xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 37b34d511dea595d3ba03a661cf3b775038ea5f8)
1 /*	$NetBSD: nfs_vnops.c,v 1.152 2002/10/18 19:08:15 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	8.19 (Berkeley) 7/31/95
39  */
40 
41 /*
42  * vnode op calls for Sun NFS version 2 and 3
43  */
44 
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.152 2002/10/18 19:08:15 thorpej Exp $");
47 
48 #include "opt_nfs.h"
49 #include "opt_uvmhist.h"
50 
51 #include <sys/param.h>
52 #include <sys/proc.h>
53 #include <sys/kernel.h>
54 #include <sys/systm.h>
55 #include <sys/resourcevar.h>
56 #include <sys/proc.h>
57 #include <sys/mount.h>
58 #include <sys/buf.h>
59 #include <sys/malloc.h>
60 #include <sys/mbuf.h>
61 #include <sys/namei.h>
62 #include <sys/vnode.h>
63 #include <sys/dirent.h>
64 #include <sys/fcntl.h>
65 #include <sys/hash.h>
66 #include <sys/lockf.h>
67 #include <sys/stat.h>
68 #include <sys/unistd.h>
69 
70 #include <uvm/uvm_extern.h>
71 #include <uvm/uvm.h>
72 
73 #include <miscfs/fifofs/fifo.h>
74 #include <miscfs/genfs/genfs.h>
75 #include <miscfs/specfs/specdev.h>
76 
77 #include <nfs/rpcv2.h>
78 #include <nfs/nfsproto.h>
79 #include <nfs/nfs.h>
80 #include <nfs/nfsnode.h>
81 #include <nfs/nfsmount.h>
82 #include <nfs/xdr_subs.h>
83 #include <nfs/nfsm_subs.h>
84 #include <nfs/nqnfs.h>
85 #include <nfs/nfs_var.h>
86 
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 
91 /* Defs */
92 #define	TRUE	1
93 #define	FALSE	0
94 
95 /*
96  * Global vfs data structures for nfs
97  */
98 int (**nfsv2_vnodeop_p) __P((void *));
99 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
100 	{ &vop_default_desc, vn_default_error },
101 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
102 	{ &vop_create_desc, nfs_create },		/* create */
103 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
104 	{ &vop_open_desc, nfs_open },			/* open */
105 	{ &vop_close_desc, nfs_close },			/* close */
106 	{ &vop_access_desc, nfs_access },		/* access */
107 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
108 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
109 	{ &vop_read_desc, nfs_read },			/* read */
110 	{ &vop_write_desc, nfs_write },			/* write */
111 	{ &vop_lease_desc, nfs_lease_check },		/* lease */
112 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
113 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
114 	{ &vop_poll_desc, nfs_poll },			/* poll */
115 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
116 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
117 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
118 	{ &vop_seek_desc, nfs_seek },			/* seek */
119 	{ &vop_remove_desc, nfs_remove },		/* remove */
120 	{ &vop_link_desc, nfs_link },			/* link */
121 	{ &vop_rename_desc, nfs_rename },		/* rename */
122 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
123 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
124 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
125 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
126 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
127 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
128 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
129 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
130 	{ &vop_lock_desc, nfs_lock },			/* lock */
131 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
132 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
133 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
134 	{ &vop_print_desc, nfs_print },			/* print */
135 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
136 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
137 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
138 	{ &vop_blkatoff_desc, nfs_blkatoff },		/* blkatoff */
139 	{ &vop_valloc_desc, nfs_valloc },		/* valloc */
140 	{ &vop_reallocblks_desc, nfs_reallocblks },	/* reallocblks */
141 	{ &vop_vfree_desc, nfs_vfree },			/* vfree */
142 	{ &vop_truncate_desc, nfs_truncate },		/* truncate */
143 	{ &vop_update_desc, nfs_update },		/* update */
144 	{ &vop_bwrite_desc, nfs_bwrite },		/* bwrite */
145 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
146 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
147 	{ NULL, NULL }
148 };
149 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
150 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
151 
152 /*
153  * Special device vnode ops
154  */
155 int (**spec_nfsv2nodeop_p) __P((void *));
156 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
157 	{ &vop_default_desc, vn_default_error },
158 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
159 	{ &vop_create_desc, spec_create },		/* create */
160 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
161 	{ &vop_open_desc, spec_open },			/* open */
162 	{ &vop_close_desc, nfsspec_close },		/* close */
163 	{ &vop_access_desc, nfsspec_access },		/* access */
164 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
165 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
166 	{ &vop_read_desc, nfsspec_read },		/* read */
167 	{ &vop_write_desc, nfsspec_write },		/* write */
168 	{ &vop_lease_desc, spec_lease_check },		/* lease */
169 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
170 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
171 	{ &vop_poll_desc, spec_poll },			/* poll */
172 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
173 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
174 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
175 	{ &vop_seek_desc, spec_seek },			/* seek */
176 	{ &vop_remove_desc, spec_remove },		/* remove */
177 	{ &vop_link_desc, spec_link },			/* link */
178 	{ &vop_rename_desc, spec_rename },		/* rename */
179 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
180 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
181 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
182 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
183 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
184 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
185 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
186 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
187 	{ &vop_lock_desc, nfs_lock },			/* lock */
188 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
189 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
190 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
191 	{ &vop_print_desc, nfs_print },			/* print */
192 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
193 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
194 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
195 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
196 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
197 	{ &vop_reallocblks_desc, spec_reallocblks },	/* reallocblks */
198 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
199 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
200 	{ &vop_update_desc, nfs_update },		/* update */
201 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
202 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
203 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
204 	{ NULL, NULL }
205 };
206 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
207 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
208 
209 int (**fifo_nfsv2nodeop_p) __P((void *));
210 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
211 	{ &vop_default_desc, vn_default_error },
212 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
213 	{ &vop_create_desc, fifo_create },		/* create */
214 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
215 	{ &vop_open_desc, fifo_open },			/* open */
216 	{ &vop_close_desc, nfsfifo_close },		/* close */
217 	{ &vop_access_desc, nfsspec_access },		/* access */
218 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
219 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
220 	{ &vop_read_desc, nfsfifo_read },		/* read */
221 	{ &vop_write_desc, nfsfifo_write },		/* write */
222 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
223 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
224 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
225 	{ &vop_poll_desc, fifo_poll },			/* poll */
226 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
227 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
228 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
229 	{ &vop_seek_desc, fifo_seek },			/* seek */
230 	{ &vop_remove_desc, fifo_remove },		/* remove */
231 	{ &vop_link_desc, fifo_link },			/* link */
232 	{ &vop_rename_desc, fifo_rename },		/* rename */
233 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
234 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
235 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
236 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
237 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
238 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
239 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
240 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
241 	{ &vop_lock_desc, nfs_lock },			/* lock */
242 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
243 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
244 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
245 	{ &vop_print_desc, nfs_print },			/* print */
246 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
247 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
248 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
249 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
250 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
251 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
252 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
253 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
254 	{ &vop_update_desc, nfs_update },		/* update */
255 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
256 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
257 	{ NULL, NULL }
258 };
259 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
260 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
261 
262 /*
263  * Global variables
264  */
265 extern u_int32_t nfs_true, nfs_false;
266 extern u_int32_t nfs_xdrneg1;
267 extern nfstype nfsv3_type[9];
268 
269 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
270 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
271 int nfs_numasync = 0;
272 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
273 
274 /*
275  * nfs null call from vfs.
276  */
277 int
278 nfs_null(vp, cred, procp)
279 	struct vnode *vp;
280 	struct ucred *cred;
281 	struct proc *procp;
282 {
283 	caddr_t bpos, dpos;
284 	int error = 0;
285 	struct mbuf *mreq, *mrep, *md, *mb;
286 
287 	nfsm_reqhead(vp, NFSPROC_NULL, 0);
288 	nfsm_request(vp, NFSPROC_NULL, procp, cred);
289 	nfsm_reqdone;
290 	return (error);
291 }
292 
293 /*
294  * nfs access vnode op.
295  * For nfs version 2, just return ok. File accesses may fail later.
296  * For nfs version 3, use the access rpc to check accessibility. If file modes
297  * are changed on the server, accesses might still fail later.
298  */
299 int
300 nfs_access(v)
301 	void *v;
302 {
303 	struct vop_access_args /* {
304 		struct vnode *a_vp;
305 		int  a_mode;
306 		struct ucred *a_cred;
307 		struct proc *a_p;
308 	} */ *ap = v;
309 	struct vnode *vp = ap->a_vp;
310 	u_int32_t *tl;
311 	caddr_t cp;
312 	int32_t t1, t2;
313 	caddr_t bpos, dpos, cp2;
314 	int error = 0, attrflag, cachevalid;
315 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
316 	u_int32_t mode, rmode;
317 	const int v3 = NFS_ISV3(vp);
318 	struct nfsnode *np = VTONFS(vp);
319 
320 	cachevalid = (np->n_accstamp != -1 &&
321 	    (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
322 	    np->n_accuid == ap->a_cred->cr_uid);
323 
324 	/*
325 	 * Check access cache first. If this request has been made for this
326 	 * uid shortly before, use the cached result.
327 	 */
328 	if (cachevalid) {
329 		if (!np->n_accerror) {
330 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
331 				return np->n_accerror;
332 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
333 			return np->n_accerror;
334 	}
335 
336 	/*
337 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
338 	 * ufs_access() locally using the vattr. This may not be correct,
339 	 * since the server may apply other access criteria such as
340 	 * client uid-->server uid mapping that we do not know about, but
341 	 * this is better than just returning anything that is lying about
342 	 * in the cache.
343 	 */
344 	if (v3) {
345 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
346 		nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
347 		nfsm_fhtom(vp, v3);
348 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
349 		if (ap->a_mode & VREAD)
350 			mode = NFSV3ACCESS_READ;
351 		else
352 			mode = 0;
353 		if (vp->v_type != VDIR) {
354 			if (ap->a_mode & VWRITE)
355 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
356 			if (ap->a_mode & VEXEC)
357 				mode |= NFSV3ACCESS_EXECUTE;
358 		} else {
359 			if (ap->a_mode & VWRITE)
360 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
361 					 NFSV3ACCESS_DELETE);
362 			if (ap->a_mode & VEXEC)
363 				mode |= NFSV3ACCESS_LOOKUP;
364 		}
365 		*tl = txdr_unsigned(mode);
366 		nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
367 		nfsm_postop_attr(vp, attrflag);
368 		if (!error) {
369 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
370 			rmode = fxdr_unsigned(u_int32_t, *tl);
371 			/*
372 			 * The NFS V3 spec does not clarify whether or not
373 			 * the returned access bits can be a superset of
374 			 * the ones requested, so...
375 			 */
376 			if ((rmode & mode) != mode)
377 				error = EACCES;
378 		}
379 		nfsm_reqdone;
380 	} else
381 		return (nfsspec_access(ap));
382 	/*
383 	 * Disallow write attempts on filesystems mounted read-only;
384 	 * unless the file is a socket, fifo, or a block or character
385 	 * device resident on the filesystem.
386 	 */
387 	if (!error && (ap->a_mode & VWRITE) &&
388 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
389 		switch (vp->v_type) {
390 		case VREG:
391 		case VDIR:
392 		case VLNK:
393 			error = EROFS;
394 		default:
395 			break;
396 		}
397 	}
398 
399 	if (!error || error == EACCES) {
400 		/*
401 		 * If we got the same result as for a previous,
402 		 * different request, OR it in. Don't update
403 		 * the timestamp in that case.
404 		 */
405 		if (cachevalid && np->n_accstamp != -1 &&
406 		    error == np->n_accerror) {
407 			if (!error)
408 				np->n_accmode |= ap->a_mode;
409 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
410 				np->n_accmode = ap->a_mode;
411 		} else {
412 			np->n_accstamp = time.tv_sec;
413 			np->n_accuid = ap->a_cred->cr_uid;
414 			np->n_accmode = ap->a_mode;
415 			np->n_accerror = error;
416 		}
417 	}
418 
419 	return (error);
420 }
421 
422 /*
423  * nfs open vnode op
424  * Check to see if the type is ok
425  * and that deletion is not in progress.
426  * For paged in text files, you will need to flush the page cache
427  * if consistency is lost.
428  */
429 /* ARGSUSED */
430 int
431 nfs_open(v)
432 	void *v;
433 {
434 	struct vop_open_args /* {
435 		struct vnode *a_vp;
436 		int  a_mode;
437 		struct ucred *a_cred;
438 		struct proc *a_p;
439 	} */ *ap = v;
440 	struct vnode *vp = ap->a_vp;
441 	struct nfsnode *np = VTONFS(vp);
442 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
443 	struct vattr vattr;
444 	int error;
445 
446 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
447 		return (EACCES);
448 	}
449 
450 	/*
451 	 * Initialize read and write creds here, for swapfiles
452 	 * and other paths that don't set the creds themselves.
453 	 */
454 
455 	if (ap->a_mode & FREAD) {
456 		if (np->n_rcred) {
457 			crfree(np->n_rcred);
458 		}
459 		np->n_rcred = ap->a_cred;
460 		crhold(np->n_rcred);
461 	}
462 	if (ap->a_mode & FWRITE) {
463 		if (np->n_wcred) {
464 			crfree(np->n_wcred);
465 		}
466 		np->n_wcred = ap->a_cred;
467 		crhold(np->n_wcred);
468 	}
469 
470 #ifndef NFS_V2_ONLY
471 	/*
472 	 * Get a valid lease. If cached data is stale, flush it.
473 	 */
474 	if (nmp->nm_flag & NFSMNT_NQNFS) {
475 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
476 		    do {
477 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
478 			    ap->a_p);
479 		    } while (error == NQNFS_EXPIRED);
480 		    if (error)
481 			return (error);
482 		    if (np->n_lrev != np->n_brev ||
483 			(np->n_flag & NQNFSNONCACHE)) {
484 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
485 				ap->a_p, 1)) == EINTR)
486 				return (error);
487 			np->n_brev = np->n_lrev;
488 		    }
489 		}
490 	} else
491 #endif
492 	{
493 		if (np->n_flag & NMODIFIED) {
494 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
495 				ap->a_p, 1)) == EINTR)
496 				return (error);
497 			np->n_attrstamp = 0;
498 			if (vp->v_type == VDIR) {
499 				nfs_invaldircache(vp, 0);
500 				np->n_direofoffset = 0;
501 			}
502 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
503 			if (error)
504 				return (error);
505 			np->n_mtime = vattr.va_mtime.tv_sec;
506 		} else {
507 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
508 			if (error)
509 				return (error);
510 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
511 				if (vp->v_type == VDIR) {
512 					nfs_invaldircache(vp, 0);
513 					np->n_direofoffset = 0;
514 				}
515 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
516 					ap->a_cred, ap->a_p, 1)) == EINTR)
517 					return (error);
518 				np->n_mtime = vattr.va_mtime.tv_sec;
519 			}
520 		}
521 	}
522 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
523 		np->n_attrstamp = 0; /* For Open/Close consistency */
524 	return (0);
525 }
526 
527 /*
528  * nfs close vnode op
529  * What an NFS client should do upon close after writing is a debatable issue.
530  * Most NFS clients push delayed writes to the server upon close, basically for
531  * two reasons:
532  * 1 - So that any write errors may be reported back to the client process
533  *     doing the close system call. By far the two most likely errors are
534  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
535  * 2 - To put a worst case upper bound on cache inconsistency between
536  *     multiple clients for the file.
537  * There is also a consistency problem for Version 2 of the protocol w.r.t.
538  * not being able to tell if other clients are writing a file concurrently,
539  * since there is no way of knowing if the changed modify time in the reply
540  * is only due to the write for this client.
541  * (NFS Version 3 provides weak cache consistency data in the reply that
542  *  should be sufficient to detect and handle this case.)
543  *
544  * The current code does the following:
545  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
546  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
547  *                     or commit them (this satisfies 1 and 2 except for the
548  *                     case where the server crashes after this close but
549  *                     before the commit RPC, which is felt to be "good
550  *                     enough". Changing the last argument to nfs_flush() to
551  *                     a 1 would force a commit operation, if it is felt a
552  *                     commit is necessary now.
553  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
554  *                     1 should be dealt with via an fsync() system call for
555  *                     cases where write errors are important.
556  */
557 /* ARGSUSED */
558 int
559 nfs_close(v)
560 	void *v;
561 {
562 	struct vop_close_args /* {
563 		struct vnodeop_desc *a_desc;
564 		struct vnode *a_vp;
565 		int  a_fflag;
566 		struct ucred *a_cred;
567 		struct proc *a_p;
568 	} */ *ap = v;
569 	struct vnode *vp = ap->a_vp;
570 	struct nfsnode *np = VTONFS(vp);
571 	int error = 0;
572 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
573 
574 	if (vp->v_type == VREG) {
575 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
576 		(np->n_flag & NMODIFIED)) {
577 		if (NFS_ISV3(vp)) {
578 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
579 		    np->n_flag &= ~NMODIFIED;
580 		} else
581 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
582 		np->n_attrstamp = 0;
583 	    }
584 	    if (np->n_flag & NWRITEERR) {
585 		np->n_flag &= ~NWRITEERR;
586 		error = np->n_error;
587 	    }
588 	}
589 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
590 	return (error);
591 }
592 
593 /*
594  * nfs getattr call from vfs.
595  */
596 int
597 nfs_getattr(v)
598 	void *v;
599 {
600 	struct vop_getattr_args /* {
601 		struct vnode *a_vp;
602 		struct vattr *a_vap;
603 		struct ucred *a_cred;
604 		struct proc *a_p;
605 	} */ *ap = v;
606 	struct vnode *vp = ap->a_vp;
607 	struct nfsnode *np = VTONFS(vp);
608 	caddr_t cp;
609 	u_int32_t *tl;
610 	int32_t t1, t2;
611 	caddr_t bpos, dpos;
612 	int error = 0;
613 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
614 	const int v3 = NFS_ISV3(vp);
615 
616 	/*
617 	 * Update local times for special files.
618 	 */
619 	if (np->n_flag & (NACC | NUPD))
620 		np->n_flag |= NCHG;
621 	/*
622 	 * First look in the cache.
623 	 */
624 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
625 		return (0);
626 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
627 	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
628 	nfsm_fhtom(vp, v3);
629 	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
630 	if (!error) {
631 		nfsm_loadattr(vp, ap->a_vap);
632 		if (vp->v_type == VDIR &&
633 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
634 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
635 	}
636 	nfsm_reqdone;
637 	return (error);
638 }
639 
640 /*
641  * nfs setattr call.
642  */
643 int
644 nfs_setattr(v)
645 	void *v;
646 {
647 	struct vop_setattr_args /* {
648 		struct vnodeop_desc *a_desc;
649 		struct vnode *a_vp;
650 		struct vattr *a_vap;
651 		struct ucred *a_cred;
652 		struct proc *a_p;
653 	} */ *ap = v;
654 	struct vnode *vp = ap->a_vp;
655 	struct nfsnode *np = VTONFS(vp);
656 	struct vattr *vap = ap->a_vap;
657 	int error = 0;
658 	u_quad_t tsize = 0;
659 
660 	/*
661 	 * Setting of flags is not supported.
662 	 */
663 	if (vap->va_flags != VNOVAL)
664 		return (EOPNOTSUPP);
665 
666 	/*
667 	 * Disallow write attempts if the filesystem is mounted read-only.
668 	 */
669   	if ((vap->va_uid != (uid_t)VNOVAL ||
670 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
671 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
672 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
673 		return (EROFS);
674 	if (vap->va_size != VNOVAL) {
675  		switch (vp->v_type) {
676  		case VDIR:
677  			return (EISDIR);
678  		case VCHR:
679  		case VBLK:
680  		case VSOCK:
681  		case VFIFO:
682 			if (vap->va_mtime.tv_sec == VNOVAL &&
683 			    vap->va_atime.tv_sec == VNOVAL &&
684 			    vap->va_mode == (mode_t)VNOVAL &&
685 			    vap->va_uid == (uid_t)VNOVAL &&
686 			    vap->va_gid == (gid_t)VNOVAL)
687 				return (0);
688  			vap->va_size = VNOVAL;
689  			break;
690  		default:
691 			/*
692 			 * Disallow write attempts if the filesystem is
693 			 * mounted read-only.
694 			 */
695 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
696 				return (EROFS);
697  			uvm_vnp_setsize(vp, vap->va_size);
698  			tsize = np->n_size;
699 			np->n_size = vap->va_size;
700  			if (vap->va_size == 0)
701  				error = nfs_vinvalbuf(vp, 0,
702  				     ap->a_cred, ap->a_p, 1);
703 			else
704 				error = nfs_vinvalbuf(vp, V_SAVE,
705 				     ap->a_cred, ap->a_p, 1);
706 			if (error) {
707 				uvm_vnp_setsize(vp, tsize);
708 				return (error);
709 			}
710  			np->n_vattr->va_size = vap->va_size;
711   		}
712   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
713 		vap->va_atime.tv_sec != VNOVAL) &&
714 		vp->v_type == VREG &&
715   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
716 		 ap->a_p, 1)) == EINTR)
717 		return (error);
718 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
719 	if (error && vap->va_size != VNOVAL) {
720 		np->n_size = np->n_vattr->va_size = tsize;
721 		uvm_vnp_setsize(vp, np->n_size);
722 	}
723 	return (error);
724 }
725 
726 /*
727  * Do an nfs setattr rpc.
728  */
729 int
730 nfs_setattrrpc(vp, vap, cred, procp)
731 	struct vnode *vp;
732 	struct vattr *vap;
733 	struct ucred *cred;
734 	struct proc *procp;
735 {
736 	struct nfsv2_sattr *sp;
737 	caddr_t cp;
738 	int32_t t1, t2;
739 	caddr_t bpos, dpos, cp2;
740 	u_int32_t *tl;
741 	int error = 0, wccflag = NFSV3_WCCRATTR;
742 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
743 	const int v3 = NFS_ISV3(vp);
744 
745 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
746 	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
747 	nfsm_fhtom(vp, v3);
748 	if (v3) {
749 		nfsm_v3attrbuild(vap, TRUE);
750 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
751 		*tl = nfs_false;
752 	} else {
753 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
754 		if (vap->va_mode == (mode_t)VNOVAL)
755 			sp->sa_mode = nfs_xdrneg1;
756 		else
757 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
758 		if (vap->va_uid == (uid_t)VNOVAL)
759 			sp->sa_uid = nfs_xdrneg1;
760 		else
761 			sp->sa_uid = txdr_unsigned(vap->va_uid);
762 		if (vap->va_gid == (gid_t)VNOVAL)
763 			sp->sa_gid = nfs_xdrneg1;
764 		else
765 			sp->sa_gid = txdr_unsigned(vap->va_gid);
766 		sp->sa_size = txdr_unsigned(vap->va_size);
767 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
768 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
769 	}
770 	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
771 	if (v3) {
772 		nfsm_wcc_data(vp, wccflag);
773 	} else
774 		nfsm_loadattr(vp, (struct vattr *)0);
775 	nfsm_reqdone;
776 	return (error);
777 }
778 
779 /*
780  * nfs lookup call, one step at a time...
781  * First look in cache
782  * If not found, unlock the directory nfsnode and do the rpc
783  *
784  * This code is full of lock/unlock statements and checks, because
785  * we continue after cache_lookup has finished (we need to check
786  * with the attr cache and do an rpc if it has timed out). This means
787  * that the locking effects of cache_lookup have to be taken into
788  * account.
789  */
790 int
791 nfs_lookup(v)
792 	void *v;
793 {
794 	struct vop_lookup_args /* {
795 		struct vnodeop_desc *a_desc;
796 		struct vnode *a_dvp;
797 		struct vnode **a_vpp;
798 		struct componentname *a_cnp;
799 	} */ *ap = v;
800 	struct componentname *cnp = ap->a_cnp;
801 	struct vnode *dvp = ap->a_dvp;
802 	struct vnode **vpp = ap->a_vpp;
803 	int flags;
804 	struct vnode *newvp;
805 	u_int32_t *tl;
806 	caddr_t cp;
807 	int32_t t1, t2;
808 	struct nfsmount *nmp;
809 	caddr_t bpos, dpos, cp2;
810 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
811 	long len;
812 	nfsfh_t *fhp;
813 	struct nfsnode *np;
814 	int lockparent, wantparent, error = 0, attrflag, fhsize;
815 	const int v3 = NFS_ISV3(dvp);
816 
817 	cnp->cn_flags &= ~PDIRUNLOCK;
818 	flags = cnp->cn_flags;
819 
820 	*vpp = NULLVP;
821 	newvp = NULLVP;
822 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
823 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
824 		return (EROFS);
825 	if (dvp->v_type != VDIR)
826 		return (ENOTDIR);
827 
828 	lockparent = flags & LOCKPARENT;
829 	wantparent = flags & (LOCKPARENT|WANTPARENT);
830 	nmp = VFSTONFS(dvp->v_mount);
831 	np = VTONFS(dvp);
832 
833 	/*
834 	 * Before tediously performing a linear scan of the directory,
835 	 * check the name cache to see if the directory/name pair
836 	 * we are looking for is known already.
837 	 * If the directory/name pair is found in the name cache,
838 	 * we have to ensure the directory has not changed from
839 	 * the time the cache entry has been created. If it has,
840 	 * the cache entry has to be ignored
841 	 */
842 	if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
843 		struct vattr vattr;
844 		int err2;
845 
846 		if (error && error != ENOENT) {
847 			*vpp = NULLVP;
848 			return error;
849 		}
850 
851 		if (cnp->cn_flags & PDIRUNLOCK) {
852 			err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
853 			if (err2 != 0) {
854 				*vpp = NULLVP;
855 				return err2;
856 			}
857 			cnp->cn_flags &= ~PDIRUNLOCK;
858 		}
859 
860 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
861 		if (err2 != 0) {
862 			if (error == 0) {
863 				if (*vpp != dvp)
864 					vput(*vpp);
865 				else
866 					vrele(*vpp);
867 			}
868 			*vpp = NULLVP;
869 			return err2;
870 		}
871 
872 		if (error == ENOENT) {
873 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
874 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
875 			    VTONFS(dvp)->n_nctime)
876 				return ENOENT;
877 			cache_purge(dvp);
878 			np->n_nctime = 0;
879 			goto dorpc;
880 		}
881 
882 		newvp = *vpp;
883 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
884 			&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
885 		{
886 			nfsstats.lookupcache_hits++;
887 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
888 				cnp->cn_flags |= SAVENAME;
889 			if ((!lockparent || !(flags & ISLASTCN)) &&
890 			     newvp != dvp)
891 				VOP_UNLOCK(dvp, 0);
892 			return (0);
893 		}
894 		cache_purge(newvp);
895 		if (newvp != dvp)
896 			vput(newvp);
897 		else
898 			vrele(newvp);
899 		*vpp = NULLVP;
900 	}
901 dorpc:
902 	error = 0;
903 	newvp = NULLVP;
904 	nfsstats.lookupcache_misses++;
905 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
906 	len = cnp->cn_namelen;
907 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
908 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
909 	nfsm_fhtom(dvp, v3);
910 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
911 	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
912 	if (error) {
913 		nfsm_postop_attr(dvp, attrflag);
914 		m_freem(mrep);
915 		goto nfsmout;
916 	}
917 	nfsm_getfh(fhp, fhsize, v3);
918 
919 	/*
920 	 * Handle RENAME case...
921 	 */
922 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
923 		if (NFS_CMPFH(np, fhp, fhsize)) {
924 			m_freem(mrep);
925 			return (EISDIR);
926 		}
927 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
928 		if (error) {
929 			m_freem(mrep);
930 			return error;
931 		}
932 		newvp = NFSTOV(np);
933 		if (v3) {
934 			nfsm_postop_attr(newvp, attrflag);
935 			nfsm_postop_attr(dvp, attrflag);
936 		} else
937 			nfsm_loadattr(newvp, (struct vattr *)0);
938 		*vpp = newvp;
939 		m_freem(mrep);
940 		cnp->cn_flags |= SAVENAME;
941 		if (!lockparent) {
942 			VOP_UNLOCK(dvp, 0);
943 			cnp->cn_flags |= PDIRUNLOCK;
944 		}
945 		return (0);
946 	}
947 
948 	/*
949 	 * The postop attr handling is duplicated for each if case,
950 	 * because it should be done while dvp is locked (unlocking
951 	 * dvp is different for each case).
952 	 */
953 
954 	if (NFS_CMPFH(np, fhp, fhsize)) {
955 		/*
956 		 * "." lookup
957 		 */
958 		VREF(dvp);
959 		newvp = dvp;
960 		if (v3) {
961 			nfsm_postop_attr(newvp, attrflag);
962 			nfsm_postop_attr(dvp, attrflag);
963 		} else
964 			nfsm_loadattr(newvp, (struct vattr *)0);
965 	} else if (flags & ISDOTDOT) {
966 		/*
967 		 * ".." lookup
968 		 */
969 		VOP_UNLOCK(dvp, 0);
970 		cnp->cn_flags |= PDIRUNLOCK;
971 
972 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
973 		if (error) {
974 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
975 				cnp->cn_flags &= ~PDIRUNLOCK;
976 			m_freem(mrep);
977 			return error;
978 		}
979 		newvp = NFSTOV(np);
980 
981 		if (v3) {
982 			nfsm_postop_attr(newvp, attrflag);
983 			nfsm_postop_attr(dvp, attrflag);
984 		} else
985 			nfsm_loadattr(newvp, (struct vattr *)0);
986 
987 		if (lockparent && (flags & ISLASTCN)) {
988 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
989 				m_freem(mrep);
990 				vput(newvp);
991 				return error;
992 			}
993 			cnp->cn_flags &= ~PDIRUNLOCK;
994 		}
995 	} else {
996 		/*
997 		 * Other lookups.
998 		 */
999 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1000 		if (error) {
1001 			m_freem(mrep);
1002 			return error;
1003 		}
1004 		newvp = NFSTOV(np);
1005 		if (v3) {
1006 			nfsm_postop_attr(newvp, attrflag);
1007 			nfsm_postop_attr(dvp, attrflag);
1008 		} else
1009 			nfsm_loadattr(newvp, (struct vattr *)0);
1010 		if (!lockparent || !(flags & ISLASTCN)) {
1011 			VOP_UNLOCK(dvp, 0);
1012 			cnp->cn_flags |= PDIRUNLOCK;
1013 		}
1014 	}
1015 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1016 		cnp->cn_flags |= SAVENAME;
1017 	if ((cnp->cn_flags & MAKEENTRY) &&
1018 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1019 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1020 		cache_enter(dvp, newvp, cnp);
1021 	}
1022 	*vpp = newvp;
1023 	nfsm_reqdone;
1024 	if (error) {
1025 		/*
1026 		 * We get here only because of errors returned by
1027 		 * the RPC. Otherwise we'll have returned above
1028 		 * (the nfsm_* macros will jump to nfsm_reqdone
1029 		 * on error).
1030 		 */
1031 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1032 		    cnp->cn_nameiop != CREATE) {
1033 			if (VTONFS(dvp)->n_nctime == 0)
1034 				VTONFS(dvp)->n_nctime =
1035 				    VTONFS(dvp)->n_vattr->va_mtime.tv_sec;
1036 			cache_enter(dvp, NULL, cnp);
1037 		}
1038 		if (newvp != NULLVP) {
1039 			vrele(newvp);
1040 			if (newvp != dvp)
1041 				VOP_UNLOCK(newvp, 0);
1042 		}
1043 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1044 		    (flags & ISLASTCN) && error == ENOENT) {
1045 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1046 				error = EROFS;
1047 			else
1048 				error = EJUSTRETURN;
1049 		}
1050 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1051 			cnp->cn_flags |= SAVENAME;
1052 		*vpp = NULL;
1053 	}
1054 
1055 	return error;
1056 }
1057 
1058 /*
1059  * nfs read call.
1060  * Just call nfs_bioread() to do the work.
1061  */
1062 int
1063 nfs_read(v)
1064 	void *v;
1065 {
1066 	struct vop_read_args /* {
1067 		struct vnode *a_vp;
1068 		struct uio *a_uio;
1069 		int  a_ioflag;
1070 		struct ucred *a_cred;
1071 	} */ *ap = v;
1072 	struct vnode *vp = ap->a_vp;
1073 
1074 	if (vp->v_type != VREG)
1075 		return (EPERM);
1076 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1077 }
1078 
1079 /*
1080  * nfs readlink call
1081  */
1082 int
1083 nfs_readlink(v)
1084 	void *v;
1085 {
1086 	struct vop_readlink_args /* {
1087 		struct vnode *a_vp;
1088 		struct uio *a_uio;
1089 		struct ucred *a_cred;
1090 	} */ *ap = v;
1091 	struct vnode *vp = ap->a_vp;
1092 
1093 	if (vp->v_type != VLNK)
1094 		return (EPERM);
1095 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1096 }
1097 
1098 /*
1099  * Do a readlink rpc.
1100  * Called by nfs_doio() from below the buffer cache.
1101  */
1102 int
1103 nfs_readlinkrpc(vp, uiop, cred)
1104 	struct vnode *vp;
1105 	struct uio *uiop;
1106 	struct ucred *cred;
1107 {
1108 	u_int32_t *tl;
1109 	caddr_t cp;
1110 	int32_t t1, t2;
1111 	caddr_t bpos, dpos, cp2;
1112 	int error = 0, len, attrflag;
1113 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1114 	const int v3 = NFS_ISV3(vp);
1115 
1116 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1117 	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1118 	nfsm_fhtom(vp, v3);
1119 	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
1120 	if (v3)
1121 		nfsm_postop_attr(vp, attrflag);
1122 	if (!error) {
1123 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1124 		nfsm_mtouio(uiop, len);
1125 	}
1126 	nfsm_reqdone;
1127 	return (error);
1128 }
1129 
1130 /*
1131  * nfs read rpc call
1132  * Ditto above
1133  */
1134 int
1135 nfs_readrpc(vp, uiop)
1136 	struct vnode *vp;
1137 	struct uio *uiop;
1138 {
1139 	u_int32_t *tl;
1140 	caddr_t cp;
1141 	int32_t t1, t2;
1142 	caddr_t bpos, dpos, cp2;
1143 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1144 	struct nfsmount *nmp;
1145 	int error = 0, len, retlen, tsiz, eof, attrflag;
1146 	const int v3 = NFS_ISV3(vp);
1147 
1148 #ifndef nolint
1149 	eof = 0;
1150 #endif
1151 	nmp = VFSTONFS(vp->v_mount);
1152 	tsiz = uiop->uio_resid;
1153 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1154 		return (EFBIG);
1155 	while (tsiz > 0) {
1156 		nfsstats.rpccnt[NFSPROC_READ]++;
1157 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1158 		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1159 		nfsm_fhtom(vp, v3);
1160 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1161 		if (v3) {
1162 			txdr_hyper(uiop->uio_offset, tl);
1163 			*(tl + 2) = txdr_unsigned(len);
1164 		} else {
1165 			*tl++ = txdr_unsigned(uiop->uio_offset);
1166 			*tl++ = txdr_unsigned(len);
1167 			*tl = 0;
1168 		}
1169 		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp,
1170 			     VTONFS(vp)->n_rcred);
1171 		if (v3) {
1172 			nfsm_postop_attr(vp, attrflag);
1173 			if (error) {
1174 				m_freem(mrep);
1175 				goto nfsmout;
1176 			}
1177 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1178 			eof = fxdr_unsigned(int, *(tl + 1));
1179 		} else
1180 			nfsm_loadattr(vp, (struct vattr *)0);
1181 		nfsm_strsiz(retlen, nmp->nm_rsize);
1182 		nfsm_mtouio(uiop, retlen);
1183 		m_freem(mrep);
1184 		tsiz -= retlen;
1185 		if (v3) {
1186 			if (eof || retlen == 0)
1187 				tsiz = 0;
1188 		} else if (retlen < len)
1189 			tsiz = 0;
1190 	}
1191 nfsmout:
1192 	return (error);
1193 }
1194 
1195 /*
1196  * nfs write call
1197  */
1198 int
1199 nfs_writerpc(vp, uiop, iomode, must_commit)
1200 	struct vnode *vp;
1201 	struct uio *uiop;
1202 	int *iomode, *must_commit;
1203 {
1204 	u_int32_t *tl;
1205 	caddr_t cp;
1206 	int32_t t1, t2, backup;
1207 	caddr_t bpos, dpos, cp2;
1208 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1209 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1210 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1211 	const int v3 = NFS_ISV3(vp);
1212 	int committed = NFSV3WRITE_FILESYNC;
1213 
1214 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1215 		panic("writerpc readonly vp %p", vp);
1216 	}
1217 
1218 #ifdef DIAGNOSTIC
1219 	if (uiop->uio_iovcnt != 1)
1220 		panic("nfs: writerpc iovcnt > 1");
1221 #endif
1222 	*must_commit = 0;
1223 	tsiz = uiop->uio_resid;
1224 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1225 		return (EFBIG);
1226 	while (tsiz > 0) {
1227 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1228 		len = min(tsiz, nmp->nm_wsize);
1229 		nfsm_reqhead(vp, NFSPROC_WRITE,
1230 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1231 		nfsm_fhtom(vp, v3);
1232 		if (v3) {
1233 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1234 			txdr_hyper(uiop->uio_offset, tl);
1235 			tl += 2;
1236 			*tl++ = txdr_unsigned(len);
1237 			*tl++ = txdr_unsigned(*iomode);
1238 			*tl = txdr_unsigned(len);
1239 		} else {
1240 			u_int32_t x;
1241 
1242 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1243 			/* Set both "begin" and "current" to non-garbage. */
1244 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1245 			*tl++ = x;      /* "begin offset" */
1246 			*tl++ = x;      /* "current offset" */
1247 			x = txdr_unsigned(len);
1248 			*tl++ = x;      /* total to this offset */
1249 			*tl = x;        /* size of this write */
1250 
1251 		}
1252 		nfsm_uiotom(uiop, len);
1253 		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp,
1254 			     VTONFS(vp)->n_wcred);
1255 		if (v3) {
1256 			wccflag = NFSV3_WCCCHK;
1257 			nfsm_wcc_data(vp, wccflag);
1258 			if (!error) {
1259 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1260 					+ NFSX_V3WRITEVERF);
1261 				rlen = fxdr_unsigned(int, *tl++);
1262 				if (rlen == 0) {
1263 					error = NFSERR_IO;
1264 					m_freem(mrep);
1265 					break;
1266 				} else if (rlen < len) {
1267 					backup = len - rlen;
1268 					uiop->uio_iov->iov_base =
1269 					    (caddr_t)uiop->uio_iov->iov_base -
1270 					    backup;
1271 					uiop->uio_iov->iov_len += backup;
1272 					uiop->uio_offset -= backup;
1273 					uiop->uio_resid += backup;
1274 					len = rlen;
1275 				}
1276 				commit = fxdr_unsigned(int, *tl++);
1277 
1278 				/*
1279 				 * Return the lowest committment level
1280 				 * obtained by any of the RPCs.
1281 				 */
1282 				if (committed == NFSV3WRITE_FILESYNC)
1283 					committed = commit;
1284 				else if (committed == NFSV3WRITE_DATASYNC &&
1285 					commit == NFSV3WRITE_UNSTABLE)
1286 					committed = commit;
1287 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1288 				    memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
1289 					NFSX_V3WRITEVERF);
1290 				    nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1291 				} else if (memcmp((caddr_t)tl,
1292 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1293 				    *must_commit = 1;
1294 				    memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
1295 					NFSX_V3WRITEVERF);
1296 				}
1297 			}
1298 		} else
1299 		    nfsm_loadattr(vp, (struct vattr *)0);
1300 		if (wccflag)
1301 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1302 		m_freem(mrep);
1303 		if (error)
1304 			break;
1305 		tsiz -= len;
1306 	}
1307 nfsmout:
1308 	*iomode = committed;
1309 	if (error)
1310 		uiop->uio_resid = tsiz;
1311 	return (error);
1312 }
1313 
1314 /*
1315  * nfs mknod rpc
1316  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1317  * mode set to specify the file type and the size field for rdev.
1318  */
1319 int
1320 nfs_mknodrpc(dvp, vpp, cnp, vap)
1321 	struct vnode *dvp;
1322 	struct vnode **vpp;
1323 	struct componentname *cnp;
1324 	struct vattr *vap;
1325 {
1326 	struct nfsv2_sattr *sp;
1327 	u_int32_t *tl;
1328 	caddr_t cp;
1329 	int32_t t1, t2;
1330 	struct vnode *newvp = (struct vnode *)0;
1331 	struct nfsnode *np;
1332 	char *cp2;
1333 	caddr_t bpos, dpos;
1334 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1335 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1336 	u_int32_t rdev;
1337 	const int v3 = NFS_ISV3(dvp);
1338 
1339 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1340 		rdev = txdr_unsigned(vap->va_rdev);
1341 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1342 		rdev = nfs_xdrneg1;
1343 	else {
1344 		VOP_ABORTOP(dvp, cnp);
1345 		vput(dvp);
1346 		return (EOPNOTSUPP);
1347 	}
1348 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1349 	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1350 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1351 	nfsm_fhtom(dvp, v3);
1352 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1353 	if (v3) {
1354 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1355 		*tl++ = vtonfsv3_type(vap->va_type);
1356 		nfsm_v3attrbuild(vap, FALSE);
1357 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1358 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1359 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1360 			*tl = txdr_unsigned(minor(vap->va_rdev));
1361 		}
1362 	} else {
1363 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1364 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1365 		sp->sa_uid = nfs_xdrneg1;
1366 		sp->sa_gid = nfs_xdrneg1;
1367 		sp->sa_size = rdev;
1368 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1369 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1370 	}
1371 	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1372 	if (!error) {
1373 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1374 		if (!gotvp) {
1375 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1376 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1377 			if (!error)
1378 				newvp = NFSTOV(np);
1379 		}
1380 	}
1381 	if (v3)
1382 		nfsm_wcc_data(dvp, wccflag);
1383 	nfsm_reqdone;
1384 	if (error) {
1385 		if (newvp)
1386 			vput(newvp);
1387 	} else {
1388 		if (cnp->cn_flags & MAKEENTRY)
1389 			cache_enter(dvp, newvp, cnp);
1390 		*vpp = newvp;
1391 	}
1392 	PNBUF_PUT(cnp->cn_pnbuf);
1393 	VTONFS(dvp)->n_flag |= NMODIFIED;
1394 	if (!wccflag)
1395 		VTONFS(dvp)->n_attrstamp = 0;
1396 	vput(dvp);
1397 	return (error);
1398 }
1399 
1400 /*
1401  * nfs mknod vop
1402  * just call nfs_mknodrpc() to do the work.
1403  */
1404 /* ARGSUSED */
1405 int
1406 nfs_mknod(v)
1407 	void *v;
1408 {
1409 	struct vop_mknod_args /* {
1410 		struct vnode *a_dvp;
1411 		struct vnode **a_vpp;
1412 		struct componentname *a_cnp;
1413 		struct vattr *a_vap;
1414 	} */ *ap = v;
1415 	int error;
1416 
1417 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1418 	return (error);
1419 }
1420 
1421 static u_long create_verf;
1422 /*
1423  * nfs file create call
1424  */
1425 int
1426 nfs_create(v)
1427 	void *v;
1428 {
1429 	struct vop_create_args /* {
1430 		struct vnode *a_dvp;
1431 		struct vnode **a_vpp;
1432 		struct componentname *a_cnp;
1433 		struct vattr *a_vap;
1434 	} */ *ap = v;
1435 	struct vnode *dvp = ap->a_dvp;
1436 	struct vattr *vap = ap->a_vap;
1437 	struct componentname *cnp = ap->a_cnp;
1438 	struct nfsv2_sattr *sp;
1439 	u_int32_t *tl;
1440 	caddr_t cp;
1441 	int32_t t1, t2;
1442 	struct nfsnode *np = (struct nfsnode *)0;
1443 	struct vnode *newvp = (struct vnode *)0;
1444 	caddr_t bpos, dpos, cp2;
1445 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1446 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1447 	const int v3 = NFS_ISV3(dvp);
1448 
1449 	/*
1450 	 * Oops, not for me..
1451 	 */
1452 	if (vap->va_type == VSOCK)
1453 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1454 
1455 #ifdef VA_EXCLUSIVE
1456 	if (vap->va_vaflags & VA_EXCLUSIVE)
1457 		fmode |= O_EXCL;
1458 #endif
1459 again:
1460 	error = 0;
1461 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1462 	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1463 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1464 	nfsm_fhtom(dvp, v3);
1465 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1466 	if (v3) {
1467 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1468 		if (fmode & O_EXCL) {
1469 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1470 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1471 #ifdef INET
1472 			if (in_ifaddr.tqh_first)
1473 				*tl++ = in_ifaddr.tqh_first->ia_addr.sin_addr.s_addr;
1474 			else
1475 				*tl++ = create_verf;
1476 #else
1477 			*tl++ = create_verf;
1478 #endif
1479 			*tl = ++create_verf;
1480 		} else {
1481 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1482 			nfsm_v3attrbuild(vap, FALSE);
1483 		}
1484 	} else {
1485 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1486 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1487 		sp->sa_uid = nfs_xdrneg1;
1488 		sp->sa_gid = nfs_xdrneg1;
1489 		sp->sa_size = 0;
1490 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1491 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1492 	}
1493 	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1494 	if (!error) {
1495 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1496 		if (!gotvp) {
1497 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1498 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1499 			if (!error)
1500 				newvp = NFSTOV(np);
1501 		}
1502 	}
1503 	if (v3)
1504 		nfsm_wcc_data(dvp, wccflag);
1505 	nfsm_reqdone;
1506 	if (error) {
1507 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1508 			fmode &= ~O_EXCL;
1509 			goto again;
1510 		}
1511 		if (newvp)
1512 			vput(newvp);
1513 	} else if (v3 && (fmode & O_EXCL))
1514 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1515 	if (!error) {
1516 		if (cnp->cn_flags & MAKEENTRY)
1517 			cache_enter(dvp, newvp, cnp);
1518 		*ap->a_vpp = newvp;
1519 	}
1520 	PNBUF_PUT(cnp->cn_pnbuf);
1521 	VTONFS(dvp)->n_flag |= NMODIFIED;
1522 	if (!wccflag)
1523 		VTONFS(dvp)->n_attrstamp = 0;
1524 	vput(dvp);
1525 	return (error);
1526 }
1527 
1528 /*
1529  * nfs file remove call
1530  * To try and make nfs semantics closer to ufs semantics, a file that has
1531  * other processes using the vnode is renamed instead of removed and then
1532  * removed later on the last close.
1533  * - If v_usecount > 1
1534  *	  If a rename is not already in the works
1535  *	     call nfs_sillyrename() to set it up
1536  *     else
1537  *	  do the remove rpc
1538  */
1539 int
1540 nfs_remove(v)
1541 	void *v;
1542 {
1543 	struct vop_remove_args /* {
1544 		struct vnodeop_desc *a_desc;
1545 		struct vnode * a_dvp;
1546 		struct vnode * a_vp;
1547 		struct componentname * a_cnp;
1548 	} */ *ap = v;
1549 	struct vnode *vp = ap->a_vp;
1550 	struct vnode *dvp = ap->a_dvp;
1551 	struct componentname *cnp = ap->a_cnp;
1552 	struct nfsnode *np = VTONFS(vp);
1553 	int error = 0;
1554 	struct vattr vattr;
1555 
1556 #ifndef DIAGNOSTIC
1557 	if ((cnp->cn_flags & HASBUF) == 0)
1558 		panic("nfs_remove: no name");
1559 	if (vp->v_usecount < 1)
1560 		panic("nfs_remove: bad v_usecount");
1561 #endif
1562 	if (vp->v_type == VDIR)
1563 		error = EPERM;
1564 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1565 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1566 	    vattr.va_nlink > 1)) {
1567 		/*
1568 		 * Purge the name cache so that the chance of a lookup for
1569 		 * the name succeeding while the remove is in progress is
1570 		 * minimized. Without node locking it can still happen, such
1571 		 * that an I/O op returns ESTALE, but since you get this if
1572 		 * another host removes the file..
1573 		 */
1574 		cache_purge(vp);
1575 		/*
1576 		 * throw away biocache buffers, mainly to avoid
1577 		 * unnecessary delayed writes later.
1578 		 */
1579 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1580 		/* Do the rpc */
1581 		if (error != EINTR)
1582 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1583 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1584 		/*
1585 		 * Kludge City: If the first reply to the remove rpc is lost..
1586 		 *   the reply to the retransmitted request will be ENOENT
1587 		 *   since the file was in fact removed
1588 		 *   Therefore, we cheat and return success.
1589 		 */
1590 		if (error == ENOENT)
1591 			error = 0;
1592 	} else if (!np->n_sillyrename)
1593 		error = nfs_sillyrename(dvp, vp, cnp);
1594 	PNBUF_PUT(cnp->cn_pnbuf);
1595 	np->n_attrstamp = 0;
1596 	if (dvp == vp)
1597 		vrele(vp);
1598 	else
1599 		vput(vp);
1600 	vput(dvp);
1601 	return (error);
1602 }
1603 
1604 /*
1605  * nfs file remove rpc called from nfs_inactive
1606  */
1607 int
1608 nfs_removeit(sp)
1609 	struct sillyrename *sp;
1610 {
1611 
1612 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1613 		(struct proc *)0));
1614 }
1615 
1616 /*
1617  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1618  */
1619 int
1620 nfs_removerpc(dvp, name, namelen, cred, proc)
1621 	struct vnode *dvp;
1622 	const char *name;
1623 	int namelen;
1624 	struct ucred *cred;
1625 	struct proc *proc;
1626 {
1627 	u_int32_t *tl;
1628 	caddr_t cp;
1629 	int32_t t1, t2;
1630 	caddr_t bpos, dpos, cp2;
1631 	int error = 0, wccflag = NFSV3_WCCRATTR;
1632 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1633 	const int v3 = NFS_ISV3(dvp);
1634 
1635 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1636 	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1637 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1638 	nfsm_fhtom(dvp, v3);
1639 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1640 	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1641 	if (v3)
1642 		nfsm_wcc_data(dvp, wccflag);
1643 	nfsm_reqdone;
1644 	VTONFS(dvp)->n_flag |= NMODIFIED;
1645 	if (!wccflag)
1646 		VTONFS(dvp)->n_attrstamp = 0;
1647 	return (error);
1648 }
1649 
1650 /*
1651  * nfs file rename call
1652  */
1653 int
1654 nfs_rename(v)
1655 	void *v;
1656 {
1657 	struct vop_rename_args  /* {
1658 		struct vnode *a_fdvp;
1659 		struct vnode *a_fvp;
1660 		struct componentname *a_fcnp;
1661 		struct vnode *a_tdvp;
1662 		struct vnode *a_tvp;
1663 		struct componentname *a_tcnp;
1664 	} */ *ap = v;
1665 	struct vnode *fvp = ap->a_fvp;
1666 	struct vnode *tvp = ap->a_tvp;
1667 	struct vnode *fdvp = ap->a_fdvp;
1668 	struct vnode *tdvp = ap->a_tdvp;
1669 	struct componentname *tcnp = ap->a_tcnp;
1670 	struct componentname *fcnp = ap->a_fcnp;
1671 	int error;
1672 
1673 #ifndef DIAGNOSTIC
1674 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1675 	    (fcnp->cn_flags & HASBUF) == 0)
1676 		panic("nfs_rename: no name");
1677 #endif
1678 	/* Check for cross-device rename */
1679 	if ((fvp->v_mount != tdvp->v_mount) ||
1680 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1681 		error = EXDEV;
1682 		goto out;
1683 	}
1684 
1685 	/*
1686 	 * If the tvp exists and is in use, sillyrename it before doing the
1687 	 * rename of the new file over it.
1688 	 */
1689 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1690 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1691 		vput(tvp);
1692 		tvp = NULL;
1693 	}
1694 
1695 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1696 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1697 		tcnp->cn_proc);
1698 
1699 	if (fvp->v_type == VDIR) {
1700 		if (tvp != NULL && tvp->v_type == VDIR)
1701 			cache_purge(tdvp);
1702 		cache_purge(fdvp);
1703 	}
1704 out:
1705 	if (tdvp == tvp)
1706 		vrele(tdvp);
1707 	else
1708 		vput(tdvp);
1709 	if (tvp)
1710 		vput(tvp);
1711 	vrele(fdvp);
1712 	vrele(fvp);
1713 	/*
1714 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1715 	 */
1716 	if (error == ENOENT)
1717 		error = 0;
1718 	return (error);
1719 }
1720 
1721 /*
1722  * nfs file rename rpc called from nfs_remove() above
1723  */
1724 int
1725 nfs_renameit(sdvp, scnp, sp)
1726 	struct vnode *sdvp;
1727 	struct componentname *scnp;
1728 	struct sillyrename *sp;
1729 {
1730 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1731 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1732 }
1733 
1734 /*
1735  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1736  */
1737 int
1738 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1739 	struct vnode *fdvp;
1740 	const char *fnameptr;
1741 	int fnamelen;
1742 	struct vnode *tdvp;
1743 	const char *tnameptr;
1744 	int tnamelen;
1745 	struct ucred *cred;
1746 	struct proc *proc;
1747 {
1748 	u_int32_t *tl;
1749 	caddr_t cp;
1750 	int32_t t1, t2;
1751 	caddr_t bpos, dpos, cp2;
1752 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1753 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1754 	const int v3 = NFS_ISV3(fdvp);
1755 
1756 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1757 	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1758 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1759 		nfsm_rndup(tnamelen));
1760 	nfsm_fhtom(fdvp, v3);
1761 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1762 	nfsm_fhtom(tdvp, v3);
1763 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1764 	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1765 	if (v3) {
1766 		nfsm_wcc_data(fdvp, fwccflag);
1767 		nfsm_wcc_data(tdvp, twccflag);
1768 	}
1769 	nfsm_reqdone;
1770 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1771 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1772 	if (!fwccflag)
1773 		VTONFS(fdvp)->n_attrstamp = 0;
1774 	if (!twccflag)
1775 		VTONFS(tdvp)->n_attrstamp = 0;
1776 	return (error);
1777 }
1778 
1779 /*
1780  * nfs hard link create call
1781  */
1782 int
1783 nfs_link(v)
1784 	void *v;
1785 {
1786 	struct vop_link_args /* {
1787 		struct vnode *a_dvp;
1788 		struct vnode *a_vp;
1789 		struct componentname *a_cnp;
1790 	} */ *ap = v;
1791 	struct vnode *vp = ap->a_vp;
1792 	struct vnode *dvp = ap->a_dvp;
1793 	struct componentname *cnp = ap->a_cnp;
1794 	u_int32_t *tl;
1795 	caddr_t cp;
1796 	int32_t t1, t2;
1797 	caddr_t bpos, dpos, cp2;
1798 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1799 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1800 	/* XXX Should be const and initialised? */
1801 	int v3;
1802 
1803 	if (dvp->v_mount != vp->v_mount) {
1804 		VOP_ABORTOP(dvp, cnp);
1805 		vput(dvp);
1806 		return (EXDEV);
1807 	}
1808 	if (dvp != vp) {
1809 		error = vn_lock(vp, LK_EXCLUSIVE);
1810 		if (error != 0) {
1811 			VOP_ABORTOP(dvp, cnp);
1812 			vput(dvp);
1813 			return error;
1814 		}
1815 	}
1816 
1817 	/*
1818 	 * Push all writes to the server, so that the attribute cache
1819 	 * doesn't get "out of sync" with the server.
1820 	 * XXX There should be a better way!
1821 	 */
1822 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1823 
1824 	v3 = NFS_ISV3(vp);
1825 	nfsstats.rpccnt[NFSPROC_LINK]++;
1826 	nfsm_reqhead(vp, NFSPROC_LINK,
1827 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1828 	nfsm_fhtom(vp, v3);
1829 	nfsm_fhtom(dvp, v3);
1830 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1831 	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1832 	if (v3) {
1833 		nfsm_postop_attr(vp, attrflag);
1834 		nfsm_wcc_data(dvp, wccflag);
1835 	}
1836 	nfsm_reqdone;
1837 	PNBUF_PUT(cnp->cn_pnbuf);
1838 	VTONFS(dvp)->n_flag |= NMODIFIED;
1839 	if (!attrflag)
1840 		VTONFS(vp)->n_attrstamp = 0;
1841 	if (!wccflag)
1842 		VTONFS(dvp)->n_attrstamp = 0;
1843 	if (dvp != vp)
1844 		VOP_UNLOCK(vp, 0);
1845 	vput(dvp);
1846 	/*
1847 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1848 	 */
1849 	if (error == EEXIST)
1850 		error = 0;
1851 	return (error);
1852 }
1853 
1854 /*
1855  * nfs symbolic link create call
1856  */
1857 int
1858 nfs_symlink(v)
1859 	void *v;
1860 {
1861 	struct vop_symlink_args /* {
1862 		struct vnode *a_dvp;
1863 		struct vnode **a_vpp;
1864 		struct componentname *a_cnp;
1865 		struct vattr *a_vap;
1866 		char *a_target;
1867 	} */ *ap = v;
1868 	struct vnode *dvp = ap->a_dvp;
1869 	struct vattr *vap = ap->a_vap;
1870 	struct componentname *cnp = ap->a_cnp;
1871 	struct nfsv2_sattr *sp;
1872 	u_int32_t *tl;
1873 	caddr_t cp;
1874 	int32_t t1, t2;
1875 	caddr_t bpos, dpos, cp2;
1876 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1877 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1878 	struct vnode *newvp = (struct vnode *)0;
1879 	const int v3 = NFS_ISV3(dvp);
1880 
1881 	*ap->a_vpp = NULL;
1882 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1883 	slen = strlen(ap->a_target);
1884 	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1885 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1886 	nfsm_fhtom(dvp, v3);
1887 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1888 	if (v3)
1889 		nfsm_v3attrbuild(vap, FALSE);
1890 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1891 	if (!v3) {
1892 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1893 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1894 		sp->sa_uid = nfs_xdrneg1;
1895 		sp->sa_gid = nfs_xdrneg1;
1896 		sp->sa_size = nfs_xdrneg1;
1897 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1898 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1899 	}
1900 	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1901 	if (v3) {
1902 		if (!error)
1903 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1904 		nfsm_wcc_data(dvp, wccflag);
1905 	}
1906 	nfsm_reqdone;
1907 	/*
1908 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1909 	 */
1910 	if (error == EEXIST)
1911 		error = 0;
1912 	if (error == 0 && newvp == NULL) {
1913 		struct nfsnode *np = NULL;
1914 
1915 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1916 		    cnp->cn_cred, cnp->cn_proc, &np);
1917 		if (error == 0)
1918 			newvp = NFSTOV(np);
1919 	}
1920 	if (error) {
1921 		if (newvp != NULL)
1922 			vput(newvp);
1923 	} else {
1924 		*ap->a_vpp = newvp;
1925 	}
1926 	PNBUF_PUT(cnp->cn_pnbuf);
1927 	VTONFS(dvp)->n_flag |= NMODIFIED;
1928 	if (!wccflag)
1929 		VTONFS(dvp)->n_attrstamp = 0;
1930 	vput(dvp);
1931 	return (error);
1932 }
1933 
1934 /*
1935  * nfs make dir call
1936  */
1937 int
1938 nfs_mkdir(v)
1939 	void *v;
1940 {
1941 	struct vop_mkdir_args /* {
1942 		struct vnode *a_dvp;
1943 		struct vnode **a_vpp;
1944 		struct componentname *a_cnp;
1945 		struct vattr *a_vap;
1946 	} */ *ap = v;
1947 	struct vnode *dvp = ap->a_dvp;
1948 	struct vattr *vap = ap->a_vap;
1949 	struct componentname *cnp = ap->a_cnp;
1950 	struct nfsv2_sattr *sp;
1951 	u_int32_t *tl;
1952 	caddr_t cp;
1953 	int32_t t1, t2;
1954 	int len;
1955 	struct nfsnode *np = (struct nfsnode *)0;
1956 	struct vnode *newvp = (struct vnode *)0;
1957 	caddr_t bpos, dpos, cp2;
1958 	int error = 0, wccflag = NFSV3_WCCRATTR;
1959 	int gotvp = 0;
1960 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1961 	const int v3 = NFS_ISV3(dvp);
1962 
1963 	len = cnp->cn_namelen;
1964 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1965 	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1966 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1967 	nfsm_fhtom(dvp, v3);
1968 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1969 	if (v3) {
1970 		nfsm_v3attrbuild(vap, FALSE);
1971 	} else {
1972 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1973 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1974 		sp->sa_uid = nfs_xdrneg1;
1975 		sp->sa_gid = nfs_xdrneg1;
1976 		sp->sa_size = nfs_xdrneg1;
1977 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1978 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1979 	}
1980 	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1981 	if (!error)
1982 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1983 	if (v3)
1984 		nfsm_wcc_data(dvp, wccflag);
1985 	nfsm_reqdone;
1986 	VTONFS(dvp)->n_flag |= NMODIFIED;
1987 	if (!wccflag)
1988 		VTONFS(dvp)->n_attrstamp = 0;
1989 	/*
1990 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1991 	 * if we can succeed in looking up the directory.
1992 	 */
1993 	if (error == EEXIST || (!error && !gotvp)) {
1994 		if (newvp) {
1995 			vput(newvp);
1996 			newvp = (struct vnode *)0;
1997 		}
1998 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1999 			cnp->cn_proc, &np);
2000 		if (!error) {
2001 			newvp = NFSTOV(np);
2002 			if (newvp->v_type != VDIR)
2003 				error = EEXIST;
2004 		}
2005 	}
2006 	if (error) {
2007 		if (newvp)
2008 			vput(newvp);
2009 	} else {
2010 		if (cnp->cn_flags & MAKEENTRY)
2011 			cache_enter(dvp, newvp, cnp);
2012 		*ap->a_vpp = newvp;
2013 	}
2014 	PNBUF_PUT(cnp->cn_pnbuf);
2015 	vput(dvp);
2016 	return (error);
2017 }
2018 
2019 /*
2020  * nfs remove directory call
2021  */
2022 int
2023 nfs_rmdir(v)
2024 	void *v;
2025 {
2026 	struct vop_rmdir_args /* {
2027 		struct vnode *a_dvp;
2028 		struct vnode *a_vp;
2029 		struct componentname *a_cnp;
2030 	} */ *ap = v;
2031 	struct vnode *vp = ap->a_vp;
2032 	struct vnode *dvp = ap->a_dvp;
2033 	struct componentname *cnp = ap->a_cnp;
2034 	u_int32_t *tl;
2035 	caddr_t cp;
2036 	int32_t t1, t2;
2037 	caddr_t bpos, dpos, cp2;
2038 	int error = 0, wccflag = NFSV3_WCCRATTR;
2039 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2040 	const int v3 = NFS_ISV3(dvp);
2041 
2042 	if (dvp == vp) {
2043 		vrele(dvp);
2044 		vput(dvp);
2045 		PNBUF_PUT(cnp->cn_pnbuf);
2046 		return (EINVAL);
2047 	}
2048 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2049 	nfsm_reqhead(dvp, NFSPROC_RMDIR,
2050 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2051 	nfsm_fhtom(dvp, v3);
2052 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2053 	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2054 	if (v3)
2055 		nfsm_wcc_data(dvp, wccflag);
2056 	nfsm_reqdone;
2057 	PNBUF_PUT(cnp->cn_pnbuf);
2058 	VTONFS(dvp)->n_flag |= NMODIFIED;
2059 	if (!wccflag)
2060 		VTONFS(dvp)->n_attrstamp = 0;
2061 	cache_purge(dvp);
2062 	cache_purge(vp);
2063 	vput(vp);
2064 	vput(dvp);
2065 	/*
2066 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2067 	 */
2068 	if (error == ENOENT)
2069 		error = 0;
2070 	return (error);
2071 }
2072 
2073 /*
2074  * nfs readdir call
2075  */
2076 int
2077 nfs_readdir(v)
2078 	void *v;
2079 {
2080 	struct vop_readdir_args /* {
2081 		struct vnode *a_vp;
2082 		struct uio *a_uio;
2083 		struct ucred *a_cred;
2084 		int *a_eofflag;
2085 		off_t **a_cookies;
2086 		int *a_ncookies;
2087 	} */ *ap = v;
2088 	struct vnode *vp = ap->a_vp;
2089 	struct uio *uio = ap->a_uio;
2090 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2091 	char *base = uio->uio_iov->iov_base;
2092 	int tresid, error;
2093 	size_t count, lost;
2094 	struct dirent *dp;
2095 	off_t *cookies = NULL;
2096 	int ncookies = 0, nc;
2097 
2098 	if (vp->v_type != VDIR)
2099 		return (EPERM);
2100 
2101 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2102 	count = uio->uio_resid - lost;
2103 	if (count <= 0)
2104 		return (EINVAL);
2105 
2106 	/*
2107 	 * Call nfs_bioread() to do the real work.
2108 	 */
2109 	tresid = uio->uio_resid = count;
2110 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2111 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2112 
2113 	if (!error && ap->a_cookies) {
2114 		ncookies = count / 16;
2115 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2116 		*ap->a_cookies = cookies;
2117 	}
2118 
2119 	if (!error && uio->uio_resid == tresid) {
2120 		uio->uio_resid += lost;
2121 		nfsstats.direofcache_misses++;
2122 		if (ap->a_cookies)
2123 			*ap->a_ncookies = 0;
2124 		*ap->a_eofflag = 1;
2125 		return (0);
2126 	}
2127 
2128 	if (!error && ap->a_cookies) {
2129 		/*
2130 		 * Only the NFS server and emulations use cookies, and they
2131 		 * load the directory block into system space, so we can
2132 		 * just look at it directly.
2133 		 */
2134 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2135 			panic("nfs_readdir: lost in space");
2136 		for (nc = 0; ncookies-- &&
2137 		     base < (char *)uio->uio_iov->iov_base; nc++){
2138 			dp = (struct dirent *) base;
2139 			if (dp->d_reclen == 0)
2140 				break;
2141 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2142 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2143 			else
2144 				*(cookies++) = NFS_GETCOOKIE(dp);
2145 			base += dp->d_reclen;
2146 		}
2147 		uio->uio_resid +=
2148 		    ((caddr_t)uio->uio_iov->iov_base - base);
2149 		uio->uio_iov->iov_len +=
2150 		    ((caddr_t)uio->uio_iov->iov_base - base);
2151 		uio->uio_iov->iov_base = base;
2152 		*ap->a_ncookies = nc;
2153 	}
2154 
2155 	uio->uio_resid += lost;
2156 	*ap->a_eofflag = 0;
2157 	return (error);
2158 }
2159 
2160 /*
2161  * Readdir rpc call.
2162  * Called from below the buffer cache by nfs_doio().
2163  */
2164 int
2165 nfs_readdirrpc(vp, uiop, cred)
2166 	struct vnode *vp;
2167 	struct uio *uiop;
2168 	struct ucred *cred;
2169 {
2170 	int len, left;
2171 	struct dirent *dp = NULL;
2172 	u_int32_t *tl;
2173 	caddr_t cp;
2174 	int32_t t1, t2;
2175 	caddr_t bpos, dpos, cp2;
2176 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2177 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2178 	struct nfsnode *dnp = VTONFS(vp);
2179 	u_quad_t fileno;
2180 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2181 	int attrflag, nrpcs = 0, reclen;
2182 	const int v3 = NFS_ISV3(vp);
2183 
2184 #ifdef DIAGNOSTIC
2185 	/*
2186 	 * Should be called from buffer cache, so only amount of
2187 	 * NFS_DIRBLKSIZ will be requested.
2188 	 */
2189 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2190 		panic("nfs readdirrpc bad uio");
2191 #endif
2192 
2193 	/*
2194 	 * Loop around doing readdir rpc's of size nm_readdirsize
2195 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2196 	 * The stopping criteria is EOF or buffer full.
2197 	 */
2198 	while (more_dirs && bigenough) {
2199 		/*
2200 		 * Heuristic: don't bother to do another RPC to further
2201 		 * fill up this block if there is not much room left. (< 50%
2202 		 * of the readdir RPC size). This wastes some buffer space
2203 		 * but can save up to 50% in RPC calls.
2204 		 */
2205 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2206 			bigenough = 0;
2207 			break;
2208 		}
2209 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2210 		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2211 			NFSX_READDIR(v3));
2212 		nfsm_fhtom(vp, v3);
2213 		if (v3) {
2214 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2215 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2216 				txdr_swapcookie3(uiop->uio_offset, tl);
2217 			} else {
2218 				txdr_cookie3(uiop->uio_offset, tl);
2219 			}
2220 			tl += 2;
2221 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2222 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2223 		} else {
2224 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2225 			*tl++ = txdr_unsigned(uiop->uio_offset);
2226 		}
2227 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2228 		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2229 		nrpcs++;
2230 		if (v3) {
2231 			nfsm_postop_attr(vp, attrflag);
2232 			if (!error) {
2233 				nfsm_dissect(tl, u_int32_t *,
2234 				    2 * NFSX_UNSIGNED);
2235 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2236 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2237 			} else {
2238 				m_freem(mrep);
2239 				goto nfsmout;
2240 			}
2241 		}
2242 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2243 		more_dirs = fxdr_unsigned(int, *tl);
2244 
2245 		/* loop thru the dir entries, doctoring them to 4bsd form */
2246 		while (more_dirs && bigenough) {
2247 			if (v3) {
2248 				nfsm_dissect(tl, u_int32_t *,
2249 				    3 * NFSX_UNSIGNED);
2250 				fileno = fxdr_hyper(tl);
2251 				len = fxdr_unsigned(int, *(tl + 2));
2252 			} else {
2253 				nfsm_dissect(tl, u_int32_t *,
2254 				    2 * NFSX_UNSIGNED);
2255 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2256 				len = fxdr_unsigned(int, *tl);
2257 			}
2258 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2259 				error = EBADRPC;
2260 				m_freem(mrep);
2261 				goto nfsmout;
2262 			}
2263 			tlen = nfsm_rndup(len);
2264 			if (tlen == len)
2265 				tlen += 4;	/* To ensure null termination */
2266 			tlen += sizeof (off_t) + sizeof (int);
2267 			reclen = ALIGN(tlen + DIRHDSIZ);
2268 			tlen = reclen - DIRHDSIZ;
2269 			left = NFS_DIRFRAGSIZ - blksiz;
2270 			if (reclen > left) {
2271 				dp->d_reclen += left;
2272 				uiop->uio_iov->iov_base =
2273 				    (caddr_t)uiop->uio_iov->iov_base + left;
2274 				uiop->uio_iov->iov_len -= left;
2275 				uiop->uio_resid -= left;
2276 				blksiz = 0;
2277 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2278 			}
2279 			if (reclen > uiop->uio_resid)
2280 				bigenough = 0;
2281 			if (bigenough) {
2282 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2283 				dp->d_fileno = (int)fileno;
2284 				dp->d_namlen = len;
2285 				dp->d_reclen = reclen;
2286 				dp->d_type = DT_UNKNOWN;
2287 				blksiz += dp->d_reclen;
2288 				if (blksiz == NFS_DIRFRAGSIZ)
2289 					blksiz = 0;
2290 				uiop->uio_resid -= DIRHDSIZ;
2291 				uiop->uio_iov->iov_base =
2292 				    (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2293 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2294 				nfsm_mtouio(uiop, len);
2295 				cp = uiop->uio_iov->iov_base;
2296 				tlen -= len;
2297 				*cp = '\0';	/* null terminate */
2298 				uiop->uio_iov->iov_base =
2299 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2300 				uiop->uio_iov->iov_len -= tlen;
2301 				uiop->uio_resid -= tlen;
2302 			} else
2303 				nfsm_adv(nfsm_rndup(len));
2304 			if (v3) {
2305 				nfsm_dissect(tl, u_int32_t *,
2306 				    3 * NFSX_UNSIGNED);
2307 			} else {
2308 				nfsm_dissect(tl, u_int32_t *,
2309 				    2 * NFSX_UNSIGNED);
2310 			}
2311 			if (bigenough) {
2312 				if (v3) {
2313 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2314 						uiop->uio_offset =
2315 						    fxdr_swapcookie3(tl);
2316 					else
2317 						uiop->uio_offset =
2318 						    fxdr_cookie3(tl);
2319 				}
2320 				else {
2321 					uiop->uio_offset =
2322 					    fxdr_unsigned(off_t, *tl);
2323 				}
2324 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2325 			}
2326 			if (v3)
2327 				tl += 2;
2328 			else
2329 				tl++;
2330 			more_dirs = fxdr_unsigned(int, *tl);
2331 		}
2332 		/*
2333 		 * If at end of rpc data, get the eof boolean
2334 		 */
2335 		if (!more_dirs) {
2336 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2337 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2338 		}
2339 		m_freem(mrep);
2340 	}
2341 	/*
2342 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2343 	 * by increasing d_reclen for the last record.
2344 	 */
2345 	if (blksiz > 0) {
2346 		left = NFS_DIRFRAGSIZ - blksiz;
2347 		dp->d_reclen += left;
2348 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2349 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2350 		    left;
2351 		uiop->uio_iov->iov_len -= left;
2352 		uiop->uio_resid -= left;
2353 	}
2354 
2355 	/*
2356 	 * We are now either at the end of the directory or have filled the
2357 	 * block.
2358 	 */
2359 	if (bigenough)
2360 		dnp->n_direofoffset = uiop->uio_offset;
2361 nfsmout:
2362 	return (error);
2363 }
2364 
2365 /*
2366  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2367  */
2368 int
2369 nfs_readdirplusrpc(vp, uiop, cred)
2370 	struct vnode *vp;
2371 	struct uio *uiop;
2372 	struct ucred *cred;
2373 {
2374 	int len, left;
2375 	struct dirent *dp = NULL;
2376 	u_int32_t *tl;
2377 	caddr_t cp;
2378 	int32_t t1, t2;
2379 	struct vnode *newvp;
2380 	caddr_t bpos, dpos, cp2;
2381 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2382 	struct nameidata nami, *ndp = &nami;
2383 	struct componentname *cnp = &ndp->ni_cnd;
2384 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2385 	struct nfsnode *dnp = VTONFS(vp), *np;
2386 	nfsfh_t *fhp;
2387 	u_quad_t fileno;
2388 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2389 	int attrflag, fhsize, nrpcs = 0, reclen;
2390 	struct nfs_fattr fattr, *fp;
2391 
2392 #ifdef DIAGNOSTIC
2393 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2394 		panic("nfs readdirplusrpc bad uio");
2395 #endif
2396 	ndp->ni_dvp = vp;
2397 	newvp = NULLVP;
2398 
2399 	/*
2400 	 * Loop around doing readdir rpc's of size nm_readdirsize
2401 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2402 	 * The stopping criteria is EOF or buffer full.
2403 	 */
2404 	while (more_dirs && bigenough) {
2405 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2406 			bigenough = 0;
2407 			break;
2408 		}
2409 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2410 		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2411 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2412 		nfsm_fhtom(vp, 1);
2413  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2414 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2415 			txdr_swapcookie3(uiop->uio_offset, tl);
2416 		} else {
2417 			txdr_cookie3(uiop->uio_offset, tl);
2418 		}
2419 		tl += 2;
2420 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2421 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2422 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2423 		*tl = txdr_unsigned(nmp->nm_rsize);
2424 		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2425 		nfsm_postop_attr(vp, attrflag);
2426 		if (error) {
2427 			m_freem(mrep);
2428 			goto nfsmout;
2429 		}
2430 		nrpcs++;
2431 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2432 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2433 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2434 		more_dirs = fxdr_unsigned(int, *tl);
2435 
2436 		/* loop thru the dir entries, doctoring them to 4bsd form */
2437 		while (more_dirs && bigenough) {
2438 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2439 			fileno = fxdr_hyper(tl);
2440 			len = fxdr_unsigned(int, *(tl + 2));
2441 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2442 				error = EBADRPC;
2443 				m_freem(mrep);
2444 				goto nfsmout;
2445 			}
2446 			tlen = nfsm_rndup(len);
2447 			if (tlen == len)
2448 				tlen += 4;	/* To ensure null termination*/
2449 			tlen += sizeof (off_t) + sizeof (int);
2450 			reclen = ALIGN(tlen + DIRHDSIZ);
2451 			tlen = reclen - DIRHDSIZ;
2452 			left = NFS_DIRFRAGSIZ - blksiz;
2453 			if (reclen > left) {
2454 				/*
2455 				 * DIRFRAGSIZ is aligned, no need to align
2456 				 * again here.
2457 				 */
2458 				dp->d_reclen += left;
2459 				uiop->uio_iov->iov_base =
2460 				    (caddr_t)uiop->uio_iov->iov_base + left;
2461 				uiop->uio_iov->iov_len -= left;
2462 				uiop->uio_resid -= left;
2463 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2464 				blksiz = 0;
2465 			}
2466 			if (reclen > uiop->uio_resid)
2467 				bigenough = 0;
2468 			if (bigenough) {
2469 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2470 				dp->d_fileno = (int)fileno;
2471 				dp->d_namlen = len;
2472 				dp->d_reclen = reclen;
2473 				dp->d_type = DT_UNKNOWN;
2474 				blksiz += dp->d_reclen;
2475 				if (blksiz == NFS_DIRFRAGSIZ)
2476 					blksiz = 0;
2477 				uiop->uio_resid -= DIRHDSIZ;
2478 				uiop->uio_iov->iov_base =
2479 				    (caddr_t)uiop->uio_iov->iov_base +
2480 				    DIRHDSIZ;
2481 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2482 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2483 				cnp->cn_namelen = len;
2484 				nfsm_mtouio(uiop, len);
2485 				cp = uiop->uio_iov->iov_base;
2486 				tlen -= len;
2487 				*cp = '\0';
2488 				uiop->uio_iov->iov_base =
2489 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2490 				uiop->uio_iov->iov_len -= tlen;
2491 				uiop->uio_resid -= tlen;
2492 			} else
2493 				nfsm_adv(nfsm_rndup(len));
2494 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2495 			if (bigenough) {
2496 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2497 					uiop->uio_offset =
2498 						fxdr_swapcookie3(tl);
2499 				else
2500 					uiop->uio_offset =
2501 						fxdr_cookie3(tl);
2502 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2503 			}
2504 			tl += 2;
2505 
2506 			/*
2507 			 * Since the attributes are before the file handle
2508 			 * (sigh), we must skip over the attributes and then
2509 			 * come back and get them.
2510 			 */
2511 			attrflag = fxdr_unsigned(int, *tl);
2512 			if (attrflag) {
2513 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2514 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2515 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2516 			    doit = fxdr_unsigned(int, *tl);
2517 			    if (doit) {
2518 				nfsm_getfh(fhp, fhsize, 1);
2519 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2520 				    VREF(vp);
2521 				    newvp = vp;
2522 				    np = dnp;
2523 				} else {
2524 				    error = nfs_nget(vp->v_mount, fhp,
2525 					fhsize, &np);
2526 				    if (!error)
2527 					newvp = NFSTOV(np);
2528 				}
2529 				if (!error) {
2530 				    const char *cp;
2531 
2532 				    nfs_loadattrcache(&newvp, &fattr, 0);
2533 				    dp->d_type =
2534 				        IFTODT(VTTOIF(np->n_vattr->va_type));
2535 				    ndp->ni_vp = newvp;
2536 				    cp = cnp->cn_nameptr + cnp->cn_namelen;
2537 				    cnp->cn_hash =
2538 					namei_hash(cnp->cn_nameptr, &cp);
2539 				    if (cnp->cn_namelen <= NCHNAMLEN)
2540 				        cache_enter(ndp->ni_dvp, ndp->ni_vp,
2541 						    cnp);
2542 				}
2543 			   }
2544 			} else {
2545 			    /* Just skip over the file handle */
2546 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2547 			    i = fxdr_unsigned(int, *tl);
2548 			    nfsm_adv(nfsm_rndup(i));
2549 			}
2550 			if (newvp != NULLVP) {
2551 			    if (newvp == vp)
2552 				vrele(newvp);
2553 			    else
2554 				vput(newvp);
2555 			    newvp = NULLVP;
2556 			}
2557 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2558 			more_dirs = fxdr_unsigned(int, *tl);
2559 		}
2560 		/*
2561 		 * If at end of rpc data, get the eof boolean
2562 		 */
2563 		if (!more_dirs) {
2564 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2565 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2566 		}
2567 		m_freem(mrep);
2568 	}
2569 	/*
2570 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2571 	 * by increasing d_reclen for the last record.
2572 	 */
2573 	if (blksiz > 0) {
2574 		left = NFS_DIRFRAGSIZ - blksiz;
2575 		dp->d_reclen += left;
2576 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2577 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2578 		    left;
2579 		uiop->uio_iov->iov_len -= left;
2580 		uiop->uio_resid -= left;
2581 	}
2582 
2583 	/*
2584 	 * We are now either at the end of the directory or have filled the
2585 	 * block.
2586 	 */
2587 	if (bigenough)
2588 		dnp->n_direofoffset = uiop->uio_offset;
2589 nfsmout:
2590 	if (newvp != NULLVP) {
2591 		if(newvp == vp)
2592 		    vrele(newvp);
2593 		else
2594 		    vput(newvp);
2595 	}
2596 	return (error);
2597 }
2598 static char hextoasc[] = "0123456789abcdef";
2599 
2600 /*
2601  * Silly rename. To make the NFS filesystem that is stateless look a little
2602  * more like the "ufs" a remove of an active vnode is translated to a rename
2603  * to a funny looking filename that is removed by nfs_inactive on the
2604  * nfsnode. There is the potential for another process on a different client
2605  * to create the same funny name between the nfs_lookitup() fails and the
2606  * nfs_rename() completes, but...
2607  */
2608 int
2609 nfs_sillyrename(dvp, vp, cnp)
2610 	struct vnode *dvp, *vp;
2611 	struct componentname *cnp;
2612 {
2613 	struct sillyrename *sp;
2614 	struct nfsnode *np;
2615 	int error;
2616 	short pid;
2617 
2618 	cache_purge(dvp);
2619 	np = VTONFS(vp);
2620 #ifndef DIAGNOSTIC
2621 	if (vp->v_type == VDIR)
2622 		panic("nfs: sillyrename dir");
2623 #endif
2624 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2625 		M_NFSREQ, M_WAITOK);
2626 	sp->s_cred = crdup(cnp->cn_cred);
2627 	sp->s_dvp = dvp;
2628 	VREF(dvp);
2629 
2630 	/* Fudge together a funny name */
2631 	pid = cnp->cn_proc->p_pid;
2632 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2633 	sp->s_namlen = 12;
2634 	sp->s_name[8] = hextoasc[pid & 0xf];
2635 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2636 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2637 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2638 
2639 	/* Try lookitups until we get one that isn't there */
2640 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2641 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2642 		sp->s_name[4]++;
2643 		if (sp->s_name[4] > 'z') {
2644 			error = EINVAL;
2645 			goto bad;
2646 		}
2647 	}
2648 	error = nfs_renameit(dvp, cnp, sp);
2649 	if (error)
2650 		goto bad;
2651 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2652 		cnp->cn_proc, &np);
2653 	np->n_sillyrename = sp;
2654 	return (0);
2655 bad:
2656 	vrele(sp->s_dvp);
2657 	crfree(sp->s_cred);
2658 	free((caddr_t)sp, M_NFSREQ);
2659 	return (error);
2660 }
2661 
2662 /*
2663  * Look up a file name and optionally either update the file handle or
2664  * allocate an nfsnode, depending on the value of npp.
2665  * npp == NULL	--> just do the lookup
2666  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2667  *			handled too
2668  * *npp != NULL --> update the file handle in the vnode
2669  */
2670 int
2671 nfs_lookitup(dvp, name, len, cred, procp, npp)
2672 	struct vnode *dvp;
2673 	const char *name;
2674 	int len;
2675 	struct ucred *cred;
2676 	struct proc *procp;
2677 	struct nfsnode **npp;
2678 {
2679 	u_int32_t *tl;
2680 	caddr_t cp;
2681 	int32_t t1, t2;
2682 	struct vnode *newvp = (struct vnode *)0;
2683 	struct nfsnode *np, *dnp = VTONFS(dvp);
2684 	caddr_t bpos, dpos, cp2;
2685 	int error = 0, fhlen, attrflag;
2686 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2687 	nfsfh_t *nfhp;
2688 	const int v3 = NFS_ISV3(dvp);
2689 
2690 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2691 	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2692 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2693 	nfsm_fhtom(dvp, v3);
2694 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2695 	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2696 	if (npp && !error) {
2697 		nfsm_getfh(nfhp, fhlen, v3);
2698 		if (*npp) {
2699 		    np = *npp;
2700 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2701 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2702 			np->n_fhp = &np->n_fh;
2703 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2704 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2705 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2706 		    np->n_fhsize = fhlen;
2707 		    newvp = NFSTOV(np);
2708 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2709 		    VREF(dvp);
2710 		    newvp = dvp;
2711 		} else {
2712 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2713 		    if (error) {
2714 			m_freem(mrep);
2715 			return (error);
2716 		    }
2717 		    newvp = NFSTOV(np);
2718 		}
2719 		if (v3) {
2720 			nfsm_postop_attr(newvp, attrflag);
2721 			if (!attrflag && *npp == NULL) {
2722 				m_freem(mrep);
2723 				vput(newvp);
2724 				return (ENOENT);
2725 			}
2726 		} else
2727 			nfsm_loadattr(newvp, (struct vattr *)0);
2728 	}
2729 	nfsm_reqdone;
2730 	if (npp && *npp == NULL) {
2731 		if (error) {
2732 			if (newvp)
2733 				vput(newvp);
2734 		} else
2735 			*npp = np;
2736 	}
2737 	return (error);
2738 }
2739 
2740 /*
2741  * Nfs Version 3 commit rpc
2742  */
2743 int
2744 nfs_commit(vp, offset, cnt, procp)
2745 	struct vnode *vp;
2746 	off_t offset;
2747 	uint32_t cnt;
2748 	struct proc *procp;
2749 {
2750 	caddr_t cp;
2751 	u_int32_t *tl;
2752 	int32_t t1, t2;
2753 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2754 	caddr_t bpos, dpos, cp2;
2755 	int error = 0, wccflag = NFSV3_WCCRATTR;
2756 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2757 
2758 #ifdef fvdl_debug
2759 	printf("commit %lu - %lu\n", (unsigned long)offset,
2760 	    (unsigned long)(offset + cnt));
2761 #endif
2762 
2763 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0)
2764 		return (0);
2765 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2766 	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2767 	nfsm_fhtom(vp, 1);
2768 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2769 	txdr_hyper(offset, tl);
2770 	tl += 2;
2771 	*tl = txdr_unsigned(cnt);
2772 	nfsm_request(vp, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2773 	nfsm_wcc_data(vp, wccflag);
2774 	if (!error) {
2775 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2776 		if (memcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2777 			NFSX_V3WRITEVERF)) {
2778 			memcpy((caddr_t)nmp->nm_verf, (caddr_t)tl,
2779 				NFSX_V3WRITEVERF);
2780 			error = NFSERR_STALEWRITEVERF;
2781 		}
2782 	}
2783 	nfsm_reqdone;
2784 	return (error);
2785 }
2786 
2787 /*
2788  * Kludge City..
2789  * - make nfs_bmap() essentially a no-op that does no translation
2790  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2791  *   (Maybe I could use the process's page mapping, but I was concerned that
2792  *    Kernel Write might not be enabled and also figured copyout() would do
2793  *    a lot more work than memcpy() and also it currently happens in the
2794  *    context of the swapper process (2).
2795  */
2796 int
2797 nfs_bmap(v)
2798 	void *v;
2799 {
2800 	struct vop_bmap_args /* {
2801 		struct vnode *a_vp;
2802 		daddr_t  a_bn;
2803 		struct vnode **a_vpp;
2804 		daddr_t *a_bnp;
2805 		int *a_runp;
2806 	} */ *ap = v;
2807 	struct vnode *vp = ap->a_vp;
2808 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2809 
2810 	if (ap->a_vpp != NULL)
2811 		*ap->a_vpp = vp;
2812 	if (ap->a_bnp != NULL)
2813 		*ap->a_bnp = ap->a_bn << bshift;
2814 	if (ap->a_runp != NULL)
2815 		*ap->a_runp = 1024 * 1024; /* XXX */
2816 	return (0);
2817 }
2818 
2819 /*
2820  * Strategy routine.
2821  * For async requests when nfsiod(s) are running, queue the request by
2822  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2823  * request.
2824  */
2825 int
2826 nfs_strategy(v)
2827 	void *v;
2828 {
2829 	struct vop_strategy_args *ap = v;
2830 	struct buf *bp = ap->a_bp;
2831 	struct proc *p;
2832 	int error = 0;
2833 
2834 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2835 		panic("nfs physio/async");
2836 	if (bp->b_flags & B_ASYNC)
2837 		p = NULL;
2838 	else
2839 		p = curproc;	/* XXX */
2840 
2841 	/*
2842 	 * If the op is asynchronous and an i/o daemon is waiting
2843 	 * queue the request, wake it up and wait for completion
2844 	 * otherwise just do it ourselves.
2845 	 */
2846 
2847 	if ((bp->b_flags & B_ASYNC) == 0 ||
2848 	    nfs_asyncio(bp))
2849 		error = nfs_doio(bp, p);
2850 	return (error);
2851 }
2852 
2853 /*
2854  * fsync vnode op. Just call nfs_flush() with commit == 1.
2855  */
2856 /* ARGSUSED */
2857 int
2858 nfs_fsync(v)
2859 	void *v;
2860 {
2861 	struct vop_fsync_args /* {
2862 		struct vnodeop_desc *a_desc;
2863 		struct vnode * a_vp;
2864 		struct ucred * a_cred;
2865 		int  a_flags;
2866 		off_t offlo;
2867 		off_t offhi;
2868 		struct proc * a_p;
2869 	} */ *ap = v;
2870 
2871 	return (nfs_flush(ap->a_vp, ap->a_cred,
2872 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
2873 }
2874 
2875 /*
2876  * Flush all the data associated with a vnode.
2877  */
2878 int
2879 nfs_flush(vp, cred, waitfor, p, commit)
2880 	struct vnode *vp;
2881 	struct ucred *cred;
2882 	int waitfor;
2883 	struct proc *p;
2884 	int commit;
2885 {
2886 	struct nfsnode *np = VTONFS(vp);
2887 	int error;
2888 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
2889 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
2890 
2891 	simple_lock(&vp->v_interlock);
2892 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
2893 	if (np->n_flag & NWRITEERR) {
2894 		error = np->n_error;
2895 		np->n_flag &= ~NWRITEERR;
2896 	}
2897 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
2898 	return (error);
2899 }
2900 
2901 /*
2902  * Return POSIX pathconf information applicable to nfs.
2903  *
2904  * N.B. The NFS V2 protocol doesn't support this RPC.
2905  */
2906 /* ARGSUSED */
2907 int
2908 nfs_pathconf(v)
2909 	void *v;
2910 {
2911 	struct vop_pathconf_args /* {
2912 		struct vnode *a_vp;
2913 		int a_name;
2914 		register_t *a_retval;
2915 	} */ *ap = v;
2916 	struct nfsv3_pathconf *pcp;
2917 	struct vnode *vp = ap->a_vp;
2918 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2919 	int32_t t1, t2;
2920 	u_int32_t *tl;
2921 	caddr_t bpos, dpos, cp, cp2;
2922 	int error = 0, attrflag;
2923 #ifndef NFS_V2_ONLY
2924 	struct nfsmount *nmp;
2925 	unsigned int l;
2926 	u_int64_t maxsize;
2927 #endif
2928 	const int v3 = NFS_ISV3(vp);
2929 
2930 	switch (ap->a_name) {
2931 		/* Names that can be resolved locally. */
2932 	case _PC_PIPE_BUF:
2933 		*ap->a_retval = PIPE_BUF;
2934 		break;
2935 	case _PC_SYNC_IO:
2936 		*ap->a_retval = 1;
2937 		break;
2938 	/* Names that cannot be resolved locally; do an RPC, if possible. */
2939 	case _PC_LINK_MAX:
2940 	case _PC_NAME_MAX:
2941 	case _PC_CHOWN_RESTRICTED:
2942 	case _PC_NO_TRUNC:
2943 		if (!v3) {
2944 			error = EINVAL;
2945 			break;
2946 		}
2947 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
2948 		nfsm_reqhead(vp, NFSPROC_PATHCONF, NFSX_FH(1));
2949 		nfsm_fhtom(vp, 1);
2950 		nfsm_request(vp, NFSPROC_PATHCONF,
2951 		    curproc, curproc->p_ucred);	/* XXX */
2952 		nfsm_postop_attr(vp, attrflag);
2953 		if (!error) {
2954 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
2955 			    NFSX_V3PATHCONF);
2956 			switch (ap->a_name) {
2957 			case _PC_LINK_MAX:
2958 				*ap->a_retval =
2959 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
2960 				break;
2961 			case _PC_NAME_MAX:
2962 				*ap->a_retval =
2963 				    fxdr_unsigned(register_t, pcp->pc_namemax);
2964 				break;
2965 			case _PC_CHOWN_RESTRICTED:
2966 				*ap->a_retval =
2967 				    (pcp->pc_chownrestricted == nfs_true);
2968 				break;
2969 			case _PC_NO_TRUNC:
2970 				*ap->a_retval =
2971 				    (pcp->pc_notrunc == nfs_true);
2972 				break;
2973 			}
2974 		}
2975 		nfsm_reqdone;
2976 		break;
2977 	case _PC_FILESIZEBITS:
2978 #ifndef NFS_V2_ONLY
2979 		if (v3) {
2980 			nmp = VFSTONFS(vp->v_mount);
2981 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
2982 				if ((error = nfs_fsinfo(nmp, vp,
2983 				    curproc->p_ucred, curproc)) != 0) /* XXX */
2984 					break;
2985 			for (l = 0, maxsize = nmp->nm_maxfilesize;
2986 			    (maxsize >> l) > 0; l++)
2987 				;
2988 			*ap->a_retval = l + 1;
2989 		} else
2990 #endif
2991 		{
2992 			*ap->a_retval = 32;	/* NFS V2 limitation */
2993 		}
2994 		break;
2995 	default:
2996 		error = EINVAL;
2997 		break;
2998 	}
2999 
3000 	return (error);
3001 }
3002 
3003 /*
3004  * NFS advisory byte-level locks.
3005  */
3006 int
3007 nfs_advlock(v)
3008 	void *v;
3009 {
3010 	struct vop_advlock_args /* {
3011 		struct vnode *a_vp;
3012 		caddr_t  a_id;
3013 		int  a_op;
3014 		struct flock *a_fl;
3015 		int  a_flags;
3016 	} */ *ap = v;
3017 	struct nfsnode *np = VTONFS(ap->a_vp);
3018 
3019 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3020 }
3021 
3022 /*
3023  * Print out the contents of an nfsnode.
3024  */
3025 int
3026 nfs_print(v)
3027 	void *v;
3028 {
3029 	struct vop_print_args /* {
3030 		struct vnode *a_vp;
3031 	} */ *ap = v;
3032 	struct vnode *vp = ap->a_vp;
3033 	struct nfsnode *np = VTONFS(vp);
3034 
3035 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3036 	    np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3037 	if (vp->v_type == VFIFO)
3038 		fifo_printinfo(vp);
3039 	printf("\n");
3040 	return (0);
3041 }
3042 
3043 /*
3044  * NFS file truncation.
3045  */
3046 int
3047 nfs_truncate(v)
3048 	void *v;
3049 {
3050 #if 0
3051 	struct vop_truncate_args /* {
3052 		struct vnode *a_vp;
3053 		off_t a_length;
3054 		int a_flags;
3055 		struct ucred *a_cred;
3056 		struct proc *a_p;
3057 	} */ *ap = v;
3058 #endif
3059 
3060 	/* Use nfs_setattr */
3061 	return (EOPNOTSUPP);
3062 }
3063 
3064 /*
3065  * NFS update.
3066  */
3067 int
3068 nfs_update(v)
3069 	void *v;
3070 #if 0
3071 	struct vop_update_args /* {
3072 		struct vnode *a_vp;
3073 		struct timespec *a_ta;
3074 		struct timespec *a_tm;
3075 		int a_waitfor;
3076 	} */ *ap = v;
3077 #endif
3078 {
3079 
3080 	/* Use nfs_setattr */
3081 	return (EOPNOTSUPP);
3082 }
3083 
3084 /*
3085  * Just call bwrite().
3086  */
3087 int
3088 nfs_bwrite(v)
3089 	void *v;
3090 {
3091 	struct vop_bwrite_args /* {
3092 		struct vnode *a_bp;
3093 	} */ *ap = v;
3094 
3095 	return (bwrite(ap->a_bp));
3096 }
3097 
3098 /*
3099  * nfs special file access vnode op.
3100  * Essentially just get vattr and then imitate iaccess() since the device is
3101  * local to the client.
3102  */
3103 int
3104 nfsspec_access(v)
3105 	void *v;
3106 {
3107 	struct vop_access_args /* {
3108 		struct vnode *a_vp;
3109 		int  a_mode;
3110 		struct ucred *a_cred;
3111 		struct proc *a_p;
3112 	} */ *ap = v;
3113 	struct vattr va;
3114 	struct vnode *vp = ap->a_vp;
3115 	int error;
3116 
3117 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3118 	if (error)
3119 		return (error);
3120 
3121         /*
3122 	 * Disallow write attempts on filesystems mounted read-only;
3123 	 * unless the file is a socket, fifo, or a block or character
3124 	 * device resident on the filesystem.
3125 	 */
3126 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3127 		switch (vp->v_type) {
3128 		case VREG:
3129 		case VDIR:
3130 		case VLNK:
3131 			return (EROFS);
3132 		default:
3133 			break;
3134 		}
3135 	}
3136 
3137 	return (vaccess(va.va_type, va.va_mode,
3138 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3139 }
3140 
3141 /*
3142  * Read wrapper for special devices.
3143  */
3144 int
3145 nfsspec_read(v)
3146 	void *v;
3147 {
3148 	struct vop_read_args /* {
3149 		struct vnode *a_vp;
3150 		struct uio *a_uio;
3151 		int  a_ioflag;
3152 		struct ucred *a_cred;
3153 	} */ *ap = v;
3154 	struct nfsnode *np = VTONFS(ap->a_vp);
3155 
3156 	/*
3157 	 * Set access flag.
3158 	 */
3159 	np->n_flag |= NACC;
3160 	np->n_atim.tv_sec = time.tv_sec;
3161 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3162 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3163 }
3164 
3165 /*
3166  * Write wrapper for special devices.
3167  */
3168 int
3169 nfsspec_write(v)
3170 	void *v;
3171 {
3172 	struct vop_write_args /* {
3173 		struct vnode *a_vp;
3174 		struct uio *a_uio;
3175 		int  a_ioflag;
3176 		struct ucred *a_cred;
3177 	} */ *ap = v;
3178 	struct nfsnode *np = VTONFS(ap->a_vp);
3179 
3180 	/*
3181 	 * Set update flag.
3182 	 */
3183 	np->n_flag |= NUPD;
3184 	np->n_mtim.tv_sec = time.tv_sec;
3185 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3186 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3187 }
3188 
3189 /*
3190  * Close wrapper for special devices.
3191  *
3192  * Update the times on the nfsnode then do device close.
3193  */
3194 int
3195 nfsspec_close(v)
3196 	void *v;
3197 {
3198 	struct vop_close_args /* {
3199 		struct vnode *a_vp;
3200 		int  a_fflag;
3201 		struct ucred *a_cred;
3202 		struct proc *a_p;
3203 	} */ *ap = v;
3204 	struct vnode *vp = ap->a_vp;
3205 	struct nfsnode *np = VTONFS(vp);
3206 	struct vattr vattr;
3207 
3208 	if (np->n_flag & (NACC | NUPD)) {
3209 		np->n_flag |= NCHG;
3210 		if (vp->v_usecount == 1 &&
3211 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3212 			VATTR_NULL(&vattr);
3213 			if (np->n_flag & NACC)
3214 				vattr.va_atime = np->n_atim;
3215 			if (np->n_flag & NUPD)
3216 				vattr.va_mtime = np->n_mtim;
3217 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3218 		}
3219 	}
3220 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3221 }
3222 
3223 /*
3224  * Read wrapper for fifos.
3225  */
3226 int
3227 nfsfifo_read(v)
3228 	void *v;
3229 {
3230 	struct vop_read_args /* {
3231 		struct vnode *a_vp;
3232 		struct uio *a_uio;
3233 		int  a_ioflag;
3234 		struct ucred *a_cred;
3235 	} */ *ap = v;
3236 	struct nfsnode *np = VTONFS(ap->a_vp);
3237 
3238 	/*
3239 	 * Set access flag.
3240 	 */
3241 	np->n_flag |= NACC;
3242 	np->n_atim.tv_sec = time.tv_sec;
3243 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3244 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3245 }
3246 
3247 /*
3248  * Write wrapper for fifos.
3249  */
3250 int
3251 nfsfifo_write(v)
3252 	void *v;
3253 {
3254 	struct vop_write_args /* {
3255 		struct vnode *a_vp;
3256 		struct uio *a_uio;
3257 		int  a_ioflag;
3258 		struct ucred *a_cred;
3259 	} */ *ap = v;
3260 	struct nfsnode *np = VTONFS(ap->a_vp);
3261 
3262 	/*
3263 	 * Set update flag.
3264 	 */
3265 	np->n_flag |= NUPD;
3266 	np->n_mtim.tv_sec = time.tv_sec;
3267 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3268 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3269 }
3270 
3271 /*
3272  * Close wrapper for fifos.
3273  *
3274  * Update the times on the nfsnode then do fifo close.
3275  */
3276 int
3277 nfsfifo_close(v)
3278 	void *v;
3279 {
3280 	struct vop_close_args /* {
3281 		struct vnode *a_vp;
3282 		int  a_fflag;
3283 		struct ucred *a_cred;
3284 		struct proc *a_p;
3285 	} */ *ap = v;
3286 	struct vnode *vp = ap->a_vp;
3287 	struct nfsnode *np = VTONFS(vp);
3288 	struct vattr vattr;
3289 
3290 	if (np->n_flag & (NACC | NUPD)) {
3291 		if (np->n_flag & NACC) {
3292 			np->n_atim.tv_sec = time.tv_sec;
3293 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3294 		}
3295 		if (np->n_flag & NUPD) {
3296 			np->n_mtim.tv_sec = time.tv_sec;
3297 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3298 		}
3299 		np->n_flag |= NCHG;
3300 		if (vp->v_usecount == 1 &&
3301 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3302 			VATTR_NULL(&vattr);
3303 			if (np->n_flag & NACC)
3304 				vattr.va_atime = np->n_atim;
3305 			if (np->n_flag & NUPD)
3306 				vattr.va_mtime = np->n_mtim;
3307 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3308 		}
3309 	}
3310 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3311 }
3312