xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: nfs_vnops.c,v 1.166 2003/05/03 16:28:59 yamt 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 1.166 2003/05/03 16:28:59 yamt 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_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
116 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
117 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
118 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
119 	{ &vop_seek_desc, nfs_seek },			/* seek */
120 	{ &vop_remove_desc, nfs_remove },		/* remove */
121 	{ &vop_link_desc, nfs_link },			/* link */
122 	{ &vop_rename_desc, nfs_rename },		/* rename */
123 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
124 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
125 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
126 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
127 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
128 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
129 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
130 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
131 	{ &vop_lock_desc, nfs_lock },			/* lock */
132 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
133 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
134 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
135 	{ &vop_print_desc, nfs_print },			/* print */
136 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
137 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
138 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
139 	{ &vop_blkatoff_desc, nfs_blkatoff },		/* blkatoff */
140 	{ &vop_valloc_desc, nfs_valloc },		/* valloc */
141 	{ &vop_reallocblks_desc, nfs_reallocblks },	/* reallocblks */
142 	{ &vop_vfree_desc, nfs_vfree },			/* vfree */
143 	{ &vop_truncate_desc, nfs_truncate },		/* truncate */
144 	{ &vop_update_desc, nfs_update },		/* update */
145 	{ &vop_bwrite_desc, nfs_bwrite },		/* bwrite */
146 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
147 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
148 	{ NULL, NULL }
149 };
150 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
151 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
152 
153 /*
154  * Special device vnode ops
155  */
156 int (**spec_nfsv2nodeop_p) __P((void *));
157 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
158 	{ &vop_default_desc, vn_default_error },
159 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
160 	{ &vop_create_desc, spec_create },		/* create */
161 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
162 	{ &vop_open_desc, spec_open },			/* open */
163 	{ &vop_close_desc, nfsspec_close },		/* close */
164 	{ &vop_access_desc, nfsspec_access },		/* access */
165 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
166 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
167 	{ &vop_read_desc, nfsspec_read },		/* read */
168 	{ &vop_write_desc, nfsspec_write },		/* write */
169 	{ &vop_lease_desc, spec_lease_check },		/* lease */
170 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
171 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
172 	{ &vop_poll_desc, spec_poll },			/* poll */
173 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
174 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
175 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
176 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
177 	{ &vop_seek_desc, spec_seek },			/* seek */
178 	{ &vop_remove_desc, spec_remove },		/* remove */
179 	{ &vop_link_desc, spec_link },			/* link */
180 	{ &vop_rename_desc, spec_rename },		/* rename */
181 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
182 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
183 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
184 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
185 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
186 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
187 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
188 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
189 	{ &vop_lock_desc, nfs_lock },			/* lock */
190 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
191 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
192 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
193 	{ &vop_print_desc, nfs_print },			/* print */
194 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
195 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
196 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
197 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
198 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
199 	{ &vop_reallocblks_desc, spec_reallocblks },	/* reallocblks */
200 	{ &vop_vfree_desc, spec_vfree },		/* vfree */
201 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
202 	{ &vop_update_desc, nfs_update },		/* update */
203 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
204 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
205 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
206 	{ NULL, NULL }
207 };
208 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
209 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
210 
211 int (**fifo_nfsv2nodeop_p) __P((void *));
212 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
213 	{ &vop_default_desc, vn_default_error },
214 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
215 	{ &vop_create_desc, fifo_create },		/* create */
216 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
217 	{ &vop_open_desc, fifo_open },			/* open */
218 	{ &vop_close_desc, nfsfifo_close },		/* close */
219 	{ &vop_access_desc, nfsspec_access },		/* access */
220 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
221 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
222 	{ &vop_read_desc, nfsfifo_read },		/* read */
223 	{ &vop_write_desc, nfsfifo_write },		/* write */
224 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
225 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
226 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
227 	{ &vop_poll_desc, fifo_poll },			/* poll */
228 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
229 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
230 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
231 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
232 	{ &vop_seek_desc, fifo_seek },			/* seek */
233 	{ &vop_remove_desc, fifo_remove },		/* remove */
234 	{ &vop_link_desc, fifo_link },			/* link */
235 	{ &vop_rename_desc, fifo_rename },		/* rename */
236 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
237 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
238 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
239 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
240 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
241 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
242 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
243 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
244 	{ &vop_lock_desc, nfs_lock },			/* lock */
245 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
246 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
247 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
248 	{ &vop_print_desc, nfs_print },			/* print */
249 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
250 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
251 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
252 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
253 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
254 	{ &vop_reallocblks_desc, fifo_reallocblks },	/* reallocblks */
255 	{ &vop_vfree_desc, fifo_vfree },		/* vfree */
256 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
257 	{ &vop_update_desc, nfs_update },		/* update */
258 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
259 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
260 	{ NULL, NULL }
261 };
262 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
263 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
264 
265 /*
266  * Global variables
267  */
268 extern u_int32_t nfs_true, nfs_false;
269 extern u_int32_t nfs_xdrneg1;
270 extern const nfstype nfsv3_type[9];
271 
272 int nfs_numasync = 0;
273 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
274 
275 /*
276  * nfs null call from vfs.
277  */
278 int
279 nfs_null(vp, cred, procp)
280 	struct vnode *vp;
281 	struct ucred *cred;
282 	struct proc *procp;
283 {
284 	caddr_t bpos, dpos;
285 	int error = 0;
286 	struct mbuf *mreq, *mrep, *md, *mb;
287 	struct nfsnode *np = VTONFS(vp);
288 
289 	nfsm_reqhead(np, NFSPROC_NULL, 0);
290 	nfsm_request(np, NFSPROC_NULL, procp, cred);
291 	nfsm_reqdone;
292 	return (error);
293 }
294 
295 /*
296  * nfs access vnode op.
297  * For nfs version 2, just return ok. File accesses may fail later.
298  * For nfs version 3, use the access rpc to check accessibility. If file modes
299  * are changed on the server, accesses might still fail later.
300  */
301 int
302 nfs_access(v)
303 	void *v;
304 {
305 	struct vop_access_args /* {
306 		struct vnode *a_vp;
307 		int  a_mode;
308 		struct ucred *a_cred;
309 		struct proc *a_p;
310 	} */ *ap = v;
311 	struct vnode *vp = ap->a_vp;
312 	u_int32_t *tl;
313 	caddr_t cp;
314 	int32_t t1, t2;
315 	caddr_t bpos, dpos, cp2;
316 	int error = 0, attrflag, cachevalid;
317 	struct mbuf *mreq, *mrep, *md, *mb;
318 	u_int32_t mode, rmode;
319 	const int v3 = NFS_ISV3(vp);
320 	struct nfsnode *np = VTONFS(vp);
321 
322 	cachevalid = (np->n_accstamp != -1 &&
323 	    (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
324 	    np->n_accuid == ap->a_cred->cr_uid);
325 
326 	/*
327 	 * Check access cache first. If this request has been made for this
328 	 * uid shortly before, use the cached result.
329 	 */
330 	if (cachevalid) {
331 		if (!np->n_accerror) {
332 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
333 				return np->n_accerror;
334 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
335 			return np->n_accerror;
336 	}
337 
338 	/*
339 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
340 	 * ufs_access() locally using the vattr. This may not be correct,
341 	 * since the server may apply other access criteria such as
342 	 * client uid-->server uid mapping that we do not know about, but
343 	 * this is better than just returning anything that is lying about
344 	 * in the cache.
345 	 */
346 	if (v3) {
347 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
348 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
349 		nfsm_fhtom(np, v3);
350 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
351 		if (ap->a_mode & VREAD)
352 			mode = NFSV3ACCESS_READ;
353 		else
354 			mode = 0;
355 		if (vp->v_type != VDIR) {
356 			if (ap->a_mode & VWRITE)
357 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
358 			if (ap->a_mode & VEXEC)
359 				mode |= NFSV3ACCESS_EXECUTE;
360 		} else {
361 			if (ap->a_mode & VWRITE)
362 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
363 					 NFSV3ACCESS_DELETE);
364 			if (ap->a_mode & VEXEC)
365 				mode |= NFSV3ACCESS_LOOKUP;
366 		}
367 		*tl = txdr_unsigned(mode);
368 		nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
369 		nfsm_postop_attr(vp, attrflag, 0);
370 		if (!error) {
371 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
372 			rmode = fxdr_unsigned(u_int32_t, *tl);
373 			/*
374 			 * The NFS V3 spec does not clarify whether or not
375 			 * the returned access bits can be a superset of
376 			 * the ones requested, so...
377 			 */
378 			if ((rmode & mode) != mode)
379 				error = EACCES;
380 		}
381 		nfsm_reqdone;
382 	} else
383 		return (nfsspec_access(ap));
384 	/*
385 	 * Disallow write attempts on filesystems mounted read-only;
386 	 * unless the file is a socket, fifo, or a block or character
387 	 * device resident on the filesystem.
388 	 */
389 	if (!error && (ap->a_mode & VWRITE) &&
390 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
391 		switch (vp->v_type) {
392 		case VREG:
393 		case VDIR:
394 		case VLNK:
395 			error = EROFS;
396 		default:
397 			break;
398 		}
399 	}
400 
401 	if (!error || error == EACCES) {
402 		/*
403 		 * If we got the same result as for a previous,
404 		 * different request, OR it in. Don't update
405 		 * the timestamp in that case.
406 		 */
407 		if (cachevalid && np->n_accstamp != -1 &&
408 		    error == np->n_accerror) {
409 			if (!error)
410 				np->n_accmode |= ap->a_mode;
411 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
412 				np->n_accmode = ap->a_mode;
413 		} else {
414 			np->n_accstamp = time.tv_sec;
415 			np->n_accuid = ap->a_cred->cr_uid;
416 			np->n_accmode = ap->a_mode;
417 			np->n_accerror = error;
418 		}
419 	}
420 
421 	return (error);
422 }
423 
424 /*
425  * nfs open vnode op
426  * Check to see if the type is ok
427  * and that deletion is not in progress.
428  * For paged in text files, you will need to flush the page cache
429  * if consistency is lost.
430  */
431 /* ARGSUSED */
432 int
433 nfs_open(v)
434 	void *v;
435 {
436 	struct vop_open_args /* {
437 		struct vnode *a_vp;
438 		int  a_mode;
439 		struct ucred *a_cred;
440 		struct proc *a_p;
441 	} */ *ap = v;
442 	struct vnode *vp = ap->a_vp;
443 	struct nfsnode *np = VTONFS(vp);
444 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
445 	struct vattr vattr;
446 	int error;
447 
448 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
449 		return (EACCES);
450 	}
451 
452 	/*
453 	 * Initialize read and write creds here, for swapfiles
454 	 * and other paths that don't set the creds themselves.
455 	 */
456 
457 	if (ap->a_mode & FREAD) {
458 		if (np->n_rcred) {
459 			crfree(np->n_rcred);
460 		}
461 		np->n_rcred = ap->a_cred;
462 		crhold(np->n_rcred);
463 	}
464 	if (ap->a_mode & FWRITE) {
465 		if (np->n_wcred) {
466 			crfree(np->n_wcred);
467 		}
468 		np->n_wcred = ap->a_cred;
469 		crhold(np->n_wcred);
470 	}
471 
472 #ifndef NFS_V2_ONLY
473 	/*
474 	 * Get a valid lease. If cached data is stale, flush it.
475 	 */
476 	if (nmp->nm_flag & NFSMNT_NQNFS) {
477 		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
478 		    do {
479 			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
480 			    ap->a_p);
481 		    } while (error == NQNFS_EXPIRED);
482 		    if (error)
483 			return (error);
484 		    if (np->n_lrev != np->n_brev ||
485 			(np->n_flag & NQNFSNONCACHE)) {
486 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
487 				ap->a_p, 1)) == EINTR)
488 				return (error);
489 			np->n_brev = np->n_lrev;
490 		    }
491 		}
492 	} else
493 #endif
494 	{
495 		if (np->n_flag & NMODIFIED) {
496 			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
497 				ap->a_p, 1)) == EINTR)
498 				return (error);
499 			np->n_attrstamp = 0;
500 			if (vp->v_type == VDIR) {
501 				nfs_invaldircache(vp, 0);
502 				np->n_direofoffset = 0;
503 			}
504 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
505 			if (error)
506 				return (error);
507 			np->n_mtime = vattr.va_mtime.tv_sec;
508 		} else {
509 			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
510 			if (error)
511 				return (error);
512 			if (np->n_mtime != vattr.va_mtime.tv_sec) {
513 				if (vp->v_type == VDIR) {
514 					nfs_invaldircache(vp, 0);
515 					np->n_direofoffset = 0;
516 				}
517 				if ((error = nfs_vinvalbuf(vp, V_SAVE,
518 					ap->a_cred, ap->a_p, 1)) == EINTR)
519 					return (error);
520 				np->n_mtime = vattr.va_mtime.tv_sec;
521 			}
522 		}
523 	}
524 	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
525 		np->n_attrstamp = 0; /* For Open/Close consistency */
526 	return (0);
527 }
528 
529 /*
530  * nfs close vnode op
531  * What an NFS client should do upon close after writing is a debatable issue.
532  * Most NFS clients push delayed writes to the server upon close, basically for
533  * two reasons:
534  * 1 - So that any write errors may be reported back to the client process
535  *     doing the close system call. By far the two most likely errors are
536  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
537  * 2 - To put a worst case upper bound on cache inconsistency between
538  *     multiple clients for the file.
539  * There is also a consistency problem for Version 2 of the protocol w.r.t.
540  * not being able to tell if other clients are writing a file concurrently,
541  * since there is no way of knowing if the changed modify time in the reply
542  * is only due to the write for this client.
543  * (NFS Version 3 provides weak cache consistency data in the reply that
544  *  should be sufficient to detect and handle this case.)
545  *
546  * The current code does the following:
547  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
548  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
549  *                     or commit them (this satisfies 1 and 2 except for the
550  *                     case where the server crashes after this close but
551  *                     before the commit RPC, which is felt to be "good
552  *                     enough". Changing the last argument to nfs_flush() to
553  *                     a 1 would force a commit operation, if it is felt a
554  *                     commit is necessary now.
555  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
556  *                     1 should be dealt with via an fsync() system call for
557  *                     cases where write errors are important.
558  */
559 /* ARGSUSED */
560 int
561 nfs_close(v)
562 	void *v;
563 {
564 	struct vop_close_args /* {
565 		struct vnodeop_desc *a_desc;
566 		struct vnode *a_vp;
567 		int  a_fflag;
568 		struct ucred *a_cred;
569 		struct proc *a_p;
570 	} */ *ap = v;
571 	struct vnode *vp = ap->a_vp;
572 	struct nfsnode *np = VTONFS(vp);
573 	int error = 0;
574 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
575 
576 	if (vp->v_type == VREG) {
577 	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
578 		(np->n_flag & NMODIFIED)) {
579 		if (NFS_ISV3(vp)) {
580 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
581 		    np->n_flag &= ~NMODIFIED;
582 		} else
583 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
584 		np->n_attrstamp = 0;
585 	    }
586 	    if (np->n_flag & NWRITEERR) {
587 		np->n_flag &= ~NWRITEERR;
588 		error = np->n_error;
589 	    }
590 	}
591 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
592 	return (error);
593 }
594 
595 /*
596  * nfs getattr call from vfs.
597  */
598 int
599 nfs_getattr(v)
600 	void *v;
601 {
602 	struct vop_getattr_args /* {
603 		struct vnode *a_vp;
604 		struct vattr *a_vap;
605 		struct ucred *a_cred;
606 		struct proc *a_p;
607 	} */ *ap = v;
608 	struct vnode *vp = ap->a_vp;
609 	struct nfsnode *np = VTONFS(vp);
610 	caddr_t cp;
611 	u_int32_t *tl;
612 	int32_t t1, t2;
613 	caddr_t bpos, dpos;
614 	int error = 0;
615 	struct mbuf *mreq, *mrep, *md, *mb;
616 	const int v3 = NFS_ISV3(vp);
617 
618 	/*
619 	 * Update local times for special files.
620 	 */
621 	if (np->n_flag & (NACC | NUPD))
622 		np->n_flag |= NCHG;
623 	/*
624 	 * First look in the cache.
625 	 */
626 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
627 		return (0);
628 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
629 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
630 	nfsm_fhtom(np, v3);
631 	nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
632 	if (!error) {
633 		nfsm_loadattr(vp, ap->a_vap, 0);
634 		if (vp->v_type == VDIR &&
635 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
636 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
637 	}
638 	nfsm_reqdone;
639 	return (error);
640 }
641 
642 /*
643  * nfs setattr call.
644  */
645 int
646 nfs_setattr(v)
647 	void *v;
648 {
649 	struct vop_setattr_args /* {
650 		struct vnodeop_desc *a_desc;
651 		struct vnode *a_vp;
652 		struct vattr *a_vap;
653 		struct ucred *a_cred;
654 		struct proc *a_p;
655 	} */ *ap = v;
656 	struct vnode *vp = ap->a_vp;
657 	struct nfsnode *np = VTONFS(vp);
658 	struct vattr *vap = ap->a_vap;
659 	int error = 0;
660 	u_quad_t tsize = 0;
661 
662 	/*
663 	 * Setting of flags is not supported.
664 	 */
665 	if (vap->va_flags != VNOVAL)
666 		return (EOPNOTSUPP);
667 
668 	/*
669 	 * Disallow write attempts if the filesystem is mounted read-only.
670 	 */
671   	if ((vap->va_uid != (uid_t)VNOVAL ||
672 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
673 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
674 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
675 		return (EROFS);
676 	if (vap->va_size != VNOVAL) {
677  		switch (vp->v_type) {
678  		case VDIR:
679  			return (EISDIR);
680  		case VCHR:
681  		case VBLK:
682  		case VSOCK:
683  		case VFIFO:
684 			if (vap->va_mtime.tv_sec == VNOVAL &&
685 			    vap->va_atime.tv_sec == VNOVAL &&
686 			    vap->va_mode == (mode_t)VNOVAL &&
687 			    vap->va_uid == (uid_t)VNOVAL &&
688 			    vap->va_gid == (gid_t)VNOVAL)
689 				return (0);
690  			vap->va_size = VNOVAL;
691  			break;
692  		default:
693 			/*
694 			 * Disallow write attempts if the filesystem is
695 			 * mounted read-only.
696 			 */
697 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
698 				return (EROFS);
699  			uvm_vnp_setsize(vp, vap->va_size);
700  			tsize = np->n_size;
701 			np->n_size = vap->va_size;
702  			if (vap->va_size == 0)
703  				error = nfs_vinvalbuf(vp, 0,
704  				     ap->a_cred, ap->a_p, 1);
705 			else
706 				error = nfs_vinvalbuf(vp, V_SAVE,
707 				     ap->a_cred, ap->a_p, 1);
708 			if (error) {
709 				uvm_vnp_setsize(vp, tsize);
710 				return (error);
711 			}
712  			np->n_vattr->va_size = vap->va_size;
713   		}
714   	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
715 		vap->va_atime.tv_sec != VNOVAL) &&
716 		vp->v_type == VREG &&
717   		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
718 		 ap->a_p, 1)) == EINTR)
719 		return (error);
720 	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
721 	if (error && vap->va_size != VNOVAL) {
722 		np->n_size = np->n_vattr->va_size = tsize;
723 		uvm_vnp_setsize(vp, np->n_size);
724 	}
725 	VN_KNOTE(vp, NOTE_ATTRIB);
726 	return (error);
727 }
728 
729 /*
730  * Do an nfs setattr rpc.
731  */
732 int
733 nfs_setattrrpc(vp, vap, cred, procp)
734 	struct vnode *vp;
735 	struct vattr *vap;
736 	struct ucred *cred;
737 	struct proc *procp;
738 {
739 	struct nfsv2_sattr *sp;
740 	caddr_t cp;
741 	int32_t t1, t2;
742 	caddr_t bpos, dpos, cp2;
743 	u_int32_t *tl;
744 	int error = 0, wccflag = NFSV3_WCCRATTR;
745 	struct mbuf *mreq, *mrep, *md, *mb;
746 	const int v3 = NFS_ISV3(vp);
747 	struct nfsnode *np = VTONFS(vp);
748 
749 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
750 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
751 	nfsm_fhtom(np, v3);
752 	if (v3) {
753 		nfsm_v3attrbuild(vap, TRUE);
754 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
755 		*tl = nfs_false;
756 	} else {
757 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
758 		if (vap->va_mode == (mode_t)VNOVAL)
759 			sp->sa_mode = nfs_xdrneg1;
760 		else
761 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
762 		if (vap->va_uid == (uid_t)VNOVAL)
763 			sp->sa_uid = nfs_xdrneg1;
764 		else
765 			sp->sa_uid = txdr_unsigned(vap->va_uid);
766 		if (vap->va_gid == (gid_t)VNOVAL)
767 			sp->sa_gid = nfs_xdrneg1;
768 		else
769 			sp->sa_gid = txdr_unsigned(vap->va_gid);
770 		sp->sa_size = txdr_unsigned(vap->va_size);
771 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
772 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
773 	}
774 	nfsm_request(np, NFSPROC_SETATTR, procp, cred);
775 	if (v3) {
776 		nfsm_wcc_data(vp, wccflag, 0);
777 	} else
778 		nfsm_loadattr(vp, (struct vattr *)0, 0);
779 	nfsm_reqdone;
780 	return (error);
781 }
782 
783 /*
784  * nfs lookup call, one step at a time...
785  * First look in cache
786  * If not found, unlock the directory nfsnode and do the rpc
787  *
788  * This code is full of lock/unlock statements and checks, because
789  * we continue after cache_lookup has finished (we need to check
790  * with the attr cache and do an rpc if it has timed out). This means
791  * that the locking effects of cache_lookup have to be taken into
792  * account.
793  */
794 int
795 nfs_lookup(v)
796 	void *v;
797 {
798 	struct vop_lookup_args /* {
799 		struct vnodeop_desc *a_desc;
800 		struct vnode *a_dvp;
801 		struct vnode **a_vpp;
802 		struct componentname *a_cnp;
803 	} */ *ap = v;
804 	struct componentname *cnp = ap->a_cnp;
805 	struct vnode *dvp = ap->a_dvp;
806 	struct vnode **vpp = ap->a_vpp;
807 	int flags;
808 	struct vnode *newvp;
809 	u_int32_t *tl;
810 	caddr_t cp;
811 	int32_t t1, t2;
812 	caddr_t bpos, dpos, cp2;
813 	struct mbuf *mreq, *mrep, *md, *mb;
814 	long len;
815 	nfsfh_t *fhp;
816 	struct nfsnode *np;
817 	int lockparent, wantparent, error = 0, attrflag, fhsize;
818 	const int v3 = NFS_ISV3(dvp);
819 
820 	cnp->cn_flags &= ~PDIRUNLOCK;
821 	flags = cnp->cn_flags;
822 
823 	*vpp = NULLVP;
824 	newvp = NULLVP;
825 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
826 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
827 		return (EROFS);
828 	if (dvp->v_type != VDIR)
829 		return (ENOTDIR);
830 
831 	lockparent = flags & LOCKPARENT;
832 	wantparent = flags & (LOCKPARENT|WANTPARENT);
833 	np = VTONFS(dvp);
834 
835 	/*
836 	 * Before tediously performing a linear scan of the directory,
837 	 * check the name cache to see if the directory/name pair
838 	 * we are looking for is known already.
839 	 * If the directory/name pair is found in the name cache,
840 	 * we have to ensure the directory has not changed from
841 	 * the time the cache entry has been created. If it has,
842 	 * the cache entry has to be ignored.
843 	 */
844 	if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
845 		struct vattr vattr;
846 		int err2;
847 
848 		if (error && error != ENOENT) {
849 			*vpp = NULLVP;
850 			return error;
851 		}
852 
853 		if (cnp->cn_flags & PDIRUNLOCK) {
854 			err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
855 			if (err2 != 0) {
856 				*vpp = NULLVP;
857 				return err2;
858 			}
859 			cnp->cn_flags &= ~PDIRUNLOCK;
860 		}
861 
862 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
863 		if (err2 != 0) {
864 			if (error == 0) {
865 				if (*vpp != dvp)
866 					vput(*vpp);
867 				else
868 					vrele(*vpp);
869 			}
870 			*vpp = NULLVP;
871 			return err2;
872 		}
873 
874 		if (error == ENOENT) {
875 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
876 			    cnp->cn_proc) && vattr.va_mtime.tv_sec ==
877 			    VTONFS(dvp)->n_nctime)
878 				return ENOENT;
879 			cache_purge(dvp);
880 			np->n_nctime = 0;
881 			goto dorpc;
882 		}
883 
884 		newvp = *vpp;
885 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
886 			&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
887 		{
888 			nfsstats.lookupcache_hits++;
889 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
890 				cnp->cn_flags |= SAVENAME;
891 			if ((!lockparent || !(flags & ISLASTCN)) &&
892 			     newvp != dvp)
893 				VOP_UNLOCK(dvp, 0);
894 			return (0);
895 		}
896 		cache_purge(newvp);
897 		if (newvp != dvp)
898 			vput(newvp);
899 		else
900 			vrele(newvp);
901 		*vpp = NULLVP;
902 	}
903 dorpc:
904 	error = 0;
905 	newvp = NULLVP;
906 	nfsstats.lookupcache_misses++;
907 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
908 	len = cnp->cn_namelen;
909 	nfsm_reqhead(np, NFSPROC_LOOKUP,
910 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
911 	nfsm_fhtom(np, v3);
912 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
913 	nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
914 	if (error) {
915 		nfsm_postop_attr(dvp, attrflag, 0);
916 		m_freem(mrep);
917 		goto nfsmout;
918 	}
919 	nfsm_getfh(fhp, fhsize, v3);
920 
921 	/*
922 	 * Handle RENAME case...
923 	 */
924 	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
925 		if (NFS_CMPFH(np, fhp, fhsize)) {
926 			m_freem(mrep);
927 			return (EISDIR);
928 		}
929 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
930 		if (error) {
931 			m_freem(mrep);
932 			return error;
933 		}
934 		newvp = NFSTOV(np);
935 		if (v3) {
936 			nfsm_postop_attr(newvp, attrflag, 0);
937 			nfsm_postop_attr(dvp, attrflag, 0);
938 		} else
939 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
940 		*vpp = newvp;
941 		m_freem(mrep);
942 		cnp->cn_flags |= SAVENAME;
943 		if (!lockparent) {
944 			VOP_UNLOCK(dvp, 0);
945 			cnp->cn_flags |= PDIRUNLOCK;
946 		}
947 		return (0);
948 	}
949 
950 	/*
951 	 * The postop attr handling is duplicated for each if case,
952 	 * because it should be done while dvp is locked (unlocking
953 	 * dvp is different for each case).
954 	 */
955 
956 	if (NFS_CMPFH(np, fhp, fhsize)) {
957 		/*
958 		 * "." lookup
959 		 */
960 		VREF(dvp);
961 		newvp = dvp;
962 		if (v3) {
963 			nfsm_postop_attr(newvp, attrflag, 0);
964 			nfsm_postop_attr(dvp, attrflag, 0);
965 		} else
966 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
967 	} else if (flags & ISDOTDOT) {
968 		/*
969 		 * ".." lookup
970 		 */
971 		VOP_UNLOCK(dvp, 0);
972 		cnp->cn_flags |= PDIRUNLOCK;
973 
974 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
975 		if (error) {
976 			if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
977 				cnp->cn_flags &= ~PDIRUNLOCK;
978 			m_freem(mrep);
979 			return error;
980 		}
981 		newvp = NFSTOV(np);
982 
983 		if (v3) {
984 			nfsm_postop_attr(newvp, attrflag, 0);
985 			nfsm_postop_attr(dvp, attrflag, 0);
986 		} else
987 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
988 
989 		if (lockparent && (flags & ISLASTCN)) {
990 			if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
991 				m_freem(mrep);
992 				vput(newvp);
993 				return error;
994 			}
995 			cnp->cn_flags &= ~PDIRUNLOCK;
996 		}
997 	} else {
998 		/*
999 		 * Other lookups.
1000 		 */
1001 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1002 		if (error) {
1003 			m_freem(mrep);
1004 			return error;
1005 		}
1006 		newvp = NFSTOV(np);
1007 		if (v3) {
1008 			nfsm_postop_attr(newvp, attrflag, 0);
1009 			nfsm_postop_attr(dvp, attrflag, 0);
1010 		} else
1011 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
1012 		if (!lockparent || !(flags & ISLASTCN)) {
1013 			VOP_UNLOCK(dvp, 0);
1014 			cnp->cn_flags |= PDIRUNLOCK;
1015 		}
1016 	}
1017 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1018 		cnp->cn_flags |= SAVENAME;
1019 	if ((cnp->cn_flags & MAKEENTRY) &&
1020 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1021 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1022 		cache_enter(dvp, newvp, cnp);
1023 	}
1024 	*vpp = newvp;
1025 	nfsm_reqdone;
1026 	if (error) {
1027 		/*
1028 		 * We get here only because of errors returned by
1029 		 * the RPC. Otherwise we'll have returned above
1030 		 * (the nfsm_* macros will jump to nfsm_reqdone
1031 		 * on error).
1032 		 */
1033 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1034 		    cnp->cn_nameiop != CREATE) {
1035 			if (VTONFS(dvp)->n_nctime == 0)
1036 				VTONFS(dvp)->n_nctime =
1037 				    VTONFS(dvp)->n_vattr->va_mtime.tv_sec;
1038 			cache_enter(dvp, NULL, cnp);
1039 		}
1040 		if (newvp != NULLVP) {
1041 			vrele(newvp);
1042 			if (newvp != dvp)
1043 				VOP_UNLOCK(newvp, 0);
1044 		}
1045 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1046 		    (flags & ISLASTCN) && error == ENOENT) {
1047 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1048 				error = EROFS;
1049 			else
1050 				error = EJUSTRETURN;
1051 		}
1052 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1053 			cnp->cn_flags |= SAVENAME;
1054 		*vpp = NULL;
1055 	}
1056 
1057 	return error;
1058 }
1059 
1060 /*
1061  * nfs read call.
1062  * Just call nfs_bioread() to do the work.
1063  */
1064 int
1065 nfs_read(v)
1066 	void *v;
1067 {
1068 	struct vop_read_args /* {
1069 		struct vnode *a_vp;
1070 		struct uio *a_uio;
1071 		int  a_ioflag;
1072 		struct ucred *a_cred;
1073 	} */ *ap = v;
1074 	struct vnode *vp = ap->a_vp;
1075 
1076 	if (vp->v_type != VREG)
1077 		return (EPERM);
1078 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1079 }
1080 
1081 /*
1082  * nfs readlink call
1083  */
1084 int
1085 nfs_readlink(v)
1086 	void *v;
1087 {
1088 	struct vop_readlink_args /* {
1089 		struct vnode *a_vp;
1090 		struct uio *a_uio;
1091 		struct ucred *a_cred;
1092 	} */ *ap = v;
1093 	struct vnode *vp = ap->a_vp;
1094 
1095 	if (vp->v_type != VLNK)
1096 		return (EPERM);
1097 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1098 }
1099 
1100 /*
1101  * Do a readlink rpc.
1102  * Called by nfs_doio() from below the buffer cache.
1103  */
1104 int
1105 nfs_readlinkrpc(vp, uiop, cred)
1106 	struct vnode *vp;
1107 	struct uio *uiop;
1108 	struct ucred *cred;
1109 {
1110 	u_int32_t *tl;
1111 	caddr_t cp;
1112 	int32_t t1, t2;
1113 	caddr_t bpos, dpos, cp2;
1114 	int error = 0, len, attrflag;
1115 	struct mbuf *mreq, *mrep, *md, *mb;
1116 	const int v3 = NFS_ISV3(vp);
1117 	struct nfsnode *np = VTONFS(vp);
1118 
1119 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1120 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1121 	nfsm_fhtom(np, v3);
1122 	nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1123 	if (v3)
1124 		nfsm_postop_attr(vp, attrflag, 0);
1125 	if (!error) {
1126 		nfsm_strsiz(len, NFS_MAXPATHLEN);
1127 		nfsm_mtouio(uiop, len);
1128 	}
1129 	nfsm_reqdone;
1130 	return (error);
1131 }
1132 
1133 /*
1134  * nfs read rpc call
1135  * Ditto above
1136  */
1137 int
1138 nfs_readrpc(vp, uiop)
1139 	struct vnode *vp;
1140 	struct uio *uiop;
1141 {
1142 	u_int32_t *tl;
1143 	caddr_t cp;
1144 	int32_t t1, t2;
1145 	caddr_t bpos, dpos, cp2;
1146 	struct mbuf *mreq, *mrep, *md, *mb;
1147 	struct nfsmount *nmp;
1148 	int error = 0, len, retlen, tsiz, eof, attrflag;
1149 	const int v3 = NFS_ISV3(vp);
1150 	struct nfsnode *np = VTONFS(vp);
1151 
1152 #ifndef nolint
1153 	eof = 0;
1154 #endif
1155 	nmp = VFSTONFS(vp->v_mount);
1156 	tsiz = uiop->uio_resid;
1157 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1158 		return (EFBIG);
1159 	while (tsiz > 0) {
1160 		nfsstats.rpccnt[NFSPROC_READ]++;
1161 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1162 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1163 		nfsm_fhtom(np, v3);
1164 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1165 		if (v3) {
1166 			txdr_hyper(uiop->uio_offset, tl);
1167 			*(tl + 2) = txdr_unsigned(len);
1168 		} else {
1169 			*tl++ = txdr_unsigned(uiop->uio_offset);
1170 			*tl++ = txdr_unsigned(len);
1171 			*tl = 0;
1172 		}
1173 		nfsm_request(np, NFSPROC_READ, uiop->uio_procp,
1174 			     VTONFS(vp)->n_rcred);
1175 		if (v3) {
1176 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1177 			if (error) {
1178 				m_freem(mrep);
1179 				goto nfsmout;
1180 			}
1181 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1182 			eof = fxdr_unsigned(int, *(tl + 1));
1183 		} else
1184 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1185 		nfsm_strsiz(retlen, nmp->nm_rsize);
1186 		nfsm_mtouio(uiop, retlen);
1187 		m_freem(mrep);
1188 		tsiz -= retlen;
1189 		if (v3) {
1190 			if (eof || retlen == 0)
1191 				tsiz = 0;
1192 		} else if (retlen < len)
1193 			tsiz = 0;
1194 	}
1195 nfsmout:
1196 	return (error);
1197 }
1198 
1199 /*
1200  * nfs write call
1201  */
1202 int
1203 nfs_writerpc(vp, uiop, iomode, stalewriteverf)
1204 	struct vnode *vp;
1205 	struct uio *uiop;
1206 	int *iomode;
1207 	boolean_t *stalewriteverf;
1208 {
1209 	u_int32_t *tl;
1210 	caddr_t cp;
1211 	int32_t t1, t2, backup;
1212 	caddr_t bpos, dpos, cp2;
1213 	struct mbuf *mreq, *mrep, *md, *mb;
1214 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1215 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1216 	const int v3 = NFS_ISV3(vp);
1217 	int committed = NFSV3WRITE_FILESYNC;
1218 	struct nfsnode *np = VTONFS(vp);
1219 
1220 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1221 		panic("writerpc readonly vp %p", vp);
1222 	}
1223 
1224 #ifdef DIAGNOSTIC
1225 	if (uiop->uio_iovcnt != 1)
1226 		panic("nfs: writerpc iovcnt > 1");
1227 #endif
1228 	tsiz = uiop->uio_resid;
1229 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1230 		return (EFBIG);
1231 	while (tsiz > 0) {
1232 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1233 		len = min(tsiz, nmp->nm_wsize);
1234 		nfsm_reqhead(np, NFSPROC_WRITE,
1235 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1236 		nfsm_fhtom(np, v3);
1237 		if (v3) {
1238 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1239 			txdr_hyper(uiop->uio_offset, tl);
1240 			tl += 2;
1241 			*tl++ = txdr_unsigned(len);
1242 			*tl++ = txdr_unsigned(*iomode);
1243 			*tl = txdr_unsigned(len);
1244 		} else {
1245 			u_int32_t x;
1246 
1247 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1248 			/* Set both "begin" and "current" to non-garbage. */
1249 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1250 			*tl++ = x;      /* "begin offset" */
1251 			*tl++ = x;      /* "current offset" */
1252 			x = txdr_unsigned(len);
1253 			*tl++ = x;      /* total to this offset */
1254 			*tl = x;        /* size of this write */
1255 
1256 		}
1257 		nfsm_uiotom(uiop, len);
1258 		nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp,
1259 			     VTONFS(vp)->n_wcred);
1260 		if (v3) {
1261 			wccflag = NFSV3_WCCCHK;
1262 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
1263 			if (!error) {
1264 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1265 					+ NFSX_V3WRITEVERF);
1266 				rlen = fxdr_unsigned(int, *tl++);
1267 				if (rlen == 0) {
1268 					error = NFSERR_IO;
1269 					m_freem(mrep);
1270 					break;
1271 				} else if (rlen < len) {
1272 					backup = len - rlen;
1273 					uiop->uio_iov->iov_base =
1274 					    (caddr_t)uiop->uio_iov->iov_base -
1275 					    backup;
1276 					uiop->uio_iov->iov_len += backup;
1277 					uiop->uio_offset -= backup;
1278 					uiop->uio_resid += backup;
1279 					len = rlen;
1280 				}
1281 				commit = fxdr_unsigned(int, *tl++);
1282 
1283 				/*
1284 				 * Return the lowest committment level
1285 				 * obtained by any of the RPCs.
1286 				 */
1287 				if (committed == NFSV3WRITE_FILESYNC)
1288 					committed = commit;
1289 				else if (committed == NFSV3WRITE_DATASYNC &&
1290 					commit == NFSV3WRITE_UNSTABLE)
1291 					committed = commit;
1292 				simple_lock(&nmp->nm_slock);
1293 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1294 					memcpy(nmp->nm_writeverf, tl,
1295 					    NFSX_V3WRITEVERF);
1296 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1297 				} else if ((nmp->nm_iflag &
1298 				    NFSMNT_STALEWRITEVERF) ||
1299 				    memcmp(tl, nmp->nm_writeverf,
1300 				    NFSX_V3WRITEVERF)) {
1301 					*stalewriteverf = TRUE;
1302 					memcpy(nmp->nm_writeverf, tl,
1303 					    NFSX_V3WRITEVERF);
1304 					nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
1305 				}
1306 				simple_unlock(&nmp->nm_slock);
1307 			}
1308 		} else
1309 		    nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1310 		if (wccflag)
1311 		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1312 		m_freem(mrep);
1313 		if (error)
1314 			break;
1315 		tsiz -= len;
1316 	}
1317 nfsmout:
1318 	*iomode = committed;
1319 	if (error)
1320 		uiop->uio_resid = tsiz;
1321 	return (error);
1322 }
1323 
1324 /*
1325  * nfs mknod rpc
1326  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1327  * mode set to specify the file type and the size field for rdev.
1328  */
1329 int
1330 nfs_mknodrpc(dvp, vpp, cnp, vap)
1331 	struct vnode *dvp;
1332 	struct vnode **vpp;
1333 	struct componentname *cnp;
1334 	struct vattr *vap;
1335 {
1336 	struct nfsv2_sattr *sp;
1337 	u_int32_t *tl;
1338 	caddr_t cp;
1339 	int32_t t1, t2;
1340 	struct vnode *newvp = (struct vnode *)0;
1341 	struct nfsnode *dnp, *np;
1342 	char *cp2;
1343 	caddr_t bpos, dpos;
1344 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1345 	struct mbuf *mreq, *mrep, *md, *mb;
1346 	u_int32_t rdev;
1347 	const int v3 = NFS_ISV3(dvp);
1348 
1349 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1350 		rdev = txdr_unsigned(vap->va_rdev);
1351 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1352 		rdev = nfs_xdrneg1;
1353 	else {
1354 		VOP_ABORTOP(dvp, cnp);
1355 		vput(dvp);
1356 		return (EOPNOTSUPP);
1357 	}
1358 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1359 	dnp = VTONFS(dvp);
1360 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1361 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1362 	nfsm_fhtom(dnp, v3);
1363 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1364 	if (v3) {
1365 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1366 		*tl++ = vtonfsv3_type(vap->va_type);
1367 		nfsm_v3attrbuild(vap, FALSE);
1368 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1369 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1370 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1371 			*tl = txdr_unsigned(minor(vap->va_rdev));
1372 		}
1373 	} else {
1374 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1375 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1376 		sp->sa_uid = nfs_xdrneg1;
1377 		sp->sa_gid = nfs_xdrneg1;
1378 		sp->sa_size = rdev;
1379 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1380 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1381 	}
1382 	nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1383 	if (!error) {
1384 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1385 		if (!gotvp) {
1386 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1387 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1388 			if (!error)
1389 				newvp = NFSTOV(np);
1390 		}
1391 	}
1392 	if (v3)
1393 		nfsm_wcc_data(dvp, wccflag, 0);
1394 	nfsm_reqdone;
1395 	if (error) {
1396 		if (newvp)
1397 			vput(newvp);
1398 	} else {
1399 		if (cnp->cn_flags & MAKEENTRY)
1400 			cache_enter(dvp, newvp, cnp);
1401 		*vpp = newvp;
1402 	}
1403 	PNBUF_PUT(cnp->cn_pnbuf);
1404 	VTONFS(dvp)->n_flag |= NMODIFIED;
1405 	if (!wccflag)
1406 		VTONFS(dvp)->n_attrstamp = 0;
1407 	vput(dvp);
1408 	return (error);
1409 }
1410 
1411 /*
1412  * nfs mknod vop
1413  * just call nfs_mknodrpc() to do the work.
1414  */
1415 /* ARGSUSED */
1416 int
1417 nfs_mknod(v)
1418 	void *v;
1419 {
1420 	struct vop_mknod_args /* {
1421 		struct vnode *a_dvp;
1422 		struct vnode **a_vpp;
1423 		struct componentname *a_cnp;
1424 		struct vattr *a_vap;
1425 	} */ *ap = v;
1426 	int error;
1427 
1428 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1429 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1430 	return (error);
1431 }
1432 
1433 static u_long create_verf;
1434 /*
1435  * nfs file create call
1436  */
1437 int
1438 nfs_create(v)
1439 	void *v;
1440 {
1441 	struct vop_create_args /* {
1442 		struct vnode *a_dvp;
1443 		struct vnode **a_vpp;
1444 		struct componentname *a_cnp;
1445 		struct vattr *a_vap;
1446 	} */ *ap = v;
1447 	struct vnode *dvp = ap->a_dvp;
1448 	struct vattr *vap = ap->a_vap;
1449 	struct componentname *cnp = ap->a_cnp;
1450 	struct nfsv2_sattr *sp;
1451 	u_int32_t *tl;
1452 	caddr_t cp;
1453 	int32_t t1, t2;
1454 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1455 	struct vnode *newvp = (struct vnode *)0;
1456 	caddr_t bpos, dpos, cp2;
1457 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1458 	struct mbuf *mreq, *mrep, *md, *mb;
1459 	const int v3 = NFS_ISV3(dvp);
1460 
1461 	/*
1462 	 * Oops, not for me..
1463 	 */
1464 	if (vap->va_type == VSOCK)
1465 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1466 
1467 #ifdef VA_EXCLUSIVE
1468 	if (vap->va_vaflags & VA_EXCLUSIVE)
1469 		fmode |= O_EXCL;
1470 #endif
1471 again:
1472 	error = 0;
1473 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1474 	dnp = VTONFS(dvp);
1475 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1476 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1477 	nfsm_fhtom(dnp, v3);
1478 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1479 	if (v3) {
1480 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1481 		if (fmode & O_EXCL) {
1482 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1483 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1484 #ifdef INET
1485 			if (TAILQ_FIRST(&in_ifaddr))
1486 				*tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr;
1487 			else
1488 				*tl++ = create_verf;
1489 #else
1490 			*tl++ = create_verf;
1491 #endif
1492 			*tl = ++create_verf;
1493 		} else {
1494 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1495 			nfsm_v3attrbuild(vap, FALSE);
1496 		}
1497 	} else {
1498 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1499 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1500 		sp->sa_uid = nfs_xdrneg1;
1501 		sp->sa_gid = nfs_xdrneg1;
1502 		sp->sa_size = 0;
1503 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1504 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1505 	}
1506 	nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1507 	if (!error) {
1508 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1509 		if (!gotvp) {
1510 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1511 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1512 			if (!error)
1513 				newvp = NFSTOV(np);
1514 		}
1515 	}
1516 	if (v3)
1517 		nfsm_wcc_data(dvp, wccflag, 0);
1518 	nfsm_reqdone;
1519 	if (error) {
1520 		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1521 			fmode &= ~O_EXCL;
1522 			goto again;
1523 		}
1524 		if (newvp)
1525 			vput(newvp);
1526 	} else if (v3 && (fmode & O_EXCL))
1527 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1528 	if (!error) {
1529 		if (cnp->cn_flags & MAKEENTRY)
1530 			cache_enter(dvp, newvp, cnp);
1531 		*ap->a_vpp = newvp;
1532 	}
1533 	PNBUF_PUT(cnp->cn_pnbuf);
1534 	VTONFS(dvp)->n_flag |= NMODIFIED;
1535 	if (!wccflag)
1536 		VTONFS(dvp)->n_attrstamp = 0;
1537 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1538 	vput(dvp);
1539 	return (error);
1540 }
1541 
1542 /*
1543  * nfs file remove call
1544  * To try and make nfs semantics closer to ufs semantics, a file that has
1545  * other processes using the vnode is renamed instead of removed and then
1546  * removed later on the last close.
1547  * - If v_usecount > 1
1548  *	  If a rename is not already in the works
1549  *	     call nfs_sillyrename() to set it up
1550  *     else
1551  *	  do the remove rpc
1552  */
1553 int
1554 nfs_remove(v)
1555 	void *v;
1556 {
1557 	struct vop_remove_args /* {
1558 		struct vnodeop_desc *a_desc;
1559 		struct vnode * a_dvp;
1560 		struct vnode * a_vp;
1561 		struct componentname * a_cnp;
1562 	} */ *ap = v;
1563 	struct vnode *vp = ap->a_vp;
1564 	struct vnode *dvp = ap->a_dvp;
1565 	struct componentname *cnp = ap->a_cnp;
1566 	struct nfsnode *np = VTONFS(vp);
1567 	int error = 0;
1568 	struct vattr vattr;
1569 
1570 #ifndef DIAGNOSTIC
1571 	if ((cnp->cn_flags & HASBUF) == 0)
1572 		panic("nfs_remove: no name");
1573 	if (vp->v_usecount < 1)
1574 		panic("nfs_remove: bad v_usecount");
1575 #endif
1576 	if (vp->v_type == VDIR)
1577 		error = EPERM;
1578 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1579 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1580 	    vattr.va_nlink > 1)) {
1581 		/*
1582 		 * Purge the name cache so that the chance of a lookup for
1583 		 * the name succeeding while the remove is in progress is
1584 		 * minimized. Without node locking it can still happen, such
1585 		 * that an I/O op returns ESTALE, but since you get this if
1586 		 * another host removes the file..
1587 		 */
1588 		cache_purge(vp);
1589 		/*
1590 		 * throw away biocache buffers, mainly to avoid
1591 		 * unnecessary delayed writes later.
1592 		 */
1593 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1594 		/* Do the rpc */
1595 		if (error != EINTR)
1596 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1597 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1598 		/*
1599 		 * Kludge City: If the first reply to the remove rpc is lost..
1600 		 *   the reply to the retransmitted request will be ENOENT
1601 		 *   since the file was in fact removed
1602 		 *   Therefore, we cheat and return success.
1603 		 */
1604 		if (error == ENOENT)
1605 			error = 0;
1606 	} else if (!np->n_sillyrename)
1607 		error = nfs_sillyrename(dvp, vp, cnp);
1608 	PNBUF_PUT(cnp->cn_pnbuf);
1609 	np->n_attrstamp = 0;
1610 	VN_KNOTE(vp, NOTE_DELETE);
1611 	VN_KNOTE(dvp, NOTE_WRITE);
1612 	if (dvp == vp)
1613 		vrele(vp);
1614 	else
1615 		vput(vp);
1616 	vput(dvp);
1617 	return (error);
1618 }
1619 
1620 /*
1621  * nfs file remove rpc called from nfs_inactive
1622  */
1623 int
1624 nfs_removeit(sp)
1625 	struct sillyrename *sp;
1626 {
1627 
1628 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1629 		(struct proc *)0));
1630 }
1631 
1632 /*
1633  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1634  */
1635 int
1636 nfs_removerpc(dvp, name, namelen, cred, proc)
1637 	struct vnode *dvp;
1638 	const char *name;
1639 	int namelen;
1640 	struct ucred *cred;
1641 	struct proc *proc;
1642 {
1643 	u_int32_t *tl;
1644 	caddr_t cp;
1645 	int32_t t1, t2;
1646 	caddr_t bpos, dpos, cp2;
1647 	int error = 0, wccflag = NFSV3_WCCRATTR;
1648 	struct mbuf *mreq, *mrep, *md, *mb;
1649 	const int v3 = NFS_ISV3(dvp);
1650 	struct nfsnode *dnp = VTONFS(dvp);
1651 
1652 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1653 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1654 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1655 	nfsm_fhtom(dnp, v3);
1656 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1657 	nfsm_request(dnp, NFSPROC_REMOVE, proc, cred);
1658 	if (v3)
1659 		nfsm_wcc_data(dvp, wccflag, 0);
1660 	nfsm_reqdone;
1661 	VTONFS(dvp)->n_flag |= NMODIFIED;
1662 	if (!wccflag)
1663 		VTONFS(dvp)->n_attrstamp = 0;
1664 	return (error);
1665 }
1666 
1667 /*
1668  * nfs file rename call
1669  */
1670 int
1671 nfs_rename(v)
1672 	void *v;
1673 {
1674 	struct vop_rename_args  /* {
1675 		struct vnode *a_fdvp;
1676 		struct vnode *a_fvp;
1677 		struct componentname *a_fcnp;
1678 		struct vnode *a_tdvp;
1679 		struct vnode *a_tvp;
1680 		struct componentname *a_tcnp;
1681 	} */ *ap = v;
1682 	struct vnode *fvp = ap->a_fvp;
1683 	struct vnode *tvp = ap->a_tvp;
1684 	struct vnode *fdvp = ap->a_fdvp;
1685 	struct vnode *tdvp = ap->a_tdvp;
1686 	struct componentname *tcnp = ap->a_tcnp;
1687 	struct componentname *fcnp = ap->a_fcnp;
1688 	int error;
1689 
1690 #ifndef DIAGNOSTIC
1691 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1692 	    (fcnp->cn_flags & HASBUF) == 0)
1693 		panic("nfs_rename: no name");
1694 #endif
1695 	/* Check for cross-device rename */
1696 	if ((fvp->v_mount != tdvp->v_mount) ||
1697 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1698 		error = EXDEV;
1699 		goto out;
1700 	}
1701 
1702 	/*
1703 	 * If the tvp exists and is in use, sillyrename it before doing the
1704 	 * rename of the new file over it.
1705 	 */
1706 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1707 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1708 		VN_KNOTE(tvp, NOTE_DELETE);
1709 		vput(tvp);
1710 		tvp = NULL;
1711 	}
1712 
1713 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1714 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1715 		tcnp->cn_proc);
1716 
1717 	VN_KNOTE(fdvp, NOTE_WRITE);
1718 	VN_KNOTE(tdvp, NOTE_WRITE);
1719 	if (fvp->v_type == VDIR) {
1720 		if (tvp != NULL && tvp->v_type == VDIR)
1721 			cache_purge(tdvp);
1722 		cache_purge(fdvp);
1723 	}
1724 out:
1725 	if (tdvp == tvp)
1726 		vrele(tdvp);
1727 	else
1728 		vput(tdvp);
1729 	if (tvp)
1730 		vput(tvp);
1731 	vrele(fdvp);
1732 	vrele(fvp);
1733 	/*
1734 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1735 	 */
1736 	if (error == ENOENT)
1737 		error = 0;
1738 	return (error);
1739 }
1740 
1741 /*
1742  * nfs file rename rpc called from nfs_remove() above
1743  */
1744 int
1745 nfs_renameit(sdvp, scnp, sp)
1746 	struct vnode *sdvp;
1747 	struct componentname *scnp;
1748 	struct sillyrename *sp;
1749 {
1750 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1751 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1752 }
1753 
1754 /*
1755  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1756  */
1757 int
1758 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1759 	struct vnode *fdvp;
1760 	const char *fnameptr;
1761 	int fnamelen;
1762 	struct vnode *tdvp;
1763 	const char *tnameptr;
1764 	int tnamelen;
1765 	struct ucred *cred;
1766 	struct proc *proc;
1767 {
1768 	u_int32_t *tl;
1769 	caddr_t cp;
1770 	int32_t t1, t2;
1771 	caddr_t bpos, dpos, cp2;
1772 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1773 	struct mbuf *mreq, *mrep, *md, *mb;
1774 	const int v3 = NFS_ISV3(fdvp);
1775 	struct nfsnode *fdnp = VTONFS(fdvp);
1776 
1777 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1778 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
1779 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1780 		nfsm_rndup(tnamelen));
1781 	nfsm_fhtom(fdnp, v3);
1782 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1783 	nfsm_fhtom(VTONFS(tdvp), v3);
1784 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1785 	nfsm_request(fdnp, NFSPROC_RENAME, proc, cred);
1786 	if (v3) {
1787 		nfsm_wcc_data(fdvp, fwccflag, 0);
1788 		nfsm_wcc_data(tdvp, twccflag, 0);
1789 	}
1790 	nfsm_reqdone;
1791 	VTONFS(fdvp)->n_flag |= NMODIFIED;
1792 	VTONFS(tdvp)->n_flag |= NMODIFIED;
1793 	if (!fwccflag)
1794 		VTONFS(fdvp)->n_attrstamp = 0;
1795 	if (!twccflag)
1796 		VTONFS(tdvp)->n_attrstamp = 0;
1797 	return (error);
1798 }
1799 
1800 /*
1801  * nfs hard link create call
1802  */
1803 int
1804 nfs_link(v)
1805 	void *v;
1806 {
1807 	struct vop_link_args /* {
1808 		struct vnode *a_dvp;
1809 		struct vnode *a_vp;
1810 		struct componentname *a_cnp;
1811 	} */ *ap = v;
1812 	struct vnode *vp = ap->a_vp;
1813 	struct vnode *dvp = ap->a_dvp;
1814 	struct componentname *cnp = ap->a_cnp;
1815 	u_int32_t *tl;
1816 	caddr_t cp;
1817 	int32_t t1, t2;
1818 	caddr_t bpos, dpos, cp2;
1819 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1820 	struct mbuf *mreq, *mrep, *md, *mb;
1821 	/* XXX Should be const and initialised? */
1822 	int v3;
1823 	struct nfsnode *np;
1824 
1825 	if (dvp->v_mount != vp->v_mount) {
1826 		VOP_ABORTOP(dvp, cnp);
1827 		vput(dvp);
1828 		return (EXDEV);
1829 	}
1830 	if (dvp != vp) {
1831 		error = vn_lock(vp, LK_EXCLUSIVE);
1832 		if (error != 0) {
1833 			VOP_ABORTOP(dvp, cnp);
1834 			vput(dvp);
1835 			return error;
1836 		}
1837 	}
1838 
1839 	/*
1840 	 * Push all writes to the server, so that the attribute cache
1841 	 * doesn't get "out of sync" with the server.
1842 	 * XXX There should be a better way!
1843 	 */
1844 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1845 
1846 	v3 = NFS_ISV3(vp);
1847 	nfsstats.rpccnt[NFSPROC_LINK]++;
1848 	np = VTONFS(vp);
1849 	nfsm_reqhead(np, NFSPROC_LINK,
1850 		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1851 	nfsm_fhtom(np, v3);
1852 	nfsm_fhtom(VTONFS(dvp), v3);
1853 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1854 	nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1855 	if (v3) {
1856 		nfsm_postop_attr(vp, attrflag, 0);
1857 		nfsm_wcc_data(dvp, wccflag, 0);
1858 	}
1859 	nfsm_reqdone;
1860 	PNBUF_PUT(cnp->cn_pnbuf);
1861 	VTONFS(dvp)->n_flag |= NMODIFIED;
1862 	if (!attrflag)
1863 		VTONFS(vp)->n_attrstamp = 0;
1864 	if (!wccflag)
1865 		VTONFS(dvp)->n_attrstamp = 0;
1866 	if (dvp != vp)
1867 		VOP_UNLOCK(vp, 0);
1868 	VN_KNOTE(vp, NOTE_LINK);
1869 	VN_KNOTE(dvp, NOTE_WRITE);
1870 	vput(dvp);
1871 	/*
1872 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1873 	 */
1874 	if (error == EEXIST)
1875 		error = 0;
1876 	return (error);
1877 }
1878 
1879 /*
1880  * nfs symbolic link create call
1881  */
1882 int
1883 nfs_symlink(v)
1884 	void *v;
1885 {
1886 	struct vop_symlink_args /* {
1887 		struct vnode *a_dvp;
1888 		struct vnode **a_vpp;
1889 		struct componentname *a_cnp;
1890 		struct vattr *a_vap;
1891 		char *a_target;
1892 	} */ *ap = v;
1893 	struct vnode *dvp = ap->a_dvp;
1894 	struct vattr *vap = ap->a_vap;
1895 	struct componentname *cnp = ap->a_cnp;
1896 	struct nfsv2_sattr *sp;
1897 	u_int32_t *tl;
1898 	caddr_t cp;
1899 	int32_t t1, t2;
1900 	caddr_t bpos, dpos, cp2;
1901 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1902 	struct mbuf *mreq, *mrep, *md, *mb;
1903 	struct vnode *newvp = (struct vnode *)0;
1904 	const int v3 = NFS_ISV3(dvp);
1905 	struct nfsnode *dnp = VTONFS(dvp);
1906 
1907 	*ap->a_vpp = NULL;
1908 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1909 	slen = strlen(ap->a_target);
1910 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1911 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1912 	nfsm_fhtom(dnp, v3);
1913 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1914 	if (v3)
1915 		nfsm_v3attrbuild(vap, FALSE);
1916 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1917 	if (!v3) {
1918 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1919 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1920 		sp->sa_uid = nfs_xdrneg1;
1921 		sp->sa_gid = nfs_xdrneg1;
1922 		sp->sa_size = nfs_xdrneg1;
1923 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1924 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1925 	}
1926 	nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1927 	if (v3) {
1928 		if (!error)
1929 			nfsm_mtofh(dvp, newvp, v3, gotvp);
1930 		nfsm_wcc_data(dvp, wccflag, 0);
1931 	}
1932 	nfsm_reqdone;
1933 	/*
1934 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1935 	 */
1936 	if (error == EEXIST)
1937 		error = 0;
1938 	if (error == 0 && newvp == NULL) {
1939 		struct nfsnode *np = NULL;
1940 
1941 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1942 		    cnp->cn_cred, cnp->cn_proc, &np);
1943 		if (error == 0)
1944 			newvp = NFSTOV(np);
1945 	}
1946 	if (error) {
1947 		if (newvp != NULL)
1948 			vput(newvp);
1949 	} else {
1950 		*ap->a_vpp = newvp;
1951 	}
1952 	PNBUF_PUT(cnp->cn_pnbuf);
1953 	VTONFS(dvp)->n_flag |= NMODIFIED;
1954 	if (!wccflag)
1955 		VTONFS(dvp)->n_attrstamp = 0;
1956 	VN_KNOTE(dvp, NOTE_WRITE);
1957 	vput(dvp);
1958 	return (error);
1959 }
1960 
1961 /*
1962  * nfs make dir call
1963  */
1964 int
1965 nfs_mkdir(v)
1966 	void *v;
1967 {
1968 	struct vop_mkdir_args /* {
1969 		struct vnode *a_dvp;
1970 		struct vnode **a_vpp;
1971 		struct componentname *a_cnp;
1972 		struct vattr *a_vap;
1973 	} */ *ap = v;
1974 	struct vnode *dvp = ap->a_dvp;
1975 	struct vattr *vap = ap->a_vap;
1976 	struct componentname *cnp = ap->a_cnp;
1977 	struct nfsv2_sattr *sp;
1978 	u_int32_t *tl;
1979 	caddr_t cp;
1980 	int32_t t1, t2;
1981 	int len;
1982 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
1983 	struct vnode *newvp = (struct vnode *)0;
1984 	caddr_t bpos, dpos, cp2;
1985 	int error = 0, wccflag = NFSV3_WCCRATTR;
1986 	int gotvp = 0;
1987 	struct mbuf *mreq, *mrep, *md, *mb;
1988 	const int v3 = NFS_ISV3(dvp);
1989 
1990 	len = cnp->cn_namelen;
1991 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1992 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
1993 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1994 	nfsm_fhtom(dnp, v3);
1995 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1996 	if (v3) {
1997 		nfsm_v3attrbuild(vap, FALSE);
1998 	} else {
1999 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2000 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2001 		sp->sa_uid = nfs_xdrneg1;
2002 		sp->sa_gid = nfs_xdrneg1;
2003 		sp->sa_size = nfs_xdrneg1;
2004 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2005 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2006 	}
2007 	nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
2008 	if (!error)
2009 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2010 	if (v3)
2011 		nfsm_wcc_data(dvp, wccflag, 0);
2012 	nfsm_reqdone;
2013 	VTONFS(dvp)->n_flag |= NMODIFIED;
2014 	if (!wccflag)
2015 		VTONFS(dvp)->n_attrstamp = 0;
2016 	/*
2017 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2018 	 * if we can succeed in looking up the directory.
2019 	 */
2020 	if (error == EEXIST || (!error && !gotvp)) {
2021 		if (newvp) {
2022 			vput(newvp);
2023 			newvp = (struct vnode *)0;
2024 		}
2025 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2026 			cnp->cn_proc, &np);
2027 		if (!error) {
2028 			newvp = NFSTOV(np);
2029 			if (newvp->v_type != VDIR)
2030 				error = EEXIST;
2031 		}
2032 	}
2033 	if (error) {
2034 		if (newvp)
2035 			vput(newvp);
2036 	} else {
2037 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2038 		if (cnp->cn_flags & MAKEENTRY)
2039 			cache_enter(dvp, newvp, cnp);
2040 		*ap->a_vpp = newvp;
2041 	}
2042 	PNBUF_PUT(cnp->cn_pnbuf);
2043 	vput(dvp);
2044 	return (error);
2045 }
2046 
2047 /*
2048  * nfs remove directory call
2049  */
2050 int
2051 nfs_rmdir(v)
2052 	void *v;
2053 {
2054 	struct vop_rmdir_args /* {
2055 		struct vnode *a_dvp;
2056 		struct vnode *a_vp;
2057 		struct componentname *a_cnp;
2058 	} */ *ap = v;
2059 	struct vnode *vp = ap->a_vp;
2060 	struct vnode *dvp = ap->a_dvp;
2061 	struct componentname *cnp = ap->a_cnp;
2062 	u_int32_t *tl;
2063 	caddr_t cp;
2064 	int32_t t1, t2;
2065 	caddr_t bpos, dpos, cp2;
2066 	int error = 0, wccflag = NFSV3_WCCRATTR;
2067 	struct mbuf *mreq, *mrep, *md, *mb;
2068 	const int v3 = NFS_ISV3(dvp);
2069 	struct nfsnode *dnp;
2070 
2071 	if (dvp == vp) {
2072 		vrele(dvp);
2073 		vput(dvp);
2074 		PNBUF_PUT(cnp->cn_pnbuf);
2075 		return (EINVAL);
2076 	}
2077 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2078 	dnp = VTONFS(dvp);
2079 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2080 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2081 	nfsm_fhtom(dnp, v3);
2082 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2083 	nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2084 	if (v3)
2085 		nfsm_wcc_data(dvp, wccflag, 0);
2086 	nfsm_reqdone;
2087 	PNBUF_PUT(cnp->cn_pnbuf);
2088 	VTONFS(dvp)->n_flag |= NMODIFIED;
2089 	if (!wccflag)
2090 		VTONFS(dvp)->n_attrstamp = 0;
2091 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2092 	VN_KNOTE(vp, NOTE_DELETE);
2093 	cache_purge(dvp);
2094 	cache_purge(vp);
2095 	vput(vp);
2096 	vput(dvp);
2097 	/*
2098 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2099 	 */
2100 	if (error == ENOENT)
2101 		error = 0;
2102 	return (error);
2103 }
2104 
2105 /*
2106  * nfs readdir call
2107  */
2108 int
2109 nfs_readdir(v)
2110 	void *v;
2111 {
2112 	struct vop_readdir_args /* {
2113 		struct vnode *a_vp;
2114 		struct uio *a_uio;
2115 		struct ucred *a_cred;
2116 		int *a_eofflag;
2117 		off_t **a_cookies;
2118 		int *a_ncookies;
2119 	} */ *ap = v;
2120 	struct vnode *vp = ap->a_vp;
2121 	struct uio *uio = ap->a_uio;
2122 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2123 	char *base = uio->uio_iov->iov_base;
2124 	int tresid, error;
2125 	size_t count, lost;
2126 	struct dirent *dp;
2127 	off_t *cookies = NULL;
2128 	int ncookies = 0, nc;
2129 
2130 	if (vp->v_type != VDIR)
2131 		return (EPERM);
2132 
2133 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2134 	count = uio->uio_resid - lost;
2135 	if (count <= 0)
2136 		return (EINVAL);
2137 
2138 	/*
2139 	 * Call nfs_bioread() to do the real work.
2140 	 */
2141 	tresid = uio->uio_resid = count;
2142 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2143 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2144 
2145 	if (!error && ap->a_cookies) {
2146 		ncookies = count / 16;
2147 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2148 		*ap->a_cookies = cookies;
2149 	}
2150 
2151 	if (!error && uio->uio_resid == tresid) {
2152 		uio->uio_resid += lost;
2153 		nfsstats.direofcache_misses++;
2154 		if (ap->a_cookies)
2155 			*ap->a_ncookies = 0;
2156 		*ap->a_eofflag = 1;
2157 		return (0);
2158 	}
2159 
2160 	if (!error && ap->a_cookies) {
2161 		/*
2162 		 * Only the NFS server and emulations use cookies, and they
2163 		 * load the directory block into system space, so we can
2164 		 * just look at it directly.
2165 		 */
2166 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2167 			panic("nfs_readdir: lost in space");
2168 		for (nc = 0; ncookies-- &&
2169 		     base < (char *)uio->uio_iov->iov_base; nc++){
2170 			dp = (struct dirent *) base;
2171 			if (dp->d_reclen == 0)
2172 				break;
2173 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2174 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2175 			else
2176 				*(cookies++) = NFS_GETCOOKIE(dp);
2177 			base += dp->d_reclen;
2178 		}
2179 		uio->uio_resid +=
2180 		    ((caddr_t)uio->uio_iov->iov_base - base);
2181 		uio->uio_iov->iov_len +=
2182 		    ((caddr_t)uio->uio_iov->iov_base - base);
2183 		uio->uio_iov->iov_base = base;
2184 		*ap->a_ncookies = nc;
2185 	}
2186 
2187 	uio->uio_resid += lost;
2188 	*ap->a_eofflag = 0;
2189 	return (error);
2190 }
2191 
2192 /*
2193  * Readdir rpc call.
2194  * Called from below the buffer cache by nfs_doio().
2195  */
2196 int
2197 nfs_readdirrpc(vp, uiop, cred)
2198 	struct vnode *vp;
2199 	struct uio *uiop;
2200 	struct ucred *cred;
2201 {
2202 	int len, left;
2203 	struct dirent *dp = NULL;
2204 	u_int32_t *tl;
2205 	caddr_t cp;
2206 	int32_t t1, t2;
2207 	caddr_t bpos, dpos, cp2;
2208 	struct mbuf *mreq, *mrep, *md, *mb;
2209 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2210 	struct nfsnode *dnp = VTONFS(vp);
2211 	u_quad_t fileno;
2212 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2213 	int attrflag, nrpcs = 0, reclen;
2214 	const int v3 = NFS_ISV3(vp);
2215 
2216 #ifdef DIAGNOSTIC
2217 	/*
2218 	 * Should be called from buffer cache, so only amount of
2219 	 * NFS_DIRBLKSIZ will be requested.
2220 	 */
2221 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2222 		panic("nfs readdirrpc bad uio");
2223 #endif
2224 
2225 	/*
2226 	 * Loop around doing readdir rpc's of size nm_readdirsize
2227 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2228 	 * The stopping criteria is EOF or buffer full.
2229 	 */
2230 	while (more_dirs && bigenough) {
2231 		/*
2232 		 * Heuristic: don't bother to do another RPC to further
2233 		 * fill up this block if there is not much room left. (< 50%
2234 		 * of the readdir RPC size). This wastes some buffer space
2235 		 * but can save up to 50% in RPC calls.
2236 		 */
2237 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2238 			bigenough = 0;
2239 			break;
2240 		}
2241 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2242 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2243 			NFSX_READDIR(v3));
2244 		nfsm_fhtom(dnp, v3);
2245 		if (v3) {
2246 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2247 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2248 				txdr_swapcookie3(uiop->uio_offset, tl);
2249 			} else {
2250 				txdr_cookie3(uiop->uio_offset, tl);
2251 			}
2252 			tl += 2;
2253 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2254 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2255 		} else {
2256 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2257 			*tl++ = txdr_unsigned(uiop->uio_offset);
2258 		}
2259 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2260 		nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2261 		nrpcs++;
2262 		if (v3) {
2263 			nfsm_postop_attr(vp, attrflag, 0);
2264 			if (!error) {
2265 				nfsm_dissect(tl, u_int32_t *,
2266 				    2 * NFSX_UNSIGNED);
2267 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2268 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2269 			} else {
2270 				m_freem(mrep);
2271 				goto nfsmout;
2272 			}
2273 		}
2274 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2275 		more_dirs = fxdr_unsigned(int, *tl);
2276 
2277 		/* loop thru the dir entries, doctoring them to 4bsd form */
2278 		while (more_dirs && bigenough) {
2279 			if (v3) {
2280 				nfsm_dissect(tl, u_int32_t *,
2281 				    3 * NFSX_UNSIGNED);
2282 				fileno = fxdr_hyper(tl);
2283 				len = fxdr_unsigned(int, *(tl + 2));
2284 			} else {
2285 				nfsm_dissect(tl, u_int32_t *,
2286 				    2 * NFSX_UNSIGNED);
2287 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2288 				len = fxdr_unsigned(int, *tl);
2289 			}
2290 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2291 				error = EBADRPC;
2292 				m_freem(mrep);
2293 				goto nfsmout;
2294 			}
2295 			tlen = nfsm_rndup(len);
2296 			if (tlen == len)
2297 				tlen += 4;	/* To ensure null termination */
2298 			tlen += sizeof (off_t) + sizeof (int);
2299 			reclen = ALIGN(tlen + DIRHDSIZ);
2300 			tlen = reclen - DIRHDSIZ;
2301 			left = NFS_DIRFRAGSIZ - blksiz;
2302 			if (reclen > left) {
2303 				dp->d_reclen += left;
2304 				uiop->uio_iov->iov_base =
2305 				    (caddr_t)uiop->uio_iov->iov_base + left;
2306 				uiop->uio_iov->iov_len -= left;
2307 				uiop->uio_resid -= left;
2308 				blksiz = 0;
2309 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2310 			}
2311 			if (reclen > uiop->uio_resid)
2312 				bigenough = 0;
2313 			if (bigenough) {
2314 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2315 				dp->d_fileno = (int)fileno;
2316 				dp->d_namlen = len;
2317 				dp->d_reclen = reclen;
2318 				dp->d_type = DT_UNKNOWN;
2319 				blksiz += dp->d_reclen;
2320 				if (blksiz == NFS_DIRFRAGSIZ)
2321 					blksiz = 0;
2322 				uiop->uio_resid -= DIRHDSIZ;
2323 				uiop->uio_iov->iov_base =
2324 				    (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2325 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2326 				nfsm_mtouio(uiop, len);
2327 				cp = uiop->uio_iov->iov_base;
2328 				tlen -= len;
2329 				*cp = '\0';	/* null terminate */
2330 				uiop->uio_iov->iov_base =
2331 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2332 				uiop->uio_iov->iov_len -= tlen;
2333 				uiop->uio_resid -= tlen;
2334 			} else
2335 				nfsm_adv(nfsm_rndup(len));
2336 			if (v3) {
2337 				nfsm_dissect(tl, u_int32_t *,
2338 				    3 * NFSX_UNSIGNED);
2339 			} else {
2340 				nfsm_dissect(tl, u_int32_t *,
2341 				    2 * NFSX_UNSIGNED);
2342 			}
2343 			if (bigenough) {
2344 				if (v3) {
2345 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2346 						uiop->uio_offset =
2347 						    fxdr_swapcookie3(tl);
2348 					else
2349 						uiop->uio_offset =
2350 						    fxdr_cookie3(tl);
2351 				}
2352 				else {
2353 					uiop->uio_offset =
2354 					    fxdr_unsigned(off_t, *tl);
2355 				}
2356 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2357 			}
2358 			if (v3)
2359 				tl += 2;
2360 			else
2361 				tl++;
2362 			more_dirs = fxdr_unsigned(int, *tl);
2363 		}
2364 		/*
2365 		 * If at end of rpc data, get the eof boolean
2366 		 */
2367 		if (!more_dirs) {
2368 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2369 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2370 		}
2371 		m_freem(mrep);
2372 	}
2373 	/*
2374 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2375 	 * by increasing d_reclen for the last record.
2376 	 */
2377 	if (blksiz > 0) {
2378 		left = NFS_DIRFRAGSIZ - blksiz;
2379 		dp->d_reclen += left;
2380 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2381 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2382 		    left;
2383 		uiop->uio_iov->iov_len -= left;
2384 		uiop->uio_resid -= left;
2385 	}
2386 
2387 	/*
2388 	 * We are now either at the end of the directory or have filled the
2389 	 * block.
2390 	 */
2391 	if (bigenough)
2392 		dnp->n_direofoffset = uiop->uio_offset;
2393 nfsmout:
2394 	return (error);
2395 }
2396 
2397 /*
2398  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2399  */
2400 int
2401 nfs_readdirplusrpc(vp, uiop, cred)
2402 	struct vnode *vp;
2403 	struct uio *uiop;
2404 	struct ucred *cred;
2405 {
2406 	int len, left;
2407 	struct dirent *dp = NULL;
2408 	u_int32_t *tl;
2409 	caddr_t cp;
2410 	int32_t t1, t2;
2411 	struct vnode *newvp;
2412 	caddr_t bpos, dpos, cp2;
2413 	struct mbuf *mreq, *mrep, *md, *mb;
2414 	struct nameidata nami, *ndp = &nami;
2415 	struct componentname *cnp = &ndp->ni_cnd;
2416 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2417 	struct nfsnode *dnp = VTONFS(vp), *np;
2418 	nfsfh_t *fhp;
2419 	u_quad_t fileno;
2420 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2421 	int attrflag, fhsize, nrpcs = 0, reclen;
2422 	struct nfs_fattr fattr, *fp;
2423 
2424 #ifdef DIAGNOSTIC
2425 	if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2426 		panic("nfs readdirplusrpc bad uio");
2427 #endif
2428 	ndp->ni_dvp = vp;
2429 	newvp = NULLVP;
2430 
2431 	/*
2432 	 * Loop around doing readdir rpc's of size nm_readdirsize
2433 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2434 	 * The stopping criteria is EOF or buffer full.
2435 	 */
2436 	while (more_dirs && bigenough) {
2437 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2438 			bigenough = 0;
2439 			break;
2440 		}
2441 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2442 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2443 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2444 		nfsm_fhtom(dnp, 1);
2445  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2446 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2447 			txdr_swapcookie3(uiop->uio_offset, tl);
2448 		} else {
2449 			txdr_cookie3(uiop->uio_offset, tl);
2450 		}
2451 		tl += 2;
2452 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2453 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2454 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2455 		*tl = txdr_unsigned(nmp->nm_rsize);
2456 		nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2457 		nfsm_postop_attr(vp, attrflag, 0);
2458 		if (error) {
2459 			m_freem(mrep);
2460 			goto nfsmout;
2461 		}
2462 		nrpcs++;
2463 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2464 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2465 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2466 		more_dirs = fxdr_unsigned(int, *tl);
2467 
2468 		/* loop thru the dir entries, doctoring them to 4bsd form */
2469 		while (more_dirs && bigenough) {
2470 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2471 			fileno = fxdr_hyper(tl);
2472 			len = fxdr_unsigned(int, *(tl + 2));
2473 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2474 				error = EBADRPC;
2475 				m_freem(mrep);
2476 				goto nfsmout;
2477 			}
2478 			tlen = nfsm_rndup(len);
2479 			if (tlen == len)
2480 				tlen += 4;	/* To ensure null termination*/
2481 			tlen += sizeof (off_t) + sizeof (int);
2482 			reclen = ALIGN(tlen + DIRHDSIZ);
2483 			tlen = reclen - DIRHDSIZ;
2484 			left = NFS_DIRFRAGSIZ - blksiz;
2485 			if (reclen > left) {
2486 				/*
2487 				 * DIRFRAGSIZ is aligned, no need to align
2488 				 * again here.
2489 				 */
2490 				dp->d_reclen += left;
2491 				uiop->uio_iov->iov_base =
2492 				    (caddr_t)uiop->uio_iov->iov_base + left;
2493 				uiop->uio_iov->iov_len -= left;
2494 				uiop->uio_resid -= left;
2495 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2496 				blksiz = 0;
2497 			}
2498 			if (reclen > uiop->uio_resid)
2499 				bigenough = 0;
2500 			if (bigenough) {
2501 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2502 				dp->d_fileno = (int)fileno;
2503 				dp->d_namlen = len;
2504 				dp->d_reclen = reclen;
2505 				dp->d_type = DT_UNKNOWN;
2506 				blksiz += dp->d_reclen;
2507 				if (blksiz == NFS_DIRFRAGSIZ)
2508 					blksiz = 0;
2509 				uiop->uio_resid -= DIRHDSIZ;
2510 				uiop->uio_iov->iov_base =
2511 				    (caddr_t)uiop->uio_iov->iov_base +
2512 				    DIRHDSIZ;
2513 				uiop->uio_iov->iov_len -= DIRHDSIZ;
2514 				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2515 				cnp->cn_namelen = len;
2516 				nfsm_mtouio(uiop, len);
2517 				cp = uiop->uio_iov->iov_base;
2518 				tlen -= len;
2519 				*cp = '\0';
2520 				uiop->uio_iov->iov_base =
2521 				    (caddr_t)uiop->uio_iov->iov_base + tlen;
2522 				uiop->uio_iov->iov_len -= tlen;
2523 				uiop->uio_resid -= tlen;
2524 			} else
2525 				nfsm_adv(nfsm_rndup(len));
2526 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2527 			if (bigenough) {
2528 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2529 					uiop->uio_offset =
2530 						fxdr_swapcookie3(tl);
2531 				else
2532 					uiop->uio_offset =
2533 						fxdr_cookie3(tl);
2534 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2535 			}
2536 			tl += 2;
2537 
2538 			/*
2539 			 * Since the attributes are before the file handle
2540 			 * (sigh), we must skip over the attributes and then
2541 			 * come back and get them.
2542 			 */
2543 			attrflag = fxdr_unsigned(int, *tl);
2544 			if (attrflag) {
2545 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2546 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2547 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2548 			    doit = fxdr_unsigned(int, *tl);
2549 			    if (doit) {
2550 				nfsm_getfh(fhp, fhsize, 1);
2551 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2552 				    VREF(vp);
2553 				    newvp = vp;
2554 				    np = dnp;
2555 				} else {
2556 				    error = nfs_nget(vp->v_mount, fhp,
2557 					fhsize, &np);
2558 				    if (!error)
2559 					newvp = NFSTOV(np);
2560 				}
2561 				if (!error) {
2562 				    const char *cp;
2563 
2564 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2565 				    dp->d_type =
2566 				        IFTODT(VTTOIF(np->n_vattr->va_type));
2567 				    ndp->ni_vp = newvp;
2568 				    cp = cnp->cn_nameptr + cnp->cn_namelen;
2569 				    cnp->cn_hash =
2570 					namei_hash(cnp->cn_nameptr, &cp);
2571 				    if (cnp->cn_namelen <= NCHNAMLEN)
2572 				        cache_enter(ndp->ni_dvp, ndp->ni_vp,
2573 						    cnp);
2574 				}
2575 			   }
2576 			} else {
2577 			    /* Just skip over the file handle */
2578 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2579 			    i = fxdr_unsigned(int, *tl);
2580 			    nfsm_adv(nfsm_rndup(i));
2581 			}
2582 			if (newvp != NULLVP) {
2583 			    if (newvp == vp)
2584 				vrele(newvp);
2585 			    else
2586 				vput(newvp);
2587 			    newvp = NULLVP;
2588 			}
2589 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2590 			more_dirs = fxdr_unsigned(int, *tl);
2591 		}
2592 		/*
2593 		 * If at end of rpc data, get the eof boolean
2594 		 */
2595 		if (!more_dirs) {
2596 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2597 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2598 		}
2599 		m_freem(mrep);
2600 	}
2601 	/*
2602 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2603 	 * by increasing d_reclen for the last record.
2604 	 */
2605 	if (blksiz > 0) {
2606 		left = NFS_DIRFRAGSIZ - blksiz;
2607 		dp->d_reclen += left;
2608 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2609 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2610 		    left;
2611 		uiop->uio_iov->iov_len -= left;
2612 		uiop->uio_resid -= left;
2613 	}
2614 
2615 	/*
2616 	 * We are now either at the end of the directory or have filled the
2617 	 * block.
2618 	 */
2619 	if (bigenough)
2620 		dnp->n_direofoffset = uiop->uio_offset;
2621 nfsmout:
2622 	if (newvp != NULLVP) {
2623 		if(newvp == vp)
2624 		    vrele(newvp);
2625 		else
2626 		    vput(newvp);
2627 	}
2628 	return (error);
2629 }
2630 static char hextoasc[] = "0123456789abcdef";
2631 
2632 /*
2633  * Silly rename. To make the NFS filesystem that is stateless look a little
2634  * more like the "ufs" a remove of an active vnode is translated to a rename
2635  * to a funny looking filename that is removed by nfs_inactive on the
2636  * nfsnode. There is the potential for another process on a different client
2637  * to create the same funny name between the nfs_lookitup() fails and the
2638  * nfs_rename() completes, but...
2639  */
2640 int
2641 nfs_sillyrename(dvp, vp, cnp)
2642 	struct vnode *dvp, *vp;
2643 	struct componentname *cnp;
2644 {
2645 	struct sillyrename *sp;
2646 	struct nfsnode *np;
2647 	int error;
2648 	short pid;
2649 
2650 	cache_purge(dvp);
2651 	np = VTONFS(vp);
2652 #ifndef DIAGNOSTIC
2653 	if (vp->v_type == VDIR)
2654 		panic("nfs: sillyrename dir");
2655 #endif
2656 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2657 		M_NFSREQ, M_WAITOK);
2658 	sp->s_cred = crdup(cnp->cn_cred);
2659 	sp->s_dvp = dvp;
2660 	VREF(dvp);
2661 
2662 	/* Fudge together a funny name */
2663 	pid = cnp->cn_proc->p_pid;
2664 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2665 	sp->s_namlen = 12;
2666 	sp->s_name[8] = hextoasc[pid & 0xf];
2667 	sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2668 	sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2669 	sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2670 
2671 	/* Try lookitups until we get one that isn't there */
2672 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2673 		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2674 		sp->s_name[4]++;
2675 		if (sp->s_name[4] > 'z') {
2676 			error = EINVAL;
2677 			goto bad;
2678 		}
2679 	}
2680 	error = nfs_renameit(dvp, cnp, sp);
2681 	if (error)
2682 		goto bad;
2683 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2684 		cnp->cn_proc, &np);
2685 	np->n_sillyrename = sp;
2686 	return (0);
2687 bad:
2688 	vrele(sp->s_dvp);
2689 	crfree(sp->s_cred);
2690 	free((caddr_t)sp, M_NFSREQ);
2691 	return (error);
2692 }
2693 
2694 /*
2695  * Look up a file name and optionally either update the file handle or
2696  * allocate an nfsnode, depending on the value of npp.
2697  * npp == NULL	--> just do the lookup
2698  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2699  *			handled too
2700  * *npp != NULL --> update the file handle in the vnode
2701  */
2702 int
2703 nfs_lookitup(dvp, name, len, cred, procp, npp)
2704 	struct vnode *dvp;
2705 	const char *name;
2706 	int len;
2707 	struct ucred *cred;
2708 	struct proc *procp;
2709 	struct nfsnode **npp;
2710 {
2711 	u_int32_t *tl;
2712 	caddr_t cp;
2713 	int32_t t1, t2;
2714 	struct vnode *newvp = (struct vnode *)0;
2715 	struct nfsnode *np, *dnp = VTONFS(dvp);
2716 	caddr_t bpos, dpos, cp2;
2717 	int error = 0, fhlen, attrflag;
2718 	struct mbuf *mreq, *mrep, *md, *mb;
2719 	nfsfh_t *nfhp;
2720 	const int v3 = NFS_ISV3(dvp);
2721 
2722 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2723 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2724 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2725 	nfsm_fhtom(dnp, v3);
2726 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2727 	nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
2728 	if (npp && !error) {
2729 		nfsm_getfh(nfhp, fhlen, v3);
2730 		if (*npp) {
2731 		    np = *npp;
2732 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2733 			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2734 			np->n_fhp = &np->n_fh;
2735 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2736 			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2737 		    memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2738 		    np->n_fhsize = fhlen;
2739 		    newvp = NFSTOV(np);
2740 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2741 		    VREF(dvp);
2742 		    newvp = dvp;
2743 		} else {
2744 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2745 		    if (error) {
2746 			m_freem(mrep);
2747 			return (error);
2748 		    }
2749 		    newvp = NFSTOV(np);
2750 		}
2751 		if (v3) {
2752 			nfsm_postop_attr(newvp, attrflag, 0);
2753 			if (!attrflag && *npp == NULL) {
2754 				m_freem(mrep);
2755 				vput(newvp);
2756 				return (ENOENT);
2757 			}
2758 		} else
2759 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
2760 	}
2761 	nfsm_reqdone;
2762 	if (npp && *npp == NULL) {
2763 		if (error) {
2764 			if (newvp)
2765 				vput(newvp);
2766 		} else
2767 			*npp = np;
2768 	}
2769 	return (error);
2770 }
2771 
2772 /*
2773  * Nfs Version 3 commit rpc
2774  */
2775 int
2776 nfs_commit(vp, offset, cnt, procp)
2777 	struct vnode *vp;
2778 	off_t offset;
2779 	uint32_t cnt;
2780 	struct proc *procp;
2781 {
2782 	caddr_t cp;
2783 	u_int32_t *tl;
2784 	int32_t t1, t2;
2785 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2786 	caddr_t bpos, dpos, cp2;
2787 	int error = 0, wccflag = NFSV3_WCCRATTR;
2788 	struct mbuf *mreq, *mrep, *md, *mb;
2789 	struct nfsnode *np;
2790 
2791 	KASSERT(NFS_ISV3(vp));
2792 
2793 #ifdef NFS_DEBUG_COMMIT
2794 	printf("commit %lu - %lu\n", (unsigned long)offset,
2795 	    (unsigned long)(offset + cnt));
2796 #endif
2797 
2798 	simple_lock(&nmp->nm_slock);
2799 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
2800 		simple_unlock(&nmp->nm_slock);
2801 		return (0);
2802 	}
2803 	simple_unlock(&nmp->nm_slock);
2804 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2805 	np = VTONFS(vp);
2806 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
2807 	nfsm_fhtom(np, 1);
2808 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2809 	txdr_hyper(offset, tl);
2810 	tl += 2;
2811 	*tl = txdr_unsigned(cnt);
2812 	nfsm_request(np, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2813 	nfsm_wcc_data(vp, wccflag, 0);
2814 	if (!error) {
2815 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2816 		simple_lock(&nmp->nm_slock);
2817 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
2818 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
2819 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
2820 			error = NFSERR_STALEWRITEVERF;
2821 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
2822 		}
2823 		simple_unlock(&nmp->nm_slock);
2824 	}
2825 	nfsm_reqdone;
2826 	return (error);
2827 }
2828 
2829 /*
2830  * Kludge City..
2831  * - make nfs_bmap() essentially a no-op that does no translation
2832  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2833  *   (Maybe I could use the process's page mapping, but I was concerned that
2834  *    Kernel Write might not be enabled and also figured copyout() would do
2835  *    a lot more work than memcpy() and also it currently happens in the
2836  *    context of the swapper process (2).
2837  */
2838 int
2839 nfs_bmap(v)
2840 	void *v;
2841 {
2842 	struct vop_bmap_args /* {
2843 		struct vnode *a_vp;
2844 		daddr_t  a_bn;
2845 		struct vnode **a_vpp;
2846 		daddr_t *a_bnp;
2847 		int *a_runp;
2848 	} */ *ap = v;
2849 	struct vnode *vp = ap->a_vp;
2850 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2851 
2852 	if (ap->a_vpp != NULL)
2853 		*ap->a_vpp = vp;
2854 	if (ap->a_bnp != NULL)
2855 		*ap->a_bnp = ap->a_bn << bshift;
2856 	if (ap->a_runp != NULL)
2857 		*ap->a_runp = 1024 * 1024; /* XXX */
2858 	return (0);
2859 }
2860 
2861 /*
2862  * Strategy routine.
2863  * For async requests when nfsiod(s) are running, queue the request by
2864  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2865  * request.
2866  */
2867 int
2868 nfs_strategy(v)
2869 	void *v;
2870 {
2871 	struct vop_strategy_args *ap = v;
2872 	struct buf *bp = ap->a_bp;
2873 	struct proc *p;
2874 	int error = 0;
2875 
2876 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2877 		panic("nfs physio/async");
2878 	if (bp->b_flags & B_ASYNC)
2879 		p = NULL;
2880 	else
2881 		p = curproc;	/* XXX */
2882 
2883 	/*
2884 	 * If the op is asynchronous and an i/o daemon is waiting
2885 	 * queue the request, wake it up and wait for completion
2886 	 * otherwise just do it ourselves.
2887 	 */
2888 
2889 	if ((bp->b_flags & B_ASYNC) == 0 ||
2890 	    nfs_asyncio(bp))
2891 		error = nfs_doio(bp, p);
2892 	return (error);
2893 }
2894 
2895 /*
2896  * fsync vnode op. Just call nfs_flush() with commit == 1.
2897  */
2898 /* ARGSUSED */
2899 int
2900 nfs_fsync(v)
2901 	void *v;
2902 {
2903 	struct vop_fsync_args /* {
2904 		struct vnodeop_desc *a_desc;
2905 		struct vnode * a_vp;
2906 		struct ucred * a_cred;
2907 		int  a_flags;
2908 		off_t offlo;
2909 		off_t offhi;
2910 		struct proc * a_p;
2911 	} */ *ap = v;
2912 
2913 	return (nfs_flush(ap->a_vp, ap->a_cred,
2914 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
2915 }
2916 
2917 /*
2918  * Flush all the data associated with a vnode.
2919  */
2920 int
2921 nfs_flush(vp, cred, waitfor, p, commit)
2922 	struct vnode *vp;
2923 	struct ucred *cred;
2924 	int waitfor;
2925 	struct proc *p;
2926 	int commit;
2927 {
2928 	struct nfsnode *np = VTONFS(vp);
2929 	int error;
2930 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
2931 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
2932 
2933 	simple_lock(&vp->v_interlock);
2934 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
2935 	if (np->n_flag & NWRITEERR) {
2936 		error = np->n_error;
2937 		np->n_flag &= ~NWRITEERR;
2938 	}
2939 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
2940 	return (error);
2941 }
2942 
2943 /*
2944  * Return POSIX pathconf information applicable to nfs.
2945  *
2946  * N.B. The NFS V2 protocol doesn't support this RPC.
2947  */
2948 /* ARGSUSED */
2949 int
2950 nfs_pathconf(v)
2951 	void *v;
2952 {
2953 	struct vop_pathconf_args /* {
2954 		struct vnode *a_vp;
2955 		int a_name;
2956 		register_t *a_retval;
2957 	} */ *ap = v;
2958 	struct nfsv3_pathconf *pcp;
2959 	struct vnode *vp = ap->a_vp;
2960 	struct mbuf *mreq, *mrep, *md, *mb;
2961 	int32_t t1, t2;
2962 	u_int32_t *tl;
2963 	caddr_t bpos, dpos, cp, cp2;
2964 	int error = 0, attrflag;
2965 #ifndef NFS_V2_ONLY
2966 	struct nfsmount *nmp;
2967 	unsigned int l;
2968 	u_int64_t maxsize;
2969 #endif
2970 	const int v3 = NFS_ISV3(vp);
2971 	struct nfsnode *np = VTONFS(vp);
2972 
2973 	switch (ap->a_name) {
2974 		/* Names that can be resolved locally. */
2975 	case _PC_PIPE_BUF:
2976 		*ap->a_retval = PIPE_BUF;
2977 		break;
2978 	case _PC_SYNC_IO:
2979 		*ap->a_retval = 1;
2980 		break;
2981 	/* Names that cannot be resolved locally; do an RPC, if possible. */
2982 	case _PC_LINK_MAX:
2983 	case _PC_NAME_MAX:
2984 	case _PC_CHOWN_RESTRICTED:
2985 	case _PC_NO_TRUNC:
2986 		if (!v3) {
2987 			error = EINVAL;
2988 			break;
2989 		}
2990 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
2991 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
2992 		nfsm_fhtom(np, 1);
2993 		nfsm_request(np, NFSPROC_PATHCONF,
2994 		    curproc, curproc->p_ucred);	/* XXX */
2995 		nfsm_postop_attr(vp, attrflag, 0);
2996 		if (!error) {
2997 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
2998 			    NFSX_V3PATHCONF);
2999 			switch (ap->a_name) {
3000 			case _PC_LINK_MAX:
3001 				*ap->a_retval =
3002 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3003 				break;
3004 			case _PC_NAME_MAX:
3005 				*ap->a_retval =
3006 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3007 				break;
3008 			case _PC_CHOWN_RESTRICTED:
3009 				*ap->a_retval =
3010 				    (pcp->pc_chownrestricted == nfs_true);
3011 				break;
3012 			case _PC_NO_TRUNC:
3013 				*ap->a_retval =
3014 				    (pcp->pc_notrunc == nfs_true);
3015 				break;
3016 			}
3017 		}
3018 		nfsm_reqdone;
3019 		break;
3020 	case _PC_FILESIZEBITS:
3021 #ifndef NFS_V2_ONLY
3022 		if (v3) {
3023 			nmp = VFSTONFS(vp->v_mount);
3024 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3025 				if ((error = nfs_fsinfo(nmp, vp,
3026 				    curproc->p_ucred, curproc)) != 0) /* XXX */
3027 					break;
3028 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3029 			    (maxsize >> l) > 0; l++)
3030 				;
3031 			*ap->a_retval = l + 1;
3032 		} else
3033 #endif
3034 		{
3035 			*ap->a_retval = 32;	/* NFS V2 limitation */
3036 		}
3037 		break;
3038 	default:
3039 		error = EINVAL;
3040 		break;
3041 	}
3042 
3043 	return (error);
3044 }
3045 
3046 /*
3047  * NFS advisory byte-level locks.
3048  */
3049 int
3050 nfs_advlock(v)
3051 	void *v;
3052 {
3053 	struct vop_advlock_args /* {
3054 		struct vnode *a_vp;
3055 		caddr_t  a_id;
3056 		int  a_op;
3057 		struct flock *a_fl;
3058 		int  a_flags;
3059 	} */ *ap = v;
3060 	struct nfsnode *np = VTONFS(ap->a_vp);
3061 
3062 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3063 }
3064 
3065 /*
3066  * Print out the contents of an nfsnode.
3067  */
3068 int
3069 nfs_print(v)
3070 	void *v;
3071 {
3072 	struct vop_print_args /* {
3073 		struct vnode *a_vp;
3074 	} */ *ap = v;
3075 	struct vnode *vp = ap->a_vp;
3076 	struct nfsnode *np = VTONFS(vp);
3077 
3078 	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3079 	    np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3080 	if (vp->v_type == VFIFO)
3081 		fifo_printinfo(vp);
3082 	printf("\n");
3083 	return (0);
3084 }
3085 
3086 /*
3087  * NFS file truncation.
3088  */
3089 int
3090 nfs_truncate(v)
3091 	void *v;
3092 {
3093 #if 0
3094 	struct vop_truncate_args /* {
3095 		struct vnode *a_vp;
3096 		off_t a_length;
3097 		int a_flags;
3098 		struct ucred *a_cred;
3099 		struct proc *a_p;
3100 	} */ *ap = v;
3101 #endif
3102 
3103 	/* Use nfs_setattr */
3104 	return (EOPNOTSUPP);
3105 }
3106 
3107 /*
3108  * NFS update.
3109  */
3110 int
3111 nfs_update(v)
3112 	void *v;
3113 #if 0
3114 	struct vop_update_args /* {
3115 		struct vnode *a_vp;
3116 		struct timespec *a_ta;
3117 		struct timespec *a_tm;
3118 		int a_waitfor;
3119 	} */ *ap = v;
3120 #endif
3121 {
3122 
3123 	/* Use nfs_setattr */
3124 	return (EOPNOTSUPP);
3125 }
3126 
3127 /*
3128  * Just call bwrite().
3129  */
3130 int
3131 nfs_bwrite(v)
3132 	void *v;
3133 {
3134 	struct vop_bwrite_args /* {
3135 		struct vnode *a_bp;
3136 	} */ *ap = v;
3137 
3138 	return (bwrite(ap->a_bp));
3139 }
3140 
3141 /*
3142  * nfs unlock wrapper.
3143  */
3144 int
3145 nfs_unlock(void *v)
3146 {
3147 	struct vop_unlock_args /* {
3148 		struct vnode *a_vp;
3149 		int a_flags;
3150 	} */ *ap = v;
3151 	struct vnode *vp = ap->a_vp;
3152 
3153 	/*
3154 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3155 	 * with v_data == 0.
3156 	 */
3157 	if (VTONFS(vp)) {
3158 		nfs_delayedtruncate(vp);
3159 	}
3160 
3161 	return genfs_unlock(v);
3162 }
3163 
3164 /*
3165  * nfs special file access vnode op.
3166  * Essentially just get vattr and then imitate iaccess() since the device is
3167  * local to the client.
3168  */
3169 int
3170 nfsspec_access(v)
3171 	void *v;
3172 {
3173 	struct vop_access_args /* {
3174 		struct vnode *a_vp;
3175 		int  a_mode;
3176 		struct ucred *a_cred;
3177 		struct proc *a_p;
3178 	} */ *ap = v;
3179 	struct vattr va;
3180 	struct vnode *vp = ap->a_vp;
3181 	int error;
3182 
3183 	error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3184 	if (error)
3185 		return (error);
3186 
3187         /*
3188 	 * Disallow write attempts on filesystems mounted read-only;
3189 	 * unless the file is a socket, fifo, or a block or character
3190 	 * device resident on the filesystem.
3191 	 */
3192 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3193 		switch (vp->v_type) {
3194 		case VREG:
3195 		case VDIR:
3196 		case VLNK:
3197 			return (EROFS);
3198 		default:
3199 			break;
3200 		}
3201 	}
3202 
3203 	return (vaccess(va.va_type, va.va_mode,
3204 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3205 }
3206 
3207 /*
3208  * Read wrapper for special devices.
3209  */
3210 int
3211 nfsspec_read(v)
3212 	void *v;
3213 {
3214 	struct vop_read_args /* {
3215 		struct vnode *a_vp;
3216 		struct uio *a_uio;
3217 		int  a_ioflag;
3218 		struct ucred *a_cred;
3219 	} */ *ap = v;
3220 	struct nfsnode *np = VTONFS(ap->a_vp);
3221 
3222 	/*
3223 	 * Set access flag.
3224 	 */
3225 	np->n_flag |= NACC;
3226 	np->n_atim.tv_sec = time.tv_sec;
3227 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3228 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3229 }
3230 
3231 /*
3232  * Write wrapper for special devices.
3233  */
3234 int
3235 nfsspec_write(v)
3236 	void *v;
3237 {
3238 	struct vop_write_args /* {
3239 		struct vnode *a_vp;
3240 		struct uio *a_uio;
3241 		int  a_ioflag;
3242 		struct ucred *a_cred;
3243 	} */ *ap = v;
3244 	struct nfsnode *np = VTONFS(ap->a_vp);
3245 
3246 	/*
3247 	 * Set update flag.
3248 	 */
3249 	np->n_flag |= NUPD;
3250 	np->n_mtim.tv_sec = time.tv_sec;
3251 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3252 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3253 }
3254 
3255 /*
3256  * Close wrapper for special devices.
3257  *
3258  * Update the times on the nfsnode then do device close.
3259  */
3260 int
3261 nfsspec_close(v)
3262 	void *v;
3263 {
3264 	struct vop_close_args /* {
3265 		struct vnode *a_vp;
3266 		int  a_fflag;
3267 		struct ucred *a_cred;
3268 		struct proc *a_p;
3269 	} */ *ap = v;
3270 	struct vnode *vp = ap->a_vp;
3271 	struct nfsnode *np = VTONFS(vp);
3272 	struct vattr vattr;
3273 
3274 	if (np->n_flag & (NACC | NUPD)) {
3275 		np->n_flag |= NCHG;
3276 		if (vp->v_usecount == 1 &&
3277 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3278 			VATTR_NULL(&vattr);
3279 			if (np->n_flag & NACC)
3280 				vattr.va_atime = np->n_atim;
3281 			if (np->n_flag & NUPD)
3282 				vattr.va_mtime = np->n_mtim;
3283 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3284 		}
3285 	}
3286 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3287 }
3288 
3289 /*
3290  * Read wrapper for fifos.
3291  */
3292 int
3293 nfsfifo_read(v)
3294 	void *v;
3295 {
3296 	struct vop_read_args /* {
3297 		struct vnode *a_vp;
3298 		struct uio *a_uio;
3299 		int  a_ioflag;
3300 		struct ucred *a_cred;
3301 	} */ *ap = v;
3302 	struct nfsnode *np = VTONFS(ap->a_vp);
3303 
3304 	/*
3305 	 * Set access flag.
3306 	 */
3307 	np->n_flag |= NACC;
3308 	np->n_atim.tv_sec = time.tv_sec;
3309 	np->n_atim.tv_nsec = time.tv_usec * 1000;
3310 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3311 }
3312 
3313 /*
3314  * Write wrapper for fifos.
3315  */
3316 int
3317 nfsfifo_write(v)
3318 	void *v;
3319 {
3320 	struct vop_write_args /* {
3321 		struct vnode *a_vp;
3322 		struct uio *a_uio;
3323 		int  a_ioflag;
3324 		struct ucred *a_cred;
3325 	} */ *ap = v;
3326 	struct nfsnode *np = VTONFS(ap->a_vp);
3327 
3328 	/*
3329 	 * Set update flag.
3330 	 */
3331 	np->n_flag |= NUPD;
3332 	np->n_mtim.tv_sec = time.tv_sec;
3333 	np->n_mtim.tv_nsec = time.tv_usec * 1000;
3334 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3335 }
3336 
3337 /*
3338  * Close wrapper for fifos.
3339  *
3340  * Update the times on the nfsnode then do fifo close.
3341  */
3342 int
3343 nfsfifo_close(v)
3344 	void *v;
3345 {
3346 	struct vop_close_args /* {
3347 		struct vnode *a_vp;
3348 		int  a_fflag;
3349 		struct ucred *a_cred;
3350 		struct proc *a_p;
3351 	} */ *ap = v;
3352 	struct vnode *vp = ap->a_vp;
3353 	struct nfsnode *np = VTONFS(vp);
3354 	struct vattr vattr;
3355 
3356 	if (np->n_flag & (NACC | NUPD)) {
3357 		if (np->n_flag & NACC) {
3358 			np->n_atim.tv_sec = time.tv_sec;
3359 			np->n_atim.tv_nsec = time.tv_usec * 1000;
3360 		}
3361 		if (np->n_flag & NUPD) {
3362 			np->n_mtim.tv_sec = time.tv_sec;
3363 			np->n_mtim.tv_nsec = time.tv_usec * 1000;
3364 		}
3365 		np->n_flag |= NCHG;
3366 		if (vp->v_usecount == 1 &&
3367 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3368 			VATTR_NULL(&vattr);
3369 			if (np->n_flag & NACC)
3370 				vattr.va_atime = np->n_atim;
3371 			if (np->n_flag & NUPD)
3372 				vattr.va_mtime = np->n_mtim;
3373 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3374 		}
3375 	}
3376 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3377 }
3378