(tl + 1));
1170 		} else
1171 			nfsm_loadattr(vp, (struct vattr *)0);
1172 		nfsm_strsiz(retlen, nmp->nm_rsize);
1173 		nfsm_mtouio(uiop, retlen);
1174 		m_freem(mrep);
1175 		tsiz -= retlen;
1176 		if (v3) {
1177 			if (eof || retlen == 0)
1178 				tsiz = 0;
1179 		} else if (retlen < len)
1180 			tsiz = 0;
1181 	}
1182 nfsmout:
1183 	return (error);
1184 }
1185 
1186 /*
1187  * nfs write call
1188  */
1189 int
1190 nfs_writerpc(vp, uiop, iomode, must_commit)
1191 	struct vnode *vp;
1192 	struct uio *uiop;
1193 	int *iomode, *must_commit;
1194 {
1195 	u_int32_t *tl;
1196 	caddr_t cp;
1197 	int32_t t1, t2, backup;
1198 	caddr_t bpos, dpos, cp2;
1199 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1200 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1201 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1202 	const int v3 = NFS_ISV3(vp);
1203 	int committed = NFSV3WRITE_FILESYNC;
1204 
1205 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1206 		panic("writerpc readonly vp %p", vp);
1207 	}
1208 
1209 #ifndef DIAGNOSTIC
1210 	if (uiop->uio_iovcnt != 1)
1211 		panic("nfs: writerpc iovcnt > 1");
1212 #endif
1213 	*must_commit = 0;
1214 	tsiz = uiop->uio_resid;
1215 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1216 		return (EFBIG);
1217 	while (tsiz > 0) {
1218 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1219 		len = min(tsiz, nmp->nm_wsize);
1220 		nfsm_reqhead(vp, NFSPROC_WRITE,
1221 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1222 		nfsm_fhtom(vp, v3);
1223 		if (v3) {
1224 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1225 			txdr_hyper(uiop->uio_offset, tl);
1226 			tl += 2;
1227 			*tl++ = txdr_unsigned(len);
1228 			*tl++ = txdr_unsigned(*iomode);
1229 			*tl = txdr_unsigned(len);
1230 		} else {
1231 			u_int32_t x;
1232 
1233 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1234 			/* Set both "begin" and "current" to non-garbage. */
1235 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1236 			*tl++ = x;      /* "begin offset" */
1237 			*tl++ = x;      /* "current offset" */
1238 			x = txdr_unsigned(len);
1239 			*tl++ = x;      /* total to this offset */
1240 			*tl = x;        /* size of this write */
1241 
1242 		}
1243 		nfsm_uiotom(uiop, len);
1244 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp,
1245 			     VTONFS(vp)->n_wcred);
1246 		if (v3) {
1247 			wccflag = NFSV3_WCCCHK;
1248 			nfsm_wcc_data(vp, wccflag);
1249 			if (!error) {
1250 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1251 					+ NFSX_V3WRITEVERF);
1252 				rlen = fxdr_unsigned(int, *tl++);
1253 				if (rlen == 0) {
1254 					error = NFSERR_IO;
1255 					m_freem(mrep);
1256 					break;
1257 				} else if (rlen < len) {
1258 					backup = len - rlen;
1259 					(caddr_t)uiop->uio_iov->iov_base -=
1260 					    backup;
1261 					uiop->uio_iov->iov_len += backup;
1262 					uiop->uio_offset -= backup;
1263 					uiop->uio_resid += backup;
1264 					len = rlen;
1265 				}
1266 				commit = fxdr_unsigned(int, *tl++);
1267 
1268 				/*
1269 				 * Return the lowest committment level
1270 				 * obtained by any of the RPCs.
1271 				 */
1272 				if (committed == NFSV3WRITE_FILESYNC)
1273 					committed = commit;
1274 				else if (committed == NFSV3WRITE_DATASYNC &&
1275 					commit == NFSV3WRITE_UNSTABLE)
1276 					committed = commit;
1277 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1278 				    memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
1279 					NFSX_V3WRITEVERF);
1280 				    nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1281 				} else if (memcmp((caddr_t)tl,
1282 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1283 				    *must_commit = 1;
1284 				    memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
1285 					NFSX_V3WRITEVERF);
1286 				}
1287 			}
1288 		} else
1289 		    nfsm_loadattr(vp, (struct vattr *)0);
1290 		if (wccflag)
1291 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1292 		m_freem(mrep);
1293 		if (error)
1294 			break;
1295 		tsiz -= len;
1296 	}
1297 nfsmout:
1298 	*iomode = committed;
1299 	if (error)
1300 		uiop->uio_resid = tsiz;
1301 	return (error);
1302 }
1303 
1304 /*
1305  * nfs mknod rpc
1306  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1307  * mode set to specify the file type and the size field for rdev.
1308  */
1309 int
1310 nfs_mknodrpc(dvp, vpp, cnp, vap)
1311 	struct vnode *dvp;
1312 	struct vnode **vpp;
1313 	struct componentname *cnp;
1314 	struct vattr *vap;
1315 {
1316 	struct nfsv2_sattr *sp;
1317 	u_int32_t *tl;
1318 	caddr_t cp;
1319 	int32_t t1, t2;
1320 	struct vnode *newvp = (struct vnode *)0;
1321 	struct nfsnode *np;
1322 	char *cp2;
1323 	caddr_t bpos, dpos;
1324 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1325 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1326 	u_int32_t rdev;
1327 	const int v3 = NFS_ISV3(dvp);
1328 
1329 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1330 		rdev = txdr_unsigned(vap->va_rdev);
1331 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1332 		rdev = nfs_xdrneg1;
1333 	else {
1334 		VOP_ABORTOP(dvp, cnp);
1335 		vput(dvp);
1336 		return (EOPNOTSUPP);
1337 	}
1338 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1339 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1340 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1341 	nfsm_fhtom(dvp, v3);
1342 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1343 	if (v3) {
1344 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1345 		*tl++ = vtonfsv3_type(vap->va_type);
1346 		nfsm_v3attrbuild(vap, FALSE);
1347 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1348 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1349 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1350 			*tl = txdr_unsigned(minor(vap->va_rdev));
1351 		}
1352 	} else {
1353 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1354 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1355 		sp->sa_uid = nfs_xdrneg1;
1356 		sp->sa_gid = nfs_xdrneg1;
1357 		sp->sa_size = rdev;
1358 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1359 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1360 	}
1361 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1362 	if (!error) {
1363 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1364 		if (!gotvp) {
1365 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1366 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1367 			if (!error)
1368 				newvp = NFSTOV(np);
1369 		}
1370 	}
1371 	if (v3)
1372 		nfsm_wcc_data(dvp, wccflag);
1373 	nfsm_reqdone;
1374 	if (error) {
1375 		if (newvp)
1376 			vput(newvp);
1377 	} else {
1378 		if (cnp->cn_flags & MAKEENTRY)
1379 			cache_enter(dvp, newvp, cnp);
1380 		*vpp = newvp;
1381 	}
1382 	PNBUF_PUT(cnp->cn_pnbuf);
1383 	VTONFS(dvp)->n_flag |= NMODIFIED;
1384 	if (!wccflag)
1385 		VTONFS(dvp)->n_attrstamp = 0;
1386 	vput(dvp);
1387 	return (error);
1388 }
1389 
1390 /*
1391  * nfs mknod vop
1392  * just call nfs_mknodrpc() to do the work.
1393  */
1394 /* ARGSUSED */
1395 int
1396 nfs_mknod(v)
1397 	void *v;
1398 {
1399 	struct vop_mknod_args /* {
1400 		struct vnode *a_dvp;
1401 		struct vnode **a_vpp;
1402 		struct componentname *a_cnp;
1403 		struct vattr *a_vap;
1404 	} */ *ap = v;
1405 	struct vnode *newvp;
1406 	int error;
1407 
1408 	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1409 	if (!error)
1410 		vput(newvp);
1411 	return (error);
1412 }
1413 
1414 static u_long create_verf;
1415 /*
1416  * nfs file create call
1417  */
1418 int
1419 nfs_create(v)
1420 	void *v;
1421 {
1422 	struct vop_create_args /* {
1423 		struct vnode *a_dvp;
1424 		struct vnode **a_vpp;
1425 		struct componentname *a_cnp;
1426 		struct vattr *a_vap;
1427 	} */ *ap = v;
1428 	struct vnode *dvp = ap->a_dvp;
1429 	struct vattr *vap = ap->a_vap;
1430 	struct componentname *cnp = ap->a_cnp;
1431 	struct nfsv2_sattr *sp;
1432 	u_int32_t *tl;
1433 	caddr_t cp;
1434 	int32_t t1, t2;
1435 	struct nfsnode *np = (struct nfsnode *)0;
1436 	struct vnode *newvp = (struct vnode *)0;
1437 	caddr_t bpos, dpos, cp2;
1438 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1439 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1440 	const int v3 = NFS_ISV3(dvp);
1441 
1442 	/*
1443 	 * Oops, not for me..
1444 	 */
1445 	if (vap->va_type == VSOCK)
1446 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1447 
1448 #ifdef VA_EXCLUSIVE
1449 	if (vap->va_vaflags & VA_EXCLUSIVE)
1450 		fmode |= O_EXCL;
1451 #endif
1452 again:
1453 	error = 0;
1454 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1455 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1456 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1457 	nfsm_fhtom(dvp, v3);
1458 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1459 	if (v3) {
1460 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1461 		if (fmode & O_EXCL) {
1462 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1463 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1464 #ifdef INET
1465 			if (in_ifaddr.tqh_first)
1466 				*tl++ = in_ifaddr.tqh_first->ia_addr.sin_addr.s_addr;
1467 			else
1468 				*tl++ = create_verf;
1469 #else
1470 			*tl++ = create_verf;
1471 #endif
1472 			*tl = ++create_verf;
1473 		} else {
1474 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1475 			nfsm_v3attrbuild(vap, FALSE);
1476 		}
1477 	} else {
1478 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1479 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1480 		sp->sa_uid = nfs_xdrneg1;
1481 		sp->sa_gid = nfs_xdrneg1;
1482 		sp->sa_size = 0;
1483 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1484 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1485 	}
1486 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1487 	if (!error) {
1488 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1489 		if (!gotvp) {
1490 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1491 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1492 			if (!error)
1493 				newvp = NFSTOV(np);
1494 		}
1495 	}
1496 	if (v3)
1497 		nfsm_wcc_data(dvp, wccflag);
1498 	nfsm_reqdone;
1499 	if (error) {
1500 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1501 			fmode &= ~O_EXCL;
1502 			goto again;
1503 		}
1504 		if (newvp)
1505 			vput(newvp);
1506 	} else if (v3 && (fmode & O_EXCL))
1507 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1508 	if (!error) {
1509 		if (cnp->cn_flags & MAKEENTRY)
1510 			cache_enter(dvp, newvp, cnp);
1511 		*ap->a_vpp = newvp;
1512 	}
1513 	PNBUF_PUT(cnp->cn_pnbuf);
1514 	VTONFS(dvp)->n_flag |= NMODIFIED;
1515 	if (!wccflag)
1516 		VTONFS(dvp)->n_attrstamp = 0;
1517 	vput(dvp);
1518 	return (error);
1519 }
1520 
1521 /*
1522  * nfs file remove call
1523  * To try and make nfs semantics closer to ufs semantics, a file that has
1524  * other processes using the vnode is renamed instead of removed and then
1525  * removed later on the last close.
1526  * - If v_usecount > 1
1527  *	  If a rename is not already in the works
1528  *	     call nfs_sillyrename() to set it up
1529  *     else
1530  *	  do the remove rpc
1531  */
1532 int
1533 nfs_remove(v)
1534 	void *v;
1535 {
1536 	struct vop_remove_args /* {
1537 		struct vnodeop_desc *a_desc;
1538 		struct vnode * a_dvp;
1539 		struct vnode * a_vp;
1540 		struct componentname * a_cnp;
1541 	} */ *ap = v;
1542 	struct vnode *vp = ap->a_vp;
1543 	struct vnode *dvp = ap->a_dvp;
1544 	struct componentname *cnp = ap->a_cnp;
1545 	struct nfsnode *np = VTONFS(vp);
1546 	int error = 0;
1547 	struct vattr vattr;
1548 
1549 #ifndef DIAGNOSTIC
1550 	if ((cnp->cn_flags & HASBUF) == 0)
1551 		panic("nfs_remove: no name");
1552 	if (vp->v_usecount < 1)
1553 		panic("nfs_remove: bad v_usecount");
1554 #endif
1555 	if (vp->v_type == VDIR)
1556 		error = EPERM;
1557 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1558 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1559 	    vattr.va_nlink > 1)) {
1560 		/*
1561 		 * Purge the name cache so that the chance of a lookup for
1562 		 * the name succeeding while the remove is in progress is
1563 		 * minimized. Without node locking it can still happen, such
1564 		 * that an I/O op returns ESTALE, but since you get this if
1565 		 * another host removes the file..
1566 		 */
1567 		cache_purge(vp);
1568 		/*
1569 		 * throw away biocache buffers, mainly to avoid
1570 		 * unnecessary delayed writes later.
1571 		 */
1572 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1573 		/* Do the rpc */
1574 		if (error != EINTR)
1575 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1576 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1577 		/*
1578 		 * Kludge City: If the first reply to the remove rpc is lost..
1579 		 *   the reply to the retransmitted request will be ENOENT
1580 		 *   since the file was in fact removed
1581 		 *   Therefore, we cheat and return success.
1582 		 */
1583 		if (error == ENOENT)
1584 			error = 0;
1585 	} else if (!np->n_sillyrename)
1586 		error = nfs_sillyrename(dvp, vp, cnp);
1587 	PNBUF_PUT(cnp->cn_pnbuf);
1588 	np->n_attrstamp = 0;
1589 	vput(dvp);
1590 	vput(vp);
1591 	return (error);
1592 }
1593 
1594 /*
1595  * nfs file remove rpc called from nfs_inactive
1596  */
1597 int
1598 nfs_removeit(sp)
1599 	struct sillyrename *sp;
1600 {
1601 
1602 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1603 		(struct proc *)0));
1604 }
1605 
1606 /*
1607  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1608  */
1609 int
1610 nfs_removerpc(dvp, name, namelen, cred, proc)
1611 	struct vnode *dvp;
1612 	const char *name;
1613 	int namelen;
1614 	struct ucred *cred;
1615 	struct proc *proc;
1616 {
1617 	u_int32_t *tl;
1618 	caddr_t cp;
1619 	int32_t t1, t2;
1620 	caddr_t bpos, dpos, cp2;
1621 	int error = 0, wccflag = NFSV3_WCCRATTR;
1622 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1623 	const int v3 = NFS_ISV3(dvp);
1624 
1625 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1626 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1627 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1628 	nfsm_fhtom(dvp, v3);
1629 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1630 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1631 	if (v3)
1632 		nfsm_wcc_data(dvp, wccflag);
1633 	nfsm_reqdone;
1634 	VTONFS(dvp)->n_flag |= NMODIFIED;
1635 	if (!wccflag)
1636 		VTONFS(dvp)->n_attrstamp = 0;
1637 	return (error);
1638 }
1639 
1640 /*
1641  * nfs file rename call
1642  */
1643 int
1644 nfs_rename(v)
1645 	void *v;
1646 {
1647 	struct vop_rename_args  /* {
1648 		struct vnode *a_fdvp;
1649 		struct vnode *a_fvp;
1650 		struct componentname *a_fcnp;
1651 		struct vnode *a_tdvp;
1652 		struct vnode *a_tvp;
1653 		struct componentname *a_tcnp;
1654 	} */ *ap = v;
1655 	struct vnode *fvp = ap->a_fvp;
1656 	struct vnode *tvp = ap->a_tvp;
1657 	struct vnode *fdvp = ap->a_fdvp;
1658 	struct vnode *tdvp = ap->a_tdvp;
1659 	struct componentname *tcnp = ap->a_tcnp;
1660 	struct componentname *fcnp = ap->a_fcnp;
1661 	int error;
1662 
1663 #ifndef DIAGNOSTIC
1664 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1665 	    (fcnp->cn_flags & HASBUF) == 0)
1666 		panic("nfs_rename: no name");
1667 #endif
1668 	/* Check for cross-device rename */
1669 	if ((fvp->v_mount != tdvp->v_mount) ||
1670 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1671 		error = EXDEV;
1672 		goto out;
1673 	}
1674 
1675 	/*
1676 	 * If the tvp exists and is in use, sillyrename it before doing the
1677 	 * rename of the new file over it.
1678 	 */
1679 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1680 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1681 		vput(tvp);
1682 		tvp = NULL;
1683 	}
1684 
1685 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1686 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1687 		tcnp->cn_proc);
1688 
1689 	if (fvp->v_type == VDIR) {
1690 		if (tvp != NULL && tvp->v_type == VDIR)
1691 			cache_purge(tdvp);
1692 		cache_purge(fdvp);
1693 	}
1694 out:
1695 	if (tdvp == tvp)
1696 		vrele(tdvp);
1697 	else
1698 		vput(tdvp);
1699 	if (tvp)
1700 		vput(tvp);
1701 	vrele(fdvp);
1702 	vrele(fvp);
1703 	/*
1704 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1705 	 */
1706 	if (error == ENOENT)
1707 		error = 0;
1708 	return (error);
1709 }
1710 
1711 /*
1712  * nfs file rename rpc called from nfs_remove() above
1713  */
1714 int
1715 nfs_renameit(sdvp, scnp, sp)
1716 	struct vnode *sdvp;
1717 	struct componentname *scnp;
1718 	struct sillyrename *sp;
1719 {
1720 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1721 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1722 }
1723 
1724 /*
1725  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1726  */
1727 int
1728 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1729 	struct vnode *fdvp;
1730 	const char *fnameptr;
1731 	int fnamelen;
1732 	struct vnode *tdvp;
1733 	const char *tnameptr;
1734 	int tnamelen;
1735 	struct ucred *cred;
1736 	struct proc *proc;
1737 {
1738 	u_int32_t *tl;
1739 	caddr_t cp;
1740 	int32_t t1, t2;
1741 	caddr_t bpos, dpos, cp2;
1742 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1743 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1744 	const int v3 = NFS_ISV3(fdvp);
1745 
1746 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1747 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1748 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1749 		nfsm_rndup(tnamelen));
1750 	nfsm_fhtom(fdvp, v3);
1751 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1752 	nfsm_fhtom(tdvp, v3);
1753 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1754 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1755 	if (v3) {
1756 		nfsm_wcc_data(fdvp, fwccflag);
1757 		nfsm_wcc_data(tdvp, twccflag);
1758 	}
1759 	nfsm_reqdone;
1760 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1761 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1762 	if (!fwccflag)
1763 		VTONFS(fdvp)->n_attrstamp = 0;
1764 	if (!twccflag)
1765 		VTONFS(tdvp)->n_attrstamp = 0;
1766 	return (error);
1767 }
1768 
1769 /*
1770  * nfs hard link create call
1771  */
1772 int
1773 nfs_link(v)
1774 	void *v;
1775 {
1776 	struct vop_link_args /* {
1777 		struct vnode *a_dvp;
1778 		struct vnode *a_vp;
1779 		struct componentname *a_cnp;
1780 	} */ *ap = v;
1781 	struct vnode *vp = ap->a_vp;
1782 	struct vnode *dvp = ap->a_dvp;
1783 	struct componentname *cnp = ap->a_cnp;
1784 	u_int32_t *tl;
1785 	caddr_t cp;
1786 	int32_t t1, t2;
1787 	caddr_t bpos, dpos, cp2;
1788 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1789 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1790 	/* XXX Should be const and initialised? */
1791 	int v3;
1792 
1793 	if (dvp->v_mount != vp->v_mount) {
1794 		VOP_ABORTOP(vp, cnp);
1795 		vput(dvp);
1796 		return (EXDEV);
1797 	}
1798 
1799 	/*
1800 	 * Push all writes to the server, so that the attribute cache
1801 	 * doesn't get "out of sync" with the server.
1802 	 * XXX There should be a better way!
1803 	 */
1804 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1805 
1806 	v3 = NFS_ISV3(vp);
1807 	nfsstats.rpccnt[NFSPROC_LINK]++;
1808 	nfsm_reqhead(vp, NFSPROC_LINK,
1809 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1810 	nfsm_fhtom(vp, v3);
1811 	nfsm_fhtom(dvp, v3);
1812 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1813 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1814 	if (v3) {
1815 		nfsm_postop_attr(vp, attrflag);
1816 		nfsm_wcc_data(dvp, wccflag);
1817 	}
1818 	nfsm_reqdone;
1819 	PNBUF_PUT(cnp->cn_pnbuf);
1820 	VTONFS(dvp)->n_flag |= NMODIFIED;
1821 	if (!attrflag)
1822 		VTONFS(vp)->n_attrstamp = 0;
1823 	if (!wccflag)
1824 		VTONFS(dvp)->n_attrstamp = 0;
1825 	vput(dvp);
1826 	/*
1827 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1828 	 */
1829 	if (error == EEXIST)
1830 		error = 0;
1831 	return (error);
1832 }
1833 
1834 /*
1835  * nfs symbolic link create call
1836  */
1837 int
1838 nfs_symlink(v)
1839 	void *v;
1840 {
1841 	struct vop_symlink_args /* {
1842 		struct vnode *a_dvp;
1843 		struct vnode **a_vpp;
1844 		struct componentname *a_cnp;
1845 		struct vattr *a_vap;
1846 		char *a_target;
1847 	} */ *ap = v;
1848 	struct vnode *dvp = ap->a_dvp;
1849 	struct vattr *vap = ap->a_vap;
1850 	struct componentname *cnp = ap->a_cnp;
1851 	struct nfsv2_sattr *sp;
1852 	u_int32_t *tl;
1853 	caddr_t cp;
1854 	int32_t t1, t2;
1855 	caddr_t bpos, dpos, cp2;
1856 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1857 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1858 	struct vnode *newvp = (struct vnode *)0;
1859 	const int v3 = NFS_ISV3(dvp);
1860 
1861 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1862 	slen = strlen(ap->a_target);
1863 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1864 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1865 	nfsm_fhtom(dvp, v3);
1866 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1867 	if (v3)
1868 		nfsm_v3attrbuild(vap, FALSE);
1869 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1870 	if (!v3) {
1871 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1872 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1873 		sp->sa_uid = nfs_xdrneg1;
1874 		sp->sa_gid = nfs_xdrneg1;
1875 		sp->sa_size = nfs_xdrneg1;
1876 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1877 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1878 	}
1879 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1880 	if (v3) {
1881 		if (!error)
1882 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1883 		nfsm_wcc_data(dvp, wccflag);
1884 	}
1885 	nfsm_reqdone;
1886 	if (newvp)
1887 		vput(newvp);
1888 	PNBUF_PUT(cnp->cn_pnbuf);
1889 	VTONFS(dvp)->n_flag |= NMODIFIED;
1890 	if (!wccflag)
1891 		VTONFS(dvp)->n_attrstamp = 0;
1892 	vput(dvp);
1893 	/*
1894 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1895 	 */
1896 	if (error == EEXIST)
1897 		error = 0;
1898 	return (error);
1899 }
1900 
1901 /*
1902  * nfs make dir call
1903  */
1904 int
1905 nfs_mkdir(v)
1906 	void *v;
1907 {
1908 	struct vop_mkdir_args /* {
1909 		struct vnode *a_dvp;
1910 		struct vnode **a_vpp;
1911 		struct componentname *a_cnp;
1912 		struct vattr *a_vap;
1913 	} */ *ap = v;
1914 	struct vnode *dvp = ap->a_dvp;
1915 	struct vattr *vap = ap->a_vap;
1916 	struct componentname *cnp = ap->a_cnp;
1917 	struct nfsv2_sattr *sp;
1918 	u_int32_t *tl;
1919 	caddr_t cp;
1920 	int32_t t1, t2;
1921 	int len;
1922 	struct nfsnode *np = (struct nfsnode *)0;
1923 	struct vnode *newvp = (struct vnode *)0;
1924 	caddr_t bpos, dpos, cp2;
1925 	int error = 0, wccflag = NFSV3_WCCRATTR;
1926 	int gotvp = 0;
1927 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1928 	const int v3 = NFS_ISV3(dvp);
1929 
1930 	len = cnp->cn_namelen;
1931 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1932 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1933 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1934 	nfsm_fhtom(dvp, v3);
1935 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1936 	if (v3) {
1937 		nfsm_v3attrbuild(vap, FALSE);
1938 	} else {
1939 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1940 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1941 		sp->sa_uid = nfs_xdrneg1;
1942 		sp->sa_gid = nfs_xdrneg1;
1943 		sp->sa_size = nfs_xdrneg1;
1944 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1945 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1946 	}
1947 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1948 	if (!error)
1949 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1950 	if (v3)
1951 		nfsm_wcc_data(dvp, wccflag);
1952 	nfsm_reqdone;
1953 	VTONFS(dvp)->n_flag |= NMODIFIED;
1954 	if (!wccflag)
1955 		VTONFS(dvp)->n_attrstamp = 0;
1956 	/*
1957 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1958 	 * if we can succeed in looking up the directory.
1959 	 */
1960 	if (error == EEXIST || (!error && !gotvp)) {
1961 		if (newvp) {
1962 			vput(newvp);
1963 			newvp = (struct vnode *)0;
1964 		}
1965 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1966 			cnp->cn_proc, &np);
1967 		if (!error) {
1968 			newvp = NFSTOV(np);
1969 			if (newvp->v_type != VDIR)
1970 				error = EEXIST;
1971 		}
1972 	}
1973 	if (error) {
1974 		if (newvp)
1975 			vput(newvp);
1976 	} else {
1977 		if (cnp->cn_flags & MAKEENTRY)
1978 			cache_enter(dvp, newvp, cnp);
1979 		*ap->a_vpp = newvp;
1980 	}
1981 	PNBUF_PUT(cnp->cn_pnbuf);
1982 	vput(dvp);
1983 	return (error);
1984 }
1985 
1986 /*
1987  * nfs remove directory call
1988  */
1989 int
1990 nfs_rmdir(v)
1991 	void *v;
1992 {
1993 	struct vop_rmdir_args /* {
1994 		struct vnode *a_dvp;
1995 		struct vnode *a_vp;
1996 		struct componentname *a_cnp;
1997 	} */ *ap = v;
1998 	struct vnode *vp = ap->a_vp;
1999 	struct vnode *dvp = ap->a_dvp;
2000 	struct componentname *cnp = ap->a_cnp;
2001 	u_int32_t *tl;
2002 	caddr_t cp;
2003 	int32_t t1, t2;
2004 	caddr_t bpos, dpos, cp2;
2005 	int error = 0, wccflag = NFSV3_WCCRATTR;
2006 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2007 	const int v3 = NFS_ISV3(dvp);
2008 
2009 	if (dvp == vp) {
2010 		vrele(dvp);
2011 		vput(dvp);
2012 		PNBUF_PUT(cnp->cn_pnbuf);
2013 		return (EINVAL);
2014 	}
2015 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2016 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
2017 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2018 	nfsm_fhtom(dvp, v3);
2019 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2020 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2021 	if (v3)
2022 		nfsm_wcc_data(dvp, wccflag);
2023 	nfsm_reqdone;
2024 	PNBUF_PUT(cnp->cn_pnbuf);
2025 	VTONFS(dvp)->n_flag |= NMODIFIED;
2026 	if (!wccflag)
2027 		VTONFS(dvp)->n_attrstamp = 0;
2028 	cache_purge(dvp);
2029 	cache_purge(vp);
2030 	vput(vp);
2031 	vput(dvp);
2032 	/*
2033 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2034 	 */
2035 	if (error == ENOENT)
2036 		error = 0;
2037 	return (error);
2038 }
2039 
2040 /*
2041  * nfs readdir call
2042  */
2043 int
2044 nfs_readdir(v)
2045 	void *v;
2046 {
2047 	struct vop_readdir_args /* {
2048 		struct vnode *a_vp;
2049 		struct uio *a_uio;
2050 		struct ucred *a_cred;
2051 		int *a_eofflag;
2052 		off_t **a_cookies;
2053 		int *a_ncookies;
2054 	} */ *ap = v;
2055 	struct vnode *vp = ap->a_vp;
2056 	struct uio *uio = ap->a_uio;
2057 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2058 	char *base = uio->uio_iov->iov_base;
2059 	int tresid, error;
2060 	size_t count, lost;
2061 	struct dirent *dp;
2062 	off_t *cookies = NULL;
2063 	int ncookies = 0, nc;
2064 
2065 	if (vp->v_type != VDIR)
2066 		return (EPERM);
2067 
2068 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2069 	count = uio->uio_resid - lost;
2070 	if (count <= 0)
2071 		return (EINVAL);
2072 
2073 	/*
2074 	 * Call nfs_bioread() to do the real work.
2075 	 */
2076 	tresid = uio->uio_resid = count;
2077 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2078 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2079 
2080 	if (!error && ap->a_cookies) {
2081 		ncookies = count / 16;
2082 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2083 		*ap->a_cookies = cookies;
2084 	}
2085 
2086 	if (!error && uio->uio_resid == tresid) {
2087 		uio->uio_resid += lost;
2088 		nfsstats.direofcache_misses++;
2089 		if (ap->a_cookies)
2090 			*ap->a_ncookies = 0;
2091 		*ap->a_eofflag = 1;
2092 		return (0);
2093 	}
2094 
2095 	if (!error && ap->a_cookies) {
2096 		/*
2097 		 * Only the NFS server and emulations use cookies, and they
2098 		 * load the directory block into system space, so we can
2099 		 * just look at it directly.
2100 		 */
2101 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2102 			panic("nfs_readdir: lost in space");
2103 		for (nc = 0; ncookies-- &&
2104 		     base < (char *)uio->uio_iov->iov_base; nc++){
2105 			dp = (struct dirent *) base;
2106 			if (dp->d_reclen == 0)
2107 				break;
2108 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2109 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2110 			else
2111 				*(cookies++) = NFS_GETCOOKIE(dp);
2112 			base += dp->d_reclen;
2113 		}
2114 		uio->uio_resid +=
2115 		    ((caddr_t)uio->uio_iov->iov_base - base);
2116 		uio->uio_iov->iov_len +=
2117 		    ((caddr_t)uio->uio_iov->iov_base - base);
2118 		uio->uio_iov->iov_base = base;
2119 		*ap->a_ncookies = nc;
2120 	}
2121 
2122 	uio->uio_resid += lost;
2123 	*ap->a_eofflag = 0;
2124 	return (error);
2125 }
2126 
2127 /*
2128  * Readdir rpc call.
2129  * Called from below the buffer cache by nfs_doio().
2130  */
2131 int
2132 nfs_readdirrpc(vp, uiop, cred)
2133 	struct vnode *vp;
2134 	struct uio *uiop;
2135 	struct ucred *cred;
2136 {
2137 	int len, left;
2138 	struct dirent *dp = NULL;
2139 	u_int32_t *tl;
2140 	caddr_t cp;
2141 	int32_t t1, t2;
2142 	caddr_t bpos, dpos, cp2;
2143 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2144 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2145 	struct nfsnode *dnp = VTONFS(vp);
2146 	u_quad_t fileno;
2147 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2148 	int attrflag, nrpcs = 0, reclen;
2149 	const int v3 = NFS_ISV3(vp);
2150 	nfsquad_t cookie;
2151 
2152 #ifdef DIAGNOSTIC
2153 	/*
2154 	 * Should be called from buffer cache, so only amount of
2155 	 * NFS_DIRBLKSIZ will be requested.
2156 	 */
2157 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2158 		panic("nfs readdirrpc bad uio");
2159 #endif
2160 
2161 	/*
2162 	 * Loop around doing readdir rpc's of size nm_readdirsize
2163 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2164 	 * The stopping criteria is EOF or buffer full.
2165 	 */
2166 	while (more_dirs && bigenough) {
2167 		/*
2168 		 * Heuristic: don't bother to do another RPC to further
2169 		 * fill up this block if there is not much room left. (< 50%
2170 		 * of the readdir RPC size). This wastes some buffer space
2171 		 * but can save up to 50% in RPC calls.
2172 		 */
2173 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2174 			bigenough = 0;
2175 			break;
2176 		}
2177 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2178 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2179 			NFSX_READDIR(v3));
2180 		nfsm_fhtom(vp, v3);
2181 		if (v3) {
2182 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2183 			cookie.qval = uiop->uio_offset;
2184 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2185 				txdr_swapcookie3(uiop->uio_offset, tl);
2186 			} else {
2187 				txdr_cookie3(uiop->uio_offset, tl);
2188 			}
2189 			tl += 2;
2190 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2191 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2192 		} else {
2193 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2194 			*tl++ = txdr_unsigned(uiop->uio_offset);
2195 		}
2196 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2197 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2198 		nrpcs++;
2199 		if (v3) {
2200 			nfsm_postop_attr(vp, attrflag);
2201 			if (!error) {
2202 				nfsm_dissect(tl, u_int32_t *,
2203 				    2 * NFSX_UNSIGNED);
2204 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2205 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2206 			} else {
2207 				m_freem(mrep);
2208 				goto nfsmout;
2209 			}
2210 		}
2211 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2212 		more_dirs = fxdr_unsigned(int, *tl);
2213 
2214 		/* loop thru the dir entries, doctoring them to 4bsd form */
2215 		while (more_dirs && bigenough) {
2216 			if (v3) {
2217 				nfsm_dissect(tl, u_int32_t *,
2218 				    3 * NFSX_UNSIGNED);
2219 				fileno = fxdr_hyper(tl);
2220 				len = fxdr_unsigned(int, *(tl + 2));
2221 			} else {
2222 				nfsm_dissect(tl, u_int32_t *,
2223 				    2 * NFSX_UNSIGNED);
2224 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2225 				len = fxdr_unsigned(int, *tl);
2226 			}
2227 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2228 				error = EBADRPC;
2229 				m_freem(mrep);
2230 				goto nfsmout;
2231 			}
2232 			tlen = nfsm_rndup(len);
2233 			if (tlen == len)
2234 				tlen += 4;	/* To ensure null termination */
2235 			tlen += sizeof (off_t) + sizeof (int);
2236 			reclen = ALIGN(tlen + DIRHDSIZ);
2237 			tlen = reclen - DIRHDSIZ;
2238 			left = NFS_DIRFRAGSIZ - blksiz;
2239 			if (reclen > left) {
2240 				dp->d_reclen += left;
2241 				(caddr_t)uiop->uio_iov->iov_base += left;
2242 				uiop->uio_iov->iov_len -= left;
2243 				uiop->uio_resid -= left;
2244 				blksiz = 0;
2245 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2246 			}
2247 			if (reclen > uiop->uio_resid)
2248 				bigenough = 0;
2249 			if (bigenough) {
2250 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2251 				dp->d_fileno = (int)fileno;
2252 				dp->d_namlen = len;
2253 				dp->d_reclen = reclen;
2254 				dp->d_type = DT_UNKNOWN;
2255 				blksiz += dp->d_reclen;
2256 				if (blksiz == NFS_DIRFRAGSIZ)
2257 					blksiz = 0;
2258 				uiop->uio_resid -= DIRHDSIZ;
2259 				(caddr_t)uiop->uio_iov->iov_base += DIRHDSIZ;
2260 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2261 				nfsm_mtouio(uiop, len);
2262 				cp = uiop->uio_iov->iov_base;
2263 				tlen -= len;
2264 				*cp = '\0';	/* null terminate */
2265 				(caddr_t)uiop->uio_iov->iov_base += tlen;
2266 				uiop->uio_iov->iov_len -= tlen;
2267 				uiop->uio_resid -= tlen;
2268 			} else
2269 				nfsm_adv(nfsm_rndup(len));
2270 			if (v3) {
2271 				nfsm_dissect(tl, u_int32_t *,
2272 				    3 * NFSX_UNSIGNED);
2273 			} else {
2274 				nfsm_dissect(tl, u_int32_t *,
2275 				    2 * NFSX_UNSIGNED);
2276 			}
2277 			if (bigenough) {
2278 				if (v3) {
2279 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2280 						uiop->uio_offset =
2281 						    fxdr_swapcookie3(tl);
2282 					else
2283 						uiop->uio_offset =
2284 						    fxdr_cookie3(tl);
2285 				}
2286 				else {
2287 					uiop->uio_offset =
2288 					    fxdr_unsigned(off_t, *tl);
2289 				}
2290 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2291 			}
2292 			if (v3)
2293 				tl += 2;
2294 			else
2295 				tl++;
2296 			more_dirs = fxdr_unsigned(int, *tl);
2297 		}
2298 		/*
2299 		 * If at end of rpc data, get the eof boolean
2300 		 */
2301 		if (!more_dirs) {
2302 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2303 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2304 		}
2305 		m_freem(mrep);
2306 	}
2307 	/*
2308 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2309 	 * by increasing d_reclen for the last record.
2310 	 */
2311 	if (blksiz > 0) {
2312 		left = NFS_DIRFRAGSIZ - blksiz;
2313 		dp->d_reclen += left;
2314 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2315 		(caddr_t)uiop->uio_iov->iov_base += left;
2316 		uiop->uio_iov->iov_len -= left;
2317 		uiop->uio_resid -= left;
2318 	}
2319 
2320 	/*
2321 	 * We are now either at the end of the directory or have filled the
2322 	 * block.
2323 	 */
2324 	if (bigenough)
2325 		dnp->n_direofoffset = uiop->uio_offset;
2326 nfsmout:
2327 	return (error);
2328 }
2329 
2330 /*
2331  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2332  */
2333 int
2334 nfs_readdirplusrpc(vp, uiop, cred)
2335 	struct vnode *vp;
2336 	struct uio *uiop;
2337 	struct ucred *cred;
2338 {
2339 	int len, left;
2340 	struct dirent *dp = NULL;
2341 	u_int32_t *tl;
2342 	caddr_t cp;
2343 	int32_t t1, t2;
2344 	struct vnode *newvp;
2345 	caddr_t bpos, dpos, cp2;
2346 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2347 	struct nameidata nami, *ndp = &nami;
2348 	struct componentname *cnp = &ndp->ni_cnd;
2349 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2350 	struct nfsnode *dnp = VTONFS(vp), *np;
2351 	const unsigned char *hcp;
2352 	nfsfh_t *fhp;
2353 	u_quad_t fileno;
2354 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2355 	int attrflag, fhsize, nrpcs = 0, reclen;
2356 	struct nfs_fattr fattr, *fp;
2357 
2358 #ifdef DIAGNOSTIC
2359 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2360 		panic("nfs readdirplusrpc bad uio");
2361 #endif
2362 	ndp->ni_dvp = vp;
2363 	newvp = NULLVP;
2364 
2365 	/*
2366 	 * Loop around doing readdir rpc's of size nm_readdirsize
2367 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2368 	 * The stopping criteria is EOF or buffer full.
2369 	 */
2370 	while (more_dirs && bigenough) {
2371 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2372 			bigenough = 0;
2373 			break;
2374 		}
2375 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2376 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2377 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2378 		nfsm_fhtom(vp, 1);
2379  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2380 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2381 			txdr_swapcookie3(uiop->uio_offset, tl);
2382 		} else {
2383 			txdr_cookie3(uiop->uio_offset, tl);
2384 		}
2385 		tl += 2;
2386 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2387 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2388 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2389 		*tl = txdr_unsigned(nmp->nm_rsize);
2390 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2391 		nfsm_postop_attr(vp, attrflag);
2392 		if (error) {
2393 			m_freem(mrep);
2394 			goto nfsmout;
2395 		}
2396 		nrpcs++;
2397 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2398 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2399 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2400 		more_dirs = fxdr_unsigned(int, *tl);
2401 
2402 		/* loop thru the dir entries, doctoring them to 4bsd form */
2403 		while (more_dirs && bigenough) {
2404 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2405 			fileno = fxdr_hyper(tl);
2406 			len = fxdr_unsigned(int, *(tl + 2));
2407 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2408 				error = EBADRPC;
2409 				m_freem(mrep);
2410 				goto nfsmout;
2411 			}
2412 			tlen = nfsm_rndup(len);
2413 			if (tlen == len)
2414 				tlen += 4;	/* To ensure null termination*/
2415 			tlen += sizeof (off_t) + sizeof (int);
2416 			reclen = ALIGN(tlen + DIRHDSIZ);
2417 			tlen = reclen - DIRHDSIZ;
2418 			left = NFS_DIRFRAGSIZ - blksiz;
2419 			if (reclen > left) {
2420 				/*
2421 				 * DIRFRAGSIZ is aligned, no need to align
2422 				 * again here.
2423 				 */
2424 				dp->d_reclen += left;
2425 				(caddr_t)uiop->uio_iov->iov_base += left;
2426 				uiop->uio_iov->iov_len -= left;
2427 				uiop->uio_resid -= left;
2428 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2429 				blksiz = 0;
2430 			}
2431 			if (reclen > uiop->uio_resid)
2432 				bigenough = 0;
2433 			if (bigenough) {
2434 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2435 				dp->d_fileno = (int)fileno;
2436 				dp->d_namlen = len;
2437 				dp->d_reclen = reclen;
2438 				dp->d_type = DT_UNKNOWN;
2439 				blksiz += dp->d_reclen;
2440 				if (blksiz == NFS_DIRFRAGSIZ)
2441 					blksiz = 0;
2442 				uiop->uio_resid -= DIRHDSIZ;
2443 				(caddr_t)uiop->uio_iov->iov_base += DIRHDSIZ;
2444 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2445 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2446 				cnp->cn_namelen = len;
2447 				nfsm_mtouio(uiop, len);
2448 				cp = uiop->uio_iov->iov_base;
2449 				tlen -= len;
2450 				*cp = '\0';
2451 				(caddr_t)uiop->uio_iov->iov_base += tlen;
2452 				uiop->uio_iov->iov_len -= tlen;
2453 				uiop->uio_resid -= tlen;
2454 			} else
2455 				nfsm_adv(nfsm_rndup(len));
2456 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2457 			if (bigenough) {
2458 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2459 					uiop->uio_offset =
2460 						fxdr_swapcookie3(tl);
2461 				else
2462 					uiop->uio_offset =
2463 						fxdr_cookie3(tl);
2464 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2465 			}
2466 			tl += 2;
2467 
2468 			/*
2469 			 * Since the attributes are before the file handle
2470 			 * (sigh), we must skip over the attributes and then
2471 			 * come back and get them.
2472 			 */
2473 			attrflag = fxdr_unsigned(int, *tl);
2474 			if (attrflag) {
2475 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2476 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2477 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2478 			    doit = fxdr_unsigned(int, *tl);
2479 			    if (doit) {
2480 				nfsm_getfh(fhp, fhsize, 1);
2481 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2482 				    VREF(vp);
2483 				    newvp = vp;
2484 				    np = dnp;
2485 				} else {
2486 				    error = nfs_nget(vp->v_mount, fhp,
2487 					fhsize, &np);
2488 				    if (!error)
2489 					newvp = NFSTOV(np);
2490 				}
2491 				if (!error) {
2492 				    nfs_loadattrcache(&newvp, &fattr, 0);
2493 				    dp->d_type =
2494 				        IFTODT(VTTOIF(np->n_vattr->va_type));
2495 				    ndp->ni_vp = newvp;
2496 				    cnp->cn_hash = 0;
2497 				    for (hcp = cnp->cn_nameptr, i = 1; i <= len;
2498 				        i++, hcp++)
2499 				        cnp->cn_hash += *hcp * i;
2500 				    if (cnp->cn_namelen <= NCHNAMLEN)
2501 				        cache_enter(ndp->ni_dvp, ndp->ni_vp,
2502 						    cnp);
2503 				}
2504 			   }
2505 			} else {
2506 			    /* Just skip over the file handle */
2507 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2508 			    i = fxdr_unsigned(int, *tl);
2509 			    nfsm_adv(nfsm_rndup(i));
2510 			}
2511 			if (newvp != NULLVP) {
2512 			    vrele(newvp);
2513 			    if (newvp != vp)
2514 				VOP_UNLOCK(vp, 0);
2515 			    newvp = NULLVP;
2516 			}
2517 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2518 			more_dirs = fxdr_unsigned(int, *tl);
2519 		}
2520 		/*
2521 		 * If at end of rpc data, get the eof boolean
2522 		 */
2523 		if (!more_dirs) {
2524 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2525 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2526 		}
2527 		m_freem(mrep);
2528 	}
2529 	/*
2530 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2531 	 * by increasing d_reclen for the last record.
2532 	 */
2533 	if (blksiz > 0) {
2534 		left = NFS_DIRFRAGSIZ - blksiz;
2535 		dp->d_reclen += left;
2536 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2537 		(caddr_t)uiop->uio_iov->iov_base += left;
2538 		uiop->uio_iov->iov_len -= left;
2539 		uiop->uio_resid -= left;
2540 	}
2541 
2542 	/*
2543 	 * We are now either at the end of the directory or have filled the
2544 	 * block.
2545 	 */
2546 	if (bigenough)
2547 		dnp->n_direofoffset = uiop->uio_offset;
2548 nfsmout:
2549 	if (newvp != NULLVP) {
2550 		vrele(newvp);
2551 		if (newvp != vp)
2552 			VOP_UNLOCK(vp, 0);
2553 	}
2554 	return (error);
2555 }
2556 static char hextoasc[] = "0123456789abcdef";
2557 
2558 /*
2559  * Silly rename. To make the NFS filesystem that is stateless look a little
2560  * more like the "ufs" a remove of an active vnode is translated to a rename
2561  * to a funny looking filename that is removed by nfs_inactive on the
2562  * nfsnode. There is the potential for another process on a different client
2563  * to create the same funny name between the nfs_lookitup() fails and the
2564  * nfs_rename() completes, but...
2565  */
2566 int
2567 nfs_sillyrename(dvp, vp, cnp)
2568 	struct vnode *dvp, *vp;
2569 	struct componentname *cnp;
2570 {
2571 	struct sillyrename *sp;
2572 	struct nfsnode *np;
2573 	int error;
2574 	short pid;
2575 
2576 	cache_purge(dvp);
2577 	np = VTONFS(vp);
2578 #ifndef DIAGNOSTIC
2579 	if (vp->v_type == VDIR)
2580 		panic("nfs: sillyrename dir");
2581 #endif
2582 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2583 		M_NFSREQ, M_WAITOK);
2584 	sp->s_cred = crdup(cnp->cn_cred);
2585 	sp->s_dvp = dvp;
2586 	VREF(dvp);
2587 
2588 	/* Fudge together a funny name */
2589 	pid = cnp->cn_proc->p_pid;
2590 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2591 	sp->s_namlen = 12;
2592 	sp->s_name[8] = hextoasc[pid & 0xf];
2593 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2594 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2595 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2596 
2597 	/* Try lookitups until we get one that isn't there */
2598 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2599 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2600 		sp->s_name[4]++;
2601 		if (sp->s_name[4] > 'z') {
2602 			error = EINVAL;
2603 			goto bad;
2604 		}
2605 	}
2606 	error = nfs_renameit(dvp, cnp, sp);
2607 	if (error)
2608 		goto bad;
2609 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2610 		cnp->cn_proc, &np);
2611 	np->n_sillyrename = sp;
2612 	return (0);
2613 bad:
2614 	vrele(sp->s_dvp);
2615 	crfree(sp->s_cred);
2616 	free((caddr_t)sp, M_NFSREQ);
2617 	return (error);
2618 }
2619 
2620 /*
2621  * Look up a file name and optionally either update the file handle or
2622  * allocate an nfsnode, depending on the value of npp.
2623  * npp == NULL	--> just do the lookup
2624  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2625  *			handled too
2626  * *npp != NULL --> update the file handle in the vnode
2627  */
2628 int
2629 nfs_lookitup(dvp, name, len, cred, procp, npp)
2630 	struct vnode *dvp;
2631 	const char *name;
2632 	int len;
2633 	struct ucred *cred;
2634 	struct proc *procp;
2635 	struct nfsnode **npp;
2636 {
2637 	u_int32_t *tl;
2638 	caddr_t cp;
2639 	int32_t t1, t2;
2640 	struct vnode *newvp = (struct vnode *)0;
2641 	struct nfsnode *np, *dnp = VTONFS(dvp);
2642 	caddr_t bpos, dpos, cp2;
2643 	int error = 0, fhlen, attrflag;
2644 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2645 	nfsfh_t *nfhp;
2646 	const int v3 = NFS_ISV3(dvp);
2647 
2648 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2649 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2650 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2651 	nfsm_fhtom(dvp, v3);
2652 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2653 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2654 	if (npp && !error) {
2655 		nfsm_getfh(nfhp, fhlen, v3);
2656 		if (*npp) {
2657 		    np = *npp;
2658 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2659 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2660 			np->n_fhp = &np->n_fh;
2661 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2662 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2663 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2664 		    np->n_fhsize = fhlen;
2665 		    newvp = NFSTOV(np);
2666 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2667 		    VREF(dvp);
2668 		    newvp = dvp;
2669 		} else {
2670 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2671 		    if (error) {
2672 			m_freem(mrep);
2673 			return (error);
2674 		    }
2675 		    newvp = NFSTOV(np);
2676 		}
2677 		if (v3) {
2678 			nfsm_postop_attr(newvp, attrflag);
2679 			if (!attrflag && *npp == NULL) {
2680 				m_freem(mrep);
2681 				vput(newvp);
2682 				return (ENOENT);
2683 			}
2684 		} else
2685 			nfsm_loadattr(newvp, (struct vattr *)0);
2686 	}
2687 	nfsm_reqdone;
2688 	if (npp && *npp == NULL) {
2689 		if (error) {
2690 			if (newvp)
2691 				vput(newvp);
2692 		} else
2693 			*npp = np;
2694 	}
2695 	return (error);
2696 }
2697 
2698 /*
2699  * Nfs Version 3 commit rpc
2700  */
2701 int
2702 nfs_commit(vp, offset, cnt, procp)
2703 	struct vnode *vp;
2704 	off_t offset;
2705 	uint32_t cnt;
2706 	struct proc *procp;
2707 {
2708 	caddr_t cp;
2709 	u_int32_t *tl;
2710 	int32_t t1, t2;
2711 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2712 	caddr_t bpos, dpos, cp2;
2713 	int error = 0, wccflag = NFSV3_WCCRATTR;
2714 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2715 
2716 #ifdef fvdl_debug
2717 	printf("commit %lu - %lu\n", (unsigned long)offset,
2718 	    (unsigned long)(offset + cnt));
2719 #endif
2720 
2721 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0)
2722 		return (0);
2723 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2724 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2725 	nfsm_fhtom(vp, 1);
2726 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2727 	txdr_hyper(offset, tl);
2728 	tl += 2;
2729 	*tl = txdr_unsigned(cnt);
2730 	nfsm_request(vp, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2731 	nfsm_wcc_data(vp, wccflag);
2732 	if (!error) {
2733 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2734 		if (memcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2735 			NFSX_V3WRITEVERF)) {
2736 			memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
2737 				NFSX_V3WRITEVERF);
2738 			error = NFSERR_STALEWRITEVERF;
2739 		}
2740 	}
2741 	nfsm_reqdone;
2742 	return (error);
2743 }
2744 
2745 /*
2746  * Kludge City..
2747  * - make nfs_bmap() essentially a no-op that does no translation
2748  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2749  *   (Maybe I could use the process's page mapping, but I was concerned that
2750  *    Kernel Write might not be enabled and also figured copyout() would do
2751  *    a lot more work than memcpy() and also it currently happens in the
2752  *    context of the swapper process (2).
2753  */
2754 int
2755 nfs_bmap(v)
2756 	void *v;
2757 {
2758 	struct vop_bmap_args /* {
2759 		struct vnode *a_vp;
2760 		daddr_t  a_bn;
2761 		struct vnode **a_vpp;
2762 		daddr_t *a_bnp;
2763 		int *a_runp;
2764 	} */ *ap = v;
2765 	struct vnode *vp = ap->a_vp;
2766 
2767 	/*
2768 	 * XXX vpp should be returned unlocked?
2769 	 */
2770 	if (ap->a_vpp != NULL)
2771 		*ap->a_vpp = vp;
2772 	if (ap->a_bnp != NULL)
2773 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2774 	return (0);
2775 }
2776 
2777 /*
2778  * Strategy routine.
2779  * For async requests when nfsiod(s) are running, queue the request by
2780  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2781  * request.
2782  */
2783 int
2784 nfs_strategy(v)
2785 	void *v;
2786 {
2787 	struct vop_strategy_args *ap = v;
2788 	struct buf *bp = ap->a_bp;
2789 	struct proc *p;
2790 	int error = 0;
2791 
2792 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2793 		panic("nfs physio/async");
2794 	if (bp->b_flags & B_ASYNC)
2795 		p = NULL;
2796 	else
2797 		p = curproc;	/* XXX */
2798 
2799 	/*
2800 	 * If the op is asynchronous and an i/o daemon is waiting
2801 	 * queue the request, wake it up and wait for completion
2802 	 * otherwise just do it ourselves.
2803 	 */
2804 
2805 	if ((bp->b_flags & B_ASYNC) == 0 ||
2806 	    nfs_asyncio(bp))
2807 		error = nfs_doio(bp, p);
2808 	return (error);
2809 }
2810 
2811 /*
2812  * Mmap a file
2813  *
2814  * NB Currently unsupported.
2815  */
2816 /* ARGSUSED */
2817 int
2818 nfs_mmap(v)
2819 	void *v;
2820 {
2821 #if 0
2822 	struct vop_mmap_args /* {
2823 		struct vnode *a_vp;
2824 		int a_fflags;
2825 		struct ucred *a_cred;
2826 		struct proc *a_p;
2827 	} */ *ap = v;
2828 #endif
2829 
2830 	return (EINVAL);
2831 }
2832 
2833 /*
2834  * fsync vnode op. Just call nfs_flush() with commit == 1.
2835  */
2836 /* ARGSUSED */
2837 int
2838 nfs_fsync(v)
2839 	void *v;
2840 {
2841 	struct vop_fsync_args /* {
2842 		struct vnodeop_desc *a_desc;
2843 		struct vnode * a_vp;
2844 		struct ucred * a_cred;
2845 		int  a_flags;
2846 		off_t offlo;
2847 		off_t offhi;
2848 		struct proc * a_p;
2849 	} */ *ap = v;
2850 
2851 	return (nfs_flush(ap->a_vp, ap->a_cred,
2852 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
2853 }
2854 
2855 /*
2856  * Flush all the data associated with a vnode.
2857  */
2858 int
2859 nfs_flush(vp, cred, waitfor, p, commit)
2860 	struct vnode *vp;
2861 	struct ucred *cred;
2862 	int waitfor;
2863 	struct proc *p;
2864 	int commit;
2865 {
2866 	struct uvm_object *uobj = &vp->v_uvm.u_obj;
2867 	struct nfsnode *np = VTONFS(vp);
2868 	int error;
2869 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
2870 	int rv;
2871 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
2872 
2873 	error = 0;
2874 	simple_lock(&uobj->vmobjlock);
2875 	rv = (uobj->pgops->pgo_flush)(uobj, 0, 0, flushflags);
2876 	simple_unlock(&uobj->vmobjlock);
2877 	if (!rv) {
2878 		error = EIO;
2879 	}
2880 	if (np->n_flag & NWRITEERR) {
2881 		error = np->n_error;
2882 		np->n_flag &= ~NWRITEERR;
2883 	}
2884 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
2885 	return (error);
2886 }
2887 
2888 /*
2889  * Return POSIX pathconf information applicable to nfs.
2890  *
2891  * N.B. The NFS V2 protocol doesn't support this RPC.
2892  */
2893 /* ARGSUSED */
2894 int
2895 nfs_pathconf(v)
2896 	void *v;
2897 {
2898 	struct vop_pathconf_args /* {
2899 		struct vnode *a_vp;
2900 		int a_name;
2901 		register_t *a_retval;
2902 	} */ *ap = v;
2903 	struct nfsv3_pathconf *pcp;
2904 	struct vnode *vp = ap->a_vp;
2905 	struct nfsmount *nmp;
2906 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2907 	int32_t t1, t2;
2908 	u_int32_t *tl;
2909 	caddr_t bpos, dpos, cp, cp2;
2910 	int error = 0, attrflag;
2911 	unsigned int l;
2912 	u_int64_t maxsize;
2913 	const int v3 = NFS_ISV3(vp);
2914 
2915 	switch (ap->a_name) {
2916 		/* Names that can be resolved locally. */
2917 	case _PC_PIPE_BUF:
2918 		*ap->a_retval = PIPE_BUF;
2919 		break;
2920 	case _PC_SYNC_IO:
2921 		*ap->a_retval = 1;
2922 		break;
2923 	/* Names that cannot be resolved locally; do an RPC, if possible. */
2924 	case _PC_LINK_MAX:
2925 	case _PC_NAME_MAX:
2926 	case _PC_CHOWN_RESTRICTED:
2927 	case _PC_NO_TRUNC:
2928 		if (!v3) {
2929 			error = EINVAL;
2930 			break;
2931 		}
2932 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
2933 		nfsm_reqhead(vp, NFSPROC_PATHCONF, NFSX_FH(1));
2934 		nfsm_fhtom(vp, 1);
2935 		nfsm_request(vp, NFSPROC_PATHCONF,
2936 		    curproc, curproc->p_ucred);	/* XXX */
2937 		nfsm_postop_attr(vp, attrflag);
2938 		if (!error) {
2939 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
2940 			    NFSX_V3PATHCONF);
2941 			switch (ap->a_name) {
2942 			case _PC_LINK_MAX:
2943 				*ap->a_retval =
2944 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
2945 				break;
2946 			case _PC_NAME_MAX:
2947 				*ap->a_retval =
2948 				    fxdr_unsigned(register_t, pcp->pc_namemax);
2949 				break;
2950 			case _PC_CHOWN_RESTRICTED:
2951 				*ap->a_retval =
2952 				    (pcp->pc_chownrestricted == nfs_true);
2953 				break;
2954 			case _PC_NO_TRUNC:
2955 				*ap->a_retval =
2956 				    (pcp->pc_notrunc == nfs_true);
2957 				break;
2958 			}
2959 		}
2960 		nfsm_reqdone;
2961 		break;
2962 	case _PC_FILESIZEBITS:
2963 		if (v3) {
2964 			nmp = VFSTONFS(vp->v_mount);
2965 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
2966 				if ((error = nfs_fsinfo(nmp, vp,
2967 				    curproc->p_ucred, curproc)) != 0) /* XXX */
2968 					break;
2969 			for (l = 0, maxsize = nmp->nm_maxfilesize;
2970 			    (maxsize >> l) > 0; l++)
2971 				;
2972 			*ap->a_retval = l + 1;
2973 		} else {
2974 			*ap->a_retval = 32;	/* NFS V2 limitation */
2975 		}
2976 		break;
2977 	default:
2978 		error = EINVAL;
2979 		break;
2980 	}
2981 
2982 	return (error);
2983 }
2984 
2985 /*
2986  * NFS advisory byte-level locks.
2987  */
2988 int
2989 nfs_advlock(v)
2990 	void *v;
2991 {
2992 	struct vop_advlock_args /* {
2993 		struct vnode *a_vp;
2994 		caddr_t  a_id;
2995 		int  a_op;
2996 		struct flock *a_fl;
2997 		int  a_flags;
2998 	} */ *ap = v;
2999 	struct nfsnode *np = VTONFS(ap->a_vp);
3000 
3001 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3002 }
3003 
3004 /*
3005  * Print out the contents of an nfsnode.
3006  */
3007 int
3008 nfs_print(v)
3009 	void *v;
3010 {
3011 	struct vop_print_args /* {
3012 		struct vnode *a_vp;
3013 	} */ *ap = v;
3014 	struct vnode *vp = ap->a_vp;
3015 	struct nfsnode *np = VTONFS(vp);
3016 
3017 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3018 	    np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3019 	if (vp->v_type == VFIFO)
3020 		fifo_printinfo(vp);
3021 	printf("\n");
3022 	return (0);
3023 }
3024 
3025 /*
3026  * NFS file truncation.
3027  */
3028 int
3029 nfs_truncate(v)
3030 	void *v;
3031 {
3032 #if 0
3033 	struct vop_truncate_args /* {
3034 		struct vnode *a_vp;
3035 		off_t a_length;
3036 		int a_flags;
3037 		struct ucred *a_cred;
3038 		struct proc *a_p;
3039 	} */ *ap = v;
3040 #endif
3041 
3042 	/* Use nfs_setattr */
3043 	return (EOPNOTSUPP);
3044 }
3045 
3046 /*
3047  * NFS update.
3048  */
3049 int
3050 nfs_update(v)
3051 	void *v;
3052 #if 0
3053 	struct vop_update_args /* {
3054 		struct vnode *a_vp;
3055 		struct timespec *a_ta;
3056 		struct timespec *a_tm;
3057 		int a_waitfor;
3058 	} */ *ap = v;
3059 #endif
3060 {
3061 
3062 	/* Use nfs_setattr */
3063 	return (EOPNOTSUPP);
3064 }
3065 
3066 /*
3067  * Just call bwrite().
3068  */
3069 int
3070 nfs_bwrite(v)
3071 	void *v;
3072 {
3073 	struct vop_bwrite_args /* {
3074 		struct vnode *a_bp;
3075 	} */ *ap = v;
3076 
3077 	return (bwrite(ap->a_bp));
3078 }
3079 
3080 /*
3081  * nfs special file access vnode op.
3082  * Essentially just get vattr and then imitate iaccess() since the device is
3083  * local to the client.
3084  */
3085 int
3086 nfsspec_access(v)
3087 	void *v;
3088 {
3089 	struct vop_access_args /* {
3090 		struct vnode *a_vp;
3091 		int  a_mode;
3092 		struct ucred *a_cred;
3093 		struct proc *a_p;
3094 	} */ *ap = v;
3095 	struct vattr va;
3096 	struct vnode *vp = ap->a_vp;
3097 	int error;
3098 
3099 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3100 	if (error)
3101 		return (error);
3102 
3103         /*
3104 	 * Disallow write attempts on filesystems mounted read-only;
3105 	 * unless the file is a socket, fifo, or a block or character
3106 	 * device resident on the filesystem.
3107 	 */
3108 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3109 		switch (vp->v_type) {
3110 		case VREG:
3111 		case VDIR:
3112 		case VLNK:
3113 			return (EROFS);
3114 		default:
3115 			break;
3116 		}
3117 	}
3118 
3119 	return (vaccess(va.va_type, va.va_mode,
3120 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3121 }
3122 
3123 /*
3124  * Read wrapper for special devices.
3125  */
3126 int
3127 nfsspec_read(v)
3128 	void *v;
3129 {
3130 	struct vop_read_args /* {
3131 		struct vnode *a_vp;
3132 		struct uio *a_uio;
3133 		int  a_ioflag;
3134 		struct ucred *a_cred;
3135 	} */ *ap = v;
3136 	struct nfsnode *np = VTONFS(ap->a_vp);
3137 
3138 	/*
3139 	 * Set access flag.
3140 	 */
3141 	np->n_flag |= NACC;
3142 	np->n_atim.tv_sec = time.tv_sec;
3143 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3144 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3145 }
3146 
3147 /*
3148  * Write wrapper for special devices.
3149  */
3150 int
3151 nfsspec_write(v)
3152 	void *v;
3153 {
3154 	struct vop_write_args /* {
3155 		struct vnode *a_vp;
3156 		struct uio *a_uio;
3157 		int  a_ioflag;
3158 		struct ucred *a_cred;
3159 	} */ *ap = v;
3160 	struct nfsnode *np = VTONFS(ap->a_vp);
3161 
3162 	/*
3163 	 * Set update flag.
3164 	 */
3165 	np->n_flag |= NUPD;
3166 	np->n_mtim.tv_sec = time.tv_sec;
3167 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3168 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3169 }
3170 
3171 /*
3172  * Close wrapper for special devices.
3173  *
3174  * Update the times on the nfsnode then do device close.
3175  */
3176 int
3177 nfsspec_close(v)
3178 	void *v;
3179 {
3180 	struct vop_close_args /* {
3181 		struct vnode *a_vp;
3182 		int  a_fflag;
3183 		struct ucred *a_cred;
3184 		struct proc *a_p;
3185 	} */ *ap = v;
3186 	struct vnode *vp = ap->a_vp;
3187 	struct nfsnode *np = VTONFS(vp);
3188 	struct vattr vattr;
3189 
3190 	if (np->n_flag & (NACC | NUPD)) {
3191 		np->n_flag |= NCHG;
3192 		if (vp->v_usecount == 1 &&
3193 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3194 			VATTR_NULL(&vattr);
3195 			if (np->n_flag & NACC)
3196 				vattr.va_atime = np->n_atim;
3197 			if (np->n_flag & NUPD)
3198 				vattr.va_mtime = np->n_mtim;
3199 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3200 		}
3201 	}
3202 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3203 }
3204 
3205 /*
3206  * Read wrapper for fifos.
3207  */
3208 int
3209 nfsfifo_read(v)
3210 	void *v;
3211 {
3212 	struct vop_read_args /* {
3213 		struct vnode *a_vp;
3214 		struct uio *a_uio;
3215 		int  a_ioflag;
3216 		struct ucred *a_cred;
3217 	} */ *ap = v;
3218 	struct nfsnode *np = VTONFS(ap->a_vp);
3219 
3220 	/*
3221 	 * Set access flag.
3222 	 */
3223 	np->n_flag |= NACC;
3224 	np->n_atim.tv_sec = time.tv_sec;
3225 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3226 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3227 }
3228 
3229 /*
3230  * Write wrapper for fifos.
3231  */
3232 int
3233 nfsfifo_write(v)
3234 	void *v;
3235 {
3236 	struct vop_write_args /* {
3237 		struct vnode *a_vp;
3238 		struct uio *a_uio;
3239 		int  a_ioflag;
3240 		struct ucred *a_cred;
3241 	} */ *ap = v;
3242 	struct nfsnode *np = VTONFS(ap->a_vp);
3243 
3244 	/*
3245 	 * Set update flag.
3246 	 */
3247 	np->n_flag |= NUPD;
3248 	np->n_mtim.tv_sec = time.tv_sec;
3249 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3250 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3251 }
3252 
3253 /*
3254  * Close wrapper for fifos.
3255  *
3256  * Update the times on the nfsnode then do fifo close.
3257  */
3258 int
3259 nfsfifo_close(v)
3260 	void *v;
3261 {
3262 	struct vop_close_args /* {
3263 		struct vnode *a_vp;
3264 		int  a_fflag;
3265 		struct ucred *a_cred;
3266 		struct proc *a_p;
3267 	} */ *ap = v;
3268 	struct vnode *vp = ap->a_vp;
3269 	struct nfsnode *np = VTONFS(vp);
3270 	struct vattr vattr;
3271 
3272 	if (np->n_flag & (NACC | NUPD)) {
3273 		if (np->n_flag & NACC) {
3274 			np->n_atim.tv_sec = time.tv_sec;
3275 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3276 		}
3277 		if (np->n_flag & NUPD) {
3278 			np->n_mtim.tv_sec = time.tv_sec;
3279 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3280 		}
3281 		np->n_flag |= NCHG;
3282 		if (vp->v_usecount == 1 &&
3283 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3284 			VATTR_NULL(&vattr);
3285 			if (np->n_flag & NACC)
3286 				vattr.va_atime = np->n_atim;
3287 			if (np->n_flag & NUPD)
3288 				vattr.va_mtime = np->n_mtim;
3289 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3290 		}
3291 	}
3292 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3293 }
3294