xref: /netbsd-src/sys/nfs/nfs_vnops.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: nfs_vnops.c,v 1.266 2008/02/13 09:51:37 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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_vnops.c	8.19 (Berkeley) 7/31/95
35  */
36 
37 /*
38  * vnode op calls for Sun NFS version 2 and 3
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.266 2008/02/13 09:51:37 yamt Exp $");
43 
44 #include "opt_inet.h"
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 
48 #include <sys/param.h>
49 #include <sys/proc.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/resourcevar.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/condvar.h>
56 #include <sys/disk.h>
57 #include <sys/malloc.h>
58 #include <sys/kmem.h>
59 #include <sys/mbuf.h>
60 #include <sys/mutex.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 #include <sys/kauth.h>
70 
71 #include <uvm/uvm_extern.h>
72 #include <uvm/uvm.h>
73 
74 #include <miscfs/fifofs/fifo.h>
75 #include <miscfs/genfs/genfs.h>
76 #include <miscfs/genfs/genfs_node.h>
77 #include <miscfs/specfs/specdev.h>
78 
79 #include <nfs/rpcv2.h>
80 #include <nfs/nfsproto.h>
81 #include <nfs/nfs.h>
82 #include <nfs/nfsnode.h>
83 #include <nfs/nfsmount.h>
84 #include <nfs/xdr_subs.h>
85 #include <nfs/nfsm_subs.h>
86 #include <nfs/nfs_var.h>
87 
88 #include <net/if.h>
89 #include <netinet/in.h>
90 #include <netinet/in_var.h>
91 
92 /*
93  * Global vfs data structures for nfs
94  */
95 int (**nfsv2_vnodeop_p) __P((void *));
96 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
97 	{ &vop_default_desc, vn_default_error },
98 	{ &vop_lookup_desc, nfs_lookup },		/* lookup */
99 	{ &vop_create_desc, nfs_create },		/* create */
100 	{ &vop_mknod_desc, nfs_mknod },			/* mknod */
101 	{ &vop_open_desc, nfs_open },			/* open */
102 	{ &vop_close_desc, nfs_close },			/* close */
103 	{ &vop_access_desc, nfs_access },		/* access */
104 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
105 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
106 	{ &vop_read_desc, nfs_read },			/* read */
107 	{ &vop_write_desc, nfs_write },			/* write */
108 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
109 	{ &vop_ioctl_desc, nfs_ioctl },			/* ioctl */
110 	{ &vop_poll_desc, nfs_poll },			/* poll */
111 	{ &vop_kqfilter_desc, nfs_kqfilter },		/* kqfilter */
112 	{ &vop_revoke_desc, nfs_revoke },		/* revoke */
113 	{ &vop_mmap_desc, nfs_mmap },			/* mmap */
114 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
115 	{ &vop_seek_desc, nfs_seek },			/* seek */
116 	{ &vop_remove_desc, nfs_remove },		/* remove */
117 	{ &vop_link_desc, nfs_link },			/* link */
118 	{ &vop_rename_desc, nfs_rename },		/* rename */
119 	{ &vop_mkdir_desc, nfs_mkdir },			/* mkdir */
120 	{ &vop_rmdir_desc, nfs_rmdir },			/* rmdir */
121 	{ &vop_symlink_desc, nfs_symlink },		/* symlink */
122 	{ &vop_readdir_desc, nfs_readdir },		/* readdir */
123 	{ &vop_readlink_desc, nfs_readlink },		/* readlink */
124 	{ &vop_abortop_desc, nfs_abortop },		/* abortop */
125 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
126 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
127 	{ &vop_lock_desc, nfs_lock },			/* lock */
128 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
129 	{ &vop_bmap_desc, nfs_bmap },			/* bmap */
130 	{ &vop_strategy_desc, nfs_strategy },		/* strategy */
131 	{ &vop_print_desc, nfs_print },			/* print */
132 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
133 	{ &vop_pathconf_desc, nfs_pathconf },		/* pathconf */
134 	{ &vop_advlock_desc, nfs_advlock },		/* advlock */
135 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
136 	{ &vop_getpages_desc, nfs_getpages },		/* getpages */
137 	{ &vop_putpages_desc, genfs_putpages },		/* putpages */
138 	{ NULL, NULL }
139 };
140 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
141 	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
142 
143 /*
144  * Special device vnode ops
145  */
146 int (**spec_nfsv2nodeop_p) __P((void *));
147 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
148 	{ &vop_default_desc, vn_default_error },
149 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
150 	{ &vop_create_desc, spec_create },		/* create */
151 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
152 	{ &vop_open_desc, spec_open },			/* open */
153 	{ &vop_close_desc, nfsspec_close },		/* close */
154 	{ &vop_access_desc, nfsspec_access },		/* access */
155 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
156 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
157 	{ &vop_read_desc, nfsspec_read },		/* read */
158 	{ &vop_write_desc, nfsspec_write },		/* write */
159 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
160 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
161 	{ &vop_poll_desc, spec_poll },			/* poll */
162 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
163 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
164 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
165 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
166 	{ &vop_seek_desc, spec_seek },			/* seek */
167 	{ &vop_remove_desc, spec_remove },		/* remove */
168 	{ &vop_link_desc, spec_link },			/* link */
169 	{ &vop_rename_desc, spec_rename },		/* rename */
170 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
171 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
172 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
173 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
174 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
175 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
176 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
177 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
178 	{ &vop_lock_desc, nfs_lock },			/* lock */
179 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
180 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
181 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
182 	{ &vop_print_desc, nfs_print },			/* print */
183 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
184 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
185 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
186 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
187 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
188 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
189 	{ NULL, NULL }
190 };
191 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
192 	{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
193 
194 int (**fifo_nfsv2nodeop_p) __P((void *));
195 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
196 	{ &vop_default_desc, vn_default_error },
197 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
198 	{ &vop_create_desc, fifo_create },		/* create */
199 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
200 	{ &vop_open_desc, fifo_open },			/* open */
201 	{ &vop_close_desc, nfsfifo_close },		/* close */
202 	{ &vop_access_desc, nfsspec_access },		/* access */
203 	{ &vop_getattr_desc, nfs_getattr },		/* getattr */
204 	{ &vop_setattr_desc, nfs_setattr },		/* setattr */
205 	{ &vop_read_desc, nfsfifo_read },		/* read */
206 	{ &vop_write_desc, nfsfifo_write },		/* write */
207 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
208 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
209 	{ &vop_poll_desc, fifo_poll },			/* poll */
210 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
211 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
212 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
213 	{ &vop_fsync_desc, nfs_fsync },			/* fsync */
214 	{ &vop_seek_desc, fifo_seek },			/* seek */
215 	{ &vop_remove_desc, fifo_remove },		/* remove */
216 	{ &vop_link_desc, fifo_link },			/* link */
217 	{ &vop_rename_desc, fifo_rename },		/* rename */
218 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
219 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
220 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
221 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
222 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
223 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
224 	{ &vop_inactive_desc, nfs_inactive },		/* inactive */
225 	{ &vop_reclaim_desc, nfs_reclaim },		/* reclaim */
226 	{ &vop_lock_desc, nfs_lock },			/* lock */
227 	{ &vop_unlock_desc, nfs_unlock },		/* unlock */
228 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
229 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
230 	{ &vop_print_desc, nfs_print },			/* print */
231 	{ &vop_islocked_desc, nfs_islocked },		/* islocked */
232 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
233 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
234 	{ &vop_bwrite_desc, genfs_badop },		/* bwrite */
235 	{ &vop_putpages_desc, fifo_putpages }, 		/* putpages */
236 	{ NULL, NULL }
237 };
238 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
239 	{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
240 
241 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
242     size_t, kauth_cred_t, struct lwp *);
243 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *);
244 
245 /*
246  * Global variables
247  */
248 extern u_int32_t nfs_true, nfs_false;
249 extern u_int32_t nfs_xdrneg1;
250 extern const nfstype nfsv3_type[9];
251 
252 int nfs_numasync = 0;
253 #define	DIRHDSIZ	_DIRENT_NAMEOFF(dp)
254 #define UIO_ADVANCE(uio, siz) \
255     (void)((uio)->uio_resid -= (siz), \
256     (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
257     (uio)->uio_iov->iov_len -= (siz))
258 
259 static void nfs_cache_enter(struct vnode *, struct vnode *,
260     struct componentname *);
261 
262 static void
263 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
264     struct componentname *cnp)
265 {
266 	struct nfsnode *dnp = VTONFS(dvp);
267 
268 	if (vp != NULL) {
269 		struct nfsnode *np = VTONFS(vp);
270 
271 		np->n_ctime = np->n_vattr->va_ctime.tv_sec;
272 	}
273 
274 	if (!timespecisset(&dnp->n_nctime))
275 		dnp->n_nctime = dnp->n_vattr->va_mtime;
276 
277 	cache_enter(dvp, vp, cnp);
278 }
279 
280 /*
281  * nfs null call from vfs.
282  */
283 int
284 nfs_null(vp, cred, l)
285 	struct vnode *vp;
286 	kauth_cred_t cred;
287 	struct lwp *l;
288 {
289 	char *bpos, *dpos;
290 	int error = 0;
291 	struct mbuf *mreq, *mrep, *md, *mb;
292 	struct nfsnode *np = VTONFS(vp);
293 
294 	nfsm_reqhead(np, NFSPROC_NULL, 0);
295 	nfsm_request(np, NFSPROC_NULL, l, cred);
296 	nfsm_reqdone;
297 	return (error);
298 }
299 
300 /*
301  * nfs access vnode op.
302  * For nfs version 2, just return ok. File accesses may fail later.
303  * For nfs version 3, use the access rpc to check accessibility. If file modes
304  * are changed on the server, accesses might still fail later.
305  */
306 int
307 nfs_access(v)
308 	void *v;
309 {
310 	struct vop_access_args /* {
311 		struct vnode *a_vp;
312 		int  a_mode;
313 		kauth_cred_t a_cred;
314 	} */ *ap = v;
315 	struct vnode *vp = ap->a_vp;
316 #ifndef NFS_V2_ONLY
317 	u_int32_t *tl;
318 	char *cp;
319 	int32_t t1, t2;
320 	char *bpos, *dpos, *cp2;
321 	int error = 0, attrflag;
322 	struct mbuf *mreq, *mrep, *md, *mb;
323 	u_int32_t mode, rmode;
324 	const int v3 = NFS_ISV3(vp);
325 #endif
326 	int cachevalid;
327 	struct nfsnode *np = VTONFS(vp);
328 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
329 
330 	cachevalid = (np->n_accstamp != -1 &&
331 	    (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
332 	    np->n_accuid == kauth_cred_geteuid(ap->a_cred));
333 
334 	/*
335 	 * Check access cache first. If this request has been made for this
336 	 * uid shortly before, use the cached result.
337 	 */
338 	if (cachevalid) {
339 		if (!np->n_accerror) {
340 			if  ((np->n_accmode & ap->a_mode) == ap->a_mode)
341 				return np->n_accerror;
342 		} else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
343 			return np->n_accerror;
344 	}
345 
346 #ifndef NFS_V2_ONLY
347 	/*
348 	 * For nfs v3, do an access rpc, otherwise you are stuck emulating
349 	 * ufs_access() locally using the vattr. This may not be correct,
350 	 * since the server may apply other access criteria such as
351 	 * client uid-->server uid mapping that we do not know about, but
352 	 * this is better than just returning anything that is lying about
353 	 * in the cache.
354 	 */
355 	if (v3) {
356 		nfsstats.rpccnt[NFSPROC_ACCESS]++;
357 		nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
358 		nfsm_fhtom(np, v3);
359 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
360 		if (ap->a_mode & VREAD)
361 			mode = NFSV3ACCESS_READ;
362 		else
363 			mode = 0;
364 		if (vp->v_type != VDIR) {
365 			if (ap->a_mode & VWRITE)
366 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
367 			if (ap->a_mode & VEXEC)
368 				mode |= NFSV3ACCESS_EXECUTE;
369 		} else {
370 			if (ap->a_mode & VWRITE)
371 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
372 					 NFSV3ACCESS_DELETE);
373 			if (ap->a_mode & VEXEC)
374 				mode |= NFSV3ACCESS_LOOKUP;
375 		}
376 		*tl = txdr_unsigned(mode);
377 		nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
378 		nfsm_postop_attr(vp, attrflag, 0);
379 		if (!error) {
380 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
381 			rmode = fxdr_unsigned(u_int32_t, *tl);
382 			/*
383 			 * The NFS V3 spec does not clarify whether or not
384 			 * the returned access bits can be a superset of
385 			 * the ones requested, so...
386 			 */
387 			if ((rmode & mode) != mode)
388 				error = EACCES;
389 		}
390 		nfsm_reqdone;
391 	} else
392 #endif
393 		return (nfsspec_access(ap));
394 #ifndef NFS_V2_ONLY
395 	/*
396 	 * Disallow write attempts on filesystems mounted read-only;
397 	 * unless the file is a socket, fifo, or a block or character
398 	 * device resident on the filesystem.
399 	 */
400 	if (!error && (ap->a_mode & VWRITE) &&
401 	    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
402 		switch (vp->v_type) {
403 		case VREG:
404 		case VDIR:
405 		case VLNK:
406 			error = EROFS;
407 		default:
408 			break;
409 		}
410 	}
411 
412 	if (!error || error == EACCES) {
413 		/*
414 		 * If we got the same result as for a previous,
415 		 * different request, OR it in. Don't update
416 		 * the timestamp in that case.
417 		 */
418 		if (cachevalid && np->n_accstamp != -1 &&
419 		    error == np->n_accerror) {
420 			if (!error)
421 				np->n_accmode |= ap->a_mode;
422 			else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
423 				np->n_accmode = ap->a_mode;
424 		} else {
425 			np->n_accstamp = time_uptime;
426 			np->n_accuid = kauth_cred_geteuid(ap->a_cred);
427 			np->n_accmode = ap->a_mode;
428 			np->n_accerror = error;
429 		}
430 	}
431 
432 	return (error);
433 #endif
434 }
435 
436 /*
437  * nfs open vnode op
438  * Check to see if the type is ok
439  * and that deletion is not in progress.
440  * For paged in text files, you will need to flush the page cache
441  * if consistency is lost.
442  */
443 /* ARGSUSED */
444 int
445 nfs_open(v)
446 	void *v;
447 {
448 	struct vop_open_args /* {
449 		struct vnode *a_vp;
450 		int  a_mode;
451 		kauth_cred_t a_cred;
452 	} */ *ap = v;
453 	struct vnode *vp = ap->a_vp;
454 	struct nfsnode *np = VTONFS(vp);
455 	int error;
456 
457 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
458 		return (EACCES);
459 	}
460 
461 	if (ap->a_mode & FREAD) {
462 		if (np->n_rcred != NULL)
463 			kauth_cred_free(np->n_rcred);
464 		np->n_rcred = ap->a_cred;
465 		kauth_cred_hold(np->n_rcred);
466 	}
467 	if (ap->a_mode & FWRITE) {
468 		if (np->n_wcred != NULL)
469 			kauth_cred_free(np->n_wcred);
470 		np->n_wcred = ap->a_cred;
471 		kauth_cred_hold(np->n_wcred);
472 	}
473 
474 	error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
475 	if (error)
476 		return error;
477 
478 	NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
479 
480 	return (0);
481 }
482 
483 /*
484  * nfs close vnode op
485  * What an NFS client should do upon close after writing is a debatable issue.
486  * Most NFS clients push delayed writes to the server upon close, basically for
487  * two reasons:
488  * 1 - So that any write errors may be reported back to the client process
489  *     doing the close system call. By far the two most likely errors are
490  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
491  * 2 - To put a worst case upper bound on cache inconsistency between
492  *     multiple clients for the file.
493  * There is also a consistency problem for Version 2 of the protocol w.r.t.
494  * not being able to tell if other clients are writing a file concurrently,
495  * since there is no way of knowing if the changed modify time in the reply
496  * is only due to the write for this client.
497  * (NFS Version 3 provides weak cache consistency data in the reply that
498  *  should be sufficient to detect and handle this case.)
499  *
500  * The current code does the following:
501  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
502  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
503  *                     or commit them (this satisfies 1 and 2 except for the
504  *                     case where the server crashes after this close but
505  *                     before the commit RPC, which is felt to be "good
506  *                     enough". Changing the last argument to nfs_flush() to
507  *                     a 1 would force a commit operation, if it is felt a
508  *                     commit is necessary now.
509  */
510 /* ARGSUSED */
511 int
512 nfs_close(v)
513 	void *v;
514 {
515 	struct vop_close_args /* {
516 		struct vnodeop_desc *a_desc;
517 		struct vnode *a_vp;
518 		int  a_fflag;
519 		kauth_cred_t a_cred;
520 	} */ *ap = v;
521 	struct vnode *vp = ap->a_vp;
522 	struct nfsnode *np = VTONFS(vp);
523 	int error = 0;
524 	UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
525 
526 	if (vp->v_type == VREG) {
527 	    if (np->n_flag & NMODIFIED) {
528 #ifndef NFS_V2_ONLY
529 		if (NFS_ISV3(vp)) {
530 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
531 		    np->n_flag &= ~NMODIFIED;
532 		} else
533 #endif
534 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
535 		NFS_INVALIDATE_ATTRCACHE(np);
536 	    }
537 	    if (np->n_flag & NWRITEERR) {
538 		np->n_flag &= ~NWRITEERR;
539 		error = np->n_error;
540 	    }
541 	}
542 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
543 	return (error);
544 }
545 
546 /*
547  * nfs getattr call from vfs.
548  */
549 int
550 nfs_getattr(v)
551 	void *v;
552 {
553 	struct vop_getattr_args /* {
554 		struct vnode *a_vp;
555 		struct vattr *a_vap;
556 		kauth_cred_t a_cred;
557 	} */ *ap = v;
558 	struct vnode *vp = ap->a_vp;
559 	struct nfsnode *np = VTONFS(vp);
560 	char *cp;
561 	u_int32_t *tl;
562 	int32_t t1, t2;
563 	char *bpos, *dpos;
564 	int error = 0;
565 	struct mbuf *mreq, *mrep, *md, *mb;
566 	const int v3 = NFS_ISV3(vp);
567 
568 	/*
569 	 * Update local times for special files.
570 	 */
571 	if (np->n_flag & (NACC | NUPD))
572 		np->n_flag |= NCHG;
573 
574 	/*
575 	 * if we have delayed truncation, do it now.
576 	 */
577 	nfs_delayedtruncate(vp);
578 
579 	/*
580 	 * First look in the cache.
581 	 */
582 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
583 		return (0);
584 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
585 	nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
586 	nfsm_fhtom(np, v3);
587 	nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
588 	if (!error) {
589 		nfsm_loadattr(vp, ap->a_vap, 0);
590 		if (vp->v_type == VDIR &&
591 		    ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
592 			ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
593 	}
594 	nfsm_reqdone;
595 	return (error);
596 }
597 
598 /*
599  * nfs setattr call.
600  */
601 int
602 nfs_setattr(v)
603 	void *v;
604 {
605 	struct vop_setattr_args /* {
606 		struct vnodeop_desc *a_desc;
607 		struct vnode *a_vp;
608 		struct vattr *a_vap;
609 		kauth_cred_t a_cred;
610 	} */ *ap = v;
611 	struct vnode *vp = ap->a_vp;
612 	struct nfsnode *np = VTONFS(vp);
613 	struct vattr *vap = ap->a_vap;
614 	int error = 0;
615 	u_quad_t tsize = 0;
616 
617 	/*
618 	 * Setting of flags is not supported.
619 	 */
620 	if (vap->va_flags != VNOVAL)
621 		return (EOPNOTSUPP);
622 
623 	/*
624 	 * Disallow write attempts if the filesystem is mounted read-only.
625 	 */
626   	if ((vap->va_uid != (uid_t)VNOVAL ||
627 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
628 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
629 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
630 		return (EROFS);
631 	if (vap->va_size != VNOVAL) {
632 		if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
633 			return EFBIG;
634 		}
635  		switch (vp->v_type) {
636  		case VDIR:
637  			return (EISDIR);
638  		case VCHR:
639  		case VBLK:
640  		case VSOCK:
641  		case VFIFO:
642 			if (vap->va_mtime.tv_sec == VNOVAL &&
643 			    vap->va_atime.tv_sec == VNOVAL &&
644 			    vap->va_mode == (mode_t)VNOVAL &&
645 			    vap->va_uid == (uid_t)VNOVAL &&
646 			    vap->va_gid == (gid_t)VNOVAL)
647 				return (0);
648  			vap->va_size = VNOVAL;
649  			break;
650  		default:
651 			/*
652 			 * Disallow write attempts if the filesystem is
653 			 * mounted read-only.
654 			 */
655 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
656 				return (EROFS);
657 			genfs_node_wrlock(vp);
658  			uvm_vnp_setsize(vp, vap->va_size);
659  			tsize = np->n_size;
660 			np->n_size = vap->va_size;
661  			if (vap->va_size == 0)
662  				error = nfs_vinvalbuf(vp, 0,
663  				     ap->a_cred, curlwp, 1);
664 			else
665 				error = nfs_vinvalbuf(vp, V_SAVE,
666 				     ap->a_cred, curlwp, 1);
667 			if (error) {
668 				uvm_vnp_setsize(vp, tsize);
669 				genfs_node_unlock(vp);
670 				return (error);
671 			}
672  			np->n_vattr->va_size = vap->va_size;
673   		}
674   	} else {
675 		/*
676 		 * flush files before setattr because a later write of
677 		 * cached data might change timestamps or reset sugid bits
678 		 */
679 		if ((vap->va_mtime.tv_sec != VNOVAL ||
680 		     vap->va_atime.tv_sec != VNOVAL ||
681 		     vap->va_mode != VNOVAL) &&
682 		    vp->v_type == VREG &&
683   		    (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
684 		 			   curlwp, 1)) == EINTR)
685 			return (error);
686 	}
687 	error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
688 	if (vap->va_size != VNOVAL) {
689 		if (error) {
690 			np->n_size = np->n_vattr->va_size = tsize;
691 			uvm_vnp_setsize(vp, np->n_size);
692 		}
693 		genfs_node_unlock(vp);
694 	}
695 	VN_KNOTE(vp, NOTE_ATTRIB);
696 	return (error);
697 }
698 
699 /*
700  * Do an nfs setattr rpc.
701  */
702 int
703 nfs_setattrrpc(vp, vap, cred, l)
704 	struct vnode *vp;
705 	struct vattr *vap;
706 	kauth_cred_t cred;
707 	struct lwp *l;
708 {
709 	struct nfsv2_sattr *sp;
710 	char *cp;
711 	int32_t t1, t2;
712 	char *bpos, *dpos;
713 	u_int32_t *tl;
714 	int error = 0;
715 	struct mbuf *mreq, *mrep, *md, *mb;
716 	const int v3 = NFS_ISV3(vp);
717 	struct nfsnode *np = VTONFS(vp);
718 #ifndef NFS_V2_ONLY
719 	int wccflag = NFSV3_WCCRATTR;
720 	char *cp2;
721 #endif
722 
723 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
724 	nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
725 	nfsm_fhtom(np, v3);
726 #ifndef NFS_V2_ONLY
727 	if (v3) {
728 		nfsm_v3attrbuild(vap, true);
729 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
730 		*tl = nfs_false;
731 	} else {
732 #endif
733 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
734 		if (vap->va_mode == (mode_t)VNOVAL)
735 			sp->sa_mode = nfs_xdrneg1;
736 		else
737 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
738 		if (vap->va_uid == (uid_t)VNOVAL)
739 			sp->sa_uid = nfs_xdrneg1;
740 		else
741 			sp->sa_uid = txdr_unsigned(vap->va_uid);
742 		if (vap->va_gid == (gid_t)VNOVAL)
743 			sp->sa_gid = nfs_xdrneg1;
744 		else
745 			sp->sa_gid = txdr_unsigned(vap->va_gid);
746 		sp->sa_size = txdr_unsigned(vap->va_size);
747 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
748 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
749 #ifndef NFS_V2_ONLY
750 	}
751 #endif
752 	nfsm_request(np, NFSPROC_SETATTR, l, cred);
753 #ifndef NFS_V2_ONLY
754 	if (v3) {
755 		nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
756 	} else
757 #endif
758 		nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
759 	nfsm_reqdone;
760 	return (error);
761 }
762 
763 /*
764  * nfs lookup call, one step at a time...
765  * First look in cache
766  * If not found, unlock the directory nfsnode and do the rpc
767  *
768  * This code is full of lock/unlock statements and checks, because
769  * we continue after cache_lookup has finished (we need to check
770  * with the attr cache and do an rpc if it has timed out). This means
771  * that the locking effects of cache_lookup have to be taken into
772  * account.
773  */
774 int
775 nfs_lookup(v)
776 	void *v;
777 {
778 	struct vop_lookup_args /* {
779 		struct vnodeop_desc *a_desc;
780 		struct vnode *a_dvp;
781 		struct vnode **a_vpp;
782 		struct componentname *a_cnp;
783 	} */ *ap = v;
784 	struct componentname *cnp = ap->a_cnp;
785 	struct vnode *dvp = ap->a_dvp;
786 	struct vnode **vpp = ap->a_vpp;
787 	int flags;
788 	struct vnode *newvp;
789 	u_int32_t *tl;
790 	char *cp;
791 	int32_t t1, t2;
792 	char *bpos, *dpos, *cp2;
793 	struct mbuf *mreq, *mrep, *md, *mb;
794 	long len;
795 	nfsfh_t *fhp;
796 	struct nfsnode *np;
797 	int error = 0, attrflag, fhsize;
798 	const int v3 = NFS_ISV3(dvp);
799 
800 	flags = cnp->cn_flags;
801 
802 	*vpp = NULLVP;
803 	newvp = NULLVP;
804 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
805 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
806 		return (EROFS);
807 	if (dvp->v_type != VDIR)
808 		return (ENOTDIR);
809 
810 	/*
811 	 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
812 	 */
813 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
814 		error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
815 		if (error)
816 			return error;
817 		if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
818 			return EISDIR;
819 		VREF(dvp);
820 		*vpp = dvp;
821 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
822 			cnp->cn_flags |= SAVENAME;
823 		return 0;
824 	}
825 
826 	np = VTONFS(dvp);
827 
828 	/*
829 	 * Before tediously performing a linear scan of the directory,
830 	 * check the name cache to see if the directory/name pair
831 	 * we are looking for is known already.
832 	 * If the directory/name pair is found in the name cache,
833 	 * we have to ensure the directory has not changed from
834 	 * the time the cache entry has been created. If it has,
835 	 * the cache entry has to be ignored.
836 	 */
837 	error = cache_lookup_raw(dvp, vpp, cnp);
838 	KASSERT(dvp != *vpp);
839 	if (error >= 0) {
840 		struct vattr vattr;
841 		int err2;
842 
843 		if (error && error != ENOENT) {
844 			*vpp = NULLVP;
845 			return error;
846 		}
847 
848 		err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
849 		if (err2 != 0) {
850 			if (error == 0)
851 				vrele(*vpp);
852 			*vpp = NULLVP;
853 			return err2;
854 		}
855 
856 		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
857 		    || timespeccmp(&vattr.va_mtime,
858 		    &VTONFS(dvp)->n_nctime, !=)) {
859 			if (error == 0) {
860 				vrele(*vpp);
861 				*vpp = NULLVP;
862 			}
863 			cache_purge1(dvp, NULL, PURGE_CHILDREN);
864 			timespecclear(&np->n_nctime);
865 			goto dorpc;
866 		}
867 
868 		if (error == ENOENT) {
869 			goto noentry;
870 		}
871 
872 		newvp = *vpp;
873 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
874 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
875 			nfsstats.lookupcache_hits++;
876 			if ((flags & ISDOTDOT) != 0) {
877 				VOP_UNLOCK(dvp, 0);
878 			}
879 			error = vn_lock(newvp, LK_EXCLUSIVE);
880 			if ((flags & ISDOTDOT) != 0) {
881 				vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
882 			}
883 			if (error) {
884 				/* newvp has been revoked. */
885 				vrele(newvp);
886 				*vpp = NULL;
887 				return error;
888 			}
889 			if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
890 				cnp->cn_flags |= SAVENAME;
891 			KASSERT(newvp->v_type != VNON);
892 			return (0);
893 		}
894 		cache_purge1(newvp, NULL, PURGE_PARENTS);
895 		vrele(newvp);
896 		*vpp = NULLVP;
897 	}
898 dorpc:
899 #if 0
900 	/*
901 	 * because nfsv3 has the same CREATE semantics as ours,
902 	 * we don't have to perform LOOKUPs beforehand.
903 	 *
904 	 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
905 	 * XXX although we have no way to know if O_EXCL is requested or not.
906 	 */
907 
908 	if (v3 && cnp->cn_nameiop == CREATE &&
909 	    (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
910 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
911 		cnp->cn_flags |= SAVENAME;
912 		return (EJUSTRETURN);
913 	}
914 #endif /* 0 */
915 
916 	error = 0;
917 	newvp = NULLVP;
918 	nfsstats.lookupcache_misses++;
919 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
920 	len = cnp->cn_namelen;
921 	nfsm_reqhead(np, NFSPROC_LOOKUP,
922 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
923 	nfsm_fhtom(np, v3);
924 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
925 	nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
926 	if (error) {
927 		nfsm_postop_attr(dvp, attrflag, 0);
928 		m_freem(mrep);
929 		goto nfsmout;
930 	}
931 	nfsm_getfh(fhp, fhsize, v3);
932 
933 	/*
934 	 * Handle RENAME case...
935 	 */
936 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
937 		if (NFS_CMPFH(np, fhp, fhsize)) {
938 			m_freem(mrep);
939 			return (EISDIR);
940 		}
941 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
942 		if (error) {
943 			m_freem(mrep);
944 			return error;
945 		}
946 		newvp = NFSTOV(np);
947 #ifndef NFS_V2_ONLY
948 		if (v3) {
949 			nfsm_postop_attr(newvp, attrflag, 0);
950 			nfsm_postop_attr(dvp, attrflag, 0);
951 		} else
952 #endif
953 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
954 		*vpp = newvp;
955 		m_freem(mrep);
956 		cnp->cn_flags |= SAVENAME;
957 		goto validate;
958 	}
959 
960 	/*
961 	 * The postop attr handling is duplicated for each if case,
962 	 * because it should be done while dvp is locked (unlocking
963 	 * dvp is different for each case).
964 	 */
965 
966 	if (NFS_CMPFH(np, fhp, fhsize)) {
967 		/*
968 		 * "." lookup
969 		 */
970 		VREF(dvp);
971 		newvp = dvp;
972 #ifndef NFS_V2_ONLY
973 		if (v3) {
974 			nfsm_postop_attr(newvp, attrflag, 0);
975 			nfsm_postop_attr(dvp, attrflag, 0);
976 		} else
977 #endif
978 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
979 	} else if (flags & ISDOTDOT) {
980 		/*
981 		 * ".." lookup
982 		 */
983 		VOP_UNLOCK(dvp, 0);
984 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
985 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
986 		if (error) {
987 			m_freem(mrep);
988 			return error;
989 		}
990 		newvp = NFSTOV(np);
991 
992 #ifndef NFS_V2_ONLY
993 		if (v3) {
994 			nfsm_postop_attr(newvp, attrflag, 0);
995 			nfsm_postop_attr(dvp, attrflag, 0);
996 		} else
997 #endif
998 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
999 	} else {
1000 		/*
1001 		 * Other lookups.
1002 		 */
1003 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1004 		if (error) {
1005 			m_freem(mrep);
1006 			return error;
1007 		}
1008 		newvp = NFSTOV(np);
1009 #ifndef NFS_V2_ONLY
1010 		if (v3) {
1011 			nfsm_postop_attr(newvp, attrflag, 0);
1012 			nfsm_postop_attr(dvp, attrflag, 0);
1013 		} else
1014 #endif
1015 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
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 		nfs_cache_enter(dvp, newvp, cnp);
1022 	}
1023 	*vpp = newvp;
1024 	nfsm_reqdone;
1025 	if (error) {
1026 		/*
1027 		 * We get here only because of errors returned by
1028 		 * the RPC. Otherwise we'll have returned above
1029 		 * (the nfsm_* macros will jump to nfsm_reqdone
1030 		 * on error).
1031 		 */
1032 		if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1033 		    cnp->cn_nameiop != CREATE) {
1034 			nfs_cache_enter(dvp, NULL, cnp);
1035 		}
1036 		if (newvp != NULLVP) {
1037 			if (newvp == dvp) {
1038 				vrele(newvp);
1039 			} else {
1040 				vput(newvp);
1041 			}
1042 		}
1043 noentry:
1044 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1045 		    (flags & ISLASTCN) && error == ENOENT) {
1046 			if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1047 				error = EROFS;
1048 			} else {
1049 				error = EJUSTRETURN;
1050 				cnp->cn_flags |= SAVENAME;
1051 			}
1052 		}
1053 		*vpp = NULL;
1054 		return error;
1055 	}
1056 
1057 validate:
1058 	/*
1059 	 * make sure we have valid type and size.
1060 	 */
1061 
1062 	newvp = *vpp;
1063 	if (newvp->v_type == VNON) {
1064 		struct vattr vattr; /* dummy */
1065 
1066 		KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1067 		error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1068 		if (error) {
1069 			vput(newvp);
1070 			*vpp = NULL;
1071 		}
1072 	}
1073 
1074 	return error;
1075 }
1076 
1077 /*
1078  * nfs read call.
1079  * Just call nfs_bioread() to do the work.
1080  */
1081 int
1082 nfs_read(v)
1083 	void *v;
1084 {
1085 	struct vop_read_args /* {
1086 		struct vnode *a_vp;
1087 		struct uio *a_uio;
1088 		int  a_ioflag;
1089 		kauth_cred_t a_cred;
1090 	} */ *ap = v;
1091 	struct vnode *vp = ap->a_vp;
1092 
1093 	if (vp->v_type != VREG)
1094 		return EISDIR;
1095 	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1096 }
1097 
1098 /*
1099  * nfs readlink call
1100  */
1101 int
1102 nfs_readlink(v)
1103 	void *v;
1104 {
1105 	struct vop_readlink_args /* {
1106 		struct vnode *a_vp;
1107 		struct uio *a_uio;
1108 		kauth_cred_t a_cred;
1109 	} */ *ap = v;
1110 	struct vnode *vp = ap->a_vp;
1111 	struct nfsnode *np = VTONFS(vp);
1112 
1113 	if (vp->v_type != VLNK)
1114 		return (EPERM);
1115 
1116 	if (np->n_rcred != NULL) {
1117 		kauth_cred_free(np->n_rcred);
1118 	}
1119 	np->n_rcred = ap->a_cred;
1120 	kauth_cred_hold(np->n_rcred);
1121 
1122 	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1123 }
1124 
1125 /*
1126  * Do a readlink rpc.
1127  * Called by nfs_doio() from below the buffer cache.
1128  */
1129 int
1130 nfs_readlinkrpc(vp, uiop, cred)
1131 	struct vnode *vp;
1132 	struct uio *uiop;
1133 	kauth_cred_t cred;
1134 {
1135 	u_int32_t *tl;
1136 	char *cp;
1137 	int32_t t1, t2;
1138 	char *bpos, *dpos, *cp2;
1139 	int error = 0;
1140 	uint32_t len;
1141 	struct mbuf *mreq, *mrep, *md, *mb;
1142 	const int v3 = NFS_ISV3(vp);
1143 	struct nfsnode *np = VTONFS(vp);
1144 #ifndef NFS_V2_ONLY
1145 	int attrflag;
1146 #endif
1147 
1148 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1149 	nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1150 	nfsm_fhtom(np, v3);
1151 	nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1152 #ifndef NFS_V2_ONLY
1153 	if (v3)
1154 		nfsm_postop_attr(vp, attrflag, 0);
1155 #endif
1156 	if (!error) {
1157 #ifndef NFS_V2_ONLY
1158 		if (v3) {
1159 			nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1160 			len = fxdr_unsigned(uint32_t, *tl);
1161 			if (len > MAXPATHLEN) {
1162 				/*
1163 				 * this pathname is too long for us.
1164 				 */
1165 				m_freem(mrep);
1166 				/* Solaris returns EINVAL. should we follow? */
1167 				error = ENAMETOOLONG;
1168 				goto nfsmout;
1169 			}
1170 		} else
1171 #endif
1172 		{
1173 			nfsm_strsiz(len, NFS_MAXPATHLEN);
1174 		}
1175 		nfsm_mtouio(uiop, len);
1176 	}
1177 	nfsm_reqdone;
1178 	return (error);
1179 }
1180 
1181 /*
1182  * nfs read rpc call
1183  * Ditto above
1184  */
1185 int
1186 nfs_readrpc(vp, uiop)
1187 	struct vnode *vp;
1188 	struct uio *uiop;
1189 {
1190 	u_int32_t *tl;
1191 	char *cp;
1192 	int32_t t1, t2;
1193 	char *bpos, *dpos, *cp2;
1194 	struct mbuf *mreq, *mrep, *md, *mb;
1195 	struct nfsmount *nmp;
1196 	int error = 0, len, retlen, tsiz, eof, byte_count;
1197 	const int v3 = NFS_ISV3(vp);
1198 	struct nfsnode *np = VTONFS(vp);
1199 #ifndef NFS_V2_ONLY
1200 	int attrflag;
1201 #endif
1202 
1203 #ifndef nolint
1204 	eof = 0;
1205 #endif
1206 	nmp = VFSTONFS(vp->v_mount);
1207 	tsiz = uiop->uio_resid;
1208 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1209 		return (EFBIG);
1210 	iostat_busy(nmp->nm_stats);
1211 	byte_count = 0; /* count bytes actually transferred */
1212 	while (tsiz > 0) {
1213 		nfsstats.rpccnt[NFSPROC_READ]++;
1214 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1215 		nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1216 		nfsm_fhtom(np, v3);
1217 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1218 #ifndef NFS_V2_ONLY
1219 		if (v3) {
1220 			txdr_hyper(uiop->uio_offset, tl);
1221 			*(tl + 2) = txdr_unsigned(len);
1222 		} else
1223 #endif
1224 		{
1225 			*tl++ = txdr_unsigned(uiop->uio_offset);
1226 			*tl++ = txdr_unsigned(len);
1227 			*tl = 0;
1228 		}
1229 		nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1230 #ifndef NFS_V2_ONLY
1231 		if (v3) {
1232 			nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1233 			if (error) {
1234 				m_freem(mrep);
1235 				goto nfsmout;
1236 			}
1237 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1238 			eof = fxdr_unsigned(int, *(tl + 1));
1239 		} else
1240 #endif
1241 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1242 		nfsm_strsiz(retlen, nmp->nm_rsize);
1243 		nfsm_mtouio(uiop, retlen);
1244 		m_freem(mrep);
1245 		tsiz -= retlen;
1246 		byte_count += retlen;
1247 #ifndef NFS_V2_ONLY
1248 		if (v3) {
1249 			if (eof || retlen == 0)
1250 				tsiz = 0;
1251 		} else
1252 #endif
1253 		if (retlen < len)
1254 			tsiz = 0;
1255 	}
1256 nfsmout:
1257 	iostat_unbusy(nmp->nm_stats, byte_count, 1);
1258 	return (error);
1259 }
1260 
1261 struct nfs_writerpc_context {
1262 	kmutex_t nwc_lock;
1263 	kcondvar_t nwc_cv;
1264 	int nwc_mbufcount;
1265 };
1266 
1267 /*
1268  * free mbuf used to refer protected pages while write rpc call.
1269  * called at splvm.
1270  */
1271 static void
1272 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1273 {
1274 	struct nfs_writerpc_context *ctx = arg;
1275 
1276 	KASSERT(m != NULL);
1277 	KASSERT(ctx != NULL);
1278 	pool_cache_put(mb_cache, m);
1279 	mutex_enter(&ctx->nwc_lock);
1280 	if (--ctx->nwc_mbufcount == 0) {
1281 		cv_signal(&ctx->nwc_cv);
1282 	}
1283 	mutex_exit(&ctx->nwc_lock);
1284 }
1285 
1286 /*
1287  * nfs write call
1288  */
1289 int
1290 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1291 	struct vnode *vp;
1292 	struct uio *uiop;
1293 	int *iomode;
1294 	bool pageprotected;
1295 	bool *stalewriteverfp;
1296 {
1297 	u_int32_t *tl;
1298 	char *cp;
1299 	int32_t t1, t2;
1300 	char *bpos, *dpos;
1301 	struct mbuf *mreq, *mrep, *md, *mb;
1302 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1303 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1304 	const int v3 = NFS_ISV3(vp);
1305 	int committed = NFSV3WRITE_FILESYNC;
1306 	struct nfsnode *np = VTONFS(vp);
1307 	struct nfs_writerpc_context ctx;
1308 	int byte_count;
1309 	struct lwp *l = NULL;
1310 	size_t origresid;
1311 #ifndef NFS_V2_ONLY
1312 	char *cp2;
1313 	int rlen, commit;
1314 #endif
1315 
1316 	mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1317 	cv_init(&ctx.nwc_cv, "nfsmblk");
1318 	ctx.nwc_mbufcount = 1;
1319 
1320 	if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1321 		panic("writerpc readonly vp %p", vp);
1322 	}
1323 
1324 #ifdef DIAGNOSTIC
1325 	if (uiop->uio_iovcnt != 1)
1326 		panic("nfs: writerpc iovcnt > 1");
1327 #endif
1328 	tsiz = uiop->uio_resid;
1329 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1330 		return (EFBIG);
1331 	if (pageprotected) {
1332 		l = curlwp;
1333 		uvm_lwp_hold(l);
1334 	}
1335 retry:
1336 	origresid = uiop->uio_resid;
1337 	KASSERT(origresid == uiop->uio_iov->iov_len);
1338 	iostat_busy(nmp->nm_stats);
1339 	byte_count = 0; /* count of bytes actually written */
1340 	while (tsiz > 0) {
1341 		uint32_t datalen; /* data bytes need to be allocated in mbuf */
1342 		uint32_t backup;
1343 		bool stalewriteverf = false;
1344 
1345 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1346 		len = min(tsiz, nmp->nm_wsize);
1347 		datalen = pageprotected ? 0 : nfsm_rndup(len);
1348 		nfsm_reqhead(np, NFSPROC_WRITE,
1349 			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1350 		nfsm_fhtom(np, v3);
1351 #ifndef NFS_V2_ONLY
1352 		if (v3) {
1353 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1354 			txdr_hyper(uiop->uio_offset, tl);
1355 			tl += 2;
1356 			*tl++ = txdr_unsigned(len);
1357 			*tl++ = txdr_unsigned(*iomode);
1358 			*tl = txdr_unsigned(len);
1359 		} else
1360 #endif
1361 		{
1362 			u_int32_t x;
1363 
1364 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1365 			/* Set both "begin" and "current" to non-garbage. */
1366 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1367 			*tl++ = x;      /* "begin offset" */
1368 			*tl++ = x;      /* "current offset" */
1369 			x = txdr_unsigned(len);
1370 			*tl++ = x;      /* total to this offset */
1371 			*tl = x;        /* size of this write */
1372 
1373 		}
1374 		if (pageprotected) {
1375 			/*
1376 			 * since we know pages can't be modified during i/o,
1377 			 * no need to copy them for us.
1378 			 */
1379 			struct mbuf *m;
1380 			struct iovec *iovp = uiop->uio_iov;
1381 
1382 			m = m_get(M_WAIT, MT_DATA);
1383 			MCLAIM(m, &nfs_mowner);
1384 			MEXTADD(m, iovp->iov_base, len, M_MBUF,
1385 			    nfs_writerpc_extfree, &ctx);
1386 			m->m_flags |= M_EXT_ROMAP;
1387 			m->m_len = len;
1388 			mb->m_next = m;
1389 			/*
1390 			 * no need to maintain mb and bpos here
1391 			 * because no one care them later.
1392 			 */
1393 #if 0
1394 			mb = m;
1395 			bpos = mtod(void *, mb) + mb->m_len;
1396 #endif
1397 			UIO_ADVANCE(uiop, len);
1398 			uiop->uio_offset += len;
1399 			mutex_enter(&ctx.nwc_lock);
1400 			ctx.nwc_mbufcount++;
1401 			mutex_exit(&ctx.nwc_lock);
1402 			nfs_zeropad(mb, 0, nfsm_padlen(len));
1403 		} else {
1404 			nfsm_uiotom(uiop, len);
1405 		}
1406 		nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1407 #ifndef NFS_V2_ONLY
1408 		if (v3) {
1409 			wccflag = NFSV3_WCCCHK;
1410 			nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1411 			if (!error) {
1412 				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1413 					+ NFSX_V3WRITEVERF);
1414 				rlen = fxdr_unsigned(int, *tl++);
1415 				if (rlen == 0) {
1416 					error = NFSERR_IO;
1417 					m_freem(mrep);
1418 					break;
1419 				} else if (rlen < len) {
1420 					backup = len - rlen;
1421 					UIO_ADVANCE(uiop, -backup);
1422 					uiop->uio_offset -= backup;
1423 					len = rlen;
1424 				}
1425 				commit = fxdr_unsigned(int, *tl++);
1426 
1427 				/*
1428 				 * Return the lowest committment level
1429 				 * obtained by any of the RPCs.
1430 				 */
1431 				if (committed == NFSV3WRITE_FILESYNC)
1432 					committed = commit;
1433 				else if (committed == NFSV3WRITE_DATASYNC &&
1434 					commit == NFSV3WRITE_UNSTABLE)
1435 					committed = commit;
1436 				mutex_enter(&nmp->nm_lock);
1437 				if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1438 					memcpy(nmp->nm_writeverf, tl,
1439 					    NFSX_V3WRITEVERF);
1440 					nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1441 				} else if ((nmp->nm_iflag &
1442 				    NFSMNT_STALEWRITEVERF) ||
1443 				    memcmp(tl, nmp->nm_writeverf,
1444 				    NFSX_V3WRITEVERF)) {
1445 					memcpy(nmp->nm_writeverf, tl,
1446 					    NFSX_V3WRITEVERF);
1447 					/*
1448 					 * note NFSMNT_STALEWRITEVERF
1449 					 * if we're the first thread to
1450 					 * notice it.
1451 					 */
1452 					if ((nmp->nm_iflag &
1453 					    NFSMNT_STALEWRITEVERF) == 0) {
1454 						stalewriteverf = true;
1455 						nmp->nm_iflag |=
1456 						    NFSMNT_STALEWRITEVERF;
1457 					}
1458 				}
1459 				mutex_exit(&nmp->nm_lock);
1460 			}
1461 		} else
1462 #endif
1463 			nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1464 		if (wccflag)
1465 			VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1466 		m_freem(mrep);
1467 		if (error)
1468 			break;
1469 		tsiz -= len;
1470 		byte_count += len;
1471 		if (stalewriteverf) {
1472 			*stalewriteverfp = true;
1473 			stalewriteverf = false;
1474 			if (committed == NFSV3WRITE_UNSTABLE &&
1475 			    len != origresid) {
1476 				/*
1477 				 * if our write requests weren't atomic but
1478 				 * unstable, datas in previous iterations
1479 				 * might have already been lost now.
1480 				 * then, we should resend them to nfsd.
1481 				 */
1482 				backup = origresid - tsiz;
1483 				UIO_ADVANCE(uiop, -backup);
1484 				uiop->uio_offset -= backup;
1485 				tsiz = origresid;
1486 				goto retry;
1487 			}
1488 		}
1489 	}
1490 nfsmout:
1491 	iostat_unbusy(nmp->nm_stats, byte_count, 0);
1492 	if (pageprotected) {
1493 		/*
1494 		 * wait until mbufs go away.
1495 		 * retransmitted mbufs can survive longer than rpc requests
1496 		 * themselves.
1497 		 */
1498 		mutex_enter(&ctx.nwc_lock);
1499 		ctx.nwc_mbufcount--;
1500 		while (ctx.nwc_mbufcount > 0) {
1501 			cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1502 		}
1503 		mutex_exit(&ctx.nwc_lock);
1504 		uvm_lwp_rele(l);
1505 	}
1506 	mutex_destroy(&ctx.nwc_lock);
1507 	cv_destroy(&ctx.nwc_cv);
1508 	*iomode = committed;
1509 	if (error)
1510 		uiop->uio_resid = tsiz;
1511 	return (error);
1512 }
1513 
1514 /*
1515  * nfs mknod rpc
1516  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1517  * mode set to specify the file type and the size field for rdev.
1518  */
1519 int
1520 nfs_mknodrpc(dvp, vpp, cnp, vap)
1521 	struct vnode *dvp;
1522 	struct vnode **vpp;
1523 	struct componentname *cnp;
1524 	struct vattr *vap;
1525 {
1526 	struct nfsv2_sattr *sp;
1527 	u_int32_t *tl;
1528 	char *cp;
1529 	int32_t t1, t2;
1530 	struct vnode *newvp = (struct vnode *)0;
1531 	struct nfsnode *dnp, *np;
1532 	char *cp2;
1533 	char *bpos, *dpos;
1534 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1535 	struct mbuf *mreq, *mrep, *md, *mb;
1536 	u_int32_t rdev;
1537 	const int v3 = NFS_ISV3(dvp);
1538 
1539 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1540 		rdev = txdr_unsigned(vap->va_rdev);
1541 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1542 		rdev = nfs_xdrneg1;
1543 	else {
1544 		VOP_ABORTOP(dvp, cnp);
1545 		vput(dvp);
1546 		return (EOPNOTSUPP);
1547 	}
1548 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1549 	dnp = VTONFS(dvp);
1550 	nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1551 		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1552 	nfsm_fhtom(dnp, v3);
1553 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1554 #ifndef NFS_V2_ONLY
1555 	if (v3) {
1556 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1557 		*tl++ = vtonfsv3_type(vap->va_type);
1558 		nfsm_v3attrbuild(vap, false);
1559 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1560 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1561 			*tl++ = txdr_unsigned(major(vap->va_rdev));
1562 			*tl = txdr_unsigned(minor(vap->va_rdev));
1563 		}
1564 	} else
1565 #endif
1566 	{
1567 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1568 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1569 		sp->sa_uid = nfs_xdrneg1;
1570 		sp->sa_gid = nfs_xdrneg1;
1571 		sp->sa_size = rdev;
1572 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1573 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1574 	}
1575 	nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1576 	if (!error) {
1577 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1578 		if (!gotvp) {
1579 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1580 			    cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1581 			if (!error)
1582 				newvp = NFSTOV(np);
1583 		}
1584 	}
1585 #ifndef NFS_V2_ONLY
1586 	if (v3)
1587 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1588 #endif
1589 	nfsm_reqdone;
1590 	if (error) {
1591 		if (newvp)
1592 			vput(newvp);
1593 	} else {
1594 		if (cnp->cn_flags & MAKEENTRY)
1595 			nfs_cache_enter(dvp, newvp, cnp);
1596 		*vpp = newvp;
1597 	}
1598 	PNBUF_PUT(cnp->cn_pnbuf);
1599 	VTONFS(dvp)->n_flag |= NMODIFIED;
1600 	if (!wccflag)
1601 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1602 	vput(dvp);
1603 	return (error);
1604 }
1605 
1606 /*
1607  * nfs mknod vop
1608  * just call nfs_mknodrpc() to do the work.
1609  */
1610 /* ARGSUSED */
1611 int
1612 nfs_mknod(v)
1613 	void *v;
1614 {
1615 	struct vop_mknod_args /* {
1616 		struct vnode *a_dvp;
1617 		struct vnode **a_vpp;
1618 		struct componentname *a_cnp;
1619 		struct vattr *a_vap;
1620 	} */ *ap = v;
1621 	struct vnode *dvp = ap->a_dvp;
1622 	struct componentname *cnp = ap->a_cnp;
1623 	int error;
1624 
1625 	error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1626 	VN_KNOTE(dvp, NOTE_WRITE);
1627 	if (error == 0 || error == EEXIST)
1628 		cache_purge1(dvp, cnp, 0);
1629 	return (error);
1630 }
1631 
1632 #ifndef NFS_V2_ONLY
1633 static u_long create_verf;
1634 #endif
1635 /*
1636  * nfs file create call
1637  */
1638 int
1639 nfs_create(v)
1640 	void *v;
1641 {
1642 	struct vop_create_args /* {
1643 		struct vnode *a_dvp;
1644 		struct vnode **a_vpp;
1645 		struct componentname *a_cnp;
1646 		struct vattr *a_vap;
1647 	} */ *ap = v;
1648 	struct vnode *dvp = ap->a_dvp;
1649 	struct vattr *vap = ap->a_vap;
1650 	struct componentname *cnp = ap->a_cnp;
1651 	struct nfsv2_sattr *sp;
1652 	u_int32_t *tl;
1653 	char *cp;
1654 	int32_t t1, t2;
1655 	struct nfsnode *dnp, *np = (struct nfsnode *)0;
1656 	struct vnode *newvp = (struct vnode *)0;
1657 	char *bpos, *dpos, *cp2;
1658 	int error, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1659 	struct mbuf *mreq, *mrep, *md, *mb;
1660 	const int v3 = NFS_ISV3(dvp);
1661 	u_int32_t excl_mode = NFSV3CREATE_UNCHECKED;
1662 
1663 	/*
1664 	 * Oops, not for me..
1665 	 */
1666 	if (vap->va_type == VSOCK)
1667 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1668 
1669 	KASSERT(vap->va_type == VREG);
1670 
1671 #ifdef VA_EXCLUSIVE
1672 	if (vap->va_vaflags & VA_EXCLUSIVE) {
1673 		excl_mode = NFSV3CREATE_EXCLUSIVE;
1674 	}
1675 #endif
1676 again:
1677 	error = 0;
1678 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1679 	dnp = VTONFS(dvp);
1680 	nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1681 		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1682 	nfsm_fhtom(dnp, v3);
1683 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1684 #ifndef NFS_V2_ONLY
1685 	if (v3) {
1686 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1687 		if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1688 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1689 			nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1690 #ifdef INET
1691 			if (TAILQ_FIRST(&in_ifaddrhead))
1692 				*tl++ = TAILQ_FIRST(&in_ifaddrhead)->
1693 				    ia_addr.sin_addr.s_addr;
1694 			else
1695 				*tl++ = create_verf;
1696 #else
1697 			*tl++ = create_verf;
1698 #endif
1699 			*tl = ++create_verf;
1700 		} else {
1701 			*tl = txdr_unsigned(excl_mode);
1702 			nfsm_v3attrbuild(vap, false);
1703 		}
1704 	} else
1705 #endif
1706 	{
1707 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1708 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1709 		sp->sa_uid = nfs_xdrneg1;
1710 		sp->sa_gid = nfs_xdrneg1;
1711 		sp->sa_size = 0;
1712 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1713 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1714 	}
1715 	nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred);
1716 	if (!error) {
1717 		nfsm_mtofh(dvp, newvp, v3, gotvp);
1718 		if (!gotvp) {
1719 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1720 			    cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1721 			if (!error)
1722 				newvp = NFSTOV(np);
1723 		}
1724 	}
1725 #ifndef NFS_V2_ONLY
1726 	if (v3)
1727 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1728 #endif
1729 	nfsm_reqdone;
1730 	if (error) {
1731 		/*
1732 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1733 		 */
1734 		if (v3 && error == ENOTSUP) {
1735 			if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1736 				excl_mode = NFSV3CREATE_GUARDED;
1737 				goto again;
1738 			} else if (excl_mode == NFSV3CREATE_GUARDED) {
1739 				excl_mode = NFSV3CREATE_UNCHECKED;
1740 				goto again;
1741 			}
1742 		}
1743 	} else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) {
1744 		struct timespec ts;
1745 
1746 		getnanotime(&ts);
1747 
1748 		/*
1749 		 * make sure that we'll update timestamps as
1750 		 * most server implementations use them to store
1751 		 * the create verifier.
1752 		 *
1753 		 * XXX it's better to use TOSERVER always.
1754 		 */
1755 
1756 		if (vap->va_atime.tv_sec == VNOVAL)
1757 			vap->va_atime = ts;
1758 		if (vap->va_mtime.tv_sec == VNOVAL)
1759 			vap->va_mtime = ts;
1760 
1761 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp);
1762 	}
1763 	if (error == 0) {
1764 		if (cnp->cn_flags & MAKEENTRY)
1765 			nfs_cache_enter(dvp, newvp, cnp);
1766 		else
1767 			cache_purge1(dvp, cnp, 0);
1768 		*ap->a_vpp = newvp;
1769 	} else {
1770 		if (newvp)
1771 			vput(newvp);
1772 		if (error == EEXIST)
1773 			cache_purge1(dvp, cnp, 0);
1774 	}
1775 	PNBUF_PUT(cnp->cn_pnbuf);
1776 	VTONFS(dvp)->n_flag |= NMODIFIED;
1777 	if (!wccflag)
1778 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1779 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1780 	vput(dvp);
1781 	return (error);
1782 }
1783 
1784 /*
1785  * nfs file remove call
1786  * To try and make nfs semantics closer to ufs semantics, a file that has
1787  * other processes using the vnode is renamed instead of removed and then
1788  * removed later on the last close.
1789  * - If v_usecount > 1
1790  *	  If a rename is not already in the works
1791  *	     call nfs_sillyrename() to set it up
1792  *     else
1793  *	  do the remove rpc
1794  */
1795 int
1796 nfs_remove(v)
1797 	void *v;
1798 {
1799 	struct vop_remove_args /* {
1800 		struct vnodeop_desc *a_desc;
1801 		struct vnode * a_dvp;
1802 		struct vnode * a_vp;
1803 		struct componentname * a_cnp;
1804 	} */ *ap = v;
1805 	struct vnode *vp = ap->a_vp;
1806 	struct vnode *dvp = ap->a_dvp;
1807 	struct componentname *cnp = ap->a_cnp;
1808 	struct nfsnode *np = VTONFS(vp);
1809 	int error = 0;
1810 	struct vattr vattr;
1811 
1812 #ifndef DIAGNOSTIC
1813 	if ((cnp->cn_flags & HASBUF) == 0)
1814 		panic("nfs_remove: no name");
1815 	if (vp->v_usecount < 1)
1816 		panic("nfs_remove: bad v_usecount");
1817 #endif
1818 	if (vp->v_type == VDIR)
1819 		error = EPERM;
1820 	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1821 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1822 	    vattr.va_nlink > 1)) {
1823 		/*
1824 		 * Purge the name cache so that the chance of a lookup for
1825 		 * the name succeeding while the remove is in progress is
1826 		 * minimized. Without node locking it can still happen, such
1827 		 * that an I/O op returns ESTALE, but since you get this if
1828 		 * another host removes the file..
1829 		 */
1830 		cache_purge(vp);
1831 		/*
1832 		 * throw away biocache buffers, mainly to avoid
1833 		 * unnecessary delayed writes later.
1834 		 */
1835 		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1);
1836 		/* Do the rpc */
1837 		if (error != EINTR)
1838 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1839 				cnp->cn_namelen, cnp->cn_cred, curlwp);
1840 	} else if (!np->n_sillyrename)
1841 		error = nfs_sillyrename(dvp, vp, cnp, false);
1842 	PNBUF_PUT(cnp->cn_pnbuf);
1843 	if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1844 	    vattr.va_nlink == 1) {
1845 		np->n_flag |= NREMOVED;
1846 	}
1847 	NFS_INVALIDATE_ATTRCACHE(np);
1848 	VN_KNOTE(vp, NOTE_DELETE);
1849 	VN_KNOTE(dvp, NOTE_WRITE);
1850 	if (dvp == vp)
1851 		vrele(vp);
1852 	else
1853 		vput(vp);
1854 	vput(dvp);
1855 	return (error);
1856 }
1857 
1858 /*
1859  * nfs file remove rpc called from nfs_inactive
1860  */
1861 int
1862 nfs_removeit(sp)
1863 	struct sillyrename *sp;
1864 {
1865 
1866 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1867 		(struct lwp *)0));
1868 }
1869 
1870 /*
1871  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1872  */
1873 int
1874 nfs_removerpc(dvp, name, namelen, cred, l)
1875 	struct vnode *dvp;
1876 	const char *name;
1877 	int namelen;
1878 	kauth_cred_t cred;
1879 	struct lwp *l;
1880 {
1881 	u_int32_t *tl;
1882 	char *cp;
1883 #ifndef NFS_V2_ONLY
1884 	int32_t t1;
1885 	char *cp2;
1886 #endif
1887 	int32_t t2;
1888 	char *bpos, *dpos;
1889 	int error = 0, wccflag = NFSV3_WCCRATTR;
1890 	struct mbuf *mreq, *mrep, *md, *mb;
1891 	const int v3 = NFS_ISV3(dvp);
1892 	int rexmit = 0;
1893 	struct nfsnode *dnp = VTONFS(dvp);
1894 
1895 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1896 	nfsm_reqhead(dnp, NFSPROC_REMOVE,
1897 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1898 	nfsm_fhtom(dnp, v3);
1899 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1900 	nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1901 #ifndef NFS_V2_ONLY
1902 	if (v3)
1903 		nfsm_wcc_data(dvp, wccflag, 0, !error);
1904 #endif
1905 	nfsm_reqdone;
1906 	VTONFS(dvp)->n_flag |= NMODIFIED;
1907 	if (!wccflag)
1908 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1909 	/*
1910 	 * Kludge City: If the first reply to the remove rpc is lost..
1911 	 *   the reply to the retransmitted request will be ENOENT
1912 	 *   since the file was in fact removed
1913 	 *   Therefore, we cheat and return success.
1914 	 */
1915 	if (rexmit && error == ENOENT)
1916 		error = 0;
1917 	return (error);
1918 }
1919 
1920 /*
1921  * nfs file rename call
1922  */
1923 int
1924 nfs_rename(v)
1925 	void *v;
1926 {
1927 	struct vop_rename_args  /* {
1928 		struct vnode *a_fdvp;
1929 		struct vnode *a_fvp;
1930 		struct componentname *a_fcnp;
1931 		struct vnode *a_tdvp;
1932 		struct vnode *a_tvp;
1933 		struct componentname *a_tcnp;
1934 	} */ *ap = v;
1935 	struct vnode *fvp = ap->a_fvp;
1936 	struct vnode *tvp = ap->a_tvp;
1937 	struct vnode *fdvp = ap->a_fdvp;
1938 	struct vnode *tdvp = ap->a_tdvp;
1939 	struct componentname *tcnp = ap->a_tcnp;
1940 	struct componentname *fcnp = ap->a_fcnp;
1941 	int error;
1942 
1943 #ifndef DIAGNOSTIC
1944 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1945 	    (fcnp->cn_flags & HASBUF) == 0)
1946 		panic("nfs_rename: no name");
1947 #endif
1948 	/* Check for cross-device rename */
1949 	if ((fvp->v_mount != tdvp->v_mount) ||
1950 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1951 		error = EXDEV;
1952 		goto out;
1953 	}
1954 
1955 	/*
1956 	 * If the tvp exists and is in use, sillyrename it before doing the
1957 	 * rename of the new file over it.
1958 	 *
1959 	 * Have sillyrename use link instead of rename if possible,
1960 	 * so that we don't lose the file if the rename fails, and so
1961 	 * that there's no window when the "to" file doesn't exist.
1962 	 */
1963 	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1964 	    tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1965 		VN_KNOTE(tvp, NOTE_DELETE);
1966 		vput(tvp);
1967 		tvp = NULL;
1968 	}
1969 
1970 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1971 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1972 		curlwp);
1973 
1974 	VN_KNOTE(fdvp, NOTE_WRITE);
1975 	VN_KNOTE(tdvp, NOTE_WRITE);
1976 	if (error == 0 || error == EEXIST) {
1977 		if (fvp->v_type == VDIR)
1978 			cache_purge(fvp);
1979 		else
1980 			cache_purge1(fdvp, fcnp, 0);
1981 		if (tvp != NULL && tvp->v_type == VDIR)
1982 			cache_purge(tvp);
1983 		else
1984 			cache_purge1(tdvp, tcnp, 0);
1985 	}
1986 out:
1987 	if (tdvp == tvp)
1988 		vrele(tdvp);
1989 	else
1990 		vput(tdvp);
1991 	if (tvp)
1992 		vput(tvp);
1993 	vrele(fdvp);
1994 	vrele(fvp);
1995 	return (error);
1996 }
1997 
1998 /*
1999  * nfs file rename rpc called from nfs_remove() above
2000  */
2001 int
2002 nfs_renameit(sdvp, scnp, sp)
2003 	struct vnode *sdvp;
2004 	struct componentname *scnp;
2005 	struct sillyrename *sp;
2006 {
2007 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2008 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
2009 }
2010 
2011 /*
2012  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2013  */
2014 int
2015 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l)
2016 	struct vnode *fdvp;
2017 	const char *fnameptr;
2018 	int fnamelen;
2019 	struct vnode *tdvp;
2020 	const char *tnameptr;
2021 	int tnamelen;
2022 	kauth_cred_t cred;
2023 	struct lwp *l;
2024 {
2025 	u_int32_t *tl;
2026 	char *cp;
2027 #ifndef NFS_V2_ONLY
2028 	int32_t t1;
2029 	char *cp2;
2030 #endif
2031 	int32_t t2;
2032 	char *bpos, *dpos;
2033 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2034 	struct mbuf *mreq, *mrep, *md, *mb;
2035 	const int v3 = NFS_ISV3(fdvp);
2036 	int rexmit = 0;
2037 	struct nfsnode *fdnp = VTONFS(fdvp);
2038 
2039 	nfsstats.rpccnt[NFSPROC_RENAME]++;
2040 	nfsm_reqhead(fdnp, NFSPROC_RENAME,
2041 		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2042 		nfsm_rndup(tnamelen));
2043 	nfsm_fhtom(fdnp, v3);
2044 	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2045 	nfsm_fhtom(VTONFS(tdvp), v3);
2046 	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2047 	nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
2048 #ifndef NFS_V2_ONLY
2049 	if (v3) {
2050 		nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2051 		nfsm_wcc_data(tdvp, twccflag, 0, !error);
2052 	}
2053 #endif
2054 	nfsm_reqdone;
2055 	VTONFS(fdvp)->n_flag |= NMODIFIED;
2056 	VTONFS(tdvp)->n_flag |= NMODIFIED;
2057 	if (!fwccflag)
2058 		NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2059 	if (!twccflag)
2060 		NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2061 	/*
2062 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2063 	 */
2064 	if (rexmit && error == ENOENT)
2065 		error = 0;
2066 	return (error);
2067 }
2068 
2069 /*
2070  * NFS link RPC, called from nfs_link.
2071  * Assumes dvp and vp locked, and leaves them that way.
2072  */
2073 
2074 static int
2075 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2076     size_t namelen, kauth_cred_t cred, struct lwp *l)
2077 {
2078 	u_int32_t *tl;
2079 	char *cp;
2080 #ifndef NFS_V2_ONLY
2081 	int32_t t1;
2082 	char *cp2;
2083 #endif
2084 	int32_t t2;
2085 	char *bpos, *dpos;
2086 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2087 	struct mbuf *mreq, *mrep, *md, *mb;
2088 	const int v3 = NFS_ISV3(dvp);
2089 	int rexmit = 0;
2090 	struct nfsnode *np = VTONFS(vp);
2091 
2092 	nfsstats.rpccnt[NFSPROC_LINK]++;
2093 	nfsm_reqhead(np, NFSPROC_LINK,
2094 	    NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2095 	nfsm_fhtom(np, v3);
2096 	nfsm_fhtom(VTONFS(dvp), v3);
2097 	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2098 	nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2099 #ifndef NFS_V2_ONLY
2100 	if (v3) {
2101 		nfsm_postop_attr(vp, attrflag, 0);
2102 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2103 	}
2104 #endif
2105 	nfsm_reqdone;
2106 
2107 	VTONFS(dvp)->n_flag |= NMODIFIED;
2108 	if (!attrflag)
2109 		NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2110 	if (!wccflag)
2111 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2112 
2113 	/*
2114 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2115 	 */
2116 	if (rexmit && error == EEXIST)
2117 		error = 0;
2118 
2119 	return error;
2120 }
2121 
2122 /*
2123  * nfs hard link create call
2124  */
2125 int
2126 nfs_link(v)
2127 	void *v;
2128 {
2129 	struct vop_link_args /* {
2130 		struct vnode *a_dvp;
2131 		struct vnode *a_vp;
2132 		struct componentname *a_cnp;
2133 	} */ *ap = v;
2134 	struct vnode *vp = ap->a_vp;
2135 	struct vnode *dvp = ap->a_dvp;
2136 	struct componentname *cnp = ap->a_cnp;
2137 	int error = 0;
2138 
2139 	if (dvp->v_mount != vp->v_mount) {
2140 		VOP_ABORTOP(dvp, cnp);
2141 		vput(dvp);
2142 		return (EXDEV);
2143 	}
2144 	if (dvp != vp) {
2145 		error = vn_lock(vp, LK_EXCLUSIVE);
2146 		if (error != 0) {
2147 			VOP_ABORTOP(dvp, cnp);
2148 			vput(dvp);
2149 			return error;
2150 		}
2151 	}
2152 
2153 	/*
2154 	 * Push all writes to the server, so that the attribute cache
2155 	 * doesn't get "out of sync" with the server.
2156 	 * XXX There should be a better way!
2157 	 */
2158 	VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0);
2159 
2160 	error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2161 	    cnp->cn_cred, curlwp);
2162 
2163 	if (error == 0)
2164 		cache_purge1(dvp, cnp, 0);
2165 	PNBUF_PUT(cnp->cn_pnbuf);
2166 	if (dvp != vp)
2167 		VOP_UNLOCK(vp, 0);
2168 	VN_KNOTE(vp, NOTE_LINK);
2169 	VN_KNOTE(dvp, NOTE_WRITE);
2170 	vput(dvp);
2171 	return (error);
2172 }
2173 
2174 /*
2175  * nfs symbolic link create call
2176  */
2177 int
2178 nfs_symlink(v)
2179 	void *v;
2180 {
2181 	struct vop_symlink_args /* {
2182 		struct vnode *a_dvp;
2183 		struct vnode **a_vpp;
2184 		struct componentname *a_cnp;
2185 		struct vattr *a_vap;
2186 		char *a_target;
2187 	} */ *ap = v;
2188 	struct vnode *dvp = ap->a_dvp;
2189 	struct vattr *vap = ap->a_vap;
2190 	struct componentname *cnp = ap->a_cnp;
2191 	struct nfsv2_sattr *sp;
2192 	u_int32_t *tl;
2193 	char *cp;
2194 	int32_t t1, t2;
2195 	char *bpos, *dpos, *cp2;
2196 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2197 	struct mbuf *mreq, *mrep, *md, *mb;
2198 	struct vnode *newvp = (struct vnode *)0;
2199 	const int v3 = NFS_ISV3(dvp);
2200 	int rexmit = 0;
2201 	struct nfsnode *dnp = VTONFS(dvp);
2202 
2203 	*ap->a_vpp = NULL;
2204 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2205 	slen = strlen(ap->a_target);
2206 	nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2207 	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2208 	nfsm_fhtom(dnp, v3);
2209 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2210 #ifndef NFS_V2_ONlY
2211 	if (v3)
2212 		nfsm_v3attrbuild(vap, false);
2213 #endif
2214 	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2215 #ifndef NFS_V2_ONlY
2216 	if (!v3) {
2217 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2218 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2219 		sp->sa_uid = nfs_xdrneg1;
2220 		sp->sa_gid = nfs_xdrneg1;
2221 		sp->sa_size = nfs_xdrneg1;
2222 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2223 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2224 	}
2225 #endif
2226 	nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2227 	    &rexmit);
2228 #ifndef NFS_V2_ONlY
2229 	if (v3) {
2230 		if (!error)
2231 			nfsm_mtofh(dvp, newvp, v3, gotvp);
2232 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2233 	}
2234 #endif
2235 	nfsm_reqdone;
2236 	/*
2237 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2238 	 */
2239 	if (rexmit && error == EEXIST)
2240 		error = 0;
2241 	if (error == 0 || error == EEXIST)
2242 		cache_purge1(dvp, cnp, 0);
2243 	if (error == 0 && newvp == NULL) {
2244 		struct nfsnode *np = NULL;
2245 
2246 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2247 		    cnp->cn_cred, curlwp, &np);
2248 		if (error == 0)
2249 			newvp = NFSTOV(np);
2250 	}
2251 	if (error) {
2252 		if (newvp != NULL)
2253 			vput(newvp);
2254 	} else {
2255 		*ap->a_vpp = newvp;
2256 	}
2257 	PNBUF_PUT(cnp->cn_pnbuf);
2258 	VTONFS(dvp)->n_flag |= NMODIFIED;
2259 	if (!wccflag)
2260 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2261 	VN_KNOTE(dvp, NOTE_WRITE);
2262 	vput(dvp);
2263 	return (error);
2264 }
2265 
2266 /*
2267  * nfs make dir call
2268  */
2269 int
2270 nfs_mkdir(v)
2271 	void *v;
2272 {
2273 	struct vop_mkdir_args /* {
2274 		struct vnode *a_dvp;
2275 		struct vnode **a_vpp;
2276 		struct componentname *a_cnp;
2277 		struct vattr *a_vap;
2278 	} */ *ap = v;
2279 	struct vnode *dvp = ap->a_dvp;
2280 	struct vattr *vap = ap->a_vap;
2281 	struct componentname *cnp = ap->a_cnp;
2282 	struct nfsv2_sattr *sp;
2283 	u_int32_t *tl;
2284 	char *cp;
2285 	int32_t t1, t2;
2286 	int len;
2287 	struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2288 	struct vnode *newvp = (struct vnode *)0;
2289 	char *bpos, *dpos, *cp2;
2290 	int error = 0, wccflag = NFSV3_WCCRATTR;
2291 	int gotvp = 0;
2292 	int rexmit = 0;
2293 	struct mbuf *mreq, *mrep, *md, *mb;
2294 	const int v3 = NFS_ISV3(dvp);
2295 
2296 	len = cnp->cn_namelen;
2297 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2298 	nfsm_reqhead(dnp, NFSPROC_MKDIR,
2299 	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2300 	nfsm_fhtom(dnp, v3);
2301 	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2302 #ifndef NFS_V2_ONLY
2303 	if (v3) {
2304 		nfsm_v3attrbuild(vap, false);
2305 	} else
2306 #endif
2307 	{
2308 		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2309 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2310 		sp->sa_uid = nfs_xdrneg1;
2311 		sp->sa_gid = nfs_xdrneg1;
2312 		sp->sa_size = nfs_xdrneg1;
2313 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2314 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2315 	}
2316 	nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2317 	if (!error)
2318 		nfsm_mtofh(dvp, newvp, v3, gotvp);
2319 	if (v3)
2320 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2321 	nfsm_reqdone;
2322 	VTONFS(dvp)->n_flag |= NMODIFIED;
2323 	if (!wccflag)
2324 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2325 	/*
2326 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2327 	 * if we can succeed in looking up the directory.
2328 	 */
2329 	if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2330 		if (newvp) {
2331 			vput(newvp);
2332 			newvp = (struct vnode *)0;
2333 		}
2334 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2335 			curlwp, &np);
2336 		if (!error) {
2337 			newvp = NFSTOV(np);
2338 			if (newvp->v_type != VDIR || newvp == dvp)
2339 				error = EEXIST;
2340 		}
2341 	}
2342 	if (error) {
2343 		if (newvp) {
2344 			if (dvp != newvp)
2345 				vput(newvp);
2346 			else
2347 				vrele(newvp);
2348 		}
2349 	} else {
2350 		VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2351 		if (cnp->cn_flags & MAKEENTRY)
2352 			nfs_cache_enter(dvp, newvp, cnp);
2353 		*ap->a_vpp = newvp;
2354 	}
2355 	PNBUF_PUT(cnp->cn_pnbuf);
2356 	vput(dvp);
2357 	return (error);
2358 }
2359 
2360 /*
2361  * nfs remove directory call
2362  */
2363 int
2364 nfs_rmdir(v)
2365 	void *v;
2366 {
2367 	struct vop_rmdir_args /* {
2368 		struct vnode *a_dvp;
2369 		struct vnode *a_vp;
2370 		struct componentname *a_cnp;
2371 	} */ *ap = v;
2372 	struct vnode *vp = ap->a_vp;
2373 	struct vnode *dvp = ap->a_dvp;
2374 	struct componentname *cnp = ap->a_cnp;
2375 	u_int32_t *tl;
2376 	char *cp;
2377 #ifndef NFS_V2_ONLY
2378 	int32_t t1;
2379 	char *cp2;
2380 #endif
2381 	int32_t t2;
2382 	char *bpos, *dpos;
2383 	int error = 0, wccflag = NFSV3_WCCRATTR;
2384 	int rexmit = 0;
2385 	struct mbuf *mreq, *mrep, *md, *mb;
2386 	const int v3 = NFS_ISV3(dvp);
2387 	struct nfsnode *dnp;
2388 
2389 	if (dvp == vp) {
2390 		vrele(dvp);
2391 		vput(dvp);
2392 		PNBUF_PUT(cnp->cn_pnbuf);
2393 		return (EINVAL);
2394 	}
2395 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2396 	dnp = VTONFS(dvp);
2397 	nfsm_reqhead(dnp, NFSPROC_RMDIR,
2398 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2399 	nfsm_fhtom(dnp, v3);
2400 	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2401 	nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2402 #ifndef NFS_V2_ONLY
2403 	if (v3)
2404 		nfsm_wcc_data(dvp, wccflag, 0, !error);
2405 #endif
2406 	nfsm_reqdone;
2407 	PNBUF_PUT(cnp->cn_pnbuf);
2408 	VTONFS(dvp)->n_flag |= NMODIFIED;
2409 	if (!wccflag)
2410 		NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2411 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2412 	VN_KNOTE(vp, NOTE_DELETE);
2413 	cache_purge(vp);
2414 	vput(vp);
2415 	vput(dvp);
2416 	/*
2417 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2418 	 */
2419 	if (rexmit && error == ENOENT)
2420 		error = 0;
2421 	return (error);
2422 }
2423 
2424 /*
2425  * nfs readdir call
2426  */
2427 int
2428 nfs_readdir(v)
2429 	void *v;
2430 {
2431 	struct vop_readdir_args /* {
2432 		struct vnode *a_vp;
2433 		struct uio *a_uio;
2434 		kauth_cred_t a_cred;
2435 		int *a_eofflag;
2436 		off_t **a_cookies;
2437 		int *a_ncookies;
2438 	} */ *ap = v;
2439 	struct vnode *vp = ap->a_vp;
2440 	struct uio *uio = ap->a_uio;
2441 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2442 	char *base = uio->uio_iov->iov_base;
2443 	int tresid, error;
2444 	size_t count, lost;
2445 	struct dirent *dp;
2446 	off_t *cookies = NULL;
2447 	int ncookies = 0, nc;
2448 
2449 	if (vp->v_type != VDIR)
2450 		return (EPERM);
2451 
2452 	lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2453 	count = uio->uio_resid - lost;
2454 	if (count <= 0)
2455 		return (EINVAL);
2456 
2457 	/*
2458 	 * Call nfs_bioread() to do the real work.
2459 	 */
2460 	tresid = uio->uio_resid = count;
2461 	error = nfs_bioread(vp, uio, 0, ap->a_cred,
2462 		    ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2463 
2464 	if (!error && ap->a_cookies) {
2465 		ncookies = count / 16;
2466 		cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2467 		*ap->a_cookies = cookies;
2468 	}
2469 
2470 	if (!error && uio->uio_resid == tresid) {
2471 		uio->uio_resid += lost;
2472 		nfsstats.direofcache_misses++;
2473 		if (ap->a_cookies)
2474 			*ap->a_ncookies = 0;
2475 		*ap->a_eofflag = 1;
2476 		return (0);
2477 	}
2478 
2479 	if (!error && ap->a_cookies) {
2480 		/*
2481 		 * Only the NFS server and emulations use cookies, and they
2482 		 * load the directory block into system space, so we can
2483 		 * just look at it directly.
2484 		 */
2485 		if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2486 		    uio->uio_iovcnt != 1)
2487 			panic("nfs_readdir: lost in space");
2488 		for (nc = 0; ncookies-- &&
2489 		     base < (char *)uio->uio_iov->iov_base; nc++){
2490 			dp = (struct dirent *) base;
2491 			if (dp->d_reclen == 0)
2492 				break;
2493 			if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2494 				*(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2495 			else
2496 				*(cookies++) = NFS_GETCOOKIE(dp);
2497 			base += dp->d_reclen;
2498 		}
2499 		uio->uio_resid +=
2500 		    ((char *)uio->uio_iov->iov_base - base);
2501 		uio->uio_iov->iov_len +=
2502 		    ((char *)uio->uio_iov->iov_base - base);
2503 		uio->uio_iov->iov_base = base;
2504 		*ap->a_ncookies = nc;
2505 	}
2506 
2507 	uio->uio_resid += lost;
2508 	*ap->a_eofflag = 0;
2509 	return (error);
2510 }
2511 
2512 /*
2513  * Readdir rpc call.
2514  * Called from below the buffer cache by nfs_doio().
2515  */
2516 int
2517 nfs_readdirrpc(vp, uiop, cred)
2518 	struct vnode *vp;
2519 	struct uio *uiop;
2520 	kauth_cred_t cred;
2521 {
2522 	int len, left;
2523 	struct dirent *dp = NULL;
2524 	u_int32_t *tl;
2525 	char *cp;
2526 	int32_t t1, t2;
2527 	char *bpos, *dpos, *cp2;
2528 	struct mbuf *mreq, *mrep, *md, *mb;
2529 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2530 	struct nfsnode *dnp = VTONFS(vp);
2531 	u_quad_t fileno;
2532 	int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2533 #ifndef NFS_V2_ONLY
2534 	int attrflag;
2535 #endif
2536 	int nrpcs = 0, reclen;
2537 	const int v3 = NFS_ISV3(vp);
2538 
2539 #ifdef DIAGNOSTIC
2540 	/*
2541 	 * Should be called from buffer cache, so only amount of
2542 	 * NFS_DIRBLKSIZ will be requested.
2543 	 */
2544 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2545 		panic("nfs readdirrpc bad uio");
2546 #endif
2547 
2548 	/*
2549 	 * Loop around doing readdir rpc's of size nm_readdirsize
2550 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2551 	 * The stopping criteria is EOF or buffer full.
2552 	 */
2553 	while (more_dirs && bigenough) {
2554 		/*
2555 		 * Heuristic: don't bother to do another RPC to further
2556 		 * fill up this block if there is not much room left. (< 50%
2557 		 * of the readdir RPC size). This wastes some buffer space
2558 		 * but can save up to 50% in RPC calls.
2559 		 */
2560 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2561 			bigenough = 0;
2562 			break;
2563 		}
2564 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2565 		nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2566 			NFSX_READDIR(v3));
2567 		nfsm_fhtom(dnp, v3);
2568 #ifndef NFS_V2_ONLY
2569 		if (v3) {
2570 			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2571 			if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2572 				txdr_swapcookie3(uiop->uio_offset, tl);
2573 			} else {
2574 				txdr_cookie3(uiop->uio_offset, tl);
2575 			}
2576 			tl += 2;
2577 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2578 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2579 		} else
2580 #endif
2581 		{
2582 			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2583 			*tl++ = txdr_unsigned(uiop->uio_offset);
2584 		}
2585 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2586 		nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2587 		nrpcs++;
2588 #ifndef NFS_V2_ONLY
2589 		if (v3) {
2590 			nfsm_postop_attr(vp, attrflag, 0);
2591 			if (!error) {
2592 				nfsm_dissect(tl, u_int32_t *,
2593 				    2 * NFSX_UNSIGNED);
2594 				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2595 				dnp->n_cookieverf.nfsuquad[1] = *tl;
2596 			} else {
2597 				m_freem(mrep);
2598 				goto nfsmout;
2599 			}
2600 		}
2601 #endif
2602 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2603 		more_dirs = fxdr_unsigned(int, *tl);
2604 
2605 		/* loop thru the dir entries, doctoring them to 4bsd form */
2606 		while (more_dirs && bigenough) {
2607 #ifndef NFS_V2_ONLY
2608 			if (v3) {
2609 				nfsm_dissect(tl, u_int32_t *,
2610 				    3 * NFSX_UNSIGNED);
2611 				fileno = fxdr_hyper(tl);
2612 				len = fxdr_unsigned(int, *(tl + 2));
2613 			} else
2614 #endif
2615 			{
2616 				nfsm_dissect(tl, u_int32_t *,
2617 				    2 * NFSX_UNSIGNED);
2618 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2619 				len = fxdr_unsigned(int, *tl);
2620 			}
2621 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2622 				error = EBADRPC;
2623 				m_freem(mrep);
2624 				goto nfsmout;
2625 			}
2626 			/* for cookie stashing */
2627 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2628 			left = NFS_DIRFRAGSIZ - blksiz;
2629 			if (reclen > left) {
2630 				memset(uiop->uio_iov->iov_base, 0, left);
2631 				dp->d_reclen += left;
2632 				UIO_ADVANCE(uiop, left);
2633 				blksiz = 0;
2634 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2635 			}
2636 			if (reclen > uiop->uio_resid)
2637 				bigenough = 0;
2638 			if (bigenough) {
2639 				int tlen;
2640 
2641 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2642 				dp->d_fileno = fileno;
2643 				dp->d_namlen = len;
2644 				dp->d_reclen = reclen;
2645 				dp->d_type = DT_UNKNOWN;
2646 				blksiz += reclen;
2647 				if (blksiz == NFS_DIRFRAGSIZ)
2648 					blksiz = 0;
2649 				UIO_ADVANCE(uiop, DIRHDSIZ);
2650 				nfsm_mtouio(uiop, len);
2651 				tlen = reclen - (DIRHDSIZ + len);
2652 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2653 				UIO_ADVANCE(uiop, tlen);
2654 			} else
2655 				nfsm_adv(nfsm_rndup(len));
2656 #ifndef NFS_V2_ONLY
2657 			if (v3) {
2658 				nfsm_dissect(tl, u_int32_t *,
2659 				    3 * NFSX_UNSIGNED);
2660 			} else
2661 #endif
2662 			{
2663 				nfsm_dissect(tl, u_int32_t *,
2664 				    2 * NFSX_UNSIGNED);
2665 			}
2666 			if (bigenough) {
2667 #ifndef NFS_V2_ONLY
2668 				if (v3) {
2669 					if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2670 						uiop->uio_offset =
2671 						    fxdr_swapcookie3(tl);
2672 					else
2673 						uiop->uio_offset =
2674 						    fxdr_cookie3(tl);
2675 				}
2676 				else
2677 #endif
2678 				{
2679 					uiop->uio_offset =
2680 					    fxdr_unsigned(off_t, *tl);
2681 				}
2682 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2683 			}
2684 			if (v3)
2685 				tl += 2;
2686 			else
2687 				tl++;
2688 			more_dirs = fxdr_unsigned(int, *tl);
2689 		}
2690 		/*
2691 		 * If at end of rpc data, get the eof boolean
2692 		 */
2693 		if (!more_dirs) {
2694 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2695 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2696 
2697 			/*
2698 			 * kludge: if we got no entries, treat it as EOF.
2699 			 * some server sometimes send a reply without any
2700 			 * entries or EOF.
2701 			 * although it might mean the server has very long name,
2702 			 * we can't handle such entries anyway.
2703 			 */
2704 
2705 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2706 				more_dirs = 0;
2707 		}
2708 		m_freem(mrep);
2709 	}
2710 	/*
2711 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2712 	 * by increasing d_reclen for the last record.
2713 	 */
2714 	if (blksiz > 0) {
2715 		left = NFS_DIRFRAGSIZ - blksiz;
2716 		memset(uiop->uio_iov->iov_base, 0, left);
2717 		dp->d_reclen += left;
2718 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2719 		UIO_ADVANCE(uiop, left);
2720 	}
2721 
2722 	/*
2723 	 * We are now either at the end of the directory or have filled the
2724 	 * block.
2725 	 */
2726 	if (bigenough) {
2727 		dnp->n_direofoffset = uiop->uio_offset;
2728 		dnp->n_flag |= NEOFVALID;
2729 	}
2730 nfsmout:
2731 	return (error);
2732 }
2733 
2734 #ifndef NFS_V2_ONLY
2735 /*
2736  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2737  */
2738 int
2739 nfs_readdirplusrpc(vp, uiop, cred)
2740 	struct vnode *vp;
2741 	struct uio *uiop;
2742 	kauth_cred_t cred;
2743 {
2744 	int len, left;
2745 	struct dirent *dp = NULL;
2746 	u_int32_t *tl;
2747 	char *cp;
2748 	int32_t t1, t2;
2749 	struct vnode *newvp;
2750 	char *bpos, *dpos, *cp2;
2751 	struct mbuf *mreq, *mrep, *md, *mb;
2752 	struct nameidata nami, *ndp = &nami;
2753 	struct componentname *cnp = &ndp->ni_cnd;
2754 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2755 	struct nfsnode *dnp = VTONFS(vp), *np;
2756 	nfsfh_t *fhp;
2757 	u_quad_t fileno;
2758 	int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2759 	int attrflag, fhsize, nrpcs = 0, reclen;
2760 	struct nfs_fattr fattr, *fp;
2761 
2762 #ifdef DIAGNOSTIC
2763 	if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2764 		panic("nfs readdirplusrpc bad uio");
2765 #endif
2766 	ndp->ni_dvp = vp;
2767 	newvp = NULLVP;
2768 
2769 	/*
2770 	 * Loop around doing readdir rpc's of size nm_readdirsize
2771 	 * truncated to a multiple of NFS_DIRFRAGSIZ.
2772 	 * The stopping criteria is EOF or buffer full.
2773 	 */
2774 	while (more_dirs && bigenough) {
2775 		if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2776 			bigenough = 0;
2777 			break;
2778 		}
2779 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2780 		nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2781 			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2782 		nfsm_fhtom(dnp, 1);
2783  		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2784 		if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2785 			txdr_swapcookie3(uiop->uio_offset, tl);
2786 		} else {
2787 			txdr_cookie3(uiop->uio_offset, tl);
2788 		}
2789 		tl += 2;
2790 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2791 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2792 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2793 		*tl = txdr_unsigned(nmp->nm_rsize);
2794 		nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2795 		nfsm_postop_attr(vp, attrflag, 0);
2796 		if (error) {
2797 			m_freem(mrep);
2798 			goto nfsmout;
2799 		}
2800 		nrpcs++;
2801 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2802 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2803 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2804 		more_dirs = fxdr_unsigned(int, *tl);
2805 
2806 		/* loop thru the dir entries, doctoring them to 4bsd form */
2807 		while (more_dirs && bigenough) {
2808 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2809 			fileno = fxdr_hyper(tl);
2810 			len = fxdr_unsigned(int, *(tl + 2));
2811 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2812 				error = EBADRPC;
2813 				m_freem(mrep);
2814 				goto nfsmout;
2815 			}
2816 			/* for cookie stashing */
2817 			reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2818 			left = NFS_DIRFRAGSIZ - blksiz;
2819 			if (reclen > left) {
2820 				/*
2821 				 * DIRFRAGSIZ is aligned, no need to align
2822 				 * again here.
2823 				 */
2824 				memset(uiop->uio_iov->iov_base, 0, left);
2825 				dp->d_reclen += left;
2826 				UIO_ADVANCE(uiop, left);
2827 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2828 				blksiz = 0;
2829 			}
2830 			if (reclen > uiop->uio_resid)
2831 				bigenough = 0;
2832 			if (bigenough) {
2833 				int tlen;
2834 
2835 				dp = (struct dirent *)uiop->uio_iov->iov_base;
2836 				dp->d_fileno = fileno;
2837 				dp->d_namlen = len;
2838 				dp->d_reclen = reclen;
2839 				dp->d_type = DT_UNKNOWN;
2840 				blksiz += reclen;
2841 				if (blksiz == NFS_DIRFRAGSIZ)
2842 					blksiz = 0;
2843 				UIO_ADVANCE(uiop, DIRHDSIZ);
2844 				nfsm_mtouio(uiop, len);
2845 				tlen = reclen - (DIRHDSIZ + len);
2846 				(void)memset(uiop->uio_iov->iov_base, 0, tlen);
2847 				UIO_ADVANCE(uiop, tlen);
2848 				cnp->cn_nameptr = dp->d_name;
2849 				cnp->cn_namelen = dp->d_namlen;
2850 			} else
2851 				nfsm_adv(nfsm_rndup(len));
2852 			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2853 			if (bigenough) {
2854 				if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2855 					uiop->uio_offset =
2856 						fxdr_swapcookie3(tl);
2857 				else
2858 					uiop->uio_offset =
2859 						fxdr_cookie3(tl);
2860 				NFS_STASHCOOKIE(dp, uiop->uio_offset);
2861 			}
2862 			tl += 2;
2863 
2864 			/*
2865 			 * Since the attributes are before the file handle
2866 			 * (sigh), we must skip over the attributes and then
2867 			 * come back and get them.
2868 			 */
2869 			attrflag = fxdr_unsigned(int, *tl);
2870 			if (attrflag) {
2871 			    nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2872 			    memcpy(&fattr, fp, NFSX_V3FATTR);
2873 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2874 			    doit = fxdr_unsigned(int, *tl);
2875 			    if (doit) {
2876 				nfsm_getfh(fhp, fhsize, 1);
2877 				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2878 				    VREF(vp);
2879 				    newvp = vp;
2880 				    np = dnp;
2881 				} else {
2882 				    error = nfs_nget1(vp->v_mount, fhp,
2883 					fhsize, &np, LK_NOWAIT);
2884 				    if (!error)
2885 					newvp = NFSTOV(np);
2886 				}
2887 				if (!error) {
2888 				    const char *xcp;
2889 
2890 				    nfs_loadattrcache(&newvp, &fattr, 0, 0);
2891 				    if (bigenough) {
2892 					dp->d_type =
2893 					   IFTODT(VTTOIF(np->n_vattr->va_type));
2894 					if (cnp->cn_namelen <= NCHNAMLEN) {
2895 					    ndp->ni_vp = newvp;
2896 					    xcp = cnp->cn_nameptr +
2897 						cnp->cn_namelen;
2898 					    cnp->cn_hash =
2899 					       namei_hash(cnp->cn_nameptr, &xcp);
2900 					    nfs_cache_enter(ndp->ni_dvp,
2901 						ndp->ni_vp, cnp);
2902 					}
2903 				    }
2904 				}
2905 				error = 0;
2906 			   }
2907 			} else {
2908 			    /* Just skip over the file handle */
2909 			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2910 			    i = fxdr_unsigned(int, *tl);
2911 			    nfsm_adv(nfsm_rndup(i));
2912 			}
2913 			if (newvp != NULLVP) {
2914 			    if (newvp == vp)
2915 				vrele(newvp);
2916 			    else
2917 				vput(newvp);
2918 			    newvp = NULLVP;
2919 			}
2920 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2921 			more_dirs = fxdr_unsigned(int, *tl);
2922 		}
2923 		/*
2924 		 * If at end of rpc data, get the eof boolean
2925 		 */
2926 		if (!more_dirs) {
2927 			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2928 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2929 
2930 			/*
2931 			 * kludge: see a comment in nfs_readdirrpc.
2932 			 */
2933 
2934 			if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2935 				more_dirs = 0;
2936 		}
2937 		m_freem(mrep);
2938 	}
2939 	/*
2940 	 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2941 	 * by increasing d_reclen for the last record.
2942 	 */
2943 	if (blksiz > 0) {
2944 		left = NFS_DIRFRAGSIZ - blksiz;
2945 		memset(uiop->uio_iov->iov_base, 0, left);
2946 		dp->d_reclen += left;
2947 		NFS_STASHCOOKIE(dp, uiop->uio_offset);
2948 		UIO_ADVANCE(uiop, left);
2949 	}
2950 
2951 	/*
2952 	 * We are now either at the end of the directory or have filled the
2953 	 * block.
2954 	 */
2955 	if (bigenough) {
2956 		dnp->n_direofoffset = uiop->uio_offset;
2957 		dnp->n_flag |= NEOFVALID;
2958 	}
2959 nfsmout:
2960 	if (newvp != NULLVP) {
2961 		if(newvp == vp)
2962 		    vrele(newvp);
2963 		else
2964 		    vput(newvp);
2965 	}
2966 	return (error);
2967 }
2968 #endif
2969 
2970 /*
2971  * Silly rename. To make the NFS filesystem that is stateless look a little
2972  * more like the "ufs" a remove of an active vnode is translated to a rename
2973  * to a funny looking filename that is removed by nfs_inactive on the
2974  * nfsnode. There is the potential for another process on a different client
2975  * to create the same funny name between the nfs_lookitup() fails and the
2976  * nfs_rename() completes, but...
2977  */
2978 int
2979 nfs_sillyrename(dvp, vp, cnp, dolink)
2980 	struct vnode *dvp, *vp;
2981 	struct componentname *cnp;
2982 	bool dolink;
2983 {
2984 	struct sillyrename *sp;
2985 	struct nfsnode *np;
2986 	int error;
2987 	short pid;
2988 
2989 	cache_purge(dvp);
2990 	np = VTONFS(vp);
2991 #ifndef DIAGNOSTIC
2992 	if (vp->v_type == VDIR)
2993 		panic("nfs: sillyrename dir");
2994 #endif
2995 	sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
2996 	sp->s_cred = kauth_cred_dup(cnp->cn_cred);
2997 	sp->s_dvp = dvp;
2998 	VREF(dvp);
2999 
3000 	/* Fudge together a funny name */
3001 	pid = curlwp->l_proc->p_pid;
3002 	memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
3003 	sp->s_namlen = 12;
3004 	sp->s_name[8] = hexdigits[pid & 0xf];
3005 	sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
3006 	sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
3007 	sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
3008 
3009 	/* Try lookitups until we get one that isn't there */
3010 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3011 		curlwp, (struct nfsnode **)0) == 0) {
3012 		sp->s_name[4]++;
3013 		if (sp->s_name[4] > 'z') {
3014 			error = EINVAL;
3015 			goto bad;
3016 		}
3017 	}
3018 	if (dolink) {
3019 		error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
3020 		    sp->s_cred, curlwp);
3021 		/*
3022 		 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
3023 		 */
3024 		if (error == ENOTSUP) {
3025 			error = nfs_renameit(dvp, cnp, sp);
3026 		}
3027 	} else {
3028 		error = nfs_renameit(dvp, cnp, sp);
3029 	}
3030 	if (error)
3031 		goto bad;
3032 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3033 		curlwp, &np);
3034 	np->n_sillyrename = sp;
3035 	return (0);
3036 bad:
3037 	vrele(sp->s_dvp);
3038 	kauth_cred_free(sp->s_cred);
3039 	kmem_free(sp, sizeof(*sp));
3040 	return (error);
3041 }
3042 
3043 /*
3044  * Look up a file name and optionally either update the file handle or
3045  * allocate an nfsnode, depending on the value of npp.
3046  * npp == NULL	--> just do the lookup
3047  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3048  *			handled too
3049  * *npp != NULL --> update the file handle in the vnode
3050  */
3051 int
3052 nfs_lookitup(dvp, name, len, cred, l, npp)
3053 	struct vnode *dvp;
3054 	const char *name;
3055 	int len;
3056 	kauth_cred_t cred;
3057 	struct lwp *l;
3058 	struct nfsnode **npp;
3059 {
3060 	u_int32_t *tl;
3061 	char *cp;
3062 	int32_t t1, t2;
3063 	struct vnode *newvp = (struct vnode *)0;
3064 	struct nfsnode *np, *dnp = VTONFS(dvp);
3065 	char *bpos, *dpos, *cp2;
3066 	int error = 0, fhlen;
3067 #ifndef NFS_V2_ONLY
3068 	int attrflag;
3069 #endif
3070 	struct mbuf *mreq, *mrep, *md, *mb;
3071 	nfsfh_t *nfhp;
3072 	const int v3 = NFS_ISV3(dvp);
3073 
3074 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3075 	nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3076 		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3077 	nfsm_fhtom(dnp, v3);
3078 	nfsm_strtom(name, len, NFS_MAXNAMLEN);
3079 	nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
3080 	if (npp && !error) {
3081 		nfsm_getfh(nfhp, fhlen, v3);
3082 		if (*npp) {
3083 		    np = *npp;
3084 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3085 			kmem_free(np->n_fhp, np->n_fhsize);
3086 			np->n_fhp = &np->n_fh;
3087 		    }
3088 #if NFS_SMALLFH < NFSX_V3FHMAX
3089 		    else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
3090 			np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
3091 #endif
3092 		    memcpy(np->n_fhp, nfhp, fhlen);
3093 		    np->n_fhsize = fhlen;
3094 		    newvp = NFSTOV(np);
3095 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3096 		    VREF(dvp);
3097 		    newvp = dvp;
3098 		    np = dnp;
3099 		} else {
3100 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3101 		    if (error) {
3102 			m_freem(mrep);
3103 			return (error);
3104 		    }
3105 		    newvp = NFSTOV(np);
3106 		}
3107 #ifndef NFS_V2_ONLY
3108 		if (v3) {
3109 			nfsm_postop_attr(newvp, attrflag, 0);
3110 			if (!attrflag && *npp == NULL) {
3111 				m_freem(mrep);
3112 				vput(newvp);
3113 				return (ENOENT);
3114 			}
3115 		} else
3116 #endif
3117 			nfsm_loadattr(newvp, (struct vattr *)0, 0);
3118 	}
3119 	nfsm_reqdone;
3120 	if (npp && *npp == NULL) {
3121 		if (error) {
3122 			if (newvp)
3123 				vput(newvp);
3124 		} else
3125 			*npp = np;
3126 	}
3127 	return (error);
3128 }
3129 
3130 #ifndef NFS_V2_ONLY
3131 /*
3132  * Nfs Version 3 commit rpc
3133  */
3134 int
3135 nfs_commit(vp, offset, cnt, l)
3136 	struct vnode *vp;
3137 	off_t offset;
3138 	uint32_t cnt;
3139 	struct lwp *l;
3140 {
3141 	char *cp;
3142 	u_int32_t *tl;
3143 	int32_t t1, t2;
3144 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3145 	char *bpos, *dpos, *cp2;
3146 	int error = 0, wccflag = NFSV3_WCCRATTR;
3147 	struct mbuf *mreq, *mrep, *md, *mb;
3148 	struct nfsnode *np;
3149 
3150 	KASSERT(NFS_ISV3(vp));
3151 
3152 #ifdef NFS_DEBUG_COMMIT
3153 	printf("commit %lu - %lu\n", (unsigned long)offset,
3154 	    (unsigned long)(offset + cnt));
3155 #endif
3156 
3157 	mutex_enter(&nmp->nm_lock);
3158 	if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3159 		mutex_exit(&nmp->nm_lock);
3160 		return (0);
3161 	}
3162 	mutex_exit(&nmp->nm_lock);
3163 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3164 	np = VTONFS(vp);
3165 	nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3166 	nfsm_fhtom(np, 1);
3167 	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3168 	txdr_hyper(offset, tl);
3169 	tl += 2;
3170 	*tl = txdr_unsigned(cnt);
3171 	nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3172 	nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3173 	if (!error) {
3174 		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3175 		mutex_enter(&nmp->nm_lock);
3176 		if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3177 		    memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3178 			memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3179 			error = NFSERR_STALEWRITEVERF;
3180 			nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3181 		}
3182 		mutex_exit(&nmp->nm_lock);
3183 	}
3184 	nfsm_reqdone;
3185 	return (error);
3186 }
3187 #endif
3188 
3189 /*
3190  * Kludge City..
3191  * - make nfs_bmap() essentially a no-op that does no translation
3192  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3193  *   (Maybe I could use the process's page mapping, but I was concerned that
3194  *    Kernel Write might not be enabled and also figured copyout() would do
3195  *    a lot more work than memcpy() and also it currently happens in the
3196  *    context of the swapper process (2).
3197  */
3198 int
3199 nfs_bmap(v)
3200 	void *v;
3201 {
3202 	struct vop_bmap_args /* {
3203 		struct vnode *a_vp;
3204 		daddr_t  a_bn;
3205 		struct vnode **a_vpp;
3206 		daddr_t *a_bnp;
3207 		int *a_runp;
3208 	} */ *ap = v;
3209 	struct vnode *vp = ap->a_vp;
3210 	int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3211 
3212 	if (ap->a_vpp != NULL)
3213 		*ap->a_vpp = vp;
3214 	if (ap->a_bnp != NULL)
3215 		*ap->a_bnp = ap->a_bn << bshift;
3216 	if (ap->a_runp != NULL)
3217 		*ap->a_runp = 1024 * 1024; /* XXX */
3218 	return (0);
3219 }
3220 
3221 /*
3222  * Strategy routine.
3223  * For async requests when nfsiod(s) are running, queue the request by
3224  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3225  * request.
3226  */
3227 int
3228 nfs_strategy(v)
3229 	void *v;
3230 {
3231 	struct vop_strategy_args *ap = v;
3232 	struct buf *bp = ap->a_bp;
3233 	int error = 0;
3234 
3235 	if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3236 		panic("nfs physio/async");
3237 
3238 	/*
3239 	 * If the op is asynchronous and an i/o daemon is waiting
3240 	 * queue the request, wake it up and wait for completion
3241 	 * otherwise just do it ourselves.
3242 	 */
3243 	if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3244 		error = nfs_doio(bp);
3245 	return (error);
3246 }
3247 
3248 /*
3249  * fsync vnode op. Just call nfs_flush() with commit == 1.
3250  */
3251 /* ARGSUSED */
3252 int
3253 nfs_fsync(v)
3254 	void *v;
3255 {
3256 	struct vop_fsync_args /* {
3257 		struct vnodeop_desc *a_desc;
3258 		struct vnode * a_vp;
3259 		kauth_cred_t  a_cred;
3260 		int  a_flags;
3261 		off_t offlo;
3262 		off_t offhi;
3263 		struct lwp * a_l;
3264 	} */ *ap = v;
3265 
3266 	struct vnode *vp = ap->a_vp;
3267 
3268 	if (vp->v_type != VREG)
3269 		return 0;
3270 
3271 	return (nfs_flush(vp, ap->a_cred,
3272 	    (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3273 }
3274 
3275 /*
3276  * Flush all the data associated with a vnode.
3277  */
3278 int
3279 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3280     int commit)
3281 {
3282 	struct nfsnode *np = VTONFS(vp);
3283 	int error;
3284 	int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3285 	UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3286 
3287 	mutex_enter(&vp->v_interlock);
3288 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3289 	if (np->n_flag & NWRITEERR) {
3290 		error = np->n_error;
3291 		np->n_flag &= ~NWRITEERR;
3292 	}
3293 	UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3294 	return (error);
3295 }
3296 
3297 /*
3298  * Return POSIX pathconf information applicable to nfs.
3299  *
3300  * N.B. The NFS V2 protocol doesn't support this RPC.
3301  */
3302 /* ARGSUSED */
3303 int
3304 nfs_pathconf(v)
3305 	void *v;
3306 {
3307 	struct vop_pathconf_args /* {
3308 		struct vnode *a_vp;
3309 		int a_name;
3310 		register_t *a_retval;
3311 	} */ *ap = v;
3312 	struct nfsv3_pathconf *pcp;
3313 	struct vnode *vp = ap->a_vp;
3314 	struct mbuf *mreq, *mrep, *md, *mb;
3315 	int32_t t1, t2;
3316 	u_int32_t *tl;
3317 	char *bpos, *dpos, *cp, *cp2;
3318 	int error = 0, attrflag;
3319 #ifndef NFS_V2_ONLY
3320 	struct nfsmount *nmp;
3321 	unsigned int l;
3322 	u_int64_t maxsize;
3323 #endif
3324 	const int v3 = NFS_ISV3(vp);
3325 	struct nfsnode *np = VTONFS(vp);
3326 
3327 	switch (ap->a_name) {
3328 		/* Names that can be resolved locally. */
3329 	case _PC_PIPE_BUF:
3330 		*ap->a_retval = PIPE_BUF;
3331 		break;
3332 	case _PC_SYNC_IO:
3333 		*ap->a_retval = 1;
3334 		break;
3335 	/* Names that cannot be resolved locally; do an RPC, if possible. */
3336 	case _PC_LINK_MAX:
3337 	case _PC_NAME_MAX:
3338 	case _PC_CHOWN_RESTRICTED:
3339 	case _PC_NO_TRUNC:
3340 		if (!v3) {
3341 			error = EINVAL;
3342 			break;
3343 		}
3344 		nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3345 		nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3346 		nfsm_fhtom(np, 1);
3347 		nfsm_request(np, NFSPROC_PATHCONF,
3348 		    curlwp, curlwp->l_cred);	/* XXX */
3349 		nfsm_postop_attr(vp, attrflag, 0);
3350 		if (!error) {
3351 			nfsm_dissect(pcp, struct nfsv3_pathconf *,
3352 			    NFSX_V3PATHCONF);
3353 			switch (ap->a_name) {
3354 			case _PC_LINK_MAX:
3355 				*ap->a_retval =
3356 				    fxdr_unsigned(register_t, pcp->pc_linkmax);
3357 				break;
3358 			case _PC_NAME_MAX:
3359 				*ap->a_retval =
3360 				    fxdr_unsigned(register_t, pcp->pc_namemax);
3361 				break;
3362 			case _PC_CHOWN_RESTRICTED:
3363 				*ap->a_retval =
3364 				    (pcp->pc_chownrestricted == nfs_true);
3365 				break;
3366 			case _PC_NO_TRUNC:
3367 				*ap->a_retval =
3368 				    (pcp->pc_notrunc == nfs_true);
3369 				break;
3370 			}
3371 		}
3372 		nfsm_reqdone;
3373 		break;
3374 	case _PC_FILESIZEBITS:
3375 #ifndef NFS_V2_ONLY
3376 		if (v3) {
3377 			nmp = VFSTONFS(vp->v_mount);
3378 			if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3379 				if ((error = nfs_fsinfo(nmp, vp,
3380 				    curlwp->l_cred, curlwp)) != 0) /* XXX */
3381 					break;
3382 			for (l = 0, maxsize = nmp->nm_maxfilesize;
3383 			    (maxsize >> l) > 0; l++)
3384 				;
3385 			*ap->a_retval = l + 1;
3386 		} else
3387 #endif
3388 		{
3389 			*ap->a_retval = 32;	/* NFS V2 limitation */
3390 		}
3391 		break;
3392 	default:
3393 		error = EINVAL;
3394 		break;
3395 	}
3396 
3397 	return (error);
3398 }
3399 
3400 /*
3401  * NFS advisory byte-level locks.
3402  */
3403 int
3404 nfs_advlock(v)
3405 	void *v;
3406 {
3407 	struct vop_advlock_args /* {
3408 		struct vnode *a_vp;
3409 		void *a_id;
3410 		int  a_op;
3411 		struct flock *a_fl;
3412 		int  a_flags;
3413 	} */ *ap = v;
3414 	struct nfsnode *np = VTONFS(ap->a_vp);
3415 
3416 	return lf_advlock(ap, &np->n_lockf, np->n_size);
3417 }
3418 
3419 /*
3420  * Print out the contents of an nfsnode.
3421  */
3422 int
3423 nfs_print(v)
3424 	void *v;
3425 {
3426 	struct vop_print_args /* {
3427 		struct vnode *a_vp;
3428 	} */ *ap = v;
3429 	struct vnode *vp = ap->a_vp;
3430 	struct nfsnode *np = VTONFS(vp);
3431 
3432 	printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3433 	    (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3434 	if (vp->v_type == VFIFO)
3435 		fifo_printinfo(vp);
3436 	printf("\n");
3437 	return (0);
3438 }
3439 
3440 /*
3441  * nfs unlock wrapper.
3442  */
3443 int
3444 nfs_unlock(void *v)
3445 {
3446 	struct vop_unlock_args /* {
3447 		struct vnode *a_vp;
3448 		int a_flags;
3449 	} */ *ap = v;
3450 	struct vnode *vp = ap->a_vp;
3451 
3452 	/*
3453 	 * VOP_UNLOCK can be called by nfs_loadattrcache
3454 	 * with v_data == 0.
3455 	 */
3456 	if (VTONFS(vp)) {
3457 		nfs_delayedtruncate(vp);
3458 	}
3459 
3460 	return genfs_unlock(v);
3461 }
3462 
3463 /*
3464  * nfs special file access vnode op.
3465  * Essentially just get vattr and then imitate iaccess() since the device is
3466  * local to the client.
3467  */
3468 int
3469 nfsspec_access(v)
3470 	void *v;
3471 {
3472 	struct vop_access_args /* {
3473 		struct vnode *a_vp;
3474 		int  a_mode;
3475 		kauth_cred_t a_cred;
3476 		struct lwp *a_l;
3477 	} */ *ap = v;
3478 	struct vattr va;
3479 	struct vnode *vp = ap->a_vp;
3480 	int error;
3481 
3482 	error = VOP_GETATTR(vp, &va, ap->a_cred);
3483 	if (error)
3484 		return (error);
3485 
3486         /*
3487 	 * Disallow write attempts on filesystems mounted read-only;
3488 	 * unless the file is a socket, fifo, or a block or character
3489 	 * device resident on the filesystem.
3490 	 */
3491 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3492 		switch (vp->v_type) {
3493 		case VREG:
3494 		case VDIR:
3495 		case VLNK:
3496 			return (EROFS);
3497 		default:
3498 			break;
3499 		}
3500 	}
3501 
3502 	return (vaccess(va.va_type, va.va_mode,
3503 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3504 }
3505 
3506 /*
3507  * Read wrapper for special devices.
3508  */
3509 int
3510 nfsspec_read(v)
3511 	void *v;
3512 {
3513 	struct vop_read_args /* {
3514 		struct vnode *a_vp;
3515 		struct uio *a_uio;
3516 		int  a_ioflag;
3517 		kauth_cred_t a_cred;
3518 	} */ *ap = v;
3519 	struct nfsnode *np = VTONFS(ap->a_vp);
3520 
3521 	/*
3522 	 * Set access flag.
3523 	 */
3524 	np->n_flag |= NACC;
3525 	getnanotime(&np->n_atim);
3526 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3527 }
3528 
3529 /*
3530  * Write wrapper for special devices.
3531  */
3532 int
3533 nfsspec_write(v)
3534 	void *v;
3535 {
3536 	struct vop_write_args /* {
3537 		struct vnode *a_vp;
3538 		struct uio *a_uio;
3539 		int  a_ioflag;
3540 		kauth_cred_t a_cred;
3541 	} */ *ap = v;
3542 	struct nfsnode *np = VTONFS(ap->a_vp);
3543 
3544 	/*
3545 	 * Set update flag.
3546 	 */
3547 	np->n_flag |= NUPD;
3548 	getnanotime(&np->n_mtim);
3549 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3550 }
3551 
3552 /*
3553  * Close wrapper for special devices.
3554  *
3555  * Update the times on the nfsnode then do device close.
3556  */
3557 int
3558 nfsspec_close(v)
3559 	void *v;
3560 {
3561 	struct vop_close_args /* {
3562 		struct vnode *a_vp;
3563 		int  a_fflag;
3564 		kauth_cred_t a_cred;
3565 		struct lwp *a_l;
3566 	} */ *ap = v;
3567 	struct vnode *vp = ap->a_vp;
3568 	struct nfsnode *np = VTONFS(vp);
3569 	struct vattr vattr;
3570 
3571 	if (np->n_flag & (NACC | NUPD)) {
3572 		np->n_flag |= NCHG;
3573 		if (vp->v_usecount == 1 &&
3574 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3575 			VATTR_NULL(&vattr);
3576 			if (np->n_flag & NACC)
3577 				vattr.va_atime = np->n_atim;
3578 			if (np->n_flag & NUPD)
3579 				vattr.va_mtime = np->n_mtim;
3580 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3581 		}
3582 	}
3583 	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3584 }
3585 
3586 /*
3587  * Read wrapper for fifos.
3588  */
3589 int
3590 nfsfifo_read(v)
3591 	void *v;
3592 {
3593 	struct vop_read_args /* {
3594 		struct vnode *a_vp;
3595 		struct uio *a_uio;
3596 		int  a_ioflag;
3597 		kauth_cred_t a_cred;
3598 	} */ *ap = v;
3599 	struct nfsnode *np = VTONFS(ap->a_vp);
3600 
3601 	/*
3602 	 * Set access flag.
3603 	 */
3604 	np->n_flag |= NACC;
3605 	getnanotime(&np->n_atim);
3606 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3607 }
3608 
3609 /*
3610  * Write wrapper for fifos.
3611  */
3612 int
3613 nfsfifo_write(v)
3614 	void *v;
3615 {
3616 	struct vop_write_args /* {
3617 		struct vnode *a_vp;
3618 		struct uio *a_uio;
3619 		int  a_ioflag;
3620 		kauth_cred_t a_cred;
3621 	} */ *ap = v;
3622 	struct nfsnode *np = VTONFS(ap->a_vp);
3623 
3624 	/*
3625 	 * Set update flag.
3626 	 */
3627 	np->n_flag |= NUPD;
3628 	getnanotime(&np->n_mtim);
3629 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3630 }
3631 
3632 /*
3633  * Close wrapper for fifos.
3634  *
3635  * Update the times on the nfsnode then do fifo close.
3636  */
3637 int
3638 nfsfifo_close(v)
3639 	void *v;
3640 {
3641 	struct vop_close_args /* {
3642 		struct vnode *a_vp;
3643 		int  a_fflag;
3644 		kauth_cred_t a_cred;
3645 		struct lwp *a_l;
3646 	} */ *ap = v;
3647 	struct vnode *vp = ap->a_vp;
3648 	struct nfsnode *np = VTONFS(vp);
3649 	struct vattr vattr;
3650 
3651 	if (np->n_flag & (NACC | NUPD)) {
3652 		struct timespec ts;
3653 
3654 		getnanotime(&ts);
3655 		if (np->n_flag & NACC)
3656 			np->n_atim = ts;
3657 		if (np->n_flag & NUPD)
3658 			np->n_mtim = ts;
3659 		np->n_flag |= NCHG;
3660 		if (vp->v_usecount == 1 &&
3661 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3662 			VATTR_NULL(&vattr);
3663 			if (np->n_flag & NACC)
3664 				vattr.va_atime = np->n_atim;
3665 			if (np->n_flag & NUPD)
3666 				vattr.va_mtime = np->n_mtim;
3667 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3668 		}
3669 	}
3670 	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3671 }
3672