xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 2a399c6883d870daece976daec6ffa7bb7f934ce)
1 /*	$NetBSD: nfs_vnops.c,v 1.86 1997/10/20 22:08:44 thorpej 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 #include <sys/stat.h>
63 
64 #include <vm/vm.h>
65 
66 #include <miscfs/fifofs/fifo.h>
67 #include <miscfs/genfs/genfs.h>
68 #include <miscfs/specfs/specdev.h>
69 
70 #include <nfs/rpcv2.h>
71 #include <nfs/nfsproto.h>
72 #include <nfs/nfs.h>
73 #include <nfs/nfsnode.h>
74 #include <nfs/nfsmount.h>
75 #include <nfs/xdr_subs.h>
76 #include <nfs/nfsm_subs.h>
77 #include <nfs/nqnfs.h>
78 #include <nfs/nfs_var.h>
79 
80 #include <net/if.h>
81 #include <netinet/in.h>
82 #include <netinet/in_var.h>
83 
84 /* Defs */
85 #define	TRUE	1
86 #define	FALSE	0
87 
88 /*
89  * Global vfs data structures for nfs
90  */
91 int (**nfsv2_vnodeop_p) __P((void *));
92 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
93 	{ &vop_default_desc, vn_default_error },
94 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
95 	{ &vop_create_desc, nfs_create },		/* create */
96 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
97 	{ &vop_open_desc, nfs_open },			/* open */
98 	{ &vop_close_desc, nfs_close },			/* close */
99 	{ &vop_access_desc, nfs_access },		/* access */
100 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
101 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
102 	{ &vop_read_desc, nfs_read },			/* read */
103 	{ &vop_write_desc, nfs_write },			/* write */
104 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
105 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
106 	{ &vop_poll_desc, nfs_poll },			/* poll */
107 #ifdef Lite2_integrated
108 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
109 #endif
110 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
111 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
112 	{ &vop_seek_desc, nfs_seek },			/* seek */
113 	{ &vop_remove_desc, nfs_remove },		/* remove */
114 	{ &vop_link_desc, nfs_link },			/* link */
115 	{ &vop_rename_desc, nfs_rename },		/* rename */
116 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
117 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
118 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
119 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
120 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
121 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
122 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
123 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
124 	{ &vop_lock_desc, nfs_lock },			/* lock */
125 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
126 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
127 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
128 	{ &vop_print_desc, nfs_print },			/* print */
129 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
130 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
131 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
132 	{ &vop_blkatoff_desc, nfs_blkatoff },		/* blkatoff */
133 	{ &vop_valloc_desc, nfs_valloc },		/* valloc */
134 	{ &vop_reallocblks_desc, nfs_reallocblks },	/* reallocblks */
135 	{ &vop_vfree_desc, nfs_vfree },			/* vfree */
136 	{ &vop_truncate_desc, nfs_truncate },		/* truncate */
137 	{ &vop_update_desc, nfs_update },		/* update */
138 	{ &vop_bwrite_desc, nfs_bwrite },		/* bwrite */
139 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
140 };
141 struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
142 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
143 
144 /*
145  * Special device vnode ops
146  */
147 int (**spec_nfsv2nodeop_p) __P((void *));
148 struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
149 	{ &vop_default_desc, vn_default_error },
150 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
151 	{ &vop_create_desc, spec_create },		/* create */
152 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
153 	{ &vop_open_desc, spec_open },			/* open */
154 	{ &vop_close_desc, nfsspec_close },		/* close */
155 	{ &vop_access_desc, nfsspec_access },		/* access */
156 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
157 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
158 	{ &vop_read_desc, nfsspec_read },		/* read */
159 	{ &vop_write_desc, nfsspec_write },		/* write */
160 	{ &vop_lease_desc, spec_lease_check },		/* lease */
161 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
162 	{ &vop_poll_desc, spec_poll },			/* poll */
163 #ifdef Lite2_integrated
164 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
165 #endif
166 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
167 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
168 	{ &vop_seek_desc, spec_seek },			/* seek */
169 	{ &vop_remove_desc, spec_remove },		/* remove */
170 	{ &vop_link_desc, spec_link },			/* link */
171 	{ &vop_rename_desc, spec_rename },		/* rename */
172 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
173 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
174 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
175 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
176 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
177 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
178 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
179 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
180 	{ &vop_lock_desc, nfs_lock },			/* lock */
181 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
182 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
183 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
184 	{ &vop_print_desc, nfs_print },			/* print */
185 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
186 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
187 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
188 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
189 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
190 	{ &vop_reallocblks_desc, spec_reallocblks },	/* reallocblks */
191 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
192 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
193 	{ &vop_update_desc, nfs_update },		/* update */
194 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
195 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
196 };
197 struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
198 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
199 
200 #ifdef FIFO
201 int (**fifo_nfsv2nodeop_p) __P((void *));
202 struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
203 	{ &vop_default_desc, vn_default_error },
204 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
205 	{ &vop_create_desc, fifo_create },		/* create */
206 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
207 	{ &vop_open_desc, fifo_open },			/* open */
208 	{ &vop_close_desc, nfsfifo_close },		/* close */
209 	{ &vop_access_desc, nfsspec_access },		/* access */
210 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
211 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
212 	{ &vop_read_desc, nfsfifo_read },		/* read */
213 	{ &vop_write_desc, nfsfifo_write },		/* write */
214 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
215 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
216 	{ &vop_poll_desc, fifo_poll },			/* poll */
217 #ifdef Lite2_integrated
218 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
219 #endif
220 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
221 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
222 	{ &vop_seek_desc, fifo_seek },			/* seek */
223 	{ &vop_remove_desc, fifo_remove },		/* remove */
224 	{ &vop_link_desc, fifo_link },			/* link */
225 	{ &vop_rename_desc, fifo_rename },		/* rename */
226 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
227 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
228 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
229 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
230 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
231 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
232 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
233 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
234 	{ &vop_lock_desc, nfs_lock },			/* lock */
235 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
236 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
237 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
238 	{ &vop_print_desc, nfs_print },			/* print */
239 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
240 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
241 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
242 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
243 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
244 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
245 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
246 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
247 	{ &vop_update_desc, nfs_update },		/* update */
248 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
249 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
250 };
251 struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
252 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
253 #endif /* FIFO */
254 
255 /*
256  * Global variables
257  */
258 extern u_int32_t nfs_true, nfs_false;
259 extern u_int32_t nfs_xdrneg1;
260 extern struct nfsstats nfsstats;
261 extern nfstype nfsv3_type[9];
262 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
263 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
264 int nfs_numasync = 0;
265 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
266 
267 /*
268  * nfs null call from vfs.
269  */
270 int
271 nfs_null(vp, cred, procp)
272 	struct vnode *vp;
273 	struct ucred *cred;
274 	struct proc *procp;
275 {
276 	caddr_t bpos, dpos;
277 	int error = 0;
278 	struct mbuf *mreq, *mrep, *md, *mb;
279 
280 	nfsm_reqhead(vp, NFSPROC_NULL, 0);
281 	nfsm_request(vp, NFSPROC_NULL, procp, cred);
282 	nfsm_reqdone;
283 	return (error);
284 }
285 
286 /*
287  * nfs access vnode op.
288  * For nfs version 2, just return ok. File accesses may fail later.
289  * For nfs version 3, use the access rpc to check accessibility. If file modes
290  * are changed on the server, accesses might still fail later.
291  */
292 int
293 nfs_access(v)
294 	void *v;
295 {
296 	struct vop_access_args /* {
297 		struct vnode *a_vp;
298 		int  a_mode;
299 		struct ucred *a_cred;
300 		struct proc *a_p;
301 	} */ *ap = v;
302 	register struct vnode *vp = ap->a_vp;
303 	register u_int32_t *tl;
304 	register caddr_t cp;
305 	register int32_t t1, t2;
306 	caddr_t bpos, dpos, cp2;
307 	int error = 0, attrflag;
308 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
309 	u_int32_t mode, rmode;
310 	int v3 = NFS_ISV3(vp);
311 
312 	/*
313 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
314 	 * ufs_access() locally using the vattr. This may not be correct,
315 	 * since the server may apply other access criteria such as
316 	 * client uid-->server uid mapping that we do not know about, but
317 	 * this is better than just returning anything that is lying about
318 	 * in the cache.
319 	 */
320 	if (v3) {
321 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
322 		nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
323 		nfsm_fhtom(vp, v3);
324 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
325 		if (ap->a_mode & VREAD)
326 			mode = NFSV3ACCESS_READ;
327 		else
328 			mode = 0;
329 		if (vp->v_type != VDIR) {
330 			if (ap->a_mode & VWRITE)
331 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
332 			if (ap->a_mode & VEXEC)
333 				mode |= NFSV3ACCESS_EXECUTE;
334 		} else {
335 			if (ap->a_mode & VWRITE)
336 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
337 					 NFSV3ACCESS_DELETE);
338 			if (ap->a_mode & VEXEC)
339 				mode |= NFSV3ACCESS_LOOKUP;
340 		}
341 		*tl = txdr_unsigned(mode);
342 		nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
343 		nfsm_postop_attr(vp, attrflag);
344 		if (!error) {
345 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
346 			rmode = fxdr_unsigned(u_int32_t, *tl);
347 			/*
348 			 * The NFS V3 spec does not clarify whether or not
349 			 * the returned access bits can be a superset of
350 			 * the ones requested, so...
351 			 */
352 			if ((rmode & mode) != mode)
353 				error = EACCES;
354 		}
355 		nfsm_reqdone;
356 		if (error)
357 			return (error);
358 	} else
359 		return (nfsspec_access(ap));
360 	/*
361 	 * Disallow write attempts on filesystems mounted read-only;
362 	 * unless the file is a socket, fifo, or a block or character
363 	 * device resident on the filesystem.
364 	 */
365 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
366 		switch (vp->v_type) {
367 		case VREG:
368 		case VDIR:
369 		case VLNK:
370 			return (EROFS);
371 		default:
372 			break;
373 		}
374 	}
375 	return (0);
376 }
377 
378 /*
379  * nfs open vnode op
380  * Check to see if the type is ok
381  * and that deletion is not in progress.
382  * For paged in text files, you will need to flush the page cache
383  * if consistency is lost.
384  */
385 /* ARGSUSED */
386 int
387 nfs_open(v)
388 	void *v;
389 {
390 	struct vop_open_args /* {
391 		struct vnode *a_vp;
392 		int  a_mode;
393 		struct ucred *a_cred;
394 		struct proc *a_p;
395 	} */ *ap = v;
396 	register struct vnode *vp = ap->a_vp;
397 	struct nfsnode *np = VTONFS(vp);
398 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
399 	struct vattr vattr;
400 	int error;
401 
402 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
403 #ifdef DIAGNOSTIC
404 		printf("open eacces vtyp=%d\n",vp->v_type);
405 #endif
406 		return (EACCES);
407 	}
408 	/*
409 	 * Get a valid lease. If cached data is stale, flush it.
410 	 */
411 	if (nmp->nm_flag & NFSMNT_NQNFS) {
412 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
413 		    do {
414 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
415 			    ap->a_p);
416 		    } while (error == NQNFS_EXPIRED);
417 		    if (error)
418 			return (error);
419 		    if (np->n_lrev != np->n_brev ||
420 			(np->n_flag & NQNFSNONCACHE)) {
421 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
422 				ap->a_p, 1)) == EINTR)
423 				return (error);
424 			(void) vnode_pager_uncache(vp);
425 			np->n_brev = np->n_lrev;
426 		    }
427 		}
428 	} else {
429 		if (np->n_flag & NMODIFIED) {
430 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
431 				ap->a_p, 1)) == EINTR)
432 				return (error);
433 			(void) vnode_pager_uncache(vp);
434 			np->n_attrstamp = 0;
435 			if (vp->v_type == VDIR) {
436 				nfs_invaldircache(vp, 0);
437 				np->n_direofoffset = 0;
438 			}
439 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
440 			if (error)
441 				return (error);
442 			np->n_mtime = vattr.va_mtime.tv_sec;
443 		} else {
444 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
445 			if (error)
446 				return (error);
447 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
448 				if (vp->v_type == VDIR) {
449 					nfs_invaldircache(vp, 0);
450 					np->n_direofoffset = 0;
451 				}
452 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
453 					ap->a_cred, ap->a_p, 1)) == EINTR)
454 					return (error);
455 				(void) vnode_pager_uncache(vp);
456 				np->n_mtime = vattr.va_mtime.tv_sec;
457 			}
458 		}
459 	}
460 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
461 		np->n_attrstamp = 0; /* For Open/Close consistency */
462 	return (0);
463 }
464 
465 /*
466  * nfs close vnode op
467  * What an NFS client should do upon close after writing is a debatable issue.
468  * Most NFS clients push delayed writes to the server upon close, basically for
469  * two reasons:
470  * 1 - So that any write errors may be reported back to the client process
471  *     doing the close system call. By far the two most likely errors are
472  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
473  * 2 - To put a worst case upper bound on cache inconsistency between
474  *     multiple clients for the file.
475  * There is also a consistency problem for Version 2 of the protocol w.r.t.
476  * not being able to tell if other clients are writing a file concurrently,
477  * since there is no way of knowing if the changed modify time in the reply
478  * is only due to the write for this client.
479  * (NFS Version 3 provides weak cache consistency data in the reply that
480  *  should be sufficient to detect and handle this case.)
481  *
482  * The current code does the following:
483  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
484  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
485  *                     or commit them (this satisfies 1 and 2 except for the
486  *                     case where the server crashes after this close but
487  *                     before the commit RPC, which is felt to be "good
488  *                     enough". Changing the last argument to nfs_flush() to
489  *                     a 1 would force a commit operation, if it is felt a
490  *                     commit is necessary now.
491  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
492  *                     1 should be dealt with via an fsync() system call for
493  *                     cases where write errors are important.
494  */
495 /* ARGSUSED */
496 int
497 nfs_close(v)
498 	void *v;
499 {
500 	struct vop_close_args /* {
501 		struct vnodeop_desc *a_desc;
502 		struct vnode *a_vp;
503 		int  a_fflag;
504 		struct ucred *a_cred;
505 		struct proc *a_p;
506 	} */ *ap = v;
507 	register struct vnode *vp = ap->a_vp;
508 	register struct nfsnode *np = VTONFS(vp);
509 	int error = 0;
510 
511 	if (vp->v_type == VREG) {
512 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
513 		(np->n_flag & NMODIFIED)) {
514 		if (NFS_ISV3(vp)) {
515 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
516 		    np->n_flag &= ~NMODIFIED;
517 		} else
518 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
519 		np->n_attrstamp = 0;
520 	    }
521 	    if (np->n_flag & NWRITEERR) {
522 		np->n_flag &= ~NWRITEERR;
523 		error = np->n_error;
524 	    }
525 	}
526 	return (error);
527 }
528 
529 /*
530  * nfs getattr call from vfs.
531  */
532 int
533 nfs_getattr(v)
534 	void *v;
535 {
536 	struct vop_getattr_args /* {
537 		struct vnode *a_vp;
538 		struct vattr *a_vap;
539 		struct ucred *a_cred;
540 		struct proc *a_p;
541 	} */ *ap = v;
542 	register struct vnode *vp = ap->a_vp;
543 	register struct nfsnode *np = VTONFS(vp);
544 	register caddr_t cp;
545 	register u_int32_t *tl;
546 	register int32_t t1, t2;
547 	caddr_t bpos, dpos;
548 	int error = 0;
549 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
550 	int v3 = NFS_ISV3(vp);
551 
552 	/*
553 	 * Update local times for special files.
554 	 */
555 	if (np->n_flag & (NACC | NUPD))
556 		np->n_flag |= NCHG;
557 	/*
558 	 * First look in the cache.
559 	 */
560 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
561 		return (0);
562 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
563 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
564 	nfsm_fhtom(vp, v3);
565 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
566 	if (!error) {
567 		nfsm_loadattr(vp, ap->a_vap);
568 		if (vp->v_type == VDIR &&
569 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
570 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
571 	}
572 	nfsm_reqdone;
573 	return (error);
574 }
575 
576 /*
577  * nfs setattr call.
578  */
579 int
580 nfs_setattr(v)
581 	void *v;
582 {
583 	struct vop_setattr_args /* {
584 		struct vnodeop_desc *a_desc;
585 		struct vnode *a_vp;
586 		struct vattr *a_vap;
587 		struct ucred *a_cred;
588 		struct proc *a_p;
589 	} */ *ap = v;
590 	register struct vnode *vp = ap->a_vp;
591 	register struct nfsnode *np = VTONFS(vp);
592 	register struct vattr *vap = ap->a_vap;
593 	int error = 0;
594 	u_quad_t tsize = 0;
595 
596 	/*
597 	 * Setting of flags is not supported.
598 	 */
599 	if (vap->va_flags != VNOVAL)
600 		return (EOPNOTSUPP);
601 
602 	/*
603 	 * Disallow write attempts if the filesystem is mounted read-only.
604 	 */
605   	if ((vap->va_uid != (uid_t)VNOVAL ||
606 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
607 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
608 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
609 		return (EROFS);
610 	if (vap->va_size != VNOVAL) {
611  		switch (vp->v_type) {
612  		case VDIR:
613  			return (EISDIR);
614  		case VCHR:
615  		case VBLK:
616  		case VSOCK:
617  		case VFIFO:
618 			if (vap->va_mtime.tv_sec == VNOVAL &&
619 			    vap->va_atime.tv_sec == VNOVAL &&
620 			    vap->va_mode == (mode_t)VNOVAL &&
621 			    vap->va_uid == (uid_t)VNOVAL &&
622 			    vap->va_gid == (gid_t)VNOVAL)
623 				return (0);
624  			vap->va_size = VNOVAL;
625  			break;
626  		default:
627 			/*
628 			 * Disallow write attempts if the filesystem is
629 			 * mounted read-only.
630 			 */
631 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
632 				return (EROFS);
633  			vnode_pager_setsize(vp, vap->va_size);
634  			if (vap->va_size == 0)
635  				error = nfs_vinvalbuf(vp, 0,
636  				     ap->a_cred, ap->a_p, 1);
637 			else
638 				error = nfs_vinvalbuf(vp, V_SAVE,
639 				     ap->a_cred, ap->a_p, 1);
640 			if (error) {
641 				vnode_pager_setsize(vp, np->n_size);
642 				return (error);
643 			}
644  			tsize = np->n_size;
645  			np->n_size = np->n_vattr->va_size = vap->va_size;
646   		}
647   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
648 		vap->va_atime.tv_sec != VNOVAL) &&
649 		vp->v_type == VREG &&
650   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
651 		 ap->a_p, 1)) == EINTR)
652 		return (error);
653 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
654 	if (error && vap->va_size != VNOVAL) {
655 		np->n_size = np->n_vattr->va_size = tsize;
656 		vnode_pager_setsize(vp, np->n_size);
657 	}
658 	return (error);
659 }
660 
661 /*
662  * Do an nfs setattr rpc.
663  */
664 int
665 nfs_setattrrpc(vp, vap, cred, procp)
666 	register struct vnode *vp;
667 	register struct vattr *vap;
668 	struct ucred *cred;
669 	struct proc *procp;
670 {
671 	register struct nfsv2_sattr *sp;
672 	register caddr_t cp;
673 	register int32_t t1, t2;
674 	caddr_t bpos, dpos, cp2;
675 	u_int32_t *tl;
676 	int error = 0, wccflag = NFSV3_WCCRATTR;
677 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
678 	int v3 = NFS_ISV3(vp);
679 
680 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
681 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
682 	nfsm_fhtom(vp, v3);
683 	if (v3) {
684 		if (vap->va_mode != (mode_t)VNOVAL) {
685 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
686 			*tl++ = nfs_true;
687 			*tl = txdr_unsigned(vap->va_mode);
688 		} else {
689 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
690 			*tl = nfs_false;
691 		}
692 		if (vap->va_uid != (uid_t)VNOVAL) {
693 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
694 			*tl++ = nfs_true;
695 			*tl = txdr_unsigned(vap->va_uid);
696 		} else {
697 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
698 			*tl = nfs_false;
699 		}
700 		if (vap->va_gid != (gid_t)VNOVAL) {
701 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
702 			*tl++ = nfs_true;
703 			*tl = txdr_unsigned(vap->va_gid);
704 		} else {
705 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
706 			*tl = nfs_false;
707 		}
708 		if (vap->va_size != VNOVAL) {
709 			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
710 			*tl++ = nfs_true;
711 			txdr_hyper(&vap->va_size, tl);
712 		} else {
713 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
714 			*tl = nfs_false;
715 		}
716 		if (vap->va_atime.tv_sec != VNOVAL) {
717 			if (vap->va_atime.tv_sec != time.tv_sec) {
718 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
719 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
720 				txdr_nfsv3time(&vap->va_atime, tl);
721 			} else {
722 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
723 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
724 			}
725 		} else {
726 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
727 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
728 		}
729 		if (vap->va_mtime.tv_sec != VNOVAL) {
730 			if (vap->va_mtime.tv_sec != time.tv_sec) {
731 				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
732 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
733 				txdr_nfsv3time(&vap->va_mtime, tl);
734 			} else {
735 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
736 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
737 			}
738 		} else {
739 			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
740 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
741 		}
742 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
743 		*tl = nfs_false;
744 	} else {
745 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
746 		if (vap->va_mode == (mode_t)VNOVAL)
747 			sp->sa_mode = nfs_xdrneg1;
748 		else
749 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
750 		if (vap->va_uid == (uid_t)VNOVAL)
751 			sp->sa_uid = nfs_xdrneg1;
752 		else
753 			sp->sa_uid = txdr_unsigned(vap->va_uid);
754 		if (vap->va_gid == (gid_t)VNOVAL)
755 			sp->sa_gid = nfs_xdrneg1;
756 		else
757 			sp->sa_gid = txdr_unsigned(vap->va_gid);
758 		sp->sa_size = txdr_unsigned(vap->va_size);
759 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
760 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
761 	}
762 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
763 	if (v3) {
764 		nfsm_wcc_data(vp, wccflag);
765 	} else
766 		nfsm_loadattr(vp, (struct vattr *)0);
767 	nfsm_reqdone;
768 	return (error);
769 }
770 
771 /*
772  * nfs lookup call, one step at a time...
773  * First look in cache
774  * If not found, unlock the directory nfsnode and do the rpc
775  */
776 int
777 nfs_lookup(v)
778 	void *v;
779 {
780 	struct vop_lookup_args /* {
781 		struct vnodeop_desc *a_desc;
782 		struct vnode *a_dvp;
783 		struct vnode **a_vpp;
784 		struct componentname *a_cnp;
785 	} */ *ap = v;
786 	register struct componentname *cnp = ap->a_cnp;
787 	register struct vnode *dvp = ap->a_dvp;
788 	register struct vnode **vpp = ap->a_vpp;
789 	register int flags = cnp->cn_flags;
790 	register struct vnode *newvp;
791 	register u_int32_t *tl;
792 	register caddr_t cp;
793 	register int32_t t1, t2;
794 	struct nfsmount *nmp;
795 	caddr_t bpos, dpos, cp2;
796 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
797 	long len;
798 	nfsfh_t *fhp;
799 	struct nfsnode *np;
800 	int lockparent, wantparent, error = 0, attrflag, fhsize;
801 	int v3 = NFS_ISV3(dvp);
802 
803 	*vpp = NULLVP;
804 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
805 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
806 		return (EROFS);
807 	if (dvp->v_type != VDIR)
808 		return (ENOTDIR);
809 	lockparent = flags & LOCKPARENT;
810 	wantparent = flags & (LOCKPARENT|WANTPARENT);
811 	nmp = VFSTONFS(dvp->v_mount);
812 	np = VTONFS(dvp);
813 	if ((error = cache_lookup(dvp, vpp, cnp)) != 0) {
814 		struct vattr vattr;
815 		int vpid;
816 
817 		if (error == ENOENT) {
818 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
819 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
820 			    VTONFS(dvp)->n_nctime)
821 				return (ENOENT);
822 			cache_purge(dvp);
823 			np->n_nctime = 0;
824 			goto dorpc;
825 		}
826 
827 		newvp = *vpp;
828 		vpid = newvp->v_id;
829 		/*
830 		 * See the comment starting `Step through' in ufs/ufs_lookup.c
831 		 * for an explanation of the locking protocol
832 		 */
833 		if (dvp == newvp) {
834 			VREF(newvp);
835 			error = 0;
836 		} else
837 #ifdef Lite2_integrated
838 			error = vget(newvp, LK_EXCLUSIVE, p);
839 #else
840 			error = vget(newvp, 1);
841 #endif
842 		if (!error) {
843 			if (vpid == newvp->v_id) {
844 			   if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
845 			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
846 				nfsstats.lookupcache_hits++;
847 				if (cnp->cn_nameiop != LOOKUP &&
848 				    (flags & ISLASTCN))
849 					cnp->cn_flags |= SAVENAME;
850 				return (0);
851 			   }
852 			   cache_purge(newvp);
853 			}
854 			vrele(newvp);
855 		}
856 		*vpp = NULLVP;
857 	}
858 dorpc:
859 	error = 0;
860 	newvp = NULLVP;
861 	nfsstats.lookupcache_misses++;
862 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
863 	len = cnp->cn_namelen;
864 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
865 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
866 	nfsm_fhtom(dvp, v3);
867 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
868 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
869 	if (error) {
870 		nfsm_postop_attr(dvp, attrflag);
871 		m_freem(mrep);
872 		goto nfsmout;
873 	}
874 	nfsm_getfh(fhp, fhsize, v3);
875 
876 	/*
877 	 * Handle RENAME case...
878 	 */
879 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
880 		if (NFS_CMPFH(np, fhp, fhsize)) {
881 			m_freem(mrep);
882 			return (EISDIR);
883 		}
884 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
885 		if (error) {
886 			m_freem(mrep);
887 			return (error);
888 		}
889 		newvp = NFSTOV(np);
890 		if (v3) {
891 			nfsm_postop_attr(newvp, attrflag);
892 			nfsm_postop_attr(dvp, attrflag);
893 		} else
894 			nfsm_loadattr(newvp, (struct vattr *)0);
895 		*vpp = newvp;
896 		m_freem(mrep);
897 		cnp->cn_flags |= SAVENAME;
898 		return (0);
899 	}
900 
901 	if (NFS_CMPFH(np, fhp, fhsize)) {
902 		VREF(dvp);
903 		newvp = dvp;
904 	} else {
905 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
906 		if (error) {
907 			m_freem(mrep);
908 			return (error);
909 		}
910 		newvp = NFSTOV(np);
911 	}
912 	if (v3) {
913 		nfsm_postop_attr(newvp, attrflag);
914 		nfsm_postop_attr(dvp, attrflag);
915 	} else
916 		nfsm_loadattr(newvp, (struct vattr *)0);
917 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
918 		cnp->cn_flags |= SAVENAME;
919 	if ((cnp->cn_flags & MAKEENTRY) &&
920 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
921 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
922 		cache_enter(dvp, newvp, cnp);
923 	}
924 	*vpp = newvp;
925 	nfsm_reqdone;
926 	if (error) {
927 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
928 		    cnp->cn_nameiop != CREATE) {
929 			if (VTONFS(dvp)->n_nctime == 0)
930 				VTONFS(dvp)->n_nctime =
931 				    VTONFS(dvp)->n_vattr->va_mtime.tv_sec;
932 			cache_enter(dvp, NULL, cnp);
933 		}
934 		if (newvp != NULLVP)
935 			vrele(newvp);
936 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
937 		    (flags & ISLASTCN) && error == ENOENT) {
938 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
939 				error = EROFS;
940 			else
941 				error = EJUSTRETURN;
942 		}
943 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
944 			cnp->cn_flags |= SAVENAME;
945 	}
946 	return (error);
947 }
948 
949 /*
950  * nfs read call.
951  * Just call nfs_bioread() to do the work.
952  */
953 int
954 nfs_read(v)
955 	void *v;
956 {
957 	struct vop_read_args /* {
958 		struct vnode *a_vp;
959 		struct uio *a_uio;
960 		int  a_ioflag;
961 		struct ucred *a_cred;
962 	} */ *ap = v;
963 	register struct vnode *vp = ap->a_vp;
964 
965 	if (vp->v_type != VREG)
966 		return (EPERM);
967 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
968 }
969 
970 /*
971  * nfs readlink call
972  */
973 int
974 nfs_readlink(v)
975 	void *v;
976 {
977 	struct vop_readlink_args /* {
978 		struct vnode *a_vp;
979 		struct uio *a_uio;
980 		struct ucred *a_cred;
981 	} */ *ap = v;
982 	register struct vnode *vp = ap->a_vp;
983 
984 	if (vp->v_type != VLNK)
985 		return (EPERM);
986 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
987 }
988 
989 /*
990  * Do a readlink rpc.
991  * Called by nfs_doio() from below the buffer cache.
992  */
993 int
994 nfs_readlinkrpc(vp, uiop, cred)
995 	register struct vnode *vp;
996 	struct uio *uiop;
997 	struct ucred *cred;
998 {
999 	register u_int32_t *tl;
1000 	register caddr_t cp;
1001 	register int32_t t1, t2;
1002 	caddr_t bpos, dpos, cp2;
1003 	int error = 0, len, attrflag;
1004 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1005 	int v3 = NFS_ISV3(vp);
1006 
1007 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1008 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1009 	nfsm_fhtom(vp, v3);
1010 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
1011 	if (v3)
1012 		nfsm_postop_attr(vp, attrflag);
1013 	if (!error) {
1014 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1015 		nfsm_mtouio(uiop, len);
1016 	}
1017 	nfsm_reqdone;
1018 	return (error);
1019 }
1020 
1021 /*
1022  * nfs read rpc call
1023  * Ditto above
1024  */
1025 int
1026 nfs_readrpc(vp, uiop, cred)
1027 	register struct vnode *vp;
1028 	struct uio *uiop;
1029 	struct ucred *cred;
1030 {
1031 	register u_int32_t *tl;
1032 	register caddr_t cp;
1033 	register int32_t t1, t2;
1034 	caddr_t bpos, dpos, cp2;
1035 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1036 	struct nfsmount *nmp;
1037 	int error = 0, len, retlen, tsiz, eof, attrflag;
1038 	int v3 = NFS_ISV3(vp);
1039 
1040 #ifndef nolint
1041 	eof = 0;
1042 #endif
1043 	nmp = VFSTONFS(vp->v_mount);
1044 	tsiz = uiop->uio_resid;
1045 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1046 		return (EFBIG);
1047 	while (tsiz > 0) {
1048 		nfsstats.rpccnt[NFSPROC_READ]++;
1049 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1050 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1051 		nfsm_fhtom(vp, v3);
1052 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1053 		if (v3) {
1054 			txdr_hyper(&uiop->uio_offset, tl);
1055 			*(tl + 2) = txdr_unsigned(len);
1056 		} else {
1057 			*tl++ = txdr_unsigned(uiop->uio_offset);
1058 			*tl++ = txdr_unsigned(len);
1059 			*tl = 0;
1060 		}
1061 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp, cred);
1062 		if (v3) {
1063 			nfsm_postop_attr(vp, attrflag);
1064 			if (error) {
1065 				m_freem(mrep);
1066 				goto nfsmout;
1067 			}
1068 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1069 			eof = fxdr_unsigned(int, *(tl + 1));
1070 		} else
1071 			nfsm_loadattr(vp, (struct vattr *)0);
1072 		nfsm_strsiz(retlen, nmp->nm_rsize);
1073 		nfsm_mtouio(uiop, retlen);
1074 		m_freem(mrep);
1075 		tsiz -= retlen;
1076 		if (v3) {
1077 			if (eof || retlen == 0)
1078 				tsiz = 0;
1079 		} else if (retlen < len)
1080 			tsiz = 0;
1081 	}
1082 nfsmout:
1083 	return (error);
1084 }
1085 
1086 /*
1087  * nfs write call
1088  */
1089 int
1090 nfs_writerpc(vp, uiop, cred, iomode, must_commit)
1091 	register struct vnode *vp;
1092 	register struct uio *uiop;
1093 	struct ucred *cred;
1094 	int *iomode, *must_commit;
1095 {
1096 	register u_int32_t *tl;
1097 	register caddr_t cp;
1098 	register int32_t t1, t2, backup;
1099 	caddr_t bpos, dpos, cp2;
1100 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1101 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1102 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1103 	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1104 
1105 #ifndef DIAGNOSTIC
1106 	if (uiop->uio_iovcnt != 1)
1107 		panic("nfs: writerpc iovcnt > 1");
1108 #endif
1109 	*must_commit = 0;
1110 	tsiz = uiop->uio_resid;
1111 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1112 		return (EFBIG);
1113 	while (tsiz > 0) {
1114 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1115 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1116 		nfsm_reqhead(vp, NFSPROC_WRITE,
1117 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1118 		nfsm_fhtom(vp, v3);
1119 		if (v3) {
1120 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1121 			txdr_hyper(&uiop->uio_offset, tl);
1122 			tl += 2;
1123 			*tl++ = txdr_unsigned(len);
1124 			*tl++ = txdr_unsigned(*iomode);
1125 			*tl = txdr_unsigned(len);
1126 		} else {
1127 			register u_int32_t x;
1128 
1129 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1130 			/* Set both "begin" and "current" to non-garbage. */
1131 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1132 			*tl++ = x;      /* "begin offset" */
1133 			*tl++ = x;      /* "current offset" */
1134 			x = txdr_unsigned(len);
1135 			*tl++ = x;      /* total to this offset */
1136 			*tl = x;        /* size of this write */
1137 
1138 		}
1139 		nfsm_uiotom(uiop, len);
1140 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp, cred);
1141 		if (v3) {
1142 			wccflag = NFSV3_WCCCHK;
1143 			nfsm_wcc_data(vp, wccflag);
1144 			if (!error) {
1145 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1146 					+ NFSX_V3WRITEVERF);
1147 				rlen = fxdr_unsigned(int, *tl++);
1148 				if (rlen == 0) {
1149 					error = NFSERR_IO;
1150 					m_freem(mrep);
1151 					break;
1152 				} else if (rlen < len) {
1153 					backup = len - rlen;
1154 					uiop->uio_iov->iov_base -= backup;
1155 					uiop->uio_iov->iov_len += backup;
1156 					uiop->uio_offset -= backup;
1157 					uiop->uio_resid += backup;
1158 					len = rlen;
1159 				}
1160 				commit = fxdr_unsigned(int, *tl++);
1161 
1162 				/*
1163 				 * Return the lowest committment level
1164 				 * obtained by any of the RPCs.
1165 				 */
1166 				if (committed == NFSV3WRITE_FILESYNC)
1167 					committed = commit;
1168 				else if (committed == NFSV3WRITE_DATASYNC &&
1169 					commit == NFSV3WRITE_UNSTABLE)
1170 					committed = commit;
1171 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1172 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1173 					NFSX_V3WRITEVERF);
1174 				    nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1175 				} else if (bcmp((caddr_t)tl,
1176 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1177 				    *must_commit = 1;
1178 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1179 					NFSX_V3WRITEVERF);
1180 				}
1181 			}
1182 		} else
1183 		    nfsm_loadattr(vp, (struct vattr *)0);
1184 		if (wccflag)
1185 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1186 		m_freem(mrep);
1187 		if (error)
1188 			break;
1189 		tsiz -= len;
1190 	}
1191 nfsmout:
1192 	*iomode = committed;
1193 	if (error)
1194 		uiop->uio_resid = tsiz;
1195 	return (error);
1196 }
1197 
1198 /*
1199  * nfs mknod rpc
1200  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1201  * mode set to specify the file type and the size field for rdev.
1202  */
1203 int
1204 nfs_mknodrpc(dvp, vpp, cnp, vap)
1205 	register struct vnode *dvp;
1206 	register struct vnode **vpp;
1207 	register struct componentname *cnp;
1208 	register struct vattr *vap;
1209 {
1210 	register struct nfsv2_sattr *sp;
1211 	register struct nfsv3_sattr *sp3;
1212 	register u_int32_t *tl;
1213 	register caddr_t cp;
1214 	register int32_t t1, t2;
1215 	struct vnode *newvp = (struct vnode *)0;
1216 	struct nfsnode *np;
1217 	char *cp2;
1218 	caddr_t bpos, dpos;
1219 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1220 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1221 	u_int32_t rdev;
1222 	int v3 = NFS_ISV3(dvp);
1223 
1224 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1225 		rdev = txdr_unsigned(vap->va_rdev);
1226 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1227 		rdev = nfs_xdrneg1;
1228 	else {
1229 		VOP_ABORTOP(dvp, cnp);
1230 		vput(dvp);
1231 		return (EOPNOTSUPP);
1232 	}
1233 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1234 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1235 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1236 	nfsm_fhtom(dvp, v3);
1237 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1238 	if (v3) {
1239 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3SRVSATTR);
1240 		*tl++ = vtonfsv3_type(vap->va_type);
1241 		sp3 = (struct nfsv3_sattr *)tl;
1242 		nfsm_v3sattr(sp3, vap);
1243 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1244 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1245 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1246 			*tl = txdr_unsigned(minor(vap->va_rdev));
1247 		}
1248 	} else {
1249 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1250 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1251 		sp->sa_uid = nfs_xdrneg1;
1252 		sp->sa_gid = nfs_xdrneg1;
1253 		sp->sa_size = rdev;
1254 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1255 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1256 	}
1257 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1258 	if (!error) {
1259 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1260 		if (!gotvp) {
1261 			if (newvp) {
1262 				vrele(newvp);
1263 				newvp = (struct vnode *)0;
1264 			}
1265 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1266 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1267 			if (!error)
1268 				newvp = NFSTOV(np);
1269 		}
1270 	}
1271 	if (v3)
1272 		nfsm_wcc_data(dvp, wccflag);
1273 	nfsm_reqdone;
1274 	if (error) {
1275 		if (newvp)
1276 			vrele(newvp);
1277 	} else {
1278 		if (cnp->cn_flags & MAKEENTRY)
1279 			cache_enter(dvp, newvp, cnp);
1280 		*vpp = newvp;
1281 	}
1282 	FREE(cnp->cn_pnbuf, M_NAMEI);
1283 	VTONFS(dvp)->n_flag |= NMODIFIED;
1284 	if (!wccflag)
1285 		VTONFS(dvp)->n_attrstamp = 0;
1286 	vrele(dvp);
1287 	return (error);
1288 }
1289 
1290 /*
1291  * nfs mknod vop
1292  * just call nfs_mknodrpc() to do the work.
1293  */
1294 /* ARGSUSED */
1295 int
1296 nfs_mknod(v)
1297 	void *v;
1298 {
1299 	struct vop_mknod_args /* {
1300 		struct vnode *a_dvp;
1301 		struct vnode **a_vpp;
1302 		struct componentname *a_cnp;
1303 		struct vattr *a_vap;
1304 	} */ *ap = v;
1305 	struct vnode *newvp;
1306 	int error;
1307 
1308 	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1309 	if (!error)
1310 		vrele(newvp);
1311 	return (error);
1312 }
1313 
1314 static u_long create_verf;
1315 /*
1316  * nfs file create call
1317  */
1318 int
1319 nfs_create(v)
1320 	void *v;
1321 {
1322 	struct vop_create_args /* {
1323 		struct vnode *a_dvp;
1324 		struct vnode **a_vpp;
1325 		struct componentname *a_cnp;
1326 		struct vattr *a_vap;
1327 	} */ *ap = v;
1328 	register struct vnode *dvp = ap->a_dvp;
1329 	register struct vattr *vap = ap->a_vap;
1330 	register struct componentname *cnp = ap->a_cnp;
1331 	register struct nfsv2_sattr *sp;
1332 	register struct nfsv3_sattr *sp3;
1333 	register u_int32_t *tl;
1334 	register caddr_t cp;
1335 	register int32_t t1, t2;
1336 	struct nfsnode *np = (struct nfsnode *)0;
1337 	struct vnode *newvp = (struct vnode *)0;
1338 	caddr_t bpos, dpos, cp2;
1339 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1340 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1341 	int v3 = NFS_ISV3(dvp);
1342 
1343 	/*
1344 	 * Oops, not for me..
1345 	 */
1346 	if (vap->va_type == VSOCK)
1347 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1348 
1349 #ifdef VA_EXCLUSIVE
1350 	if (vap->va_vaflags & VA_EXCLUSIVE)
1351 		fmode |= O_EXCL;
1352 #endif
1353 again:
1354 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1355 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1356 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1357 	nfsm_fhtom(dvp, v3);
1358 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1359 	if (v3) {
1360 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1361 		if (fmode & O_EXCL) {
1362 		    *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1363 		    nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1364 		    if (in_ifaddr.tqh_first)
1365 			*tl++ = in_ifaddr.tqh_first->ia_addr.sin_addr.s_addr;
1366 		    else
1367 			*tl++ = create_verf;
1368 		    *tl = ++create_verf;
1369 		} else {
1370 		    *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1371 		    nfsm_build(tl, u_int32_t *, NFSX_V3SRVSATTR);
1372 		    sp3 = (struct nfsv3_sattr *)tl;
1373 		    nfsm_v3sattr(sp3, vap);
1374 		}
1375 	} else {
1376 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1377 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1378 		sp->sa_uid = nfs_xdrneg1;
1379 		sp->sa_gid = nfs_xdrneg1;
1380 		sp->sa_size = 0;
1381 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1382 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1383 	}
1384 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1385 	if (!error) {
1386 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1387 		if (!gotvp) {
1388 			if (newvp) {
1389 				vrele(newvp);
1390 				newvp = (struct vnode *)0;
1391 			}
1392 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1393 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1394 			if (!error)
1395 				newvp = NFSTOV(np);
1396 		}
1397 	}
1398 	if (v3)
1399 		nfsm_wcc_data(dvp, wccflag);
1400 	nfsm_reqdone;
1401 	if (error) {
1402 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1403 			fmode &= ~O_EXCL;
1404 			goto again;
1405 		}
1406 		if (newvp)
1407 			vrele(newvp);
1408 	} else if (v3 && (fmode & O_EXCL))
1409 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1410 	if (!error) {
1411 		if (cnp->cn_flags & MAKEENTRY)
1412 			cache_enter(dvp, newvp, cnp);
1413 		*ap->a_vpp = newvp;
1414 	}
1415 	FREE(cnp->cn_pnbuf, M_NAMEI);
1416 	VTONFS(dvp)->n_flag |= NMODIFIED;
1417 	if (!wccflag)
1418 		VTONFS(dvp)->n_attrstamp = 0;
1419 	vrele(dvp);
1420 	return (error);
1421 }
1422 
1423 /*
1424  * nfs file remove call
1425  * To try and make nfs semantics closer to ufs semantics, a file that has
1426  * other processes using the vnode is renamed instead of removed and then
1427  * removed later on the last close.
1428  * - If v_usecount > 1
1429  *	  If a rename is not already in the works
1430  *	     call nfs_sillyrename() to set it up
1431  *     else
1432  *	  do the remove rpc
1433  */
1434 int
1435 nfs_remove(v)
1436 	void *v;
1437 {
1438 	struct vop_remove_args /* {
1439 		struct vnodeop_desc *a_desc;
1440 		struct vnode * a_dvp;
1441 		struct vnode * a_vp;
1442 		struct componentname * a_cnp;
1443 	} */ *ap = v;
1444 	register struct vnode *vp = ap->a_vp;
1445 	register struct vnode *dvp = ap->a_dvp;
1446 	register struct componentname *cnp = ap->a_cnp;
1447 	register struct nfsnode *np = VTONFS(vp);
1448 	int error = 0;
1449 	struct vattr vattr;
1450 
1451 #ifndef DIAGNOSTIC
1452 	if ((cnp->cn_flags & HASBUF) == 0)
1453 		panic("nfs_remove: no name");
1454 	if (vp->v_usecount < 1)
1455 		panic("nfs_remove: bad v_usecount");
1456 #endif
1457 	if (vp->v_type == VDIR)
1458 		error = EPERM;
1459 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1460 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1461 	    vattr.va_nlink > 1)) {
1462 		/*
1463 		 * Purge the name cache so that the chance of a lookup for
1464 		 * the name succeeding while the remove is in progress is
1465 		 * minimized. Without node locking it can still happen, such
1466 		 * that an I/O op returns ESTALE, but since you get this if
1467 		 * another host removes the file..
1468 		 */
1469 		cache_purge(vp);
1470 		/*
1471 		 * throw away biocache buffers, mainly to avoid
1472 		 * unnecessary delayed writes later.
1473 		 */
1474 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1475 		/* Do the rpc */
1476 		if (error != EINTR)
1477 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1478 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1479 		/*
1480 		 * Kludge City: If the first reply to the remove rpc is lost..
1481 		 *   the reply to the retransmitted request will be ENOENT
1482 		 *   since the file was in fact removed
1483 		 *   Therefore, we cheat and return success.
1484 		 */
1485 		if (error == ENOENT)
1486 			error = 0;
1487 	} else if (!np->n_sillyrename)
1488 		error = nfs_sillyrename(dvp, vp, cnp);
1489 	FREE(cnp->cn_pnbuf, M_NAMEI);
1490 	np->n_attrstamp = 0;
1491 	vrele(dvp);
1492 	vrele(vp);
1493 	return (error);
1494 }
1495 
1496 /*
1497  * nfs file remove rpc called from nfs_inactive
1498  */
1499 int
1500 nfs_removeit(sp)
1501 	register struct sillyrename *sp;
1502 {
1503 
1504 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1505 		(struct proc *)0));
1506 }
1507 
1508 /*
1509  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1510  */
1511 int
1512 nfs_removerpc(dvp, name, namelen, cred, proc)
1513 	register struct vnode *dvp;
1514 	const char *name;
1515 	int namelen;
1516 	struct ucred *cred;
1517 	struct proc *proc;
1518 {
1519 	register u_int32_t *tl;
1520 	register caddr_t cp;
1521 	register int32_t t1, t2;
1522 	caddr_t bpos, dpos, cp2;
1523 	int error = 0, wccflag = NFSV3_WCCRATTR;
1524 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1525 	int v3 = NFS_ISV3(dvp);
1526 
1527 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1528 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1529 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1530 	nfsm_fhtom(dvp, v3);
1531 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1532 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1533 	if (v3)
1534 		nfsm_wcc_data(dvp, wccflag);
1535 	nfsm_reqdone;
1536 	VTONFS(dvp)->n_flag |= NMODIFIED;
1537 	if (!wccflag)
1538 		VTONFS(dvp)->n_attrstamp = 0;
1539 	return (error);
1540 }
1541 
1542 /*
1543  * nfs file rename call
1544  */
1545 int
1546 nfs_rename(v)
1547 	void *v;
1548 {
1549 	struct vop_rename_args  /* {
1550 		struct vnode *a_fdvp;
1551 		struct vnode *a_fvp;
1552 		struct componentname *a_fcnp;
1553 		struct vnode *a_tdvp;
1554 		struct vnode *a_tvp;
1555 		struct componentname *a_tcnp;
1556 	} */ *ap = v;
1557 	register struct vnode *fvp = ap->a_fvp;
1558 	register struct vnode *tvp = ap->a_tvp;
1559 	register struct vnode *fdvp = ap->a_fdvp;
1560 	register struct vnode *tdvp = ap->a_tdvp;
1561 	register struct componentname *tcnp = ap->a_tcnp;
1562 	register struct componentname *fcnp = ap->a_fcnp;
1563 	int error;
1564 
1565 #ifndef DIAGNOSTIC
1566 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1567 	    (fcnp->cn_flags & HASBUF) == 0)
1568 		panic("nfs_rename: no name");
1569 #endif
1570 	/* Check for cross-device rename */
1571 	if ((fvp->v_mount != tdvp->v_mount) ||
1572 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1573 		error = EXDEV;
1574 		goto out;
1575 	}
1576 
1577 	/*
1578 	 * If the tvp exists and is in use, sillyrename it before doing the
1579 	 * rename of the new file over it.
1580 	 */
1581 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1582 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1583 		vrele(tvp);
1584 		tvp = NULL;
1585 	}
1586 
1587 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1588 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1589 		tcnp->cn_proc);
1590 
1591 	if (fvp->v_type == VDIR) {
1592 		if (tvp != NULL && tvp->v_type == VDIR)
1593 			cache_purge(tdvp);
1594 		cache_purge(fdvp);
1595 	}
1596 out:
1597 	if (tdvp == tvp)
1598 		vrele(tdvp);
1599 	else
1600 		vput(tdvp);
1601 	if (tvp)
1602 		vput(tvp);
1603 	vrele(fdvp);
1604 	vrele(fvp);
1605 	/*
1606 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1607 	 */
1608 	if (error == ENOENT)
1609 		error = 0;
1610 	return (error);
1611 }
1612 
1613 /*
1614  * nfs file rename rpc called from nfs_remove() above
1615  */
1616 int
1617 nfs_renameit(sdvp, scnp, sp)
1618 	struct vnode *sdvp;
1619 	struct componentname *scnp;
1620 	register struct sillyrename *sp;
1621 {
1622 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1623 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1624 }
1625 
1626 /*
1627  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1628  */
1629 int
1630 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1631 	register struct vnode *fdvp;
1632 	const char *fnameptr;
1633 	int fnamelen;
1634 	register struct vnode *tdvp;
1635 	const char *tnameptr;
1636 	int tnamelen;
1637 	struct ucred *cred;
1638 	struct proc *proc;
1639 {
1640 	register u_int32_t *tl;
1641 	register caddr_t cp;
1642 	register int32_t t1, t2;
1643 	caddr_t bpos, dpos, cp2;
1644 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1645 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1646 	int v3 = NFS_ISV3(fdvp);
1647 
1648 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1649 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1650 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1651 		nfsm_rndup(tnamelen));
1652 	nfsm_fhtom(fdvp, v3);
1653 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1654 	nfsm_fhtom(tdvp, v3);
1655 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1656 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1657 	if (v3) {
1658 		nfsm_wcc_data(fdvp, fwccflag);
1659 		nfsm_wcc_data(tdvp, twccflag);
1660 	}
1661 	nfsm_reqdone;
1662 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1663 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1664 	if (!fwccflag)
1665 		VTONFS(fdvp)->n_attrstamp = 0;
1666 	if (!twccflag)
1667 		VTONFS(tdvp)->n_attrstamp = 0;
1668 	return (error);
1669 }
1670 
1671 /*
1672  * nfs hard link create call
1673  */
1674 int
1675 nfs_link(v)
1676 	void *v;
1677 {
1678 	struct vop_link_args /* {
1679 		struct vnode *a_dvp;
1680 		struct vnode *a_vp;
1681 		struct componentname *a_cnp;
1682 	} */ *ap = v;
1683 	register struct vnode *vp = ap->a_vp;
1684 	register struct vnode *dvp = ap->a_dvp;
1685 	register struct componentname *cnp = ap->a_cnp;
1686 	register u_int32_t *tl;
1687 	register caddr_t cp;
1688 	register int32_t t1, t2;
1689 	caddr_t bpos, dpos, cp2;
1690 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1691 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1692 	int v3;
1693 
1694 	if (dvp->v_mount != vp->v_mount) {
1695 		VOP_ABORTOP(vp, cnp);
1696 		vput(dvp);
1697 		return (EXDEV);
1698 	}
1699 
1700 	/*
1701 	 * Push all writes to the server, so that the attribute cache
1702 	 * doesn't get "out of sync" with the server.
1703 	 * XXX There should be a better way!
1704 	 */
1705 	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1706 
1707 	v3 = NFS_ISV3(vp);
1708 	nfsstats.rpccnt[NFSPROC_LINK]++;
1709 	nfsm_reqhead(vp, NFSPROC_LINK,
1710 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1711 	nfsm_fhtom(vp, v3);
1712 	nfsm_fhtom(dvp, v3);
1713 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1714 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1715 	if (v3) {
1716 		nfsm_postop_attr(vp, attrflag);
1717 		nfsm_wcc_data(dvp, wccflag);
1718 	}
1719 	nfsm_reqdone;
1720 	FREE(cnp->cn_pnbuf, M_NAMEI);
1721 	VTONFS(dvp)->n_flag |= NMODIFIED;
1722 	if (!attrflag)
1723 		VTONFS(vp)->n_attrstamp = 0;
1724 	if (!wccflag)
1725 		VTONFS(dvp)->n_attrstamp = 0;
1726 	vput(dvp);
1727 	/*
1728 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1729 	 */
1730 	if (error == EEXIST)
1731 		error = 0;
1732 	return (error);
1733 }
1734 
1735 /*
1736  * nfs symbolic link create call
1737  */
1738 int
1739 nfs_symlink(v)
1740 	void *v;
1741 {
1742 	struct vop_symlink_args /* {
1743 		struct vnode *a_dvp;
1744 		struct vnode **a_vpp;
1745 		struct componentname *a_cnp;
1746 		struct vattr *a_vap;
1747 		char *a_target;
1748 	} */ *ap = v;
1749 	register struct vnode *dvp = ap->a_dvp;
1750 	register struct vattr *vap = ap->a_vap;
1751 	register struct componentname *cnp = ap->a_cnp;
1752 	register struct nfsv2_sattr *sp;
1753 	register struct nfsv3_sattr *sp3;
1754 	register u_int32_t *tl;
1755 	register caddr_t cp;
1756 	register int32_t t1, t2;
1757 	caddr_t bpos, dpos, cp2;
1758 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1759 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1760 	struct vnode *newvp = (struct vnode *)0;
1761 	int v3 = NFS_ISV3(dvp);
1762 
1763 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1764 	slen = strlen(ap->a_target);
1765 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1766 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1767 	nfsm_fhtom(dvp, v3);
1768 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1769 	if (v3) {
1770 		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1771 		nfsm_v3sattr(sp3, vap);
1772 	}
1773 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1774 	if (!v3) {
1775 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1776 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1777 		sp->sa_uid = nfs_xdrneg1;
1778 		sp->sa_gid = nfs_xdrneg1;
1779 		sp->sa_size = nfs_xdrneg1;
1780 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1781 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1782 	}
1783 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1784 	if (v3) {
1785 		if (!error)
1786 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1787 		nfsm_wcc_data(dvp, wccflag);
1788 	}
1789 	nfsm_reqdone;
1790 	if (newvp)
1791 		vrele(newvp);
1792 	FREE(cnp->cn_pnbuf, M_NAMEI);
1793 	VTONFS(dvp)->n_flag |= NMODIFIED;
1794 	if (!wccflag)
1795 		VTONFS(dvp)->n_attrstamp = 0;
1796 	vrele(dvp);
1797 	/*
1798 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1799 	 */
1800 	if (error == EEXIST)
1801 		error = 0;
1802 	return (error);
1803 }
1804 
1805 /*
1806  * nfs make dir call
1807  */
1808 int
1809 nfs_mkdir(v)
1810 	void *v;
1811 {
1812 	struct vop_mkdir_args /* {
1813 		struct vnode *a_dvp;
1814 		struct vnode **a_vpp;
1815 		struct componentname *a_cnp;
1816 		struct vattr *a_vap;
1817 	} */ *ap = v;
1818 	register struct vnode *dvp = ap->a_dvp;
1819 	register struct vattr *vap = ap->a_vap;
1820 	register struct componentname *cnp = ap->a_cnp;
1821 	register struct nfsv2_sattr *sp;
1822 	register struct nfsv3_sattr *sp3;
1823 	register u_int32_t *tl;
1824 	register caddr_t cp;
1825 	register int32_t t1, t2;
1826 	register int len;
1827 	struct nfsnode *np = (struct nfsnode *)0;
1828 	struct vnode *newvp = (struct vnode *)0;
1829 	caddr_t bpos, dpos, cp2;
1830 	int error = 0, wccflag = NFSV3_WCCRATTR;
1831 	int gotvp = 0;
1832 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1833 	int v3 = NFS_ISV3(dvp);
1834 
1835 	len = cnp->cn_namelen;
1836 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1837 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1838 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1839 	nfsm_fhtom(dvp, v3);
1840 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1841 	if (v3) {
1842 		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1843 		nfsm_v3sattr(sp3, vap);
1844 	} else {
1845 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1846 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1847 		sp->sa_uid = nfs_xdrneg1;
1848 		sp->sa_gid = nfs_xdrneg1;
1849 		sp->sa_size = nfs_xdrneg1;
1850 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1851 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1852 	}
1853 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1854 	if (!error)
1855 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1856 	if (v3)
1857 		nfsm_wcc_data(dvp, wccflag);
1858 	nfsm_reqdone;
1859 	VTONFS(dvp)->n_flag |= NMODIFIED;
1860 	if (!wccflag)
1861 		VTONFS(dvp)->n_attrstamp = 0;
1862 	/*
1863 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1864 	 * if we can succeed in looking up the directory.
1865 	 */
1866 	if (error == EEXIST || (!error && !gotvp)) {
1867 		if (newvp) {
1868 			vrele(newvp);
1869 			newvp = (struct vnode *)0;
1870 		}
1871 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1872 			cnp->cn_proc, &np);
1873 		if (!error) {
1874 			newvp = NFSTOV(np);
1875 			if (newvp->v_type != VDIR)
1876 				error = EEXIST;
1877 		}
1878 	}
1879 	if (error) {
1880 		if (newvp)
1881 			vrele(newvp);
1882 	} else {
1883 		if (cnp->cn_flags & MAKEENTRY)
1884 			cache_enter(dvp, newvp, cnp);
1885 		*ap->a_vpp = newvp;
1886 	}
1887 	FREE(cnp->cn_pnbuf, M_NAMEI);
1888 	vrele(dvp);
1889 	return (error);
1890 }
1891 
1892 /*
1893  * nfs remove directory call
1894  */
1895 int
1896 nfs_rmdir(v)
1897 	void *v;
1898 {
1899 	struct vop_rmdir_args /* {
1900 		struct vnode *a_dvp;
1901 		struct vnode *a_vp;
1902 		struct componentname *a_cnp;
1903 	} */ *ap = v;
1904 	register struct vnode *vp = ap->a_vp;
1905 	register struct vnode *dvp = ap->a_dvp;
1906 	register struct componentname *cnp = ap->a_cnp;
1907 	register u_int32_t *tl;
1908 	register caddr_t cp;
1909 	register int32_t t1, t2;
1910 	caddr_t bpos, dpos, cp2;
1911 	int error = 0, wccflag = NFSV3_WCCRATTR;
1912 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1913 	int v3 = NFS_ISV3(dvp);
1914 
1915 	if (dvp == vp) {
1916 		vrele(dvp);
1917 		vrele(dvp);
1918 		FREE(cnp->cn_pnbuf, M_NAMEI);
1919 		return (EINVAL);
1920 	}
1921 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1922 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
1923 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1924 	nfsm_fhtom(dvp, v3);
1925 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1926 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
1927 	if (v3)
1928 		nfsm_wcc_data(dvp, wccflag);
1929 	nfsm_reqdone;
1930 	FREE(cnp->cn_pnbuf, M_NAMEI);
1931 	VTONFS(dvp)->n_flag |= NMODIFIED;
1932 	if (!wccflag)
1933 		VTONFS(dvp)->n_attrstamp = 0;
1934 	cache_purge(dvp);
1935 	cache_purge(vp);
1936 	vrele(vp);
1937 	vrele(dvp);
1938 	/*
1939 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1940 	 */
1941 	if (error == ENOENT)
1942 		error = 0;
1943 	return (error);
1944 }
1945 
1946 /*
1947  * nfs readdir call
1948  */
1949 int
1950 nfs_readdir(v)
1951 	void *v;
1952 {
1953 	struct vop_readdir_args /* {
1954 		struct vnode *a_vp;
1955 		struct uio *a_uio;
1956 		struct ucred *a_cred;
1957 		int *a_eofflag;
1958 		off_t *a_cookies;
1959 		int a_ncookies;
1960 	} */ *ap = v;
1961 	register struct vnode *vp = ap->a_vp;
1962 	register struct uio *uio = ap->a_uio;
1963 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1964 	char *base = uio->uio_iov->iov_base;
1965 	int tresid, error;
1966 	size_t count, lost;
1967 
1968 	if (vp->v_type != VDIR)
1969 		return (EPERM);
1970 
1971 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
1972 	count = uio->uio_resid - lost;
1973 	if (count <= 0)
1974 		return (EINVAL);
1975 
1976 	/*
1977 	 * Call nfs_bioread() to do the real work.
1978 	 */
1979 	tresid = uio->uio_resid = count;
1980 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
1981 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
1982 
1983 	if (!error && uio->uio_resid == tresid) {
1984 		uio->uio_resid += lost;
1985 		nfsstats.direofcache_misses++;
1986 		*ap->a_eofflag = 1;
1987 		return (0);
1988 	}
1989 
1990 	if (!error && ap->a_cookies) {
1991 		struct dirent *dp;
1992 		off_t *cookies = ap->a_cookies;
1993 		int ncookies = ap->a_ncookies;
1994 
1995 		/*
1996 		 * Only the NFS server and emulations use cookies, and they
1997 		 * load the directory block into system space, so we can
1998 		 * just look at it directly.
1999 		 */
2000 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2001 			panic("nfs_readdir: lost in space");
2002 		while (ncookies-- && base < uio->uio_iov->iov_base) {
2003 			dp = (struct dirent *) base;
2004 			if (dp->d_reclen == 0)
2005 				break;
2006 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2007 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2008 			else
2009 				*(cookies++) = NFS_GETCOOKIE(dp);
2010 			base += dp->d_reclen;
2011 		}
2012 		uio->uio_resid += (uio->uio_iov->iov_base - base);
2013 		uio->uio_iov->iov_len += (uio->uio_iov->iov_base - base);
2014 		uio->uio_iov->iov_base = base;
2015 	}
2016 
2017 	uio->uio_resid += lost;
2018 	*ap->a_eofflag = 0;
2019 	return (error);
2020 }
2021 
2022 /*
2023  * Readdir rpc call.
2024  * Called from below the buffer cache by nfs_doio().
2025  */
2026 int
2027 nfs_readdirrpc(vp, uiop, cred)
2028 	struct vnode *vp;
2029 	register struct uio *uiop;
2030 	struct ucred *cred;
2031 {
2032 	register int len, left;
2033 	register struct dirent *dp = NULL;
2034 	register u_int32_t *tl;
2035 	register caddr_t cp;
2036 	register int32_t t1, t2;
2037 	caddr_t bpos, dpos, cp2;
2038 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2039 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2040 	struct nfsnode *dnp = VTONFS(vp);
2041 	u_quad_t fileno;
2042 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2043 	int attrflag, nrpcs = 0, reclen;
2044 	int v3 = NFS_ISV3(vp);
2045 	nfsquad_t cookie;
2046 
2047 #ifdef DIAGNOSTIC
2048 	/*
2049 	 * Should be called from buffer cache, so only amount of
2050 	 * NFS_DIRBLKSIZ will be requested.
2051 	 */
2052 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2053 		panic("nfs readdirrpc bad uio");
2054 #endif
2055 
2056 	/*
2057 	 * Loop around doing readdir rpc's of size nm_readdirsize
2058 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2059 	 * The stopping criteria is EOF or buffer full.
2060 	 */
2061 	while (more_dirs && bigenough) {
2062 		/*
2063 		 * Heuristic: don't bother to do another RPC to further
2064 		 * fill up this block if there is not much room left. (< 50%
2065 		 * of the readdir RPC size). This wastes some buffer space
2066 		 * but can save up to 50% in RPC calls.
2067 		 */
2068 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2069 			bigenough = 0;
2070 			break;
2071 		}
2072 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2073 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2074 			NFSX_READDIR(v3));
2075 		nfsm_fhtom(vp, v3);
2076 		if (v3) {
2077 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2078 			cookie.qval = uiop->uio_offset;
2079 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2080 				txdr_swapcookie3(uiop->uio_offset, tl);
2081 			} else {
2082 				txdr_cookie3(uiop->uio_offset, tl);
2083 			}
2084 			tl += 2;
2085 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2086 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2087 		} else {
2088 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2089 			*tl++ = txdr_unsigned(uiop->uio_offset);
2090 		}
2091 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2092 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2093 		nrpcs++;
2094 		if (v3) {
2095 			nfsm_postop_attr(vp, attrflag);
2096 			if (!error) {
2097 				nfsm_dissect(tl, u_int32_t *,
2098 				    2 * NFSX_UNSIGNED);
2099 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2100 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2101 			} else {
2102 				m_freem(mrep);
2103 				goto nfsmout;
2104 			}
2105 		}
2106 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2107 		more_dirs = fxdr_unsigned(int, *tl);
2108 
2109 		/* loop thru the dir entries, doctoring them to 4bsd form */
2110 		while (more_dirs && bigenough) {
2111 			if (v3) {
2112 				nfsm_dissect(tl, u_int32_t *,
2113 				    3 * NFSX_UNSIGNED);
2114 				fxdr_hyper(tl, &fileno);
2115 				len = fxdr_unsigned(int, *(tl + 2));
2116 			} else {
2117 				nfsm_dissect(tl, u_int32_t *,
2118 				    2 * NFSX_UNSIGNED);
2119 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2120 				len = fxdr_unsigned(int, *tl);
2121 			}
2122 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2123 				error = EBADRPC;
2124 				m_freem(mrep);
2125 				goto nfsmout;
2126 			}
2127 			tlen = nfsm_rndup(len);
2128 			if (tlen == len)
2129 				tlen += 4;	/* To ensure null termination */
2130 			tlen += sizeof (off_t) + sizeof (int);
2131 			reclen = ALIGN(tlen + DIRHDSIZ);
2132 			tlen = reclen - DIRHDSIZ;
2133 			left = NFS_DIRFRAGSIZ - blksiz;
2134 			if (reclen > left) {
2135 				dp->d_reclen += left;
2136 				uiop->uio_iov->iov_base += left;
2137 				uiop->uio_iov->iov_len -= left;
2138 				uiop->uio_resid -= left;
2139 				blksiz = 0;
2140 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2141 			}
2142 			if (reclen > uiop->uio_resid)
2143 				bigenough = 0;
2144 			if (bigenough) {
2145 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2146 				dp->d_fileno = (int)fileno;
2147 				dp->d_namlen = len;
2148 				dp->d_reclen = reclen;
2149 				dp->d_type = DT_UNKNOWN;
2150 				blksiz += dp->d_reclen;
2151 				if (blksiz == NFS_DIRFRAGSIZ)
2152 					blksiz = 0;
2153 				uiop->uio_resid -= DIRHDSIZ;
2154 				uiop->uio_iov->iov_base += DIRHDSIZ;
2155 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2156 				nfsm_mtouio(uiop, len);
2157 				cp = uiop->uio_iov->iov_base;
2158 				tlen -= len;
2159 				*cp = '\0';	/* null terminate */
2160 				uiop->uio_iov->iov_base += tlen;
2161 				uiop->uio_iov->iov_len -= tlen;
2162 				uiop->uio_resid -= tlen;
2163 			} else
2164 				nfsm_adv(nfsm_rndup(len));
2165 			if (v3) {
2166 				nfsm_dissect(tl, u_int32_t *,
2167 				    3 * NFSX_UNSIGNED);
2168 			} else {
2169 				nfsm_dissect(tl, u_int32_t *,
2170 				    2 * NFSX_UNSIGNED);
2171 			}
2172 			if (bigenough) {
2173 				if (v3) {
2174 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2175 						uiop->uio_offset =
2176 						    fxdr_swapcookie3(tl);
2177 					else
2178 						uiop->uio_offset =
2179 						    fxdr_cookie3(tl);
2180 				}
2181 				else {
2182 					uiop->uio_offset =
2183 					    fxdr_unsigned(off_t, *tl);
2184 				}
2185 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2186 			}
2187 			if (v3)
2188 				tl += 2;
2189 			else
2190 				tl++;
2191 			more_dirs = fxdr_unsigned(int, *tl);
2192 		}
2193 		/*
2194 		 * If at end of rpc data, get the eof boolean
2195 		 */
2196 		if (!more_dirs) {
2197 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2198 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2199 		}
2200 		m_freem(mrep);
2201 	}
2202 	/*
2203 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2204 	 * by increasing d_reclen for the last record.
2205 	 */
2206 	if (blksiz > 0) {
2207 		left = NFS_DIRFRAGSIZ - blksiz;
2208 		dp->d_reclen += left;
2209 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2210 		uiop->uio_iov->iov_base += left;
2211 		uiop->uio_iov->iov_len -= left;
2212 		uiop->uio_resid -= left;
2213 	}
2214 
2215 	/*
2216 	 * We are now either at the end of the directory or have filled the
2217 	 * block.
2218 	 */
2219 	if (bigenough)
2220 		dnp->n_direofoffset = uiop->uio_offset;
2221 nfsmout:
2222 	return (error);
2223 }
2224 
2225 /*
2226  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2227  */
2228 int
2229 nfs_readdirplusrpc(vp, uiop, cred)
2230 	struct vnode *vp;
2231 	register struct uio *uiop;
2232 	struct ucred *cred;
2233 {
2234 	register int len, left;
2235 	register struct dirent *dp = NULL;
2236 	register u_int32_t *tl;
2237 	register caddr_t cp;
2238 	register int32_t t1, t2;
2239 	struct vnode *newvp;
2240 	caddr_t bpos, dpos, cp2;
2241 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2242 	struct nameidata nami, *ndp = &nami;
2243 	struct componentname *cnp = &ndp->ni_cnd;
2244 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2245 	struct nfsnode *dnp = VTONFS(vp), *np;
2246 	const unsigned char *hcp;
2247 	nfsfh_t *fhp;
2248 	u_quad_t fileno;
2249 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2250 	int attrflag, fhsize, nrpcs = 0, reclen;
2251 	struct nfs_fattr fattr, *fp;
2252 
2253 #ifdef DIAGNOSTIC
2254 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2255 		panic("nfs readdirplusrpc bad uio");
2256 #endif
2257 	ndp->ni_dvp = vp;
2258 	newvp = NULLVP;
2259 
2260 	/*
2261 	 * Loop around doing readdir rpc's of size nm_readdirsize
2262 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2263 	 * The stopping criteria is EOF or buffer full.
2264 	 */
2265 	while (more_dirs && bigenough) {
2266 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2267 			bigenough = 0;
2268 			break;
2269 		}
2270 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2271 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2272 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2273 		nfsm_fhtom(vp, 1);
2274  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2275 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2276 			txdr_swapcookie3(uiop->uio_offset, tl);
2277 		} else {
2278 			txdr_cookie3(uiop->uio_offset, tl);
2279 		}
2280 		tl += 2;
2281 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2282 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2283 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2284 		*tl = txdr_unsigned(nmp->nm_rsize);
2285 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2286 		nfsm_postop_attr(vp, attrflag);
2287 		if (error) {
2288 			m_freem(mrep);
2289 			goto nfsmout;
2290 		}
2291 		nrpcs++;
2292 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2293 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2294 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2295 		more_dirs = fxdr_unsigned(int, *tl);
2296 
2297 		/* loop thru the dir entries, doctoring them to 4bsd form */
2298 		while (more_dirs && bigenough) {
2299 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2300 			fxdr_hyper(tl, &fileno);
2301 			len = fxdr_unsigned(int, *(tl + 2));
2302 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2303 				error = EBADRPC;
2304 				m_freem(mrep);
2305 				goto nfsmout;
2306 			}
2307 			tlen = nfsm_rndup(len);
2308 			if (tlen == len)
2309 				tlen += 4;	/* To ensure null termination*/
2310 			tlen += sizeof (off_t) + sizeof (int);
2311 			reclen = ALIGN(tlen + DIRHDSIZ);
2312 			tlen = reclen - DIRHDSIZ;
2313 			left = NFS_DIRFRAGSIZ - blksiz;
2314 			if (reclen > left) {
2315 				/*
2316 				 * DIRFRAGSIZ is aligned, no need to align
2317 				 * again here.
2318 				 */
2319 				dp->d_reclen += left;
2320 				uiop->uio_iov->iov_base += left;
2321 				uiop->uio_iov->iov_len -= left;
2322 				uiop->uio_resid -= left;
2323 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2324 				blksiz = 0;
2325 			}
2326 			if (reclen > uiop->uio_resid)
2327 				bigenough = 0;
2328 			if (bigenough) {
2329 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2330 				dp->d_fileno = (int)fileno;
2331 				dp->d_namlen = len;
2332 				dp->d_reclen = reclen;
2333 				dp->d_type = DT_UNKNOWN;
2334 				blksiz += dp->d_reclen;
2335 				if (blksiz == NFS_DIRFRAGSIZ)
2336 					blksiz = 0;
2337 				uiop->uio_resid -= DIRHDSIZ;
2338 				uiop->uio_iov->iov_base += DIRHDSIZ;
2339 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2340 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2341 				cnp->cn_namelen = len;
2342 				nfsm_mtouio(uiop, len);
2343 				cp = uiop->uio_iov->iov_base;
2344 				tlen -= len;
2345 				*cp = '\0';
2346 				uiop->uio_iov->iov_base += tlen;
2347 				uiop->uio_iov->iov_len -= tlen;
2348 				uiop->uio_resid -= tlen;
2349 			} else
2350 				nfsm_adv(nfsm_rndup(len));
2351 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2352 			if (bigenough) {
2353 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2354 					uiop->uio_offset =
2355 						fxdr_swapcookie3(tl);
2356 				else
2357 					uiop->uio_offset =
2358 						fxdr_cookie3(tl);
2359 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2360 			}
2361 			tl += 2;
2362 
2363 			/*
2364 			 * Since the attributes are before the file handle
2365 			 * (sigh), we must skip over the attributes and then
2366 			 * come back and get them.
2367 			 */
2368 			attrflag = fxdr_unsigned(int, *tl);
2369 			if (attrflag) {
2370 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2371 			    bcopy(fp, &fattr, NFSX_V3FATTR);
2372 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2373 			    doit = fxdr_unsigned(int, *tl);
2374 			    if (doit) {
2375 				nfsm_getfh(fhp, fhsize, 1);
2376 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2377 				    VREF(vp);
2378 				    newvp = vp;
2379 				    np = dnp;
2380 				} else {
2381 				    error = nfs_nget(vp->v_mount, fhp,
2382 					fhsize, &np);
2383 				    if (!error)
2384 					newvp = NFSTOV(np);
2385 				}
2386 				if (!error) {
2387 				    nfs_loadattrcache(&newvp, &fattr, 0);
2388 				    dp->d_type =
2389 				        IFTODT(VTTOIF(np->n_vattr->va_type));
2390 				    ndp->ni_vp = newvp;
2391 				    cnp->cn_hash = 0;
2392 				    for (hcp = cnp->cn_nameptr, i = 1; i <= len;
2393 				        i++, hcp++)
2394 				        cnp->cn_hash += *hcp * i;
2395 				    if (cnp->cn_namelen <= NCHNAMLEN)
2396 				        cache_enter(ndp->ni_dvp, ndp->ni_vp,
2397 						    cnp);
2398 				}
2399 			   }
2400 			} else {
2401 			    /* Just skip over the file handle */
2402 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2403 			    i = fxdr_unsigned(int, *tl);
2404 			    nfsm_adv(nfsm_rndup(i));
2405 			}
2406 			if (newvp != NULLVP) {
2407 			    vrele(newvp);
2408 			    newvp = NULLVP;
2409 			}
2410 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2411 			more_dirs = fxdr_unsigned(int, *tl);
2412 		}
2413 		/*
2414 		 * If at end of rpc data, get the eof boolean
2415 		 */
2416 		if (!more_dirs) {
2417 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2418 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2419 		}
2420 		m_freem(mrep);
2421 	}
2422 	/*
2423 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2424 	 * by increasing d_reclen for the last record.
2425 	 */
2426 	if (blksiz > 0) {
2427 		left = NFS_DIRFRAGSIZ - blksiz;
2428 		dp->d_reclen += left;
2429 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2430 		uiop->uio_iov->iov_base += left;
2431 		uiop->uio_iov->iov_len -= left;
2432 		uiop->uio_resid -= left;
2433 	}
2434 
2435 	/*
2436 	 * We are now either at the end of the directory or have filled the
2437 	 * block.
2438 	 */
2439 	if (bigenough)
2440 		dnp->n_direofoffset = uiop->uio_offset;
2441 nfsmout:
2442 	if (newvp != NULLVP)
2443 		vrele(newvp);
2444 	return (error);
2445 }
2446 static char hextoasc[] = "0123456789abcdef";
2447 
2448 /*
2449  * Silly rename. To make the NFS filesystem that is stateless look a little
2450  * more like the "ufs" a remove of an active vnode is translated to a rename
2451  * to a funny looking filename that is removed by nfs_inactive on the
2452  * nfsnode. There is the potential for another process on a different client
2453  * to create the same funny name between the nfs_lookitup() fails and the
2454  * nfs_rename() completes, but...
2455  */
2456 int
2457 nfs_sillyrename(dvp, vp, cnp)
2458 	struct vnode *dvp, *vp;
2459 	struct componentname *cnp;
2460 {
2461 	register struct sillyrename *sp;
2462 	struct nfsnode *np;
2463 	int error;
2464 	short pid;
2465 
2466 	cache_purge(dvp);
2467 	np = VTONFS(vp);
2468 #ifndef DIAGNOSTIC
2469 	if (vp->v_type == VDIR)
2470 		panic("nfs: sillyrename dir");
2471 #endif
2472 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2473 		M_NFSREQ, M_WAITOK);
2474 	sp->s_cred = crdup(cnp->cn_cred);
2475 	sp->s_dvp = dvp;
2476 	VREF(dvp);
2477 
2478 	/* Fudge together a funny name */
2479 	pid = cnp->cn_proc->p_pid;
2480 	bcopy(".nfsAxxxx4.4", sp->s_name, 13);
2481 	sp->s_namlen = 12;
2482 	sp->s_name[8] = hextoasc[pid & 0xf];
2483 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2484 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2485 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2486 
2487 	/* Try lookitups until we get one that isn't there */
2488 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2489 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2490 		sp->s_name[4]++;
2491 		if (sp->s_name[4] > 'z') {
2492 			error = EINVAL;
2493 			goto bad;
2494 		}
2495 	}
2496 	error = nfs_renameit(dvp, cnp, sp);
2497 	if (error)
2498 		goto bad;
2499 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2500 		cnp->cn_proc, &np);
2501 	np->n_sillyrename = sp;
2502 	return (0);
2503 bad:
2504 	vrele(sp->s_dvp);
2505 	crfree(sp->s_cred);
2506 	free((caddr_t)sp, M_NFSREQ);
2507 	return (error);
2508 }
2509 
2510 /*
2511  * Look up a file name and optionally either update the file handle or
2512  * allocate an nfsnode, depending on the value of npp.
2513  * npp == NULL	--> just do the lookup
2514  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2515  *			handled too
2516  * *npp != NULL --> update the file handle in the vnode
2517  */
2518 int
2519 nfs_lookitup(dvp, name, len, cred, procp, npp)
2520 	register struct vnode *dvp;
2521 	const char *name;
2522 	int len;
2523 	struct ucred *cred;
2524 	struct proc *procp;
2525 	struct nfsnode **npp;
2526 {
2527 	register u_int32_t *tl;
2528 	register caddr_t cp;
2529 	register int32_t t1, t2;
2530 	struct vnode *newvp = (struct vnode *)0;
2531 	struct nfsnode *np, *dnp = VTONFS(dvp);
2532 	caddr_t bpos, dpos, cp2;
2533 	int error = 0, fhlen, attrflag;
2534 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2535 	nfsfh_t *nfhp;
2536 	int v3 = NFS_ISV3(dvp);
2537 
2538 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2539 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2540 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2541 	nfsm_fhtom(dvp, v3);
2542 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2543 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2544 	if (npp && !error) {
2545 		nfsm_getfh(nfhp, fhlen, v3);
2546 		if (*npp) {
2547 		    np = *npp;
2548 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2549 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2550 			np->n_fhp = &np->n_fh;
2551 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2552 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2553 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2554 		    np->n_fhsize = fhlen;
2555 		    newvp = NFSTOV(np);
2556 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2557 		    VREF(dvp);
2558 		    newvp = dvp;
2559 		} else {
2560 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2561 		    if (error) {
2562 			m_freem(mrep);
2563 			return (error);
2564 		    }
2565 		    newvp = NFSTOV(np);
2566 		}
2567 		if (v3) {
2568 			nfsm_postop_attr(newvp, attrflag);
2569 			if (!attrflag && *npp == NULL) {
2570 				m_freem(mrep);
2571 				vrele(newvp);
2572 				return (ENOENT);
2573 			}
2574 		} else
2575 			nfsm_loadattr(newvp, (struct vattr *)0);
2576 	}
2577 	nfsm_reqdone;
2578 	if (npp && *npp == NULL) {
2579 		if (error) {
2580 			if (newvp)
2581 				vrele(newvp);
2582 		} else
2583 			*npp = np;
2584 	}
2585 	return (error);
2586 }
2587 
2588 /*
2589  * Nfs Version 3 commit rpc
2590  */
2591 int
2592 nfs_commit(vp, offset, cnt, cred, procp)
2593 	register struct vnode *vp;
2594 	u_quad_t offset;
2595 	int cnt;
2596 	struct ucred *cred;
2597 	struct proc *procp;
2598 {
2599 	register caddr_t cp;
2600 	register u_int32_t *tl;
2601 	register int32_t t1, t2;
2602 	register struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2603 	caddr_t bpos, dpos, cp2;
2604 	int error = 0, wccflag = NFSV3_WCCRATTR;
2605 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2606 
2607 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0)
2608 		return (0);
2609 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2610 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2611 	nfsm_fhtom(vp, 1);
2612 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2613 	txdr_hyper(&offset, tl);
2614 	tl += 2;
2615 	*tl = txdr_unsigned(cnt);
2616 	nfsm_request(vp, NFSPROC_COMMIT, procp, cred);
2617 	nfsm_wcc_data(vp, wccflag);
2618 	if (!error) {
2619 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2620 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2621 			NFSX_V3WRITEVERF)) {
2622 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2623 				NFSX_V3WRITEVERF);
2624 			error = NFSERR_STALEWRITEVERF;
2625 		}
2626 	}
2627 	nfsm_reqdone;
2628 	return (error);
2629 }
2630 
2631 /*
2632  * Kludge City..
2633  * - make nfs_bmap() essentially a no-op that does no translation
2634  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2635  *   (Maybe I could use the process's page mapping, but I was concerned that
2636  *    Kernel Write might not be enabled and also figured copyout() would do
2637  *    a lot more work than bcopy() and also it currently happens in the
2638  *    context of the swapper process (2).
2639  */
2640 int
2641 nfs_bmap(v)
2642 	void *v;
2643 {
2644 	struct vop_bmap_args /* {
2645 		struct vnode *a_vp;
2646 		daddr_t  a_bn;
2647 		struct vnode **a_vpp;
2648 		daddr_t *a_bnp;
2649 		int *a_runp;
2650 	} */ *ap = v;
2651 	register struct vnode *vp = ap->a_vp;
2652 
2653 	if (ap->a_vpp != NULL)
2654 		*ap->a_vpp = vp;
2655 	if (ap->a_bnp != NULL)
2656 		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2657 	return (0);
2658 }
2659 
2660 /*
2661  * Strategy routine.
2662  * For async requests when nfsiod(s) are running, queue the request by
2663  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2664  * request.
2665  */
2666 int
2667 nfs_strategy(v)
2668 	void *v;
2669 {
2670 	struct vop_strategy_args *ap = v;
2671 	register struct buf *bp = ap->a_bp;
2672 	struct ucred *cr;
2673 	struct proc *p;
2674 	int error = 0;
2675 
2676 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2677 		panic("nfs physio/async");
2678 	if (bp->b_flags & B_ASYNC)
2679 		p = (struct proc *)0;
2680 	else
2681 		p = curproc;	/* XXX */
2682 	if (bp->b_flags & B_READ)
2683 		cr = bp->b_rcred;
2684 	else
2685 		cr = bp->b_wcred;
2686 	/*
2687 	 * If the op is asynchronous and an i/o daemon is waiting
2688 	 * queue the request, wake it up and wait for completion
2689 	 * otherwise just do it ourselves.
2690 	 */
2691 	if ((bp->b_flags & B_ASYNC) == 0 ||
2692 		nfs_asyncio(bp, NOCRED))
2693 		error = nfs_doio(bp, cr, p);
2694 	return (error);
2695 }
2696 
2697 /*
2698  * Mmap a file
2699  *
2700  * NB Currently unsupported.
2701  */
2702 /* ARGSUSED */
2703 int
2704 nfs_mmap(v)
2705 	void *v;
2706 {
2707 #if 0
2708 	struct vop_mmap_args /* {
2709 		struct vnode *a_vp;
2710 		int a_fflags;
2711 		struct ucred *a_cred;
2712 		struct proc *a_p;
2713 	} */ *ap = v;
2714 #endif
2715 
2716 	return (EINVAL);
2717 }
2718 
2719 /*
2720  * fsync vnode op. Just call nfs_flush() with commit == 1.
2721  */
2722 /* ARGSUSED */
2723 int
2724 nfs_fsync(v)
2725 	void *v;
2726 {
2727 	struct vop_fsync_args /* {
2728 		struct vnodeop_desc *a_desc;
2729 		struct vnode * a_vp;
2730 		struct ucred * a_cred;
2731 		int  a_waitfor;
2732 		struct proc * a_p;
2733 	} */ *ap = v;
2734 
2735 	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2736 }
2737 
2738 /*
2739  * Flush all the blocks associated with a vnode.
2740  * 	Walk through the buffer pool and push any dirty pages
2741  *	associated with the vnode.
2742  */
2743 int
2744 nfs_flush(vp, cred, waitfor, p, commit)
2745 	register struct vnode *vp;
2746 	struct ucred *cred;
2747 	int waitfor;
2748 	struct proc *p;
2749 	int commit;
2750 {
2751 	register struct nfsnode *np = VTONFS(vp);
2752 	register struct buf *bp;
2753 	register int i;
2754 	struct buf *nbp;
2755 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2756 	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2757 	int passone = 1;
2758 	u_quad_t off = (u_quad_t)-1, endoff = 0, toff;
2759 	struct ucred* wcred = NULL;
2760 #ifndef NFS_COMMITBVECSIZ
2761 #define NFS_COMMITBVECSIZ	20
2762 #endif
2763 	struct buf *bvec[NFS_COMMITBVECSIZ];
2764 
2765 	if (nmp->nm_flag & NFSMNT_INT)
2766 		slpflag = PCATCH;
2767 	if (!commit)
2768 		passone = 0;
2769 	/*
2770 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2771 	 * server, but nas not been committed to stable storage on the server
2772 	 * yet. On the first pass, the byte range is worked out and the commit
2773 	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2774 	 * job.
2775 	 */
2776 again:
2777 	bvecpos = 0;
2778 	if (NFS_ISV3(vp) && commit) {
2779 		s = splbio();
2780 		for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2781 			nbp = bp->b_vnbufs.le_next;
2782 			if (bvecpos >= NFS_COMMITBVECSIZ)
2783 				break;
2784 			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2785 				!= (B_DELWRI | B_NEEDCOMMIT))
2786 				continue;
2787 			bremfree(bp);
2788 			/*
2789 			 * Work out if all buffers are using the same cred
2790 			 * so we can deal with them all with one commit.
2791 			 */
2792 			if (wcred == NULL)
2793 				wcred = bp->b_wcred;
2794 			else if (wcred != bp->b_wcred)
2795 				wcred = NOCRED;
2796 			bp->b_flags |= (B_BUSY | B_WRITEINPROG);
2797 			/*
2798 			 * A list of these buffers is kept so that the
2799 			 * second loop knows which buffers have actually
2800 			 * been committed. This is necessary, since there
2801 			 * may be a race between the commit rpc and new
2802 			 * uncommitted writes on the file.
2803 			 */
2804 			bvec[bvecpos++] = bp;
2805 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2806 				bp->b_dirtyoff;
2807 			if (toff < off)
2808 				off = toff;
2809 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2810 			if (toff > endoff)
2811 				endoff = toff;
2812 		}
2813 		splx(s);
2814 	}
2815 	if (bvecpos > 0) {
2816 		/*
2817 		 * Commit data on the server, as required.
2818 		 * If all bufs are using the same wcred, then use that with
2819 		 * one call for all of them, otherwise commit each one
2820 		 * separately.
2821 		 */
2822 		if (wcred != NOCRED)
2823 			retv = nfs_commit(vp, off, (int)(endoff - off),
2824 					  wcred, p);
2825 		else {
2826 			retv = 0;
2827 			for (i = 0; i < bvecpos; i++) {
2828 				off_t off, size;
2829 				bp = bvec[i];
2830 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2831 					bp->b_dirtyoff;
2832 				size = (u_quad_t)(bp->b_dirtyend
2833 						  - bp->b_dirtyoff);
2834 				retv = nfs_commit(vp, off, (int)size,
2835 						  bp->b_wcred, p);
2836 				if (retv) break;
2837 			}
2838 		}
2839 
2840 		if (retv == NFSERR_STALEWRITEVERF)
2841 			nfs_clearcommit(vp->v_mount);
2842 		/*
2843 		 * Now, either mark the blocks I/O done or mark the
2844 		 * blocks dirty, depending on whether the commit
2845 		 * succeeded.
2846 		 */
2847 		for (i = 0; i < bvecpos; i++) {
2848 			bp = bvec[i];
2849 			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG);
2850 			if (retv)
2851 			    brelse(bp);
2852 			else {
2853 			    vp->v_numoutput++;
2854 			    bp->b_flags |= B_ASYNC;
2855 			    bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
2856 			    bp->b_dirtyoff = bp->b_dirtyend = 0;
2857 			    reassignbuf(bp, vp);
2858 			    biodone(bp);
2859 			}
2860 		}
2861 	}
2862 
2863 	/*
2864 	 * Start/do any write(s) that are required.
2865 	 */
2866 loop:
2867 	s = splbio();
2868 	for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2869 		nbp = bp->b_vnbufs.le_next;
2870 		if (bp->b_flags & B_BUSY) {
2871 			if (waitfor != MNT_WAIT || passone)
2872 				continue;
2873 			bp->b_flags |= B_WANTED;
2874 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
2875 				"nfsfsync", slptimeo);
2876 			splx(s);
2877 			if (error) {
2878 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
2879 				return (EINTR);
2880 			    if (slpflag == PCATCH) {
2881 				slpflag = 0;
2882 				slptimeo = 2 * hz;
2883 			    }
2884 			}
2885 			goto loop;
2886 		}
2887 		if ((bp->b_flags & B_DELWRI) == 0)
2888 			panic("nfs_fsync: not dirty");
2889 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT))
2890 			continue;
2891 		bremfree(bp);
2892 		if (passone || !commit)
2893 		    bp->b_flags |= (B_BUSY|B_ASYNC);
2894 		else
2895 		    bp->b_flags |= (B_BUSY|B_ASYNC|B_WRITEINPROG|B_NEEDCOMMIT);
2896 		splx(s);
2897 		VOP_BWRITE(bp);
2898 		goto loop;
2899 	}
2900 	splx(s);
2901 	if (passone) {
2902 		passone = 0;
2903 		goto again;
2904 	}
2905 	if (waitfor == MNT_WAIT) {
2906 		while (vp->v_numoutput) {
2907 			vp->v_flag |= VBWAIT;
2908 			error = tsleep((caddr_t)&vp->v_numoutput,
2909 				slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
2910 			if (error) {
2911 			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
2912 				return (EINTR);
2913 			    if (slpflag == PCATCH) {
2914 				slpflag = 0;
2915 				slptimeo = 2 * hz;
2916 			    }
2917 			}
2918 		}
2919 		if (vp->v_dirtyblkhd.lh_first && commit) {
2920 #if 0
2921 			vprint("nfs_fsync: dirty", vp);
2922 #endif
2923 			goto loop;
2924 		}
2925 	}
2926 	if (np->n_flag & NWRITEERR) {
2927 		error = np->n_error;
2928 		np->n_flag &= ~NWRITEERR;
2929 	}
2930 	return (error);
2931 }
2932 
2933 /*
2934  * Return POSIX pathconf information applicable to nfs.
2935  *
2936  * The NFS V2 protocol doesn't support this, so just return EINVAL
2937  * for V2.
2938  */
2939 /* ARGSUSED */
2940 int
2941 nfs_pathconf(v)
2942 	void *v;
2943 {
2944 #if 0
2945 	struct vop_pathconf_args /* {
2946 		struct vnode *a_vp;
2947 		int a_name;
2948 		register_t *a_retval;
2949 	} */ *ap = v;
2950 #endif
2951 
2952 	return (EINVAL);
2953 }
2954 
2955 /*
2956  * NFS advisory byte-level locks.
2957  */
2958 int
2959 nfs_advlock(v)
2960 	void *v;
2961 {
2962 	struct vop_advlock_args /* {
2963 		struct vnode *a_vp;
2964 		caddr_t  a_id;
2965 		int  a_op;
2966 		struct flock *a_fl;
2967 		int  a_flags;
2968 	} */ *ap = v;
2969 	register struct nfsnode *np = VTONFS(ap->a_vp);
2970 
2971 	return (lf_advlock(&np->n_lockf, np->n_size, ap->a_id, ap->a_op,
2972 	    ap->a_fl, ap->a_flags));
2973 }
2974 
2975 /*
2976  * Print out the contents of an nfsnode.
2977  */
2978 int
2979 nfs_print(v)
2980 	void *v;
2981 {
2982 	struct vop_print_args /* {
2983 		struct vnode *a_vp;
2984 	} */ *ap = v;
2985 	register struct vnode *vp = ap->a_vp;
2986 	register struct nfsnode *np = VTONFS(vp);
2987 
2988 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
2989 	    np->n_vattr->va_fileid, np->n_vattr->va_fsid);
2990 #ifdef FIFO
2991 	if (vp->v_type == VFIFO)
2992 		fifo_printinfo(vp);
2993 #endif
2994 	printf("\n");
2995 	return (0);
2996 }
2997 
2998 /*
2999  * NFS file truncation.
3000  */
3001 int
3002 nfs_truncate(v)
3003 	void *v;
3004 {
3005 #if 0
3006 	struct vop_truncate_args /* {
3007 		struct vnode *a_vp;
3008 		off_t a_length;
3009 		int a_flags;
3010 		struct ucred *a_cred;
3011 		struct proc *a_p;
3012 	} */ *ap = v;
3013 #endif
3014 
3015 	/* Use nfs_setattr */
3016 	return (EOPNOTSUPP);
3017 }
3018 
3019 /*
3020  * NFS update.
3021  */
3022 int
3023 nfs_update(v)
3024 	void *v;
3025 #if 0
3026 	struct vop_update_args /* {
3027 		struct vnode *a_vp;
3028 		struct timespec *a_ta;
3029 		struct timespec *a_tm;
3030 		int a_waitfor;
3031 	} */ *ap = v;
3032 #endif
3033 {
3034 
3035 	/* Use nfs_setattr */
3036 	return (EOPNOTSUPP);
3037 }
3038 
3039 /*
3040  * Just call nfs_writebp() with the force argument set to 1.
3041  */
3042 int
3043 nfs_bwrite(v)
3044 	void *v;
3045 {
3046 	struct vop_bwrite_args /* {
3047 		struct vnode *a_bp;
3048 	} */ *ap = v;
3049 
3050 	return (nfs_writebp(ap->a_bp, 1));
3051 }
3052 
3053 /*
3054  * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3055  * the force flag is one and it also handles the B_NEEDCOMMIT flag.
3056  */
3057 int
3058 nfs_writebp(bp, force)
3059 	register struct buf *bp;
3060 	int force;
3061 {
3062 	register int oldflags = bp->b_flags, retv = 1;
3063 	register struct proc *p = curproc;	/* XXX */
3064 	off_t off;
3065 
3066 	if(!(bp->b_flags & B_BUSY))
3067 		panic("bwrite: buffer is not busy???");
3068 
3069 #ifdef fvdl_debug
3070 	printf("nfs_writebp(%x): vp %x voff %d vend %d doff %d dend %d\n",
3071 	    bp, bp->b_vp, bp->b_validoff, bp->b_validend, bp->b_dirtyoff,
3072 	    bp->b_dirtyend);
3073 #endif
3074 	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI|B_AGE);
3075 
3076 	if (oldflags & B_ASYNC) {
3077 		if (oldflags & B_DELWRI) {
3078 			reassignbuf(bp, bp->b_vp);
3079 		} else if (p) {
3080 			++p->p_stats->p_ru.ru_oublock;
3081 		}
3082 	}
3083 	bp->b_vp->v_numoutput++;
3084 
3085 	/*
3086 	 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not
3087 	 * an actual write will have to be scheduled via. VOP_STRATEGY().
3088 	 * If B_WRITEINPROG is already set, then push it with a write anyhow.
3089 	 */
3090 	if ((oldflags & (B_NEEDCOMMIT | B_WRITEINPROG)) == B_NEEDCOMMIT) {
3091 		off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
3092 		bp->b_flags |= B_WRITEINPROG;
3093 		retv = nfs_commit(bp->b_vp, off, bp->b_dirtyend-bp->b_dirtyoff,
3094 			bp->b_wcred, bp->b_proc);
3095 		bp->b_flags &= ~B_WRITEINPROG;
3096 		if (!retv) {
3097 			bp->b_dirtyoff = bp->b_dirtyend = 0;
3098 			bp->b_flags &= ~B_NEEDCOMMIT;
3099 			biodone(bp);
3100 		} else if (retv == NFSERR_STALEWRITEVERF)
3101 			nfs_clearcommit(bp->b_vp->v_mount);
3102 	}
3103 	if (retv) {
3104 		if (force)
3105 			bp->b_flags |= B_WRITEINPROG;
3106 		VOP_STRATEGY(bp);
3107 	}
3108 
3109 	if( (oldflags & B_ASYNC) == 0) {
3110 		int rtval = biowait(bp);
3111 		if (oldflags & B_DELWRI) {
3112 			reassignbuf(bp, bp->b_vp);
3113 		} else if (p) {
3114 			++p->p_stats->p_ru.ru_oublock;
3115 		}
3116 		brelse(bp);
3117 		return (rtval);
3118 	}
3119 
3120 	return (0);
3121 }
3122 
3123 /*
3124  * nfs special file access vnode op.
3125  * Essentially just get vattr and then imitate iaccess() since the device is
3126  * local to the client.
3127  */
3128 int
3129 nfsspec_access(v)
3130 	void *v;
3131 {
3132 	struct vop_access_args /* {
3133 		struct vnode *a_vp;
3134 		int  a_mode;
3135 		struct ucred *a_cred;
3136 		struct proc *a_p;
3137 	} */ *ap = v;
3138 	struct vattr va;
3139 	struct vnode *vp = ap->a_vp;
3140 	int error;
3141 
3142 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3143 	if (error)
3144 		return (error);
3145 
3146         /*
3147 	 * Disallow write attempts on filesystems mounted read-only;
3148 	 * unless the file is a socket, fifo, or a block or character
3149 	 * device resident on the filesystem.
3150 	 */
3151 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3152 		switch (vp->v_type) {
3153 		case VREG:
3154 		case VDIR:
3155 		case VLNK:
3156 			return (EROFS);
3157 		default:
3158 			break;
3159 		}
3160 	}
3161 
3162 	return (vaccess(va.va_type, va.va_mode,
3163 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3164 }
3165 
3166 /*
3167  * Read wrapper for special devices.
3168  */
3169 int
3170 nfsspec_read(v)
3171 	void *v;
3172 {
3173 	struct vop_read_args /* {
3174 		struct vnode *a_vp;
3175 		struct uio *a_uio;
3176 		int  a_ioflag;
3177 		struct ucred *a_cred;
3178 	} */ *ap = v;
3179 	register struct nfsnode *np = VTONFS(ap->a_vp);
3180 
3181 	/*
3182 	 * Set access flag.
3183 	 */
3184 	np->n_flag |= NACC;
3185 	np->n_atim.tv_sec = time.tv_sec;
3186 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3187 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3188 }
3189 
3190 /*
3191  * Write wrapper for special devices.
3192  */
3193 int
3194 nfsspec_write(v)
3195 	void *v;
3196 {
3197 	struct vop_write_args /* {
3198 		struct vnode *a_vp;
3199 		struct uio *a_uio;
3200 		int  a_ioflag;
3201 		struct ucred *a_cred;
3202 	} */ *ap = v;
3203 	register struct nfsnode *np = VTONFS(ap->a_vp);
3204 
3205 	/*
3206 	 * Set update flag.
3207 	 */
3208 	np->n_flag |= NUPD;
3209 	np->n_mtim.tv_sec = time.tv_sec;
3210 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3211 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3212 }
3213 
3214 /*
3215  * Close wrapper for special devices.
3216  *
3217  * Update the times on the nfsnode then do device close.
3218  */
3219 int
3220 nfsspec_close(v)
3221 	void *v;
3222 {
3223 	struct vop_close_args /* {
3224 		struct vnode *a_vp;
3225 		int  a_fflag;
3226 		struct ucred *a_cred;
3227 		struct proc *a_p;
3228 	} */ *ap = v;
3229 	register struct vnode *vp = ap->a_vp;
3230 	register struct nfsnode *np = VTONFS(vp);
3231 	struct vattr vattr;
3232 
3233 	if (np->n_flag & (NACC | NUPD)) {
3234 		np->n_flag |= NCHG;
3235 		if (vp->v_usecount == 1 &&
3236 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3237 			VATTR_NULL(&vattr);
3238 			if (np->n_flag & NACC)
3239 				vattr.va_atime = np->n_atim;
3240 			if (np->n_flag & NUPD)
3241 				vattr.va_mtime = np->n_mtim;
3242 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3243 		}
3244 	}
3245 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3246 }
3247 
3248 #ifdef FIFO
3249 /*
3250  * Read wrapper for fifos.
3251  */
3252 int
3253 nfsfifo_read(v)
3254 	void *v;
3255 {
3256 	struct vop_read_args /* {
3257 		struct vnode *a_vp;
3258 		struct uio *a_uio;
3259 		int  a_ioflag;
3260 		struct ucred *a_cred;
3261 	} */ *ap = v;
3262 	extern int (**fifo_vnodeop_p) __P((void *));
3263 	register struct nfsnode *np = VTONFS(ap->a_vp);
3264 
3265 	/*
3266 	 * Set access flag.
3267 	 */
3268 	np->n_flag |= NACC;
3269 	np->n_atim.tv_sec = time.tv_sec;
3270 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3271 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3272 }
3273 
3274 /*
3275  * Write wrapper for fifos.
3276  */
3277 int
3278 nfsfifo_write(v)
3279 	void *v;
3280 {
3281 	struct vop_write_args /* {
3282 		struct vnode *a_vp;
3283 		struct uio *a_uio;
3284 		int  a_ioflag;
3285 		struct ucred *a_cred;
3286 	} */ *ap = v;
3287 	extern int (**fifo_vnodeop_p) __P((void *));
3288 	register struct nfsnode *np = VTONFS(ap->a_vp);
3289 
3290 	/*
3291 	 * Set update flag.
3292 	 */
3293 	np->n_flag |= NUPD;
3294 	np->n_mtim.tv_sec = time.tv_sec;
3295 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3296 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3297 }
3298 
3299 /*
3300  * Close wrapper for fifos.
3301  *
3302  * Update the times on the nfsnode then do fifo close.
3303  */
3304 int
3305 nfsfifo_close(v)
3306 	void *v;
3307 {
3308 	struct vop_close_args /* {
3309 		struct vnode *a_vp;
3310 		int  a_fflag;
3311 		struct ucred *a_cred;
3312 		struct proc *a_p;
3313 	} */ *ap = v;
3314 	register struct vnode *vp = ap->a_vp;
3315 	register struct nfsnode *np = VTONFS(vp);
3316 	struct vattr vattr;
3317 	extern int (**fifo_vnodeop_p) __P((void *));
3318 
3319 	if (np->n_flag & (NACC | NUPD)) {
3320 		if (np->n_flag & NACC) {
3321 			np->n_atim.tv_sec = time.tv_sec;
3322 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3323 		}
3324 		if (np->n_flag & NUPD) {
3325 			np->n_mtim.tv_sec = time.tv_sec;
3326 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3327 		}
3328 		np->n_flag |= NCHG;
3329 		if (vp->v_usecount == 1 &&
3330 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3331 			VATTR_NULL(&vattr);
3332 			if (np->n_flag & NACC)
3333 				vattr.va_atime = np->n_atim;
3334 			if (np->n_flag & NUPD)
3335 				vattr.va_mtime = np->n_mtim;
3336 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3337 		}
3338 	}
3339 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3340 }
3341 #endif /* ! FIFO */
3342