xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 76dfffe33547c37f8bdd446e3e4ab0f3c16cea4b)
1 /*	$NetBSD: nfs_vnops.c,v 1.68 1996/10/25 23:14:12 cgd 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	8.16 (Berkeley) 5/27/95
39  */
40 
41 
42 /*
43  * vnode op calls for Sun NFS version 2 and 3
44  */
45 
46 #include <sys/param.h>
47 #include <sys/proc.h>
48 #include <sys/kernel.h>
49 #include <sys/systm.h>
50 #include <sys/resourcevar.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/buf.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/conf.h>
57 #include <sys/namei.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 
63 #include <vm/vm.h>
64 
65 #include <miscfs/fifofs/fifo.h>
66 #include <miscfs/genfs/genfs.h>
67 #include <miscfs/specfs/specdev.h>
68 
69 #include <nfs/rpcv2.h>
70 #include <nfs/nfsproto.h>
71 #include <nfs/nfs.h>
72 #include <nfs/nfsnode.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nqnfs.h>
77 #include <nfs/nfs_var.h>
78 
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 
83 /* Defs */
84 #define	TRUE	1
85 #define	FALSE	0
86 
87 /*
88  * Global vfs data structures for nfs
89  */
90 int (**nfsv2_vnodeop_p) __P((void *));
91 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
92 	{ &vop_default_desc, vn_default_error },
93 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
94 	{ &vop_create_desc, nfs_create },		/* create */
95 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
96 	{ &vop_open_desc, nfs_open },			/* open */
97 	{ &vop_close_desc, nfs_close },			/* close */
98 	{ &vop_access_desc, nfs_access },		/* access */
99 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
100 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
101 	{ &vop_read_desc, nfs_read },			/* read */
102 	{ &vop_write_desc, nfs_write },			/* write */
103 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
104 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
105 	{ &vop_poll_desc, nfs_poll },			/* poll */
106 #ifdef Lite2_integrated
107 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
108 #endif
109 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
110 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
111 	{ &vop_seek_desc, nfs_seek },			/* seek */
112 	{ &vop_remove_desc, nfs_remove },		/* remove */
113 	{ &vop_link_desc, nfs_link },			/* link */
114 	{ &vop_rename_desc, nfs_rename },		/* rename */
115 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
116 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
117 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
118 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
119 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
120 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
121 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
122 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
123 	{ &vop_lock_desc, nfs_lock },			/* lock */
124 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
125 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
126 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
127 	{ &vop_print_desc, nfs_print },			/* print */
128 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
129 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
130 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
131 	{ &vop_blkatoff_desc, nfs_blkatoff },		/* blkatoff */
132 	{ &vop_valloc_desc, nfs_valloc },		/* valloc */
133 	{ &vop_reallocblks_desc, nfs_reallocblks },	/* reallocblks */
134 	{ &vop_vfree_desc, nfs_vfree },			/* vfree */
135 	{ &vop_truncate_desc, nfs_truncate },		/* truncate */
136 	{ &vop_update_desc, nfs_update },		/* update */
137 	{ &vop_bwrite_desc, nfs_bwrite },		/* bwrite */
138 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
139 };
140 struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
141 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
142 
143 /*
144  * Special device vnode ops
145  */
146 int (**spec_nfsv2nodeop_p) __P((void *));
147 struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
148 	{ &vop_default_desc, vn_default_error },
149 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
150 	{ &vop_create_desc, spec_create },		/* create */
151 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
152 	{ &vop_open_desc, spec_open },			/* open */
153 	{ &vop_close_desc, nfsspec_close },		/* close */
154 	{ &vop_access_desc, nfsspec_access },		/* access */
155 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
156 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
157 	{ &vop_read_desc, nfsspec_read },		/* read */
158 	{ &vop_write_desc, nfsspec_write },		/* write */
159 	{ &vop_lease_desc, spec_lease_check },		/* lease */
160 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
161 	{ &vop_poll_desc, spec_poll },			/* poll */
162 #ifdef Lite2_integrated
163 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
164 #endif
165 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
166 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
167 	{ &vop_seek_desc, spec_seek },			/* seek */
168 	{ &vop_remove_desc, spec_remove },		/* remove */
169 	{ &vop_link_desc, spec_link },			/* link */
170 	{ &vop_rename_desc, spec_rename },		/* rename */
171 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
172 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
173 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
174 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
175 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
176 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
177 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
178 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
179 	{ &vop_lock_desc, nfs_lock },			/* lock */
180 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
181 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
182 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
183 	{ &vop_print_desc, nfs_print },			/* print */
184 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
185 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
186 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
187 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
188 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
189 	{ &vop_reallocblks_desc, spec_reallocblks },	/* reallocblks */
190 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
191 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
192 	{ &vop_update_desc, nfs_update },		/* update */
193 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
194 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
195 };
196 struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
197 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
198 
199 #ifdef FIFO
200 int (**fifo_nfsv2nodeop_p) __P((void *));
201 struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
202 	{ &vop_default_desc, vn_default_error },
203 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
204 	{ &vop_create_desc, fifo_create },		/* create */
205 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
206 	{ &vop_open_desc, fifo_open },			/* open */
207 	{ &vop_close_desc, nfsfifo_close },		/* close */
208 	{ &vop_access_desc, nfsspec_access },		/* access */
209 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
210 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
211 	{ &vop_read_desc, nfsfifo_read },		/* read */
212 	{ &vop_write_desc, nfsfifo_write },		/* write */
213 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
214 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
215 	{ &vop_poll_desc, fifo_poll },			/* poll */
216 #ifdef Lite2_integrated
217 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
218 #endif
219 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
220 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
221 	{ &vop_seek_desc, fifo_seek },			/* seek */
222 	{ &vop_remove_desc, fifo_remove },		/* remove */
223 	{ &vop_link_desc, fifo_link },			/* link */
224 	{ &vop_rename_desc, fifo_rename },		/* rename */
225 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
226 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
227 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
228 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
229 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
230 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
231 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
232 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
233 	{ &vop_lock_desc, nfs_lock },			/* lock */
234 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
235 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
236 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
237 	{ &vop_print_desc, nfs_print },			/* print */
238 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
239 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
240 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
241 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
242 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
243 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
244 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
245 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
246 	{ &vop_update_desc, nfs_update },		/* update */
247 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
248 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
249 };
250 struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
251 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
252 #endif /* FIFO */
253 
254 /*
255  * Global variables
256  */
257 extern u_int32_t nfs_true, nfs_false;
258 extern u_int32_t nfs_xdrneg1;
259 extern struct nfsstats nfsstats;
260 extern nfstype nfsv3_type[9];
261 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
262 int nfs_numasync = 0;
263 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
264 
265 /*
266  * nfs null call from vfs.
267  */
268 int
269 nfs_null(vp, cred, procp)
270 	struct vnode *vp;
271 	struct ucred *cred;
272 	struct proc *procp;
273 {
274 	caddr_t bpos, dpos;
275 	int error = 0;
276 	struct mbuf *mreq, *mrep, *md, *mb;
277 
278 	nfsm_reqhead(vp, NFSPROC_NULL, 0);
279 	nfsm_request(vp, NFSPROC_NULL, procp, cred);
280 	nfsm_reqdone;
281 	return (error);
282 }
283 
284 /*
285  * nfs access vnode op.
286  * For nfs version 2, just return ok. File accesses may fail later.
287  * For nfs version 3, use the access rpc to check accessibility. If file modes
288  * are changed on the server, accesses might still fail later.
289  */
290 int
291 nfs_access(v)
292 	void *v;
293 {
294 	struct vop_access_args /* {
295 		struct vnode *a_vp;
296 		int  a_mode;
297 		struct ucred *a_cred;
298 		struct proc *a_p;
299 	} */ *ap = v;
300 	register struct vnode *vp = ap->a_vp;
301 	register u_int32_t *tl;
302 	register caddr_t cp;
303 	register int32_t t1, t2;
304 	caddr_t bpos, dpos, cp2;
305 	int error = 0, attrflag;
306 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
307 	u_int32_t mode, rmode;
308 	int v3 = NFS_ISV3(vp);
309 
310 	/*
311 	 * Disallow write attempts on filesystems mounted read-only;
312 	 * unless the file is a socket, fifo, or a block or character
313 	 * device resident on the filesystem.
314 	 */
315 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
316 		switch (vp->v_type) {
317 		case VREG:
318 		case VDIR:
319 		case VLNK:
320 			return (EROFS);
321 		default:
322 			break;
323 		}
324 	}
325 	/*
326 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
327 	 * ufs_access() locally using the vattr. This may not be correct,
328 	 * since the server may apply other access criteria such as
329 	 * client uid-->server uid mapping that we do not know about, but
330 	 * this is better than just returning anything that is lying about
331 	 * in the cache.
332 	 */
333 	if (v3) {
334 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
335 		nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
336 		nfsm_fhtom(vp, v3);
337 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
338 		if (ap->a_mode & VREAD)
339 			mode = NFSV3ACCESS_READ;
340 		else
341 			mode = 0;
342 		if (vp->v_type == VDIR) {
343 			if (ap->a_mode & VWRITE)
344 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
345 					 NFSV3ACCESS_DELETE);
346 			if (ap->a_mode & VEXEC)
347 				mode |= NFSV3ACCESS_LOOKUP;
348 		} else {
349 			if (ap->a_mode & VWRITE)
350 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
351 			if (ap->a_mode & VEXEC)
352 				mode |= NFSV3ACCESS_EXECUTE;
353 		}
354 		*tl = txdr_unsigned(mode);
355 		nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
356 		nfsm_postop_attr(vp, attrflag);
357 		if (!error) {
358 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
359 			rmode = fxdr_unsigned(u_int32_t, *tl);
360 			/*
361 			 * The NFS V3 spec does not clarify whether or not
362 			 * the returned access bits can be a superset of
363 			 * the ones requested, so...
364 			 */
365 			if ((rmode & mode) != mode)
366 				error = EACCES;
367 		}
368 		nfsm_reqdone;
369 		return (error);
370 	} else
371 		return (nfsspec_access(ap));
372 }
373 
374 /*
375  * nfs open vnode op
376  * Check to see if the type is ok
377  * and that deletion is not in progress.
378  * For paged in text files, you will need to flush the page cache
379  * if consistency is lost.
380  */
381 /* ARGSUSED */
382 int
383 nfs_open(v)
384 	void *v;
385 {
386 	struct vop_open_args /* {
387 		struct vnode *a_vp;
388 		int  a_mode;
389 		struct ucred *a_cred;
390 		struct proc *a_p;
391 	} */ *ap = v;
392 	register struct vnode *vp = ap->a_vp;
393 	struct nfsnode *np = VTONFS(vp);
394 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
395 	struct vattr vattr;
396 	int error;
397 
398 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
399 #ifdef DIAGNOSTIC
400 		printf("open eacces vtyp=%d\n",vp->v_type);
401 #endif
402 		return (EACCES);
403 	}
404 	/*
405 	 * Get a valid lease. If cached data is stale, flush it.
406 	 */
407 	if (nmp->nm_flag & NFSMNT_NQNFS) {
408 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
409 		    do {
410 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
411 			    ap->a_p);
412 		    } while (error == NQNFS_EXPIRED);
413 		    if (error)
414 			return (error);
415 		    if (np->n_lrev != np->n_brev ||
416 			(np->n_flag & NQNFSNONCACHE)) {
417 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
418 				ap->a_p, 1)) == EINTR)
419 				return (error);
420 			(void) vnode_pager_uncache(vp);
421 			np->n_brev = np->n_lrev;
422 		    }
423 		}
424 	} else {
425 		if (np->n_flag & NMODIFIED) {
426 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
427 				ap->a_p, 1)) == EINTR)
428 				return (error);
429 			(void) vnode_pager_uncache(vp);
430 			np->n_attrstamp = 0;
431 			if (vp->v_type == VDIR)
432 				np->n_direofoffset = 0;
433 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
434 			if (error)
435 				return (error);
436 			np->n_mtime = vattr.va_mtime.tv_sec;
437 		} else {
438 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
439 			if (error)
440 				return (error);
441 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
442 				if (vp->v_type == VDIR)
443 					np->n_direofoffset = 0;
444 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
445 					ap->a_cred, ap->a_p, 1)) == EINTR)
446 					return (error);
447 				(void) vnode_pager_uncache(vp);
448 				np->n_mtime = vattr.va_mtime.tv_sec;
449 			}
450 		}
451 	}
452 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
453 		np->n_attrstamp = 0; /* For Open/Close consistency */
454 	return (0);
455 }
456 
457 /*
458  * nfs close vnode op
459  * What an NFS client should do upon close after writing is a debatable issue.
460  * Most NFS clients push delayed writes to the server upon close, basically for
461  * two reasons:
462  * 1 - So that any write errors may be reported back to the client process
463  *     doing the close system call. By far the two most likely errors are
464  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
465  * 2 - To put a worst case upper bound on cache inconsistency between
466  *     multiple clients for the file.
467  * There is also a consistency problem for Version 2 of the protocol w.r.t.
468  * not being able to tell if other clients are writing a file concurrently,
469  * since there is no way of knowing if the changed modify time in the reply
470  * is only due to the write for this client.
471  * (NFS Version 3 provides weak cache consistency data in the reply that
472  *  should be sufficient to detect and handle this case.)
473  *
474  * The current code does the following:
475  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
476  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
477  *                     or commit them (this satisfies 1 and 2 except for the
478  *                     case where the server crashes after this close but
479  *                     before the commit RPC, which is felt to be "good
480  *                     enough". Changing the last argument to nfs_flush() to
481  *                     a 1 would force a commit operation, if it is felt a
482  *                     commit is necessary now.
483  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
484  *                     1 should be dealt with via an fsync() system call for
485  *                     cases where write errors are important.
486  */
487 /* ARGSUSED */
488 int
489 nfs_close(v)
490 	void *v;
491 {
492 	struct vop_close_args /* {
493 		struct vnodeop_desc *a_desc;
494 		struct vnode *a_vp;
495 		int  a_fflag;
496 		struct ucred *a_cred;
497 		struct proc *a_p;
498 	} */ *ap = v;
499 	register struct vnode *vp = ap->a_vp;
500 	register struct nfsnode *np = VTONFS(vp);
501 	int error = 0;
502 
503 	if (vp->v_type == VREG) {
504 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
505 		(np->n_flag & NMODIFIED)) {
506 		if (NFS_ISV3(vp)) {
507 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
508 		    np->n_flag &= ~NMODIFIED;
509 		} else
510 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
511 		np->n_attrstamp = 0;
512 	    }
513 	    if (np->n_flag & NWRITEERR) {
514 		np->n_flag &= ~NWRITEERR;
515 		error = np->n_error;
516 	    }
517 	}
518 	return (error);
519 }
520 
521 /*
522  * nfs getattr call from vfs.
523  */
524 int
525 nfs_getattr(v)
526 	void *v;
527 {
528 	struct vop_getattr_args /* {
529 		struct vnode *a_vp;
530 		struct vattr *a_vap;
531 		struct ucred *a_cred;
532 		struct proc *a_p;
533 	} */ *ap = v;
534 	register struct vnode *vp = ap->a_vp;
535 	register struct nfsnode *np = VTONFS(vp);
536 	register caddr_t cp;
537 	register u_int32_t *tl;
538 	register int32_t t1, t2;
539 	caddr_t bpos, dpos;
540 	int error = 0;
541 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
542 	int v3 = NFS_ISV3(vp);
543 
544 	/*
545 	 * Update local times for special files.
546 	 */
547 	if (np->n_flag & (NACC | NUPD))
548 		np->n_flag |= NCHG;
549 	/*
550 	 * First look in the cache.
551 	 */
552 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
553 		return (0);
554 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
555 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
556 	nfsm_fhtom(vp, v3);
557 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
558 	if (!error)
559 		nfsm_loadattr(vp, ap->a_vap);
560 	nfsm_reqdone;
561 	return (error);
562 }
563 
564 /*
565  * nfs setattr call.
566  */
567 int
568 nfs_setattr(v)
569 	void *v;
570 {
571 	struct vop_setattr_args /* {
572 		struct vnodeop_desc *a_desc;
573 		struct vnode *a_vp;
574 		struct vattr *a_vap;
575 		struct ucred *a_cred;
576 		struct proc *a_p;
577 	} */ *ap = v;
578 	register struct vnode *vp = ap->a_vp;
579 	register struct nfsnode *np = VTONFS(vp);
580 	register struct vattr *vap = ap->a_vap;
581 	int error = 0;
582 	u_quad_t tsize = 0;
583 
584 	/*
585 	 * Disallow write attempts if the filesystem is mounted read-only.
586 	 */
587   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
588 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
589 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
590 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
591 		return (EROFS);
592 	if (vap->va_size != VNOVAL) {
593  		switch (vp->v_type) {
594  		case VDIR:
595  			return (EISDIR);
596  		case VCHR:
597  		case VBLK:
598  		case VSOCK:
599  		case VFIFO:
600 			if (vap->va_mtime.tv_sec == VNOVAL &&
601 			    vap->va_atime.tv_sec == VNOVAL &&
602 			    vap->va_mode == (u_short)VNOVAL &&
603 			    vap->va_uid == (uid_t)VNOVAL &&
604 			    vap->va_gid == (gid_t)VNOVAL)
605 				return (0);
606  			vap->va_size = VNOVAL;
607  			break;
608  		default:
609 			/*
610 			 * Disallow write attempts if the filesystem is
611 			 * mounted read-only.
612 			 */
613 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
614 				return (EROFS);
615  			if (vap->va_size == 0)
616  				error = nfs_vinvalbuf(vp, 0,
617  				     ap->a_cred, ap->a_p, 1);
618 			else
619 				error = nfs_vinvalbuf(vp, V_SAVE,
620 				     ap->a_cred, ap->a_p, 1);
621 			if (error)
622 				return (error);
623  			tsize = np->n_size;
624  			np->n_size = np->n_vattr.va_size = vap->va_size;
625  			vnode_pager_setsize(vp, (u_long)np->n_size);
626   		};
627   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
628 		vap->va_atime.tv_sec != VNOVAL) &&
629 		vp->v_type == VREG &&
630   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
631 		 ap->a_p, 1)) == EINTR)
632 		return (error);
633 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
634 	if (error && vap->va_size != VNOVAL) {
635 		np->n_size = np->n_vattr.va_size = tsize;
636 		vnode_pager_setsize(vp, (u_long)np->n_size);
637 	}
638 	return (error);
639 }
640 
641 /*
642  * Do an nfs setattr rpc.
643  */
644 int
645 nfs_setattrrpc(vp, vap, cred, procp)
646 	register struct vnode *vp;
647 	register struct vattr *vap;
648 	struct ucred *cred;
649 	struct proc *procp;
650 {
651 	register struct nfsv2_sattr *sp;
652 	register caddr_t cp;
653 	register int32_t t1, t2;
654 	caddr_t bpos, dpos, cp2;
655 	u_int32_t *tl;
656 	int error = 0, wccflag = NFSV3_WCCRATTR;
657 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
658 	int v3 = NFS_ISV3(vp);
659 
660 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
661 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
662 	nfsm_fhtom(vp, v3);
663 	if (v3) {
664 		if (vap->va_mode != (u_short)VNOVAL) {
665 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
666 			*tl++ = nfs_true;
667 			*tl = txdr_unsigned(vap->va_mode);
668 		} else {
669 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
670 			*tl = nfs_false;
671 		}
672 		if (vap->va_uid != (uid_t)VNOVAL) {
673 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
674 			*tl++ = nfs_true;
675 			*tl = txdr_unsigned(vap->va_uid);
676 		} else {
677 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
678 			*tl = nfs_false;
679 		}
680 		if (vap->va_gid != (gid_t)VNOVAL) {
681 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
682 			*tl++ = nfs_true;
683 			*tl = txdr_unsigned(vap->va_gid);
684 		} else {
685 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
686 			*tl = nfs_false;
687 		}
688 		if (vap->va_size != VNOVAL) {
689 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
690 			*tl++ = nfs_true;
691 			txdr_hyper(&vap->va_size, tl);
692 		} else {
693 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
694 			*tl = nfs_false;
695 		}
696 		if (vap->va_atime.tv_sec != VNOVAL) {
697 			if (vap->va_atime.tv_sec != time.tv_sec) {
698 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
699 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
700 				txdr_nfsv3time(&vap->va_atime, tl);
701 			} else {
702 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
703 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
704 			}
705 		} else {
706 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
707 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
708 		}
709 		if (vap->va_mtime.tv_sec != VNOVAL) {
710 			if (vap->va_mtime.tv_sec != time.tv_sec) {
711 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
712 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
713 				txdr_nfsv3time(&vap->va_mtime, tl);
714 			} else {
715 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
716 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
717 			}
718 		} else {
719 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
720 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
721 		}
722 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
723 		*tl = nfs_false;
724 	} else {
725 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
726 		if (vap->va_mode == (u_short)VNOVAL)
727 			sp->sa_mode = nfs_xdrneg1;
728 		else
729 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
730 		if (vap->va_uid == (uid_t)VNOVAL)
731 			sp->sa_uid = nfs_xdrneg1;
732 		else
733 			sp->sa_uid = txdr_unsigned(vap->va_uid);
734 		if (vap->va_gid == (gid_t)VNOVAL)
735 			sp->sa_gid = nfs_xdrneg1;
736 		else
737 			sp->sa_gid = txdr_unsigned(vap->va_gid);
738 		sp->sa_size = txdr_unsigned(vap->va_size);
739 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
740 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
741 	}
742 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
743 	if (v3) {
744 		nfsm_wcc_data(vp, wccflag);
745 	} else
746 		nfsm_loadattr(vp, (struct vattr *)0);
747 	nfsm_reqdone;
748 	return (error);
749 }
750 
751 /*
752  * nfs lookup call, one step at a time...
753  * First look in cache
754  * If not found, unlock the directory nfsnode and do the rpc
755  */
756 int
757 nfs_lookup(v)
758 	void *v;
759 {
760 	struct vop_lookup_args /* {
761 		struct vnodeop_desc *a_desc;
762 		struct vnode *a_dvp;
763 		struct vnode **a_vpp;
764 		struct componentname *a_cnp;
765 	} */ *ap = v;
766 	register struct componentname *cnp = ap->a_cnp;
767 	register struct vnode *dvp = ap->a_dvp;
768 	register struct vnode **vpp = ap->a_vpp;
769 	register int flags = cnp->cn_flags;
770 	register struct vnode *newvp;
771 	register u_int32_t *tl;
772 	register caddr_t cp;
773 	register int32_t t1, t2;
774 	struct nfsmount *nmp;
775 	caddr_t bpos, dpos, cp2;
776 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
777 	long len;
778 	nfsfh_t *fhp;
779 	struct nfsnode *np;
780 	int lockparent, wantparent, error = 0, attrflag, fhsize;
781 	int v3 = NFS_ISV3(dvp);
782 
783 	*vpp = NULLVP;
784 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
785 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
786 		return (EROFS);
787 	if (dvp->v_type != VDIR)
788 		return (ENOTDIR);
789 	lockparent = flags & LOCKPARENT;
790 	wantparent = flags & (LOCKPARENT|WANTPARENT);
791 	nmp = VFSTONFS(dvp->v_mount);
792 	np = VTONFS(dvp);
793 	if ((error = cache_lookup(dvp, vpp, cnp)) != 0 && error != ENOENT) {
794 		struct vattr vattr;
795 		int vpid;
796 
797 		newvp = *vpp;
798 		vpid = newvp->v_id;
799 		/*
800 		 * See the comment starting `Step through' in ufs/ufs_lookup.c
801 		 * for an explanation of the locking protocol
802 		 */
803 		if (dvp == newvp) {
804 			VREF(newvp);
805 			error = 0;
806 		} else
807 #ifdef Lite2_integrated
808 			error = vget(newvp, LK_EXCLUSIVE, p);
809 #else
810 			error = vget(newvp, 1);
811 #endif
812 		if (!error) {
813 			if (vpid == newvp->v_id) {
814 			   if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
815 			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
816 				nfsstats.lookupcache_hits++;
817 				if (cnp->cn_nameiop != LOOKUP &&
818 				    (flags & ISLASTCN))
819 					cnp->cn_flags |= SAVENAME;
820 				return (0);
821 			   }
822 			   cache_purge(newvp);
823 			}
824 			vrele(newvp);
825 		}
826 		*vpp = NULLVP;
827 	}
828 	error = 0;
829 	newvp = NULLVP;
830 	nfsstats.lookupcache_misses++;
831 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
832 	len = cnp->cn_namelen;
833 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
834 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
835 	nfsm_fhtom(dvp, v3);
836 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
837 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
838 	if (error) {
839 		nfsm_postop_attr(dvp, attrflag);
840 		m_freem(mrep);
841 		goto nfsmout;
842 	}
843 	nfsm_getfh(fhp, fhsize, v3);
844 
845 	/*
846 	 * Handle RENAME case...
847 	 */
848 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
849 		if (NFS_CMPFH(np, fhp, fhsize)) {
850 			m_freem(mrep);
851 			return (EISDIR);
852 		}
853 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
854 		if (error) {
855 			m_freem(mrep);
856 			return (error);
857 		}
858 		newvp = NFSTOV(np);
859 		if (v3) {
860 			nfsm_postop_attr(newvp, attrflag);
861 			nfsm_postop_attr(dvp, attrflag);
862 		} else
863 			nfsm_loadattr(newvp, (struct vattr *)0);
864 		*vpp = newvp;
865 		m_freem(mrep);
866 		cnp->cn_flags |= SAVENAME;
867 		return (0);
868 	}
869 
870 	if (NFS_CMPFH(np, fhp, fhsize)) {
871 		VREF(dvp);
872 		newvp = dvp;
873 	} else {
874 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
875 		if (error) {
876 			m_freem(mrep);
877 			return (error);
878 		}
879 		newvp = NFSTOV(np);
880 	}
881 	if (v3) {
882 		nfsm_postop_attr(newvp, attrflag);
883 		nfsm_postop_attr(dvp, attrflag);
884 	} else
885 		nfsm_loadattr(newvp, (struct vattr *)0);
886 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
887 		cnp->cn_flags |= SAVENAME;
888 	if ((cnp->cn_flags & MAKEENTRY) &&
889 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
890 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
891 		cache_enter(dvp, newvp, cnp);
892 	}
893 	*vpp = newvp;
894 	nfsm_reqdone;
895 	if (error) {
896 		if (newvp != NULLVP)
897 			vrele(newvp);
898 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
899 		    (flags & ISLASTCN) && error == ENOENT) {
900 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
901 				error = EROFS;
902 			else
903 				error = EJUSTRETURN;
904 		}
905 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
906 			cnp->cn_flags |= SAVENAME;
907 	}
908 	return (error);
909 }
910 
911 /*
912  * nfs read call.
913  * Just call nfs_bioread() to do the work.
914  */
915 int
916 nfs_read(v)
917 	void *v;
918 {
919 	struct vop_read_args /* {
920 		struct vnode *a_vp;
921 		struct uio *a_uio;
922 		int  a_ioflag;
923 		struct ucred *a_cred;
924 	} */ *ap = v;
925 	register struct vnode *vp = ap->a_vp;
926 
927 	if (vp->v_type != VREG)
928 		return (EPERM);
929 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
930 }
931 
932 /*
933  * nfs readlink call
934  */
935 int
936 nfs_readlink(v)
937 	void *v;
938 {
939 	struct vop_readlink_args /* {
940 		struct vnode *a_vp;
941 		struct uio *a_uio;
942 		struct ucred *a_cred;
943 	} */ *ap = v;
944 	register struct vnode *vp = ap->a_vp;
945 
946 	if (vp->v_type != VLNK)
947 		return (EPERM);
948 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
949 }
950 
951 /*
952  * Do a readlink rpc.
953  * Called by nfs_doio() from below the buffer cache.
954  */
955 int
956 nfs_readlinkrpc(vp, uiop, cred)
957 	register struct vnode *vp;
958 	struct uio *uiop;
959 	struct ucred *cred;
960 {
961 	register u_int32_t *tl;
962 	register caddr_t cp;
963 	register int32_t t1, t2;
964 	caddr_t bpos, dpos, cp2;
965 	int error = 0, len, attrflag;
966 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
967 	int v3 = NFS_ISV3(vp);
968 
969 	nfsstats.rpccnt[NFSPROC_READLINK]++;
970 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
971 	nfsm_fhtom(vp, v3);
972 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
973 	if (v3)
974 		nfsm_postop_attr(vp, attrflag);
975 	if (!error) {
976 		nfsm_strsiz(len, NFS_MAXPATHLEN);
977 		nfsm_mtouio(uiop, len);
978 	}
979 	nfsm_reqdone;
980 	return (error);
981 }
982 
983 /*
984  * nfs read rpc call
985  * Ditto above
986  */
987 int
988 nfs_readrpc(vp, uiop, cred)
989 	register struct vnode *vp;
990 	struct uio *uiop;
991 	struct ucred *cred;
992 {
993 	register u_int32_t *tl;
994 	register caddr_t cp;
995 	register int32_t t1, t2;
996 	caddr_t bpos, dpos, cp2;
997 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
998 	struct nfsmount *nmp;
999 	int error = 0, len, retlen, tsiz, eof, attrflag;
1000 	int v3 = NFS_ISV3(vp);
1001 
1002 #ifndef nolint
1003 	eof = 0;
1004 #endif
1005 	nmp = VFSTONFS(vp->v_mount);
1006 	tsiz = uiop->uio_resid;
1007 	if (uiop->uio_offset + tsiz > 0xffffffff && !v3)
1008 		return (EFBIG);
1009 	while (tsiz > 0) {
1010 		nfsstats.rpccnt[NFSPROC_READ]++;
1011 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1012 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1013 		nfsm_fhtom(vp, v3);
1014 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1015 		if (v3) {
1016 			txdr_hyper(&uiop->uio_offset, tl);
1017 			*(tl + 2) = txdr_unsigned(len);
1018 		} else {
1019 			*tl++ = txdr_unsigned(uiop->uio_offset);
1020 			*tl++ = txdr_unsigned(len);
1021 			*tl = 0;
1022 		}
1023 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp, cred);
1024 		if (v3) {
1025 			nfsm_postop_attr(vp, attrflag);
1026 			if (error) {
1027 				m_freem(mrep);
1028 				goto nfsmout;
1029 			}
1030 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1031 			eof = fxdr_unsigned(int, *(tl + 1));
1032 		} else
1033 			nfsm_loadattr(vp, (struct vattr *)0);
1034 		nfsm_strsiz(retlen, nmp->nm_rsize);
1035 		nfsm_mtouio(uiop, retlen);
1036 		m_freem(mrep);
1037 		tsiz -= retlen;
1038 		if (v3) {
1039 			if (eof || retlen == 0)
1040 				tsiz = 0;
1041 		} else if (retlen < len)
1042 			tsiz = 0;
1043 	}
1044 nfsmout:
1045 	return (error);
1046 }
1047 
1048 /*
1049  * nfs write call
1050  */
1051 int
1052 nfs_writerpc(vp, uiop, cred, iomode, must_commit)
1053 	register struct vnode *vp;
1054 	register struct uio *uiop;
1055 	struct ucred *cred;
1056 	int *iomode, *must_commit;
1057 {
1058 	register u_int32_t *tl;
1059 	register caddr_t cp;
1060 	register int32_t t1, t2, backup;
1061 	caddr_t bpos, dpos, cp2;
1062 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1063 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1064 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1065 	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1066 
1067 #ifndef DIAGNOSTIC
1068 	if (uiop->uio_iovcnt != 1)
1069 		panic("nfs: writerpc iovcnt > 1");
1070 #endif
1071 	*must_commit = 0;
1072 	tsiz = uiop->uio_resid;
1073 	if (uiop->uio_offset + tsiz > 0xffffffff && !v3)
1074 		return (EFBIG);
1075 	while (tsiz > 0) {
1076 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1077 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1078 		nfsm_reqhead(vp, NFSPROC_WRITE,
1079 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1080 		nfsm_fhtom(vp, v3);
1081 		if (v3) {
1082 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1083 			txdr_hyper(&uiop->uio_offset, tl);
1084 			tl += 2;
1085 			*tl++ = txdr_unsigned(len);
1086 			*tl++ = txdr_unsigned(*iomode);
1087 			*tl = txdr_unsigned(len);
1088 		} else {
1089 			register u_int32_t x;
1090 
1091 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1092 			/* Set both "begin" and "current" to non-garbage. */
1093 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1094 			*tl++ = x;      /* "begin offset" */
1095 			*tl++ = x;      /* "current offset" */
1096 			x = txdr_unsigned(len);
1097 			*tl++ = x;      /* total to this offset */
1098 			*tl = x;        /* size of this write */
1099 
1100 		}
1101 		nfsm_uiotom(uiop, len);
1102 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp, cred);
1103 		if (v3) {
1104 			wccflag = NFSV3_WCCCHK;
1105 			nfsm_wcc_data(vp, wccflag);
1106 			if (!error) {
1107 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1108 					+ NFSX_V3WRITEVERF);
1109 				rlen = fxdr_unsigned(int, *tl++);
1110 				if (rlen == 0) {
1111 					error = NFSERR_IO;
1112 					break;
1113 				} else if (rlen < len) {
1114 					backup = len - rlen;
1115 					uiop->uio_iov->iov_base -= backup;
1116 					uiop->uio_iov->iov_len += backup;
1117 					uiop->uio_offset -= backup;
1118 					uiop->uio_resid += backup;
1119 					len = rlen;
1120 				}
1121 				commit = fxdr_unsigned(int, *tl++);
1122 
1123 				/*
1124 				 * Return the lowest committment level
1125 				 * obtained by any of the RPCs.
1126 				 */
1127 				if (committed == NFSV3WRITE_FILESYNC)
1128 					committed = commit;
1129 				else if (committed == NFSV3WRITE_DATASYNC &&
1130 					commit == NFSV3WRITE_UNSTABLE)
1131 					committed = commit;
1132 				if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0) {
1133 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1134 					NFSX_V3WRITEVERF);
1135 				    nmp->nm_flag |= NFSMNT_HASWRITEVERF;
1136 				} else if (bcmp((caddr_t)tl,
1137 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1138 				    *must_commit = 1;
1139 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1140 					NFSX_V3WRITEVERF);
1141 				}
1142 			}
1143 		} else
1144 		    nfsm_loadattr(vp, (struct vattr *)0);
1145 		if (wccflag)
1146 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1147 		m_freem(mrep);
1148 		tsiz -= len;
1149 	}
1150 nfsmout:
1151 	*iomode = committed;
1152 	if (error)
1153 		uiop->uio_resid = tsiz;
1154 	return (error);
1155 }
1156 
1157 /*
1158  * nfs mknod rpc
1159  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1160  * mode set to specify the file type and the size field for rdev.
1161  */
1162 int
1163 nfs_mknodrpc(dvp, vpp, cnp, vap)
1164 	register struct vnode *dvp;
1165 	register struct vnode **vpp;
1166 	register struct componentname *cnp;
1167 	register struct vattr *vap;
1168 {
1169 	register struct nfsv2_sattr *sp;
1170 	register struct nfsv3_sattr *sp3;
1171 	register u_int32_t *tl;
1172 	register caddr_t cp;
1173 	register int32_t t1, t2;
1174 	struct vnode *newvp = (struct vnode *)0;
1175 	struct nfsnode *np;
1176 	char *cp2;
1177 	caddr_t bpos, dpos;
1178 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1179 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1180 	u_int32_t rdev;
1181 	int v3 = NFS_ISV3(dvp);
1182 
1183 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1184 		rdev = txdr_unsigned(vap->va_rdev);
1185 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1186 		rdev = nfs_xdrneg1;
1187 	else {
1188 		VOP_ABORTOP(dvp, cnp);
1189 		vput(dvp);
1190 		return (EOPNOTSUPP);
1191 	}
1192 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1193 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1194 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1195 	nfsm_fhtom(dvp, v3);
1196 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1197 	if (v3) {
1198 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3SRVSATTR);
1199 		*tl++ = vtonfsv3_type(vap->va_type);
1200 		sp3 = (struct nfsv3_sattr *)tl;
1201 		nfsm_v3sattr(sp3, vap);
1202 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1203 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1204 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1205 			*tl = txdr_unsigned(minor(vap->va_rdev));
1206 		}
1207 	} else {
1208 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1209 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1210 		sp->sa_uid = nfs_xdrneg1;
1211 		sp->sa_gid = nfs_xdrneg1;
1212 		sp->sa_size = rdev;
1213 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1214 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1215 	}
1216 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1217 	if (!error) {
1218 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1219 		if (!gotvp) {
1220 			if (newvp) {
1221 				vrele(newvp);
1222 				newvp = (struct vnode *)0;
1223 			}
1224 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1225 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1226 			if (!error)
1227 				newvp = NFSTOV(np);
1228 		}
1229 	}
1230 	if (v3)
1231 		nfsm_wcc_data(dvp, wccflag);
1232 	nfsm_reqdone;
1233 	if (error) {
1234 		if (newvp)
1235 			vrele(newvp);
1236 	} else {
1237 		if (cnp->cn_flags & MAKEENTRY)
1238 			cache_enter(dvp, newvp, cnp);
1239 		*vpp = newvp;
1240 	}
1241 	FREE(cnp->cn_pnbuf, M_NAMEI);
1242 	VTONFS(dvp)->n_flag |= NMODIFIED;
1243 	if (!wccflag)
1244 		VTONFS(dvp)->n_attrstamp = 0;
1245 	vrele(dvp);
1246 	return (error);
1247 }
1248 
1249 /*
1250  * nfs mknod vop
1251  * just call nfs_mknodrpc() to do the work.
1252  */
1253 /* ARGSUSED */
1254 int
1255 nfs_mknod(v)
1256 	void *v;
1257 {
1258 	struct vop_mknod_args /* {
1259 		struct vnode *a_dvp;
1260 		struct vnode **a_vpp;
1261 		struct componentname *a_cnp;
1262 		struct vattr *a_vap;
1263 	} */ *ap = v;
1264 	struct vnode *newvp;
1265 	int error;
1266 
1267 	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1268 	if (!error)
1269 		vrele(newvp);
1270 	return (error);
1271 }
1272 
1273 static u_long create_verf;
1274 /*
1275  * nfs file create call
1276  */
1277 int
1278 nfs_create(v)
1279 	void *v;
1280 {
1281 	struct vop_create_args /* {
1282 		struct vnode *a_dvp;
1283 		struct vnode **a_vpp;
1284 		struct componentname *a_cnp;
1285 		struct vattr *a_vap;
1286 	} */ *ap = v;
1287 	register struct vnode *dvp = ap->a_dvp;
1288 	register struct vattr *vap = ap->a_vap;
1289 	register struct componentname *cnp = ap->a_cnp;
1290 	register struct nfsv2_sattr *sp;
1291 	register struct nfsv3_sattr *sp3;
1292 	register u_int32_t *tl;
1293 	register caddr_t cp;
1294 	register int32_t t1, t2;
1295 	struct nfsnode *np = (struct nfsnode *)0;
1296 	struct vnode *newvp = (struct vnode *)0;
1297 	caddr_t bpos, dpos, cp2;
1298 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1299 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1300 	int v3 = NFS_ISV3(dvp);
1301 
1302 	/*
1303 	 * Oops, not for me..
1304 	 */
1305 	if (vap->va_type == VSOCK)
1306 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1307 
1308 #ifdef VA_EXCLUSIVE
1309 	if (vap->va_vaflags & VA_EXCLUSIVE)
1310 		fmode |= O_EXCL;
1311 #endif
1312 again:
1313 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1314 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1315 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1316 	nfsm_fhtom(dvp, v3);
1317 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1318 	if (v3) {
1319 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1320 		if (fmode & O_EXCL) {
1321 		    *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1322 		    nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1323 		    if (in_ifaddr.tqh_first)
1324 			*tl++ = in_ifaddr.tqh_first->ia_addr.sin_addr.s_addr;
1325 		    else
1326 			*tl++ = create_verf;
1327 		    *tl = ++create_verf;
1328 		} else {
1329 		    *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1330 		    nfsm_build(tl, u_int32_t *, NFSX_V3SRVSATTR);
1331 		    sp3 = (struct nfsv3_sattr *)tl;
1332 		    nfsm_v3sattr(sp3, vap);
1333 		}
1334 	} else {
1335 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1336 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1337 		sp->sa_uid = nfs_xdrneg1;
1338 		sp->sa_gid = nfs_xdrneg1;
1339 		sp->sa_size = 0;
1340 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1341 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1342 	}
1343 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1344 	if (!error) {
1345 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1346 		if (!gotvp) {
1347 			if (newvp) {
1348 				vrele(newvp);
1349 				newvp = (struct vnode *)0;
1350 			}
1351 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1352 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1353 			if (!error)
1354 				newvp = NFSTOV(np);
1355 		}
1356 	}
1357 	if (v3)
1358 		nfsm_wcc_data(dvp, wccflag);
1359 	nfsm_reqdone;
1360 	if (error) {
1361 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1362 			fmode &= ~O_EXCL;
1363 			goto again;
1364 		}
1365 		if (newvp)
1366 			vrele(newvp);
1367 	} else if (v3 && (fmode & O_EXCL))
1368 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1369 	if (!error) {
1370 		if (cnp->cn_flags & MAKEENTRY)
1371 			cache_enter(dvp, newvp, cnp);
1372 		*ap->a_vpp = newvp;
1373 	}
1374 	FREE(cnp->cn_pnbuf, M_NAMEI);
1375 	VTONFS(dvp)->n_flag |= NMODIFIED;
1376 	if (!wccflag)
1377 		VTONFS(dvp)->n_attrstamp = 0;
1378 	vrele(dvp);
1379 	return (error);
1380 }
1381 
1382 /*
1383  * nfs file remove call
1384  * To try and make nfs semantics closer to ufs semantics, a file that has
1385  * other processes using the vnode is renamed instead of removed and then
1386  * removed later on the last close.
1387  * - If v_usecount > 1
1388  *	  If a rename is not already in the works
1389  *	     call nfs_sillyrename() to set it up
1390  *     else
1391  *	  do the remove rpc
1392  */
1393 int
1394 nfs_remove(v)
1395 	void *v;
1396 {
1397 	struct vop_remove_args /* {
1398 		struct vnodeop_desc *a_desc;
1399 		struct vnode * a_dvp;
1400 		struct vnode * a_vp;
1401 		struct componentname * a_cnp;
1402 	} */ *ap = v;
1403 	register struct vnode *vp = ap->a_vp;
1404 	register struct vnode *dvp = ap->a_dvp;
1405 	register struct componentname *cnp = ap->a_cnp;
1406 	register struct nfsnode *np = VTONFS(vp);
1407 	int error = 0;
1408 	struct vattr vattr;
1409 
1410 #ifndef DIAGNOSTIC
1411 	if ((cnp->cn_flags & HASBUF) == 0)
1412 		panic("nfs_remove: no name");
1413 	if (vp->v_usecount < 1)
1414 		panic("nfs_remove: bad v_usecount");
1415 #endif
1416 	if (vp->v_usecount == 1 || (np->n_sillyrename &&
1417 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1418 	    vattr.va_nlink > 1)) {
1419 		/*
1420 		 * Purge the name cache so that the chance of a lookup for
1421 		 * the name succeeding while the remove is in progress is
1422 		 * minimized. Without node locking it can still happen, such
1423 		 * that an I/O op returns ESTALE, but since you get this if
1424 		 * another host removes the file..
1425 		 */
1426 		cache_purge(vp);
1427 		/*
1428 		 * throw away biocache buffers, mainly to avoid
1429 		 * unnecessary delayed writes later.
1430 		 */
1431 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1432 		/* Do the rpc */
1433 		if (error != EINTR)
1434 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1435 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1436 		/*
1437 		 * Kludge City: If the first reply to the remove rpc is lost..
1438 		 *   the reply to the retransmitted request will be ENOENT
1439 		 *   since the file was in fact removed
1440 		 *   Therefore, we cheat and return success.
1441 		 */
1442 		if (error == ENOENT)
1443 			error = 0;
1444 	} else if (!np->n_sillyrename)
1445 		error = nfs_sillyrename(dvp, vp, cnp);
1446 	FREE(cnp->cn_pnbuf, M_NAMEI);
1447 	np->n_attrstamp = 0;
1448 	vrele(dvp);
1449 	vrele(vp);
1450 	return (error);
1451 }
1452 
1453 /*
1454  * nfs file remove rpc called from nfs_inactive
1455  */
1456 int
1457 nfs_removeit(sp)
1458 	register struct sillyrename *sp;
1459 {
1460 
1461 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1462 		(struct proc *)0));
1463 }
1464 
1465 /*
1466  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1467  */
1468 int
1469 nfs_removerpc(dvp, name, namelen, cred, proc)
1470 	register struct vnode *dvp;
1471 	const char *name;
1472 	int namelen;
1473 	struct ucred *cred;
1474 	struct proc *proc;
1475 {
1476 	register u_int32_t *tl;
1477 	register caddr_t cp;
1478 	register int32_t t1, t2;
1479 	caddr_t bpos, dpos, cp2;
1480 	int error = 0, wccflag = NFSV3_WCCRATTR;
1481 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1482 	int v3 = NFS_ISV3(dvp);
1483 
1484 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1485 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1486 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1487 	nfsm_fhtom(dvp, v3);
1488 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1489 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1490 	if (v3)
1491 		nfsm_wcc_data(dvp, wccflag);
1492 	nfsm_reqdone;
1493 	VTONFS(dvp)->n_flag |= NMODIFIED;
1494 	if (!wccflag)
1495 		VTONFS(dvp)->n_attrstamp = 0;
1496 	return (error);
1497 }
1498 
1499 /*
1500  * nfs file rename call
1501  */
1502 int
1503 nfs_rename(v)
1504 	void *v;
1505 {
1506 	struct vop_rename_args  /* {
1507 		struct vnode *a_fdvp;
1508 		struct vnode *a_fvp;
1509 		struct componentname *a_fcnp;
1510 		struct vnode *a_tdvp;
1511 		struct vnode *a_tvp;
1512 		struct componentname *a_tcnp;
1513 	} */ *ap = v;
1514 	register struct vnode *fvp = ap->a_fvp;
1515 	register struct vnode *tvp = ap->a_tvp;
1516 	register struct vnode *fdvp = ap->a_fdvp;
1517 	register struct vnode *tdvp = ap->a_tdvp;
1518 	register struct componentname *tcnp = ap->a_tcnp;
1519 	register struct componentname *fcnp = ap->a_fcnp;
1520 	int error;
1521 
1522 #ifndef DIAGNOSTIC
1523 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1524 	    (fcnp->cn_flags & HASBUF) == 0)
1525 		panic("nfs_rename: no name");
1526 #endif
1527 	/* Check for cross-device rename */
1528 	if ((fvp->v_mount != tdvp->v_mount) ||
1529 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1530 		error = EXDEV;
1531 		goto out;
1532 	}
1533 
1534 	/*
1535 	 * If the tvp exists and is in use, sillyrename it before doing the
1536 	 * rename of the new file over it.
1537 	 */
1538 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1539 		!nfs_sillyrename(tdvp, tvp, tcnp)) {
1540 		vrele(tvp);
1541 		tvp = NULL;
1542 	}
1543 
1544 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1545 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1546 		tcnp->cn_proc);
1547 
1548 	if (fvp->v_type == VDIR) {
1549 		if (tvp != NULL && tvp->v_type == VDIR)
1550 			cache_purge(tdvp);
1551 		cache_purge(fdvp);
1552 	}
1553 out:
1554 	if (tdvp == tvp)
1555 		vrele(tdvp);
1556 	else
1557 		vput(tdvp);
1558 	if (tvp)
1559 		vput(tvp);
1560 	vrele(fdvp);
1561 	vrele(fvp);
1562 	/*
1563 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1564 	 */
1565 	if (error == ENOENT)
1566 		error = 0;
1567 	return (error);
1568 }
1569 
1570 /*
1571  * nfs file rename rpc called from nfs_remove() above
1572  */
1573 int
1574 nfs_renameit(sdvp, scnp, sp)
1575 	struct vnode *sdvp;
1576 	struct componentname *scnp;
1577 	register struct sillyrename *sp;
1578 {
1579 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1580 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1581 }
1582 
1583 /*
1584  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1585  */
1586 int
1587 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1588 	register struct vnode *fdvp;
1589 	const char *fnameptr;
1590 	int fnamelen;
1591 	register struct vnode *tdvp;
1592 	const char *tnameptr;
1593 	int tnamelen;
1594 	struct ucred *cred;
1595 	struct proc *proc;
1596 {
1597 	register u_int32_t *tl;
1598 	register caddr_t cp;
1599 	register int32_t t1, t2;
1600 	caddr_t bpos, dpos, cp2;
1601 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1602 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1603 	int v3 = NFS_ISV3(fdvp);
1604 
1605 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1606 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1607 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1608 		nfsm_rndup(tnamelen));
1609 	nfsm_fhtom(fdvp, v3);
1610 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1611 	nfsm_fhtom(tdvp, v3);
1612 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1613 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1614 	if (v3) {
1615 		nfsm_wcc_data(fdvp, fwccflag);
1616 		nfsm_wcc_data(tdvp, twccflag);
1617 	}
1618 	nfsm_reqdone;
1619 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1620 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1621 	if (!fwccflag)
1622 		VTONFS(fdvp)->n_attrstamp = 0;
1623 	if (!twccflag)
1624 		VTONFS(tdvp)->n_attrstamp = 0;
1625 	return (error);
1626 }
1627 
1628 /*
1629  * nfs hard link create call
1630  */
1631 int
1632 nfs_link(v)
1633 	void *v;
1634 {
1635 	struct vop_link_args /* {
1636 		struct vnode *a_dvp;
1637 		struct vnode *a_vp;
1638 		struct componentname *a_cnp;
1639 	} */ *ap = v;
1640 	register struct vnode *vp = ap->a_vp;
1641 	register struct vnode *dvp = ap->a_dvp;
1642 	register struct componentname *cnp = ap->a_cnp;
1643 	register u_int32_t *tl;
1644 	register caddr_t cp;
1645 	register int32_t t1, t2;
1646 	caddr_t bpos, dpos, cp2;
1647 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1648 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1649 	int v3 = NFS_ISV3(vp);
1650 
1651 
1652 	if (dvp->v_mount != vp->v_mount) {
1653 		FREE(cnp->cn_pnbuf, M_NAMEI);
1654 		if (vp == dvp)
1655 			vrele(dvp);
1656 		else
1657 			vput(dvp);
1658 		return (EXDEV);
1659 	}
1660 
1661 	/*
1662 	 * Push all writes to the server, so that the attribute cache
1663 	 * doesn't get "out of sync" with the server.
1664 	 * XXX There should be a better way!
1665 	 */
1666 	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1667 
1668 	nfsstats.rpccnt[NFSPROC_LINK]++;
1669 	nfsm_reqhead(vp, NFSPROC_LINK,
1670 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1671 	nfsm_fhtom(vp, v3);
1672 	nfsm_fhtom(dvp, v3);
1673 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1674 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1675 	if (v3) {
1676 		nfsm_postop_attr(vp, attrflag);
1677 		nfsm_wcc_data(dvp, wccflag);
1678 	}
1679 	nfsm_reqdone;
1680 	FREE(cnp->cn_pnbuf, M_NAMEI);
1681 	VTONFS(dvp)->n_flag |= NMODIFIED;
1682 	if (!attrflag)
1683 		VTONFS(vp)->n_attrstamp = 0;
1684 	if (!wccflag)
1685 		VTONFS(dvp)->n_attrstamp = 0;
1686 	vput(dvp);
1687 	/*
1688 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1689 	 */
1690 	if (error == EEXIST)
1691 		error = 0;
1692 	return (error);
1693 }
1694 
1695 /*
1696  * nfs symbolic link create call
1697  */
1698 int
1699 nfs_symlink(v)
1700 	void *v;
1701 {
1702 	struct vop_symlink_args /* {
1703 		struct vnode *a_dvp;
1704 		struct vnode **a_vpp;
1705 		struct componentname *a_cnp;
1706 		struct vattr *a_vap;
1707 		char *a_target;
1708 	} */ *ap = v;
1709 	register struct vnode *dvp = ap->a_dvp;
1710 	register struct vattr *vap = ap->a_vap;
1711 	register struct componentname *cnp = ap->a_cnp;
1712 	register struct nfsv2_sattr *sp;
1713 	register struct nfsv3_sattr *sp3;
1714 	register u_int32_t *tl;
1715 	register caddr_t cp;
1716 	register int32_t t1, t2;
1717 	caddr_t bpos, dpos, cp2;
1718 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1719 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1720 	struct vnode *newvp = (struct vnode *)0;
1721 	int v3 = NFS_ISV3(dvp);
1722 
1723 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1724 	slen = strlen(ap->a_target);
1725 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1726 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1727 	nfsm_fhtom(dvp, v3);
1728 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1729 	if (v3) {
1730 		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1731 		nfsm_v3sattr(sp3, vap);
1732 	}
1733 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1734 	if (!v3) {
1735 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1736 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1737 		sp->sa_uid = nfs_xdrneg1;
1738 		sp->sa_gid = nfs_xdrneg1;
1739 		sp->sa_size = nfs_xdrneg1;
1740 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1741 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1742 	}
1743 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1744 	if (v3) {
1745 		if (!error)
1746 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1747 		nfsm_wcc_data(dvp, wccflag);
1748 	}
1749 	nfsm_reqdone;
1750 	if (newvp)
1751 		vrele(newvp);
1752 	FREE(cnp->cn_pnbuf, M_NAMEI);
1753 	VTONFS(dvp)->n_flag |= NMODIFIED;
1754 	if (!wccflag)
1755 		VTONFS(dvp)->n_attrstamp = 0;
1756 	vrele(dvp);
1757 	/*
1758 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1759 	 */
1760 	if (error == EEXIST)
1761 		error = 0;
1762 	return (error);
1763 }
1764 
1765 /*
1766  * nfs make dir call
1767  */
1768 int
1769 nfs_mkdir(v)
1770 	void *v;
1771 {
1772 	struct vop_mkdir_args /* {
1773 		struct vnode *a_dvp;
1774 		struct vnode **a_vpp;
1775 		struct componentname *a_cnp;
1776 		struct vattr *a_vap;
1777 	} */ *ap = v;
1778 	register struct vnode *dvp = ap->a_dvp;
1779 	register struct vattr *vap = ap->a_vap;
1780 	register struct componentname *cnp = ap->a_cnp;
1781 	register struct nfsv2_sattr *sp;
1782 	register struct nfsv3_sattr *sp3;
1783 	register u_int32_t *tl;
1784 	register caddr_t cp;
1785 	register int32_t t1, t2;
1786 	register int len;
1787 	struct nfsnode *np = (struct nfsnode *)0;
1788 	struct vnode *newvp = (struct vnode *)0;
1789 	caddr_t bpos, dpos, cp2;
1790 	int error = 0, wccflag = NFSV3_WCCRATTR;
1791 	int gotvp = 0;
1792 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1793 	int v3 = NFS_ISV3(dvp);
1794 
1795 	len = cnp->cn_namelen;
1796 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1797 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1798 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1799 	nfsm_fhtom(dvp, v3);
1800 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1801 	if (v3) {
1802 		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1803 		nfsm_v3sattr(sp3, vap);
1804 	} else {
1805 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1806 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1807 		sp->sa_uid = nfs_xdrneg1;
1808 		sp->sa_gid = nfs_xdrneg1;
1809 		sp->sa_size = nfs_xdrneg1;
1810 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1811 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1812 	}
1813 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1814 	if (!error)
1815 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1816 	if (v3)
1817 		nfsm_wcc_data(dvp, wccflag);
1818 	nfsm_reqdone;
1819 	VTONFS(dvp)->n_flag |= NMODIFIED;
1820 	if (!wccflag)
1821 		VTONFS(dvp)->n_attrstamp = 0;
1822 	/*
1823 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1824 	 * if we can succeed in looking up the directory.
1825 	 */
1826 	if (error == EEXIST || (!error && !gotvp)) {
1827 		if (newvp) {
1828 			vrele(newvp);
1829 			newvp = (struct vnode *)0;
1830 		}
1831 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1832 			cnp->cn_proc, &np);
1833 		if (!error) {
1834 			newvp = NFSTOV(np);
1835 			if (newvp->v_type != VDIR)
1836 				error = EEXIST;
1837 		}
1838 	}
1839 	if (error) {
1840 		if (newvp)
1841 			vrele(newvp);
1842 	} else
1843 		*ap->a_vpp = newvp;
1844 	FREE(cnp->cn_pnbuf, M_NAMEI);
1845 	vrele(dvp);
1846 	return (error);
1847 }
1848 
1849 /*
1850  * nfs remove directory call
1851  */
1852 int
1853 nfs_rmdir(v)
1854 	void *v;
1855 {
1856 	struct vop_rmdir_args /* {
1857 		struct vnode *a_dvp;
1858 		struct vnode *a_vp;
1859 		struct componentname *a_cnp;
1860 	} */ *ap = v;
1861 	register struct vnode *vp = ap->a_vp;
1862 	register struct vnode *dvp = ap->a_dvp;
1863 	register struct componentname *cnp = ap->a_cnp;
1864 	register u_int32_t *tl;
1865 	register caddr_t cp;
1866 	register int32_t t1, t2;
1867 	caddr_t bpos, dpos, cp2;
1868 	int error = 0, wccflag = NFSV3_WCCRATTR;
1869 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1870 	int v3 = NFS_ISV3(dvp);
1871 
1872 	if (dvp == vp) {
1873 		vrele(dvp);
1874 		vrele(dvp);
1875 		FREE(cnp->cn_pnbuf, M_NAMEI);
1876 		return (EINVAL);
1877 	}
1878 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1879 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
1880 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1881 	nfsm_fhtom(dvp, v3);
1882 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1883 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
1884 	if (v3)
1885 		nfsm_wcc_data(dvp, wccflag);
1886 	nfsm_reqdone;
1887 	FREE(cnp->cn_pnbuf, M_NAMEI);
1888 	VTONFS(dvp)->n_flag |= NMODIFIED;
1889 	if (!wccflag)
1890 		VTONFS(dvp)->n_attrstamp = 0;
1891 	cache_purge(dvp);
1892 	cache_purge(vp);
1893 	vrele(vp);
1894 	vrele(dvp);
1895 	/*
1896 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1897 	 */
1898 	if (error == ENOENT)
1899 		error = 0;
1900 	return (error);
1901 }
1902 
1903 /*
1904  * nfs readdir call
1905  */
1906 int
1907 nfs_readdir(v)
1908 	void *v;
1909 {
1910 	struct vop_readdir_args /* {
1911 		struct vnode *a_vp;
1912 		struct uio *a_uio;
1913 		struct ucred *a_cred;
1914 		int *a_eofflag;
1915 		u_long *a_cookies;
1916 		int a_ncookies;
1917 	} */ *ap = v;
1918 	register struct vnode *vp = ap->a_vp;
1919 	register struct nfsnode *np = VTONFS(vp);
1920 	register struct uio *uio = ap->a_uio;
1921 	char *base = uio->uio_iov->iov_base;
1922 	off_t off = uio->uio_offset;
1923 	int tresid, error;
1924 	struct vattr vattr;
1925 
1926 	if (vp->v_type != VDIR)
1927 		return (EPERM);
1928 	/*
1929 	 * First, check for hit on the EOF offset cache
1930 	 */
1931 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1932 	    (np->n_flag & NMODIFIED) == 0) {
1933 		if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
1934 			if (NQNFS_CKCACHABLE(vp, ND_READ)) {
1935 				nfsstats.direofcache_hits++;
1936 				*ap->a_eofflag = 1;
1937 				return (0);
1938 			}
1939 		} else if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_procp) == 0 &&
1940 			np->n_mtime == vattr.va_mtime.tv_sec) {
1941 			nfsstats.direofcache_hits++;
1942 			*ap->a_eofflag = 1;
1943 			return (0);
1944 		}
1945 	}
1946 
1947 	/*
1948 	 * Call nfs_bioread() to do the real work.
1949 	 */
1950 	tresid = uio->uio_resid;
1951 	error = nfs_bioread(vp, uio, 0, ap->a_cred);
1952 
1953 	if (!error && uio->uio_resid == tresid) {
1954 		nfsstats.direofcache_misses++;
1955 		*ap->a_eofflag = 1;
1956 		return (0);
1957 	}
1958 
1959 	if (!error && ap->a_cookies) {
1960 		struct dirent *dp;
1961 		u_long *cookies = ap->a_cookies;
1962 		int ncookies = ap->a_ncookies;
1963 
1964 		/*
1965 		 * Only the NFS server and emulations use cookies, and they
1966 		 * load the directory block into system space, so we can
1967 		 * just look at it directly.
1968 		 */
1969 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1970 			panic("nfs_readdir: lost in space");
1971 		while (ncookies-- && base < uio->uio_iov->iov_base) {
1972 			dp = (struct dirent *) base;
1973 			if (dp->d_reclen == 0)
1974 				break;
1975 			off += dp->d_reclen;
1976 			*(cookies++) = off;
1977 			base += dp->d_reclen;
1978 		}
1979 		uio->uio_resid += (uio->uio_iov->iov_base - base);
1980 		uio->uio_iov->iov_len += (uio->uio_iov->iov_base - base);
1981 		uio->uio_iov->iov_base = base;
1982 	}
1983 
1984 	*ap->a_eofflag = 0;
1985 	return (error);
1986 }
1987 
1988 /*
1989  * Readdir rpc call.
1990  * Called from below the buffer cache by nfs_doio().
1991  */
1992 int
1993 nfs_readdirrpc(vp, uiop, cred)
1994 	struct vnode *vp;
1995 	register struct uio *uiop;
1996 	struct ucred *cred;
1997 {
1998 	register int len, left;
1999 	register struct dirent *dp = NULL;
2000 	register u_int32_t *tl;
2001 	register caddr_t cp;
2002 	register int32_t t1, t2;
2003 	register nfsuint64 *cookiep;
2004 	caddr_t bpos, dpos, cp2;
2005 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2006 	nfsuint64 cookie;
2007 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2008 	struct nfsnode *dnp = VTONFS(vp);
2009 	u_quad_t fileno;
2010 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2011 	int attrflag;
2012 	int v3 = NFS_ISV3(vp);
2013 
2014 #ifndef DIAGNOSTIC
2015 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (NFS_DIRBLKSIZ - 1)) ||
2016 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2017 		panic("nfs readdirrpc bad uio");
2018 #endif
2019 
2020 	/*
2021 	 * If there is no cookie, assume end of directory.
2022 	 */
2023 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2024 	if (cookiep)
2025 		cookie = *cookiep;
2026 	else
2027 		return (0);
2028 	/*
2029 	 * Loop around doing readdir rpc's of size nm_readdirsize
2030 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2031 	 * The stopping criteria is EOF or buffer full.
2032 	 */
2033 	while (more_dirs && bigenough) {
2034 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2035 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2036 			NFSX_READDIR(v3));
2037 		nfsm_fhtom(vp, v3);
2038 		if (v3) {
2039 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2040 			*tl++ = cookie.nfsuquad[0];
2041 			*tl++ = cookie.nfsuquad[1];
2042 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2043 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2044 		} else {
2045 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2046 			*tl++ = cookie.nfsuquad[0];
2047 		}
2048 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2049 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2050 		if (v3) {
2051 			nfsm_postop_attr(vp, attrflag);
2052 			if (!error) {
2053 				nfsm_dissect(tl, u_int32_t *,
2054 				    2 * NFSX_UNSIGNED);
2055 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2056 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2057 			} else {
2058 				m_freem(mrep);
2059 				goto nfsmout;
2060 			}
2061 		}
2062 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2063 		more_dirs = fxdr_unsigned(int, *tl);
2064 
2065 		/* loop thru the dir entries, doctoring them to 4bsd form */
2066 		while (more_dirs && bigenough) {
2067 			if (v3) {
2068 				nfsm_dissect(tl, u_int32_t *,
2069 				    3 * NFSX_UNSIGNED);
2070 				fxdr_hyper(tl, &fileno);
2071 				len = fxdr_unsigned(int, *(tl + 2));
2072 			} else {
2073 				nfsm_dissect(tl, u_int32_t *,
2074 				    2 * NFSX_UNSIGNED);
2075 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2076 				len = fxdr_unsigned(int, *tl);
2077 			}
2078 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2079 				error = EBADRPC;
2080 				m_freem(mrep);
2081 				goto nfsmout;
2082 			}
2083 			tlen = nfsm_rndup(len);
2084 			if (tlen == len)
2085 				tlen += 4;	/* To ensure null termination */
2086 			left = NFS_READDIRBLKSIZ - blksiz;
2087 			if ((tlen + DIRHDSIZ) > left) {
2088 				dp->d_reclen += left;
2089 				uiop->uio_iov->iov_base += left;
2090 				uiop->uio_iov->iov_len -= left;
2091 				uiop->uio_offset += left;
2092 				uiop->uio_resid -= left;
2093 				blksiz = 0;
2094 			}
2095 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2096 				bigenough = 0;
2097 			if (bigenough) {
2098 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2099 				dp->d_fileno = (int)fileno;
2100 				dp->d_namlen = len;
2101 				dp->d_reclen = tlen + DIRHDSIZ;
2102 				dp->d_type = DT_UNKNOWN;
2103 				blksiz += dp->d_reclen;
2104 				if (blksiz == NFS_READDIRBLKSIZ)
2105 					blksiz = 0;
2106 				uiop->uio_offset += DIRHDSIZ;
2107 				uiop->uio_resid -= DIRHDSIZ;
2108 				uiop->uio_iov->iov_base += DIRHDSIZ;
2109 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2110 				nfsm_mtouio(uiop, len);
2111 				cp = uiop->uio_iov->iov_base;
2112 				tlen -= len;
2113 				*cp = '\0';	/* null terminate */
2114 				uiop->uio_iov->iov_base += tlen;
2115 				uiop->uio_iov->iov_len -= tlen;
2116 				uiop->uio_offset += tlen;
2117 				uiop->uio_resid -= tlen;
2118 			} else
2119 				nfsm_adv(nfsm_rndup(len));
2120 			if (v3) {
2121 				nfsm_dissect(tl, u_int32_t *,
2122 				    3 * NFSX_UNSIGNED);
2123 			} else {
2124 				nfsm_dissect(tl, u_int32_t *,
2125 				    2 * NFSX_UNSIGNED);
2126 			}
2127 			if (bigenough) {
2128 				cookie.nfsuquad[0] = *tl++;
2129 				if (v3)
2130 					cookie.nfsuquad[1] = *tl++;
2131 			} else if (v3)
2132 				tl += 2;
2133 			else
2134 				tl++;
2135 			more_dirs = fxdr_unsigned(int, *tl);
2136 		}
2137 		/*
2138 		 * If at end of rpc data, get the eof boolean
2139 		 */
2140 		if (!more_dirs) {
2141 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2142 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2143 		}
2144 		m_freem(mrep);
2145 	}
2146 	/*
2147 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2148 	 * by increasing d_reclen for the last record.
2149 	 */
2150 	if (blksiz > 0) {
2151 		left = NFS_READDIRBLKSIZ - blksiz;
2152 		dp->d_reclen += left;
2153 		uiop->uio_iov->iov_base += left;
2154 		uiop->uio_iov->iov_len -= left;
2155 		uiop->uio_offset += left;
2156 		uiop->uio_resid -= left;
2157 	}
2158 
2159 	/*
2160 	 * We are now either at the end of the directory or have filled the
2161 	 * block.
2162 	 */
2163 	if (bigenough)
2164 		dnp->n_direofoffset = uiop->uio_offset;
2165 	else {
2166 		if (uiop->uio_resid > 0)
2167 			printf("EEK! readdirrpc resid > 0\n");
2168 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2169 		*cookiep = cookie;
2170 	}
2171 nfsmout:
2172 	return (error);
2173 }
2174 
2175 /*
2176  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2177  */
2178 int
2179 nfs_readdirplusrpc(vp, uiop, cred)
2180 	struct vnode *vp;
2181 	register struct uio *uiop;
2182 	struct ucred *cred;
2183 {
2184 	register int len, left;
2185 	register struct dirent *dp = NULL;
2186 	register u_int32_t *tl;
2187 	register caddr_t cp;
2188 	register int32_t t1, t2;
2189 	register struct vnode *newvp;
2190 	register nfsuint64 *cookiep;
2191 	caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2192 	struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2193 	struct nameidata nami, *ndp = &nami;
2194 	struct componentname *cnp = &ndp->ni_cnd;
2195 	nfsuint64 cookie;
2196 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2197 	struct nfsnode *dnp = VTONFS(vp), *np;
2198 	const unsigned char *hcp;
2199 	nfsfh_t *fhp;
2200 	u_quad_t fileno;
2201 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2202 	int attrflag, fhsize;
2203 
2204 #ifndef DIAGNOSTIC
2205 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (NFS_DIRBLKSIZ - 1)) ||
2206 		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2207 		panic("nfs readdirplusrpc bad uio");
2208 #endif
2209 	ndp->ni_dvp = vp;
2210 	newvp = NULLVP;
2211 
2212 	/*
2213 	 * If there is no cookie, assume end of directory.
2214 	 */
2215 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2216 	if (cookiep)
2217 		cookie = *cookiep;
2218 	else
2219 		return (0);
2220 	/*
2221 	 * Loop around doing readdir rpc's of size nm_readdirsize
2222 	 * truncated to a multiple of NFS_READDIRBLKSIZ.
2223 	 * The stopping criteria is EOF or buffer full.
2224 	 */
2225 	while (more_dirs && bigenough) {
2226 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2227 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2228 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2229 		nfsm_fhtom(vp, 1);
2230  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2231 		*tl++ = cookie.nfsuquad[0];
2232 		*tl++ = cookie.nfsuquad[1];
2233 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2234 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2235 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2236 		*tl = txdr_unsigned(nmp->nm_rsize);
2237 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2238 		nfsm_postop_attr(vp, attrflag);
2239 		if (error) {
2240 			m_freem(mrep);
2241 			goto nfsmout;
2242 		}
2243 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2244 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2245 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2246 		more_dirs = fxdr_unsigned(int, *tl);
2247 
2248 		/* loop thru the dir entries, doctoring them to 4bsd form */
2249 		while (more_dirs && bigenough) {
2250 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2251 			fxdr_hyper(tl, &fileno);
2252 			len = fxdr_unsigned(int, *(tl + 2));
2253 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2254 				error = EBADRPC;
2255 				m_freem(mrep);
2256 				goto nfsmout;
2257 			}
2258 			tlen = nfsm_rndup(len);
2259 			if (tlen == len)
2260 				tlen += 4;	/* To ensure null termination*/
2261 			left = NFS_READDIRBLKSIZ - blksiz;
2262 			if ((tlen + DIRHDSIZ) > left) {
2263 				dp->d_reclen += left;
2264 				uiop->uio_iov->iov_base += left;
2265 				uiop->uio_iov->iov_len -= left;
2266 				uiop->uio_offset += left;
2267 				uiop->uio_resid -= left;
2268 				blksiz = 0;
2269 			}
2270 			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2271 				bigenough = 0;
2272 			if (bigenough) {
2273 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2274 				dp->d_fileno = (int)fileno;
2275 				dp->d_namlen = len;
2276 				dp->d_reclen = tlen + DIRHDSIZ;
2277 				dp->d_type = DT_UNKNOWN;
2278 				blksiz += dp->d_reclen;
2279 				if (blksiz == NFS_READDIRBLKSIZ)
2280 					blksiz = 0;
2281 				uiop->uio_offset += DIRHDSIZ;
2282 				uiop->uio_resid -= DIRHDSIZ;
2283 				uiop->uio_iov->iov_base += DIRHDSIZ;
2284 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2285 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2286 				cnp->cn_namelen = len;
2287 				nfsm_mtouio(uiop, len);
2288 				cp = uiop->uio_iov->iov_base;
2289 				tlen -= len;
2290 				*cp = '\0';
2291 				uiop->uio_iov->iov_base += tlen;
2292 				uiop->uio_iov->iov_len -= tlen;
2293 				uiop->uio_offset += tlen;
2294 				uiop->uio_resid -= tlen;
2295 			} else
2296 				nfsm_adv(nfsm_rndup(len));
2297 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2298 			if (bigenough) {
2299 				cookie.nfsuquad[0] = *tl++;
2300 				cookie.nfsuquad[1] = *tl++;
2301 			} else
2302 				tl += 2;
2303 
2304 			/*
2305 			 * Since the attributes are before the file handle
2306 			 * (sigh), we must skip over the attributes and then
2307 			 * come back and get them.
2308 			 */
2309 			attrflag = fxdr_unsigned(int, *tl);
2310 			if (attrflag) {
2311 			    dpossav1 = dpos;
2312 			    mdsav1 = md;
2313 			    nfsm_adv(NFSX_V3FATTR);
2314 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2315 			    doit = fxdr_unsigned(int, *tl);
2316 			    if (doit) {
2317 				nfsm_getfh(fhp, fhsize, 1);
2318 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2319 				    VREF(vp);
2320 				    newvp = vp;
2321 				    np = dnp;
2322 				} else {
2323 				    error = nfs_nget(vp->v_mount, fhp,
2324 					fhsize, &np);
2325 				    if (error)
2326 					doit = 0;
2327 				    else
2328 					newvp = NFSTOV(np);
2329 				}
2330 			    }
2331 			    if (doit) {
2332 				dpossav2 = dpos;
2333 				dpos = dpossav1;
2334 				mdsav2 = md;
2335 				md = mdsav1;
2336 				nfsm_loadattr(newvp, (struct vattr *)0);
2337 				dpos = dpossav2;
2338 				md = mdsav2;
2339 				dp->d_type =
2340 				    IFTODT(VTTOIF(np->n_vattr.va_type));
2341 				ndp->ni_vp = newvp;
2342 				cnp->cn_hash = 0;
2343 				for (hcp = cnp->cn_nameptr, i = 1; i <= len;
2344 				    i++, hcp++)
2345 				    cnp->cn_hash += *hcp * i;
2346 				if (cnp->cn_namelen <= NCHNAMLEN)
2347 				    cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2348 			    }
2349 			} else {
2350 			    /* Just skip over the file handle */
2351 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2352 			    i = fxdr_unsigned(int, *tl);
2353 			    nfsm_adv(nfsm_rndup(i));
2354 			}
2355 			if (newvp != NULLVP) {
2356 			    vrele(newvp);
2357 			    newvp = NULLVP;
2358 			}
2359 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2360 			more_dirs = fxdr_unsigned(int, *tl);
2361 		}
2362 		/*
2363 		 * If at end of rpc data, get the eof boolean
2364 		 */
2365 		if (!more_dirs) {
2366 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2367 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2368 		}
2369 		m_freem(mrep);
2370 	}
2371 	/*
2372 	 * Fill last record, iff any, out to a multiple of NFS_READDIRBLKSIZ
2373 	 * by increasing d_reclen for the last record.
2374 	 */
2375 	if (blksiz > 0) {
2376 		left = NFS_READDIRBLKSIZ - blksiz;
2377 		dp->d_reclen += left;
2378 		uiop->uio_iov->iov_base += left;
2379 		uiop->uio_iov->iov_len -= left;
2380 		uiop->uio_offset += left;
2381 		uiop->uio_resid -= left;
2382 	}
2383 
2384 	/*
2385 	 * We are now either at the end of the directory or have filled the
2386 	 * block.
2387 	 */
2388 	if (bigenough)
2389 		dnp->n_direofoffset = uiop->uio_offset;
2390 	else {
2391 		if (uiop->uio_resid > 0)
2392 			printf("EEK! readdirplusrpc resid > 0\n");
2393 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2394 		*cookiep = cookie;
2395 	}
2396 nfsmout:
2397 	if (newvp != NULLVP)
2398 		vrele(newvp);
2399 	return (error);
2400 }
2401 static char hextoasc[] = "0123456789abcdef";
2402 
2403 /*
2404  * Silly rename. To make the NFS filesystem that is stateless look a little
2405  * more like the "ufs" a remove of an active vnode is translated to a rename
2406  * to a funny looking filename that is removed by nfs_inactive on the
2407  * nfsnode. There is the potential for another process on a different client
2408  * to create the same funny name between the nfs_lookitup() fails and the
2409  * nfs_rename() completes, but...
2410  */
2411 int
2412 nfs_sillyrename(dvp, vp, cnp)
2413 	struct vnode *dvp, *vp;
2414 	struct componentname *cnp;
2415 {
2416 	register struct sillyrename *sp;
2417 	struct nfsnode *np;
2418 	int error;
2419 	short pid;
2420 
2421 	cache_purge(dvp);
2422 	np = VTONFS(vp);
2423 #ifndef DIAGNOSTIC
2424 	if (vp->v_type == VDIR)
2425 		panic("nfs: sillyrename dir");
2426 #endif
2427 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2428 		M_NFSREQ, M_WAITOK);
2429 	sp->s_cred = crdup(cnp->cn_cred);
2430 	sp->s_dvp = dvp;
2431 	VREF(dvp);
2432 
2433 	/* Fudge together a funny name */
2434 	pid = cnp->cn_proc->p_pid;
2435 	bcopy(".nfsAxxxx4.4", sp->s_name, 13);
2436 	sp->s_namlen = 12;
2437 	sp->s_name[8] = hextoasc[pid & 0xf];
2438 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2439 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2440 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2441 
2442 	/* Try lookitups until we get one that isn't there */
2443 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2444 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2445 		sp->s_name[4]++;
2446 		if (sp->s_name[4] > 'z') {
2447 			error = EINVAL;
2448 			goto bad;
2449 		}
2450 	}
2451 	error = nfs_renameit(dvp, cnp, sp);
2452 	if (error)
2453 		goto bad;
2454 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2455 		cnp->cn_proc, &np);
2456 	np->n_sillyrename = sp;
2457 	return (0);
2458 bad:
2459 	vrele(sp->s_dvp);
2460 	crfree(sp->s_cred);
2461 	free((caddr_t)sp, M_NFSREQ);
2462 	return (error);
2463 }
2464 
2465 /*
2466  * Look up a file name and optionally either update the file handle or
2467  * allocate an nfsnode, depending on the value of npp.
2468  * npp == NULL	--> just do the lookup
2469  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2470  *			handled too
2471  * *npp != NULL --> update the file handle in the vnode
2472  */
2473 int
2474 nfs_lookitup(dvp, name, len, cred, procp, npp)
2475 	register struct vnode *dvp;
2476 	const char *name;
2477 	int len;
2478 	struct ucred *cred;
2479 	struct proc *procp;
2480 	struct nfsnode **npp;
2481 {
2482 	register u_int32_t *tl;
2483 	register caddr_t cp;
2484 	register int32_t t1, t2;
2485 	struct vnode *newvp = (struct vnode *)0;
2486 	struct nfsnode *np, *dnp = VTONFS(dvp);
2487 	caddr_t bpos, dpos, cp2;
2488 	int error = 0, fhlen, attrflag;
2489 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2490 	nfsfh_t *nfhp;
2491 	int v3 = NFS_ISV3(dvp);
2492 
2493 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2494 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2495 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2496 	nfsm_fhtom(dvp, v3);
2497 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2498 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2499 	if (npp && !error) {
2500 		nfsm_getfh(nfhp, fhlen, v3);
2501 		if (*npp) {
2502 		    np = *npp;
2503 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2504 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2505 			np->n_fhp = &np->n_fh;
2506 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2507 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2508 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2509 		    np->n_fhsize = fhlen;
2510 		    newvp = NFSTOV(np);
2511 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2512 		    VREF(dvp);
2513 		    newvp = dvp;
2514 		} else {
2515 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2516 		    if (error) {
2517 			m_freem(mrep);
2518 			return (error);
2519 		    }
2520 		    newvp = NFSTOV(np);
2521 		}
2522 		if (v3) {
2523 			nfsm_postop_attr(newvp, attrflag);
2524 			if (!attrflag && *npp == NULL) {
2525 				m_freem(mrep);
2526 				vrele(newvp);
2527 				return (ENOENT);
2528 			}
2529 		} else
2530 			nfsm_loadattr(newvp, (struct vattr *)0);
2531 	}
2532 	nfsm_reqdone;
2533 	if (npp && *npp == NULL) {
2534 		if (error) {
2535 			if (newvp)
2536 				vrele(newvp);
2537 		} else
2538 			*npp = np;
2539 	}
2540 	return (error);
2541 }
2542 
2543 /*
2544  * Nfs Version 3 commit rpc
2545  */
2546 int
2547 nfs_commit(vp, offset, cnt, cred, procp)
2548 	register struct vnode *vp;
2549 	u_quad_t offset;
2550 	int cnt;
2551 	struct ucred *cred;
2552 	struct proc *procp;
2553 {
2554 	register caddr_t cp;
2555 	register u_int32_t *tl;
2556 	register int32_t t1, t2;
2557 	register struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2558 	caddr_t bpos, dpos, cp2;
2559 	int error = 0, wccflag = NFSV3_WCCRATTR;
2560 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2561 
2562 	if ((nmp->nm_flag & NFSMNT_HASWRITEVERF) == 0)
2563 		return (0);
2564 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2565 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2566 	nfsm_fhtom(vp, 1);
2567 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2568 	txdr_hyper(&offset, tl);
2569 	tl += 2;
2570 	*tl = txdr_unsigned(cnt);
2571 	nfsm_request(vp, NFSPROC_COMMIT, procp, cred);
2572 	nfsm_wcc_data(vp, wccflag);
2573 	if (!error) {
2574 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2575 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2576 			NFSX_V3WRITEVERF)) {
2577 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2578 				NFSX_V3WRITEVERF);
2579 			error = NFSERR_STALEWRITEVERF;
2580 		}
2581 	}
2582 	nfsm_reqdone;
2583 	return (error);
2584 }
2585 
2586 /*
2587  * Kludge City..
2588  * - make nfs_bmap() essentially a no-op that does no translation
2589  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2590  *   (Maybe I could use the process's page mapping, but I was concerned that
2591  *    Kernel Write might not be enabled and also figured copyout() would do
2592  *    a lot more work than bcopy() and also it currently happens in the
2593  *    context of the swapper process (2).
2594  */
2595 int
2596 nfs_bmap(v)
2597 	void *v;
2598 {
2599 	struct vop_bmap_args /* {
2600 		struct vnode *a_vp;
2601 		daddr_t  a_bn;
2602 		struct vnode **a_vpp;
2603 		daddr_t *a_bnp;
2604 		int *a_runp;
2605 	} */ *ap = v;
2606 	register struct vnode *vp = ap->a_vp;
2607 
2608 	if (ap->a_vpp != NULL)
2609 		*ap->a_vpp = vp;
2610 	if (ap->a_bnp != NULL)
2611 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2612 	return (0);
2613 }
2614 
2615 /*
2616  * Strategy routine.
2617  * For async requests when nfsiod(s) are running, queue the request by
2618  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2619  * request.
2620  */
2621 int
2622 nfs_strategy(v)
2623 	void *v;
2624 {
2625 	struct vop_strategy_args *ap = v;
2626 	register struct buf *bp = ap->a_bp;
2627 	struct ucred *cr;
2628 	struct proc *p;
2629 	int error = 0;
2630 
2631 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2632 		panic("nfs physio/async");
2633 	if (bp->b_flags & B_ASYNC)
2634 		p = (struct proc *)0;
2635 	else
2636 		p = curproc;	/* XXX */
2637 	if (bp->b_flags & B_READ)
2638 		cr = bp->b_rcred;
2639 	else
2640 		cr = bp->b_wcred;
2641 	/*
2642 	 * If the op is asynchronous and an i/o daemon is waiting
2643 	 * queue the request, wake it up and wait for completion
2644 	 * otherwise just do it ourselves.
2645 	 */
2646 	if ((bp->b_flags & B_ASYNC) == 0 ||
2647 		nfs_asyncio(bp, NOCRED))
2648 		error = nfs_doio(bp, cr, p);
2649 	return (error);
2650 }
2651 
2652 /*
2653  * Mmap a file
2654  *
2655  * NB Currently unsupported.
2656  */
2657 /* ARGSUSED */
2658 int
2659 nfs_mmap(v)
2660 	void *v;
2661 {
2662 #if 0
2663 	struct vop_mmap_args /* {
2664 		struct vnode *a_vp;
2665 		int a_fflags;
2666 		struct ucred *a_cred;
2667 		struct proc *a_p;
2668 	} */ *ap = v;
2669 #endif
2670 
2671 	return (EINVAL);
2672 }
2673 
2674 /*
2675  * fsync vnode op. Just call nfs_flush() with commit == 1.
2676  */
2677 /* ARGSUSED */
2678 int
2679 nfs_fsync(v)
2680 	void *v;
2681 {
2682 	struct vop_fsync_args /* {
2683 		struct vnodeop_desc *a_desc;
2684 		struct vnode * a_vp;
2685 		struct ucred * a_cred;
2686 		int  a_waitfor;
2687 		struct proc * a_p;
2688 	} */ *ap = v;
2689 
2690 	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2691 }
2692 
2693 /*
2694  * Flush all the blocks associated with a vnode.
2695  * 	Walk through the buffer pool and push any dirty pages
2696  *	associated with the vnode.
2697  */
2698 int
2699 nfs_flush(vp, cred, waitfor, p, commit)
2700 	register struct vnode *vp;
2701 	struct ucred *cred;
2702 	int waitfor;
2703 	struct proc *p;
2704 	int commit;
2705 {
2706 	register struct nfsnode *np = VTONFS(vp);
2707 	register struct buf *bp;
2708 	register int i;
2709 	struct buf *nbp;
2710 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2711 	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2712 	int passone = 1;
2713 	u_quad_t off = (u_quad_t)-1, endoff = 0, toff;
2714 	struct ucred* wcred = NULL;
2715 #ifndef NFS_COMMITBVECSIZ
2716 #define NFS_COMMITBVECSIZ	20
2717 #endif
2718 	struct buf *bvec[NFS_COMMITBVECSIZ];
2719 
2720 	if (nmp->nm_flag & NFSMNT_INT)
2721 		slpflag = PCATCH;
2722 	if (!commit)
2723 		passone = 0;
2724 	/*
2725 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2726 	 * server, but nas not been committed to stable storage on the server
2727 	 * yet. On the first pass, the byte range is worked out and the commit
2728 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2729 	 * job.
2730 	 */
2731 again:
2732 	bvecpos = 0;
2733 	if (NFS_ISV3(vp) && commit) {
2734 		s = splbio();
2735 		for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2736 			nbp = bp->b_vnbufs.le_next;
2737 			if (bvecpos >= NFS_COMMITBVECSIZ)
2738 				break;
2739 			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2740 				!= (B_DELWRI | B_NEEDCOMMIT))
2741 				continue;
2742 			bremfree(bp);
2743 			/*
2744 			 * Work out if all buffers are using the same cred
2745 			 * so we can deal with them all with one commit.
2746 			 */
2747 			if (wcred == NULL)
2748 				wcred = bp->b_wcred;
2749 			else if (wcred != bp->b_wcred)
2750 				wcred = NOCRED;
2751 			bp->b_flags |= (B_BUSY | B_WRITEINPROG);
2752 			/*
2753 			 * A list of these buffers is kept so that the
2754 			 * second loop knows which buffers have actually
2755 			 * been committed. This is necessary, since there
2756 			 * may be a race between the commit rpc and new
2757 			 * uncommitted writes on the file.
2758 			 */
2759 			bvec[bvecpos++] = bp;
2760 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2761 				bp->b_dirtyoff;
2762 			if (toff < off)
2763 				off = toff;
2764 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2765 			if (toff > endoff)
2766 				endoff = toff;
2767 		}
2768 		splx(s);
2769 	}
2770 	if (bvecpos > 0) {
2771 		/*
2772 		 * Commit data on the server, as required.
2773 		 * If all bufs are using the same wcred, then use that with
2774 		 * one call for all of them, otherwise commit each one
2775 		 * separately.
2776 		 */
2777 		if (wcred != NOCRED)
2778 			retv = nfs_commit(vp, off, (int)(endoff - off),
2779 					  wcred, p);
2780 		else {
2781 			retv = 0;
2782 			for (i = 0; i < bvecpos; i++) {
2783 				off_t off, size;
2784 				bp = bvec[i];
2785 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2786 					bp->b_dirtyoff;
2787 				size = (u_quad_t)(bp->b_dirtyend
2788 						  - bp->b_dirtyoff);
2789 				retv = nfs_commit(vp, off, (int)size,
2790 						  bp->b_wcred, p);
2791 				if (retv) break;
2792 			}
2793 		}
2794 
2795 		if (retv == NFSERR_STALEWRITEVERF)
2796 			nfs_clearcommit(vp->v_mount);
2797 		/*
2798 		 * Now, either mark the blocks I/O done or mark the
2799 		 * blocks dirty, depending on whether the commit
2800 		 * succeeded.
2801 		 */
2802 		for (i = 0; i < bvecpos; i++) {
2803 			bp = bvec[i];
2804 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG);
2805 			if (retv)
2806 			    brelse(bp);
2807 			else {
2808 			    vp->v_numoutput++;
2809 			    bp->b_flags |= B_ASYNC;
2810 			    bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
2811 			    bp->b_dirtyoff = bp->b_dirtyend = 0;
2812 			    reassignbuf(bp, vp);
2813 			    biodone(bp);
2814 			}
2815 		}
2816 	}
2817 
2818 	/*
2819 	 * Start/do any write(s) that are required.
2820 	 */
2821 loop:
2822 	s = splbio();
2823 	for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2824 		nbp = bp->b_vnbufs.le_next;
2825 		if (bp->b_flags & B_BUSY) {
2826 			if (waitfor != MNT_WAIT || passone)
2827 				continue;
2828 			bp->b_flags |= B_WANTED;
2829 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
2830 				"nfsfsync", slptimeo);
2831 			splx(s);
2832 			if (error) {
2833 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
2834 				return (EINTR);
2835 			    if (slpflag == PCATCH) {
2836 				slpflag = 0;
2837 				slptimeo = 2 * hz;
2838 			    }
2839 			}
2840 			goto loop;
2841 		}
2842 		if ((bp->b_flags & B_DELWRI) == 0)
2843 			panic("nfs_fsync: not dirty");
2844 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT))
2845 			continue;
2846 		bremfree(bp);
2847 		if (passone || !commit)
2848 		    bp->b_flags |= (B_BUSY|B_ASYNC);
2849 		else
2850 		    bp->b_flags |= (B_BUSY|B_ASYNC|B_WRITEINPROG|B_NEEDCOMMIT);
2851 		splx(s);
2852 		VOP_BWRITE(bp);
2853 		goto loop;
2854 	}
2855 	splx(s);
2856 	if (passone) {
2857 		passone = 0;
2858 		goto again;
2859 	}
2860 	if (waitfor == MNT_WAIT) {
2861 		while (vp->v_numoutput) {
2862 			vp->v_flag |= VBWAIT;
2863 			error = tsleep((caddr_t)&vp->v_numoutput,
2864 				slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
2865 			if (error) {
2866 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
2867 				return (EINTR);
2868 			    if (slpflag == PCATCH) {
2869 				slpflag = 0;
2870 				slptimeo = 2 * hz;
2871 			    }
2872 			}
2873 		}
2874 		if (vp->v_dirtyblkhd.lh_first && commit) {
2875 #if 0
2876 			vprint("nfs_fsync: dirty", vp);
2877 #endif
2878 			goto loop;
2879 		}
2880 	}
2881 	if (np->n_flag & NWRITEERR) {
2882 		error = np->n_error;
2883 		np->n_flag &= ~NWRITEERR;
2884 	}
2885 	return (error);
2886 }
2887 
2888 /*
2889  * Return POSIX pathconf information applicable to nfs.
2890  *
2891  * The NFS V2 protocol doesn't support this, so just return EINVAL
2892  * for V2.
2893  */
2894 /* ARGSUSED */
2895 int
2896 nfs_pathconf(v)
2897 	void *v;
2898 {
2899 #if 0
2900 	struct vop_pathconf_args /* {
2901 		struct vnode *a_vp;
2902 		int a_name;
2903 		register_t *a_retval;
2904 	} */ *ap = v;
2905 #endif
2906 
2907 	return (EINVAL);
2908 }
2909 
2910 /*
2911  * NFS advisory byte-level locks.
2912  */
2913 int
2914 nfs_advlock(v)
2915 	void *v;
2916 {
2917 	struct vop_advlock_args /* {
2918 		struct vnode *a_vp;
2919 		caddr_t  a_id;
2920 		int  a_op;
2921 		struct flock *a_fl;
2922 		int  a_flags;
2923 	} */ *ap = v;
2924 	register struct nfsnode *np = VTONFS(ap->a_vp);
2925 
2926 	return (lf_advlock(&np->n_lockf, np->n_size, ap->a_id, ap->a_op,
2927 	    ap->a_fl, ap->a_flags));
2928 }
2929 
2930 /*
2931  * Print out the contents of an nfsnode.
2932  */
2933 int
2934 nfs_print(v)
2935 	void *v;
2936 {
2937 	struct vop_print_args /* {
2938 		struct vnode *a_vp;
2939 	} */ *ap = v;
2940 	register struct vnode *vp = ap->a_vp;
2941 	register struct nfsnode *np = VTONFS(vp);
2942 
2943 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
2944 	    np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2945 #ifdef FIFO
2946 	if (vp->v_type == VFIFO)
2947 		fifo_printinfo(vp);
2948 #endif
2949 	printf("\n");
2950 	return (0);
2951 }
2952 
2953 /*
2954  * NFS file truncation.
2955  */
2956 int
2957 nfs_truncate(v)
2958 	void *v;
2959 {
2960 #if 0
2961 	struct vop_truncate_args /* {
2962 		struct vnode *a_vp;
2963 		off_t a_length;
2964 		int a_flags;
2965 		struct ucred *a_cred;
2966 		struct proc *a_p;
2967 	} */ *ap = v;
2968 #endif
2969 
2970 	/* Use nfs_setattr */
2971 	return (EOPNOTSUPP);
2972 }
2973 
2974 /*
2975  * NFS update.
2976  */
2977 int
2978 nfs_update(v)
2979 	void *v;
2980 #if 0
2981 	struct vop_update_args /* {
2982 		struct vnode *a_vp;
2983 		struct timespec *a_ta;
2984 		struct timespec *a_tm;
2985 		int a_waitfor;
2986 	} */ *ap = v;
2987 #endif
2988 {
2989 
2990 	/* Use nfs_setattr */
2991 	return (EOPNOTSUPP);
2992 }
2993 
2994 /*
2995  * Just call nfs_writebp() with the force argument set to 1.
2996  */
2997 int
2998 nfs_bwrite(v)
2999 	void *v;
3000 {
3001 	struct vop_bwrite_args /* {
3002 		struct vnode *a_bp;
3003 	} */ *ap = v;
3004 
3005 	return (nfs_writebp(ap->a_bp, 1));
3006 }
3007 
3008 /*
3009  * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3010  * the force flag is one and it also handles the B_NEEDCOMMIT flag.
3011  */
3012 int
3013 nfs_writebp(bp, force)
3014 	register struct buf *bp;
3015 	int force;
3016 {
3017 	register int oldflags = bp->b_flags, retv = 1;
3018 	register struct proc *p = curproc;	/* XXX */
3019 	off_t off;
3020 
3021 	if(!(bp->b_flags & B_BUSY))
3022 		panic("bwrite: buffer is not busy???");
3023 
3024 #ifdef fvdl_debug
3025 	printf("nfs_writebp(%x): vp %x voff %d vend %d doff %d dend %d\n",
3026 	    bp, bp->b_vp, bp->b_validoff, bp->b_validend, bp->b_dirtyoff,
3027 	    bp->b_dirtyend);
3028 #endif
3029 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
3030 
3031 	if (oldflags & B_ASYNC) {
3032 		if (oldflags & B_DELWRI) {
3033 			reassignbuf(bp, bp->b_vp);
3034 		} else if (p) {
3035 			++p->p_stats->p_ru.ru_oublock;
3036 		}
3037 	}
3038 	bp->b_vp->v_numoutput++;
3039 
3040 	/*
3041 	 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not
3042 	 * an actual write will have to be scheduled via. VOP_STRATEGY().
3043 	 * If B_WRITEINPROG is already set, then push it with a write anyhow.
3044 	 */
3045 	if ((oldflags & (B_NEEDCOMMIT | B_WRITEINPROG)) == B_NEEDCOMMIT) {
3046 		off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
3047 		bp->b_flags |= B_WRITEINPROG;
3048 		retv = nfs_commit(bp->b_vp, off, bp->b_dirtyend-bp->b_dirtyoff,
3049 			bp->b_wcred, bp->b_proc);
3050 		bp->b_flags &= ~B_WRITEINPROG;
3051 		if (!retv) {
3052 			bp->b_dirtyoff = bp->b_dirtyend = 0;
3053 			bp->b_flags &= ~B_NEEDCOMMIT;
3054 			biodone(bp);
3055 		} else if (retv == NFSERR_STALEWRITEVERF)
3056 			nfs_clearcommit(bp->b_vp->v_mount);
3057 	}
3058 	if (retv) {
3059 		if (force)
3060 			bp->b_flags |= B_WRITEINPROG;
3061 		VOP_STRATEGY(bp);
3062 	}
3063 
3064 	if( (oldflags & B_ASYNC) == 0) {
3065 		int rtval = biowait(bp);
3066 		if (oldflags & B_DELWRI) {
3067 			reassignbuf(bp, bp->b_vp);
3068 		} else if (p) {
3069 			++p->p_stats->p_ru.ru_oublock;
3070 		}
3071 		brelse(bp);
3072 		return (rtval);
3073 	}
3074 
3075 	return (0);
3076 }
3077 
3078 /*
3079  * nfs special file access vnode op.
3080  * Essentially just get vattr and then imitate iaccess() since the device is
3081  * local to the client.
3082  */
3083 int
3084 nfsspec_access(v)
3085 	void *v;
3086 {
3087 	struct vop_access_args /* {
3088 		struct vnode *a_vp;
3089 		int  a_mode;
3090 		struct ucred *a_cred;
3091 		struct proc *a_p;
3092 	} */ *ap = v;
3093 	struct vattr va;
3094 	struct vnode *vp = ap->a_vp;
3095 	int error;
3096 
3097         /*
3098 	 * Disallow write attempts on filesystems mounted read-only;
3099 	 * unless the file is a socket, fifo, or a block or character
3100 	 * device resident on the filesystem.
3101 	 */
3102 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3103 		switch (vp->v_type) {
3104 		case VREG:
3105 		case VDIR:
3106 		case VLNK:
3107 			return (EROFS);
3108 		default:
3109 			break;
3110 		}
3111 	}
3112 
3113 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3114 	if (error)
3115 		return (error);
3116 
3117 	return (vaccess(va.va_mode, va.va_uid, va.va_gid, ap->a_mode,
3118 	    ap->a_cred));
3119 }
3120 
3121 /*
3122  * Read wrapper for special devices.
3123  */
3124 int
3125 nfsspec_read(v)
3126 	void *v;
3127 {
3128 	struct vop_read_args /* {
3129 		struct vnode *a_vp;
3130 		struct uio *a_uio;
3131 		int  a_ioflag;
3132 		struct ucred *a_cred;
3133 	} */ *ap = v;
3134 	register struct nfsnode *np = VTONFS(ap->a_vp);
3135 
3136 	/*
3137 	 * Set access flag.
3138 	 */
3139 	np->n_flag |= NACC;
3140 	np->n_atim.tv_sec = time.tv_sec;
3141 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3142 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3143 }
3144 
3145 /*
3146  * Write wrapper for special devices.
3147  */
3148 int
3149 nfsspec_write(v)
3150 	void *v;
3151 {
3152 	struct vop_write_args /* {
3153 		struct vnode *a_vp;
3154 		struct uio *a_uio;
3155 		int  a_ioflag;
3156 		struct ucred *a_cred;
3157 	} */ *ap = v;
3158 	register struct nfsnode *np = VTONFS(ap->a_vp);
3159 
3160 	/*
3161 	 * Set update flag.
3162 	 */
3163 	np->n_flag |= NUPD;
3164 	np->n_mtim.tv_sec = time.tv_sec;
3165 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3166 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3167 }
3168 
3169 /*
3170  * Close wrapper for special devices.
3171  *
3172  * Update the times on the nfsnode then do device close.
3173  */
3174 int
3175 nfsspec_close(v)
3176 	void *v;
3177 {
3178 	struct vop_close_args /* {
3179 		struct vnode *a_vp;
3180 		int  a_fflag;
3181 		struct ucred *a_cred;
3182 		struct proc *a_p;
3183 	} */ *ap = v;
3184 	register struct vnode *vp = ap->a_vp;
3185 	register struct nfsnode *np = VTONFS(vp);
3186 	struct vattr vattr;
3187 
3188 	if (np->n_flag & (NACC | NUPD)) {
3189 		np->n_flag |= NCHG;
3190 		if (vp->v_usecount == 1 &&
3191 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3192 			VATTR_NULL(&vattr);
3193 			if (np->n_flag & NACC)
3194 				vattr.va_atime = np->n_atim;
3195 			if (np->n_flag & NUPD)
3196 				vattr.va_mtime = np->n_mtim;
3197 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3198 		}
3199 	}
3200 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3201 }
3202 
3203 #ifdef FIFO
3204 /*
3205  * Read wrapper for fifos.
3206  */
3207 int
3208 nfsfifo_read(v)
3209 	void *v;
3210 {
3211 	struct vop_read_args /* {
3212 		struct vnode *a_vp;
3213 		struct uio *a_uio;
3214 		int  a_ioflag;
3215 		struct ucred *a_cred;
3216 	} */ *ap = v;
3217 	extern int (**fifo_vnodeop_p) __P((void *));
3218 	register 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 	extern int (**fifo_vnodeop_p) __P((void *));
3243 	register struct nfsnode *np = VTONFS(ap->a_vp);
3244 
3245 	/*
3246 	 * Set update flag.
3247 	 */
3248 	np->n_flag |= NUPD;
3249 	np->n_mtim.tv_sec = time.tv_sec;
3250 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3251 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3252 }
3253 
3254 /*
3255  * Close wrapper for fifos.
3256  *
3257  * Update the times on the nfsnode then do fifo close.
3258  */
3259 int
3260 nfsfifo_close(v)
3261 	void *v;
3262 {
3263 	struct vop_close_args /* {
3264 		struct vnode *a_vp;
3265 		int  a_fflag;
3266 		struct ucred *a_cred;
3267 		struct proc *a_p;
3268 	} */ *ap = v;
3269 	register struct vnode *vp = ap->a_vp;
3270 	register struct nfsnode *np = VTONFS(vp);
3271 	struct vattr vattr;
3272 	extern int (**fifo_vnodeop_p) __P((void *));
3273 
3274 	if (np->n_flag & (NACC | NUPD)) {
3275 		if (np->n_flag & NACC) {
3276 			np->n_atim.tv_sec = time.tv_sec;
3277 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3278 		}
3279 		if (np->n_flag & NUPD) {
3280 			np->n_mtim.tv_sec = time.tv_sec;
3281 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3282 		}
3283 		np->n_flag |= NCHG;
3284 		if (vp->v_usecount == 1 &&
3285 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3286 			VATTR_NULL(&vattr);
3287 			if (np->n_flag & NACC)
3288 				vattr.va_atime = np->n_atim;
3289 			if (np->n_flag & NUPD)
3290 				vattr.va_mtime = np->n_mtim;
3291 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3292 		}
3293 	}
3294 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3295 }
3296 #endif /* ! FIFO */
3297