nfs_vnops.c revision 84827
1/*
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 *    must display the following acknowledgement:
18 *	This product includes software developed by the University of
19 *	California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 *    may be used to endorse or promote products derived from this software
22 *    without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD: head/sys/nfsclient/nfs_vnops.c 84827 2001-10-11 23:38:17Z jhb $");
41
42/*
43 * vnode op calls for Sun NFS version 2 and 3
44 */
45
46#include "opt_inet.h"
47
48#include <sys/param.h>
49#include <sys/kernel.h>
50#include <sys/systm.h>
51#include <sys/resourcevar.h>
52#include <sys/proc.h>
53#include <sys/mount.h>
54#include <sys/bio.h>
55#include <sys/buf.h>
56#include <sys/malloc.h>
57#include <sys/mbuf.h>
58#include <sys/namei.h>
59#include <sys/socket.h>
60#include <sys/vnode.h>
61#include <sys/dirent.h>
62#include <sys/fcntl.h>
63#include <sys/lockf.h>
64#include <sys/stat.h>
65#include <sys/sysctl.h>
66
67#include <vm/vm.h>
68#include <vm/vm_extern.h>
69
70#include <fs/fifofs/fifo.h>
71
72#include <nfs/rpcv2.h>
73#include <nfs/nfsproto.h>
74#include <nfsclient/nfs.h>
75#include <nfsclient/nfsnode.h>
76#include <nfsclient/nfsmount.h>
77#include <nfsclient/nfs_lock.h>
78#include <nfs/xdr_subs.h>
79#include <nfsclient/nfsm_subs.h>
80
81#include <net/if.h>
82#include <netinet/in.h>
83#include <netinet/in_var.h>
84
85/* Defs */
86#define	TRUE	1
87#define	FALSE	0
88
89/*
90 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91 * calls are not in getblk() and brelse() so that they would not be necessary
92 * here.
93 */
94#ifndef B_VMIO
95#define vfs_busy_pages(bp, f)
96#endif
97
98static int	nfsspec_read(struct vop_read_args *);
99static int	nfsspec_write(struct vop_write_args *);
100static int	nfsfifo_read(struct vop_read_args *);
101static int	nfsfifo_write(struct vop_write_args *);
102static int	nfsspec_close(struct vop_close_args *);
103static int	nfsfifo_close(struct vop_close_args *);
104static int	nfs_flush(struct vnode *, struct ucred *, int, struct thread *,
105		    int);
106static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
107		    struct thread *);
108static	int	nfs_lookup(struct vop_lookup_args *);
109static	int	nfs_create(struct vop_create_args *);
110static	int	nfs_mknod(struct vop_mknod_args *);
111static	int	nfs_open(struct vop_open_args *);
112static	int	nfs_close(struct vop_close_args *);
113static	int	nfs_access(struct vop_access_args *);
114static	int	nfs_getattr(struct vop_getattr_args *);
115static	int	nfs_setattr(struct vop_setattr_args *);
116static	int	nfs_read(struct vop_read_args *);
117static	int	nfs_fsync(struct vop_fsync_args *);
118static	int	nfs_remove(struct vop_remove_args *);
119static	int	nfs_link(struct vop_link_args *);
120static	int	nfs_rename(struct vop_rename_args *);
121static	int	nfs_mkdir(struct vop_mkdir_args *);
122static	int	nfs_rmdir(struct vop_rmdir_args *);
123static	int	nfs_symlink(struct vop_symlink_args *);
124static	int	nfs_readdir(struct vop_readdir_args *);
125static	int	nfs_strategy(struct vop_strategy_args *);
126static	int	nfs_lookitup(struct vnode *, const char *, int,
127		    struct ucred *, struct thread *, struct nfsnode **);
128static	int	nfs_sillyrename(struct vnode *, struct vnode *,
129		    struct componentname *);
130static int	nfsspec_access(struct vop_access_args *);
131static int	nfs_readlink(struct vop_readlink_args *);
132static int	nfs_print(struct vop_print_args *);
133static int	nfs_advlock(struct vop_advlock_args *);
134
135/*
136 * Global vfs data structures for nfs
137 */
138vop_t **nfsv2_vnodeop_p;
139static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
140	{ &vop_default_desc,		(vop_t *) vop_defaultop },
141	{ &vop_access_desc,		(vop_t *) nfs_access },
142	{ &vop_advlock_desc,		(vop_t *) nfs_advlock },
143	{ &vop_close_desc,		(vop_t *) nfs_close },
144	{ &vop_create_desc,		(vop_t *) nfs_create },
145	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
146	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
147	{ &vop_getpages_desc,		(vop_t *) nfs_getpages },
148	{ &vop_putpages_desc,		(vop_t *) nfs_putpages },
149	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
150	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
151	{ &vop_lease_desc,		(vop_t *) vop_null },
152	{ &vop_link_desc,		(vop_t *) nfs_link },
153	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
154	{ &vop_lookup_desc,		(vop_t *) nfs_lookup },
155	{ &vop_mkdir_desc,		(vop_t *) nfs_mkdir },
156	{ &vop_mknod_desc,		(vop_t *) nfs_mknod },
157	{ &vop_open_desc,		(vop_t *) nfs_open },
158	{ &vop_print_desc,		(vop_t *) nfs_print },
159	{ &vop_read_desc,		(vop_t *) nfs_read },
160	{ &vop_readdir_desc,		(vop_t *) nfs_readdir },
161	{ &vop_readlink_desc,		(vop_t *) nfs_readlink },
162	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
163	{ &vop_remove_desc,		(vop_t *) nfs_remove },
164	{ &vop_rename_desc,		(vop_t *) nfs_rename },
165	{ &vop_rmdir_desc,		(vop_t *) nfs_rmdir },
166	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
167	{ &vop_strategy_desc,		(vop_t *) nfs_strategy },
168	{ &vop_symlink_desc,		(vop_t *) nfs_symlink },
169	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
170	{ &vop_write_desc,		(vop_t *) nfs_write },
171	{ NULL, NULL }
172};
173static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
174	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
175VNODEOP_SET(nfsv2_vnodeop_opv_desc);
176
177/*
178 * Special device vnode ops
179 */
180vop_t **spec_nfsv2nodeop_p;
181static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
182	{ &vop_default_desc,		(vop_t *) spec_vnoperate },
183	{ &vop_access_desc,		(vop_t *) nfsspec_access },
184	{ &vop_close_desc,		(vop_t *) nfsspec_close },
185	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
186	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
187	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
188	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
189	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
190	{ &vop_print_desc,		(vop_t *) nfs_print },
191	{ &vop_read_desc,		(vop_t *) nfsspec_read },
192	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
193	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
194	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
195	{ &vop_write_desc,		(vop_t *) nfsspec_write },
196	{ NULL, NULL }
197};
198static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
199	{ &spec_nfsv2nodeop_p, nfsv2_specop_entries };
200VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
201
202vop_t **fifo_nfsv2nodeop_p;
203static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
204	{ &vop_default_desc,		(vop_t *) fifo_vnoperate },
205	{ &vop_access_desc,		(vop_t *) nfsspec_access },
206	{ &vop_close_desc,		(vop_t *) nfsfifo_close },
207	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
208	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
209	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
210	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
211	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
212	{ &vop_print_desc,		(vop_t *) nfs_print },
213	{ &vop_read_desc,		(vop_t *) nfsfifo_read },
214	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
215	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
216	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
217	{ &vop_write_desc,		(vop_t *) nfsfifo_write },
218	{ NULL, NULL }
219};
220static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
221	{ &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
222VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
223
224static int	nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
225			     struct componentname *cnp, struct vattr *vap);
226static int	nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
227			      struct ucred *cred, struct thread *td);
228static int	nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
229			      int fnamelen, struct vnode *tdvp,
230			      const char *tnameptr, int tnamelen,
231			      struct ucred *cred, struct thread *td);
232static int	nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
233			     struct sillyrename *sp);
234
235/*
236 * Global variables
237 */
238struct proc	*nfs_iodwant[NFS_MAXASYNCDAEMON];
239struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
240int		 nfs_numasync = 0;
241#define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
242
243SYSCTL_DECL(_vfs_nfs);
244
245static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
246SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
247	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
248
249static int	nfsv3_commit_on_close = 0;
250SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
251	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
252#if 0
253SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
254	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
255
256SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
257	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
258#endif
259
260#define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
261			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
262			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
263static int
264nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
265    struct ucred *cred)
266{
267	const int v3 = 1;
268	u_int32_t *tl;
269	int error = 0, attrflag;
270
271	struct mbuf *mreq, *mrep, *md, *mb;
272	caddr_t bpos, dpos;
273	u_int32_t rmode;
274	struct nfsnode *np = VTONFS(vp);
275
276	nfsstats.rpccnt[NFSPROC_ACCESS]++;
277	mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
278	mb = mreq;
279	bpos = mtod(mb, caddr_t);
280	nfsm_fhtom(vp, v3);
281	tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
282	*tl = txdr_unsigned(wmode);
283	nfsm_request(vp, NFSPROC_ACCESS, td, cred);
284	nfsm_postop_attr(vp, attrflag);
285	if (!error) {
286		tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
287		rmode = fxdr_unsigned(u_int32_t, *tl);
288		np->n_mode = rmode;
289		np->n_modeuid = cred->cr_uid;
290		np->n_modestamp = time_second;
291	}
292	m_freem(mrep);
293nfsmout:
294	return error;
295}
296
297/*
298 * nfs access vnode op.
299 * For nfs version 2, just return ok. File accesses may fail later.
300 * For nfs version 3, use the access rpc to check accessibility. If file modes
301 * are changed on the server, accesses might still fail later.
302 */
303static int
304nfs_access(struct vop_access_args *ap)
305{
306	struct vnode *vp = ap->a_vp;
307	int error = 0;
308	u_int32_t mode, wmode;
309	int v3 = NFS_ISV3(vp);
310	struct nfsnode *np = VTONFS(vp);
311
312	/*
313	 * Disallow write attempts on filesystems mounted read-only;
314	 * unless the file is a socket, fifo, or a block or character
315	 * device resident on the filesystem.
316	 */
317	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
318		switch (vp->v_type) {
319		case VREG:
320		case VDIR:
321		case VLNK:
322			return (EROFS);
323		default:
324			break;
325		}
326	}
327	/*
328	 * For nfs v3, check to see if we have done this recently, and if
329	 * so return our cached result instead of making an ACCESS call.
330	 * If not, do an access rpc, otherwise you are stuck emulating
331	 * ufs_access() locally using the vattr. This may not be correct,
332	 * since the server may apply other access criteria such as
333	 * client uid-->server uid mapping that we do not know about.
334	 */
335	if (v3) {
336		if (ap->a_mode & VREAD)
337			mode = NFSV3ACCESS_READ;
338		else
339			mode = 0;
340		if (vp->v_type != VDIR) {
341			if (ap->a_mode & VWRITE)
342				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
343			if (ap->a_mode & VEXEC)
344				mode |= NFSV3ACCESS_EXECUTE;
345		} else {
346			if (ap->a_mode & VWRITE)
347				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
348					 NFSV3ACCESS_DELETE);
349			if (ap->a_mode & VEXEC)
350				mode |= NFSV3ACCESS_LOOKUP;
351		}
352		/* XXX safety belt, only make blanket request if caching */
353		if (nfsaccess_cache_timeout > 0) {
354			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
355				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
356				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
357		} else {
358			wmode = mode;
359		}
360
361		/*
362		 * Does our cached result allow us to give a definite yes to
363		 * this request?
364		 */
365		if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
366		    (ap->a_cred->cr_uid == np->n_modeuid) &&
367		    ((np->n_mode & mode) == mode)) {
368			nfsstats.accesscache_hits++;
369		} else {
370			/*
371			 * Either a no, or a don't know.  Go to the wire.
372			 */
373			nfsstats.accesscache_misses++;
374		        error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
375			if (!error) {
376				if ((np->n_mode & mode) != mode) {
377					error = EACCES;
378				}
379			}
380		}
381		return (error);
382	} else {
383		if ((error = nfsspec_access(ap)) != 0)
384			return (error);
385
386		/*
387		 * Attempt to prevent a mapped root from accessing a file
388		 * which it shouldn't.  We try to read a byte from the file
389		 * if the user is root and the file is not zero length.
390		 * After calling nfsspec_access, we should have the correct
391		 * file size cached.
392		 */
393		if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
394		    && VTONFS(vp)->n_size > 0) {
395			struct iovec aiov;
396			struct uio auio;
397			char buf[1];
398
399			aiov.iov_base = buf;
400			aiov.iov_len = 1;
401			auio.uio_iov = &aiov;
402			auio.uio_iovcnt = 1;
403			auio.uio_offset = 0;
404			auio.uio_resid = 1;
405			auio.uio_segflg = UIO_SYSSPACE;
406			auio.uio_rw = UIO_READ;
407			auio.uio_td = ap->a_td;
408
409			if (vp->v_type == VREG)
410				error = nfs_readrpc(vp, &auio, ap->a_cred);
411			else if (vp->v_type == VDIR) {
412				char* bp;
413				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
414				aiov.iov_base = bp;
415				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
416				error = nfs_readdirrpc(vp, &auio, ap->a_cred);
417				free(bp, M_TEMP);
418			} else if (vp->v_type == VLNK)
419				error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
420			else
421				error = EACCES;
422		}
423		return (error);
424	}
425}
426
427/*
428 * nfs open vnode op
429 * Check to see if the type is ok
430 * and that deletion is not in progress.
431 * For paged in text files, you will need to flush the page cache
432 * if consistency is lost.
433 */
434/* ARGSUSED */
435static int
436nfs_open(struct vop_open_args *ap)
437{
438	struct vnode *vp = ap->a_vp;
439	struct nfsnode *np = VTONFS(vp);
440	struct vattr vattr;
441	int error;
442
443	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
444#ifdef DIAGNOSTIC
445		printf("open eacces vtyp=%d\n", vp->v_type);
446#endif
447		return (EACCES);
448	}
449	/*
450	 * Get a valid lease. If cached data is stale, flush it.
451	 */
452	if (np->n_flag & NMODIFIED) {
453		if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
454			ap->a_td, 1)) == EINTR)
455			return (error);
456		np->n_attrstamp = 0;
457		if (vp->v_type == VDIR)
458			np->n_direofoffset = 0;
459		error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
460		if (error)
461			return (error);
462		np->n_mtime = vattr.va_mtime.tv_sec;
463	} else {
464		error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
465		if (error)
466			return (error);
467		if (np->n_mtime != vattr.va_mtime.tv_sec) {
468			if (vp->v_type == VDIR)
469				np->n_direofoffset = 0;
470			if ((error = nfs_vinvalbuf(vp, V_SAVE,
471				ap->a_cred, ap->a_td, 1)) == EINTR)
472				return (error);
473			np->n_mtime = vattr.va_mtime.tv_sec;
474		}
475	}
476	np->n_attrstamp = 0; /* For Open/Close consistency */
477	return (0);
478}
479
480/*
481 * nfs close vnode op
482 * What an NFS client should do upon close after writing is a debatable issue.
483 * Most NFS clients push delayed writes to the server upon close, basically for
484 * two reasons:
485 * 1 - So that any write errors may be reported back to the client process
486 *     doing the close system call. By far the two most likely errors are
487 *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
488 * 2 - To put a worst case upper bound on cache inconsistency between
489 *     multiple clients for the file.
490 * There is also a consistency problem for Version 2 of the protocol w.r.t.
491 * not being able to tell if other clients are writing a file concurrently,
492 * since there is no way of knowing if the changed modify time in the reply
493 * is only due to the write for this client.
494 * (NFS Version 3 provides weak cache consistency data in the reply that
495 *  should be sufficient to detect and handle this case.)
496 *
497 * The current code does the following:
498 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
499 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
500 *                     or commit them (this satisfies 1 and 2 except for the
501 *                     case where the server crashes after this close but
502 *                     before the commit RPC, which is felt to be "good
503 *                     enough". Changing the last argument to nfs_flush() to
504 *                     a 1 would force a commit operation, if it is felt a
505 *                     commit is necessary now.
506 */
507/* ARGSUSED */
508static int
509nfs_close(struct vop_close_args *ap)
510{
511	struct vnode *vp = ap->a_vp;
512	struct nfsnode *np = VTONFS(vp);
513	int error = 0;
514
515	if (vp->v_type == VREG) {
516	    if (np->n_flag & NMODIFIED) {
517		if (NFS_ISV3(vp)) {
518		    /*
519		     * Under NFSv3 we have dirty buffers to dispose of.  We
520		     * must flush them to the NFS server.  We have the option
521		     * of waiting all the way through the commit rpc or just
522		     * waiting for the initial write.  The default is to only
523		     * wait through the initial write so the data is in the
524		     * server's cache, which is roughly similar to the state
525		     * a standard disk subsystem leaves the file in on close().
526		     *
527		     * We cannot clear the NMODIFIED bit in np->n_flag due to
528		     * potential races with other processes, and certainly
529		     * cannot clear it if we don't commit.
530		     */
531		    int cm = nfsv3_commit_on_close ? 1 : 0;
532		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm);
533		    /* np->n_flag &= ~NMODIFIED; */
534		} else {
535		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
536		}
537		np->n_attrstamp = 0;
538	    }
539	    if (np->n_flag & NWRITEERR) {
540		np->n_flag &= ~NWRITEERR;
541		error = np->n_error;
542	    }
543	}
544	return (error);
545}
546
547/*
548 * nfs getattr call from vfs.
549 */
550static int
551nfs_getattr(struct vop_getattr_args *ap)
552{
553	struct vnode *vp = ap->a_vp;
554	struct nfsnode *np = VTONFS(vp);
555	u_int32_t *tl;
556	caddr_t bpos, dpos;
557	int error = 0;
558	struct mbuf *mreq, *mrep, *md, *mb;
559	int v3 = NFS_ISV3(vp);
560
561	/*
562	 * Update local times for special files.
563	 */
564	if (np->n_flag & (NACC | NUPD))
565		np->n_flag |= NCHG;
566	/*
567	 * First look in the cache.
568	 */
569	if (nfs_getattrcache(vp, ap->a_vap) == 0)
570		return (0);
571
572	if (v3 && nfsaccess_cache_timeout > 0) {
573		nfsstats.accesscache_misses++;
574		nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
575		if (nfs_getattrcache(vp, ap->a_vap) == 0)
576			return (0);
577	}
578
579	nfsstats.rpccnt[NFSPROC_GETATTR]++;
580	mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
581	mb = mreq;
582	bpos = mtod(mb, caddr_t);
583	nfsm_fhtom(vp, v3);
584	nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
585	if (!error) {
586		nfsm_loadattr(vp, ap->a_vap);
587	}
588	m_freem(mrep);
589nfsmout:
590	return (error);
591}
592
593/*
594 * nfs setattr call.
595 */
596static int
597nfs_setattr(struct vop_setattr_args *ap)
598{
599	struct vnode *vp = ap->a_vp;
600	struct nfsnode *np = VTONFS(vp);
601	struct vattr *vap = ap->a_vap;
602	int error = 0;
603	u_quad_t tsize;
604
605#ifndef nolint
606	tsize = (u_quad_t)0;
607#endif
608
609	/*
610	 * Setting of flags is not supported.
611	 */
612	if (vap->va_flags != VNOVAL)
613		return (EOPNOTSUPP);
614
615	/*
616	 * Disallow write attempts if the filesystem is mounted read-only.
617	 */
618  	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
619	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
620	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
621	    (vp->v_mount->mnt_flag & MNT_RDONLY))
622		return (EROFS);
623	if (vap->va_size != VNOVAL) {
624 		switch (vp->v_type) {
625 		case VDIR:
626 			return (EISDIR);
627 		case VCHR:
628 		case VBLK:
629 		case VSOCK:
630 		case VFIFO:
631			if (vap->va_mtime.tv_sec == VNOVAL &&
632			    vap->va_atime.tv_sec == VNOVAL &&
633			    vap->va_mode == (mode_t)VNOVAL &&
634			    vap->va_uid == (uid_t)VNOVAL &&
635			    vap->va_gid == (gid_t)VNOVAL)
636				return (0);
637 			vap->va_size = VNOVAL;
638 			break;
639 		default:
640			/*
641			 * Disallow write attempts if the filesystem is
642			 * mounted read-only.
643			 */
644			if (vp->v_mount->mnt_flag & MNT_RDONLY)
645				return (EROFS);
646			vnode_pager_setsize(vp, vap->va_size);
647 			if (np->n_flag & NMODIFIED) {
648 			    if (vap->va_size == 0)
649 				error = nfs_vinvalbuf(vp, 0,
650 					ap->a_cred, ap->a_td, 1);
651 			    else
652 				error = nfs_vinvalbuf(vp, V_SAVE,
653 					ap->a_cred, ap->a_td, 1);
654 			    if (error) {
655				vnode_pager_setsize(vp, np->n_size);
656 				return (error);
657			    }
658 			}
659 			tsize = np->n_size;
660 			np->n_size = np->n_vattr.va_size = vap->va_size;
661  		};
662  	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
663		vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
664		vp->v_type == VREG &&
665  		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
666		 ap->a_td, 1)) == EINTR)
667		return (error);
668	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
669	if (error && vap->va_size != VNOVAL) {
670		np->n_size = np->n_vattr.va_size = tsize;
671		vnode_pager_setsize(vp, np->n_size);
672	}
673	return (error);
674}
675
676/*
677 * Do an nfs setattr rpc.
678 */
679static int
680nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
681    struct thread *td)
682{
683	struct nfsv2_sattr *sp;
684	caddr_t bpos, dpos;
685	u_int32_t *tl;
686	int error = 0, wccflag = NFSV3_WCCRATTR;
687	struct mbuf *mreq, *mrep, *md, *mb;
688	int v3 = NFS_ISV3(vp);
689
690	nfsstats.rpccnt[NFSPROC_SETATTR]++;
691	mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
692	mb = mreq;
693	bpos = mtod(mb, caddr_t);
694	nfsm_fhtom(vp, v3);
695	if (v3) {
696		nfsm_v3attrbuild(vap, TRUE);
697		tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
698		*tl = nfs_false;
699	} else {
700		sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
701		if (vap->va_mode == (mode_t)VNOVAL)
702			sp->sa_mode = nfs_xdrneg1;
703		else
704			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
705		if (vap->va_uid == (uid_t)VNOVAL)
706			sp->sa_uid = nfs_xdrneg1;
707		else
708			sp->sa_uid = txdr_unsigned(vap->va_uid);
709		if (vap->va_gid == (gid_t)VNOVAL)
710			sp->sa_gid = nfs_xdrneg1;
711		else
712			sp->sa_gid = txdr_unsigned(vap->va_gid);
713		sp->sa_size = txdr_unsigned(vap->va_size);
714		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
715		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
716	}
717	nfsm_request(vp, NFSPROC_SETATTR, td, cred);
718	if (v3) {
719		nfsm_wcc_data(vp, wccflag);
720	} else
721		nfsm_loadattr(vp, (struct vattr *)0);
722	m_freem(mrep);
723nfsmout:
724	return (error);
725}
726
727/*
728 * nfs lookup call, one step at a time...
729 * First look in cache
730 * If not found, unlock the directory nfsnode and do the rpc
731 */
732static int
733nfs_lookup(struct vop_lookup_args *ap)
734{
735	struct componentname *cnp = ap->a_cnp;
736	struct vnode *dvp = ap->a_dvp;
737	struct vnode **vpp = ap->a_vpp;
738	int flags = cnp->cn_flags;
739	struct vnode *newvp;
740	u_int32_t *tl;
741	struct nfsmount *nmp;
742	caddr_t bpos, dpos;
743	struct mbuf *mreq, *mrep, *md, *mb;
744	long len;
745	nfsfh_t *fhp;
746	struct nfsnode *np;
747	int lockparent, wantparent, error = 0, attrflag, fhsize;
748	int v3 = NFS_ISV3(dvp);
749	struct thread *td = cnp->cn_thread;
750
751	*vpp = NULLVP;
752	cnp->cn_flags &= ~PDIRUNLOCK;
753	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
754	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
755		return (EROFS);
756	if (dvp->v_type != VDIR)
757		return (ENOTDIR);
758	lockparent = flags & LOCKPARENT;
759	wantparent = flags & (LOCKPARENT|WANTPARENT);
760	nmp = VFSTONFS(dvp->v_mount);
761	np = VTONFS(dvp);
762	if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
763		struct vattr vattr;
764		int vpid;
765
766		if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
767			*vpp = NULLVP;
768			return (error);
769		}
770
771		newvp = *vpp;
772		vpid = newvp->v_id;
773		/*
774		 * See the comment starting `Step through' in ufs/ufs_lookup.c
775		 * for an explanation of the locking protocol
776		 */
777		if (dvp == newvp) {
778			VREF(newvp);
779			error = 0;
780		} else if (flags & ISDOTDOT) {
781			VOP_UNLOCK(dvp, 0, td);
782			cnp->cn_flags |= PDIRUNLOCK;
783			error = vget(newvp, LK_EXCLUSIVE, td);
784			if (!error && lockparent && (flags & ISLASTCN)) {
785				error = vn_lock(dvp, LK_EXCLUSIVE, td);
786				if (error == 0)
787					cnp->cn_flags &= ~PDIRUNLOCK;
788			}
789		} else {
790			error = vget(newvp, LK_EXCLUSIVE, td);
791			if (!lockparent || error || !(flags & ISLASTCN)) {
792				VOP_UNLOCK(dvp, 0, td);
793				cnp->cn_flags |= PDIRUNLOCK;
794			}
795		}
796		if (!error) {
797			if (vpid == newvp->v_id) {
798			   if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
799			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
800				nfsstats.lookupcache_hits++;
801				if (cnp->cn_nameiop != LOOKUP &&
802				    (flags & ISLASTCN))
803					cnp->cn_flags |= SAVENAME;
804				return (0);
805			   }
806			   cache_purge(newvp);
807			}
808			vput(newvp);
809			if (lockparent && dvp != newvp && (flags & ISLASTCN))
810				VOP_UNLOCK(dvp, 0, td);
811		}
812		error = vn_lock(dvp, LK_EXCLUSIVE, td);
813		*vpp = NULLVP;
814		if (error) {
815			cnp->cn_flags |= PDIRUNLOCK;
816			return (error);
817		}
818		cnp->cn_flags &= ~PDIRUNLOCK;
819	}
820	error = 0;
821	newvp = NULLVP;
822	nfsstats.lookupcache_misses++;
823	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
824	len = cnp->cn_namelen;
825	mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
826		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
827	mb = mreq;
828	bpos = mtod(mb, caddr_t);
829	nfsm_fhtom(dvp, v3);
830	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
831	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
832	if (error) {
833		nfsm_postop_attr(dvp, attrflag);
834		m_freem(mrep);
835		goto nfsmout;
836	}
837	nfsm_getfh(fhp, fhsize, v3);
838
839	/*
840	 * Handle RENAME case...
841	 */
842	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
843		if (NFS_CMPFH(np, fhp, fhsize)) {
844			m_freem(mrep);
845			return (EISDIR);
846		}
847		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
848		if (error) {
849			m_freem(mrep);
850			return (error);
851		}
852		newvp = NFSTOV(np);
853		if (v3) {
854			nfsm_postop_attr(newvp, attrflag);
855			nfsm_postop_attr(dvp, attrflag);
856		} else
857			nfsm_loadattr(newvp, (struct vattr *)0);
858		*vpp = newvp;
859		m_freem(mrep);
860		cnp->cn_flags |= SAVENAME;
861		if (!lockparent) {
862			VOP_UNLOCK(dvp, 0, td);
863			cnp->cn_flags |= PDIRUNLOCK;
864		}
865		return (0);
866	}
867
868	if (flags & ISDOTDOT) {
869		VOP_UNLOCK(dvp, 0, td);
870		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
871		if (error) {
872			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
873			return (error);
874		}
875		newvp = NFSTOV(np);
876		if (lockparent && (flags & ISLASTCN)) {
877			error = vn_lock(dvp, LK_EXCLUSIVE, td);
878			if (error) {
879				cnp->cn_flags |= PDIRUNLOCK;
880		    		vput(newvp);
881				return (error);
882			}
883		} else
884			cnp->cn_flags |= PDIRUNLOCK;
885	} else if (NFS_CMPFH(np, fhp, fhsize)) {
886		VREF(dvp);
887		newvp = dvp;
888	} else {
889		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
890		if (error) {
891			m_freem(mrep);
892			return (error);
893		}
894		if (!lockparent || !(flags & ISLASTCN)) {
895			cnp->cn_flags |= PDIRUNLOCK;
896			VOP_UNLOCK(dvp, 0, td);
897		}
898		newvp = NFSTOV(np);
899	}
900	if (v3) {
901		nfsm_postop_attr(newvp, attrflag);
902		nfsm_postop_attr(dvp, attrflag);
903	} else
904		nfsm_loadattr(newvp, (struct vattr *)0);
905	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
906		cnp->cn_flags |= SAVENAME;
907	if ((cnp->cn_flags & MAKEENTRY) &&
908	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
909		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
910		cache_enter(dvp, newvp, cnp);
911	}
912	*vpp = newvp;
913	m_freem(mrep);
914nfsmout:
915	if (error) {
916		if (newvp != NULLVP) {
917			vrele(newvp);
918			*vpp = NULLVP;
919		}
920		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
921		    (flags & ISLASTCN) && error == ENOENT) {
922			if (!lockparent) {
923				VOP_UNLOCK(dvp, 0, td);
924				cnp->cn_flags |= PDIRUNLOCK;
925			}
926			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
927				error = EROFS;
928			else
929				error = EJUSTRETURN;
930		}
931		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
932			cnp->cn_flags |= SAVENAME;
933	}
934	return (error);
935}
936
937/*
938 * nfs read call.
939 * Just call nfs_bioread() to do the work.
940 */
941static int
942nfs_read(struct vop_read_args *ap)
943{
944	struct vnode *vp = ap->a_vp;
945
946	if (vp->v_type != VREG)
947		return (EPERM);
948	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
949}
950
951/*
952 * nfs readlink call
953 */
954static int
955nfs_readlink(struct vop_readlink_args *ap)
956{
957	struct vnode *vp = ap->a_vp;
958
959	if (vp->v_type != VLNK)
960		return (EINVAL);
961	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
962}
963
964/*
965 * Do a readlink rpc.
966 * Called by nfs_doio() from below the buffer cache.
967 */
968int
969nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
970{
971	u_int32_t *tl;
972	caddr_t bpos, dpos;
973	int error = 0, len, attrflag;
974	struct mbuf *mreq, *mrep, *md, *mb;
975	int v3 = NFS_ISV3(vp);
976
977	nfsstats.rpccnt[NFSPROC_READLINK]++;
978	mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
979	mb = mreq;
980	bpos = mtod(mb, caddr_t);
981	nfsm_fhtom(vp, v3);
982	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
983	if (v3)
984		nfsm_postop_attr(vp, attrflag);
985	if (!error) {
986		nfsm_strsiz(len, NFS_MAXPATHLEN);
987		if (len == NFS_MAXPATHLEN) {
988			struct nfsnode *np = VTONFS(vp);
989			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
990				len = np->n_size;
991		}
992		nfsm_mtouio(uiop, len);
993	}
994	m_freem(mrep);
995nfsmout:
996	return (error);
997}
998
999/*
1000 * nfs read rpc call
1001 * Ditto above
1002 */
1003int
1004nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1005{
1006	u_int32_t *tl;
1007	caddr_t bpos, dpos;
1008	struct mbuf *mreq, *mrep, *md, *mb;
1009	struct nfsmount *nmp;
1010	int error = 0, len, retlen, tsiz, eof, attrflag;
1011	int v3 = NFS_ISV3(vp);
1012
1013#ifndef nolint
1014	eof = 0;
1015#endif
1016	nmp = VFSTONFS(vp->v_mount);
1017	tsiz = uiop->uio_resid;
1018	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1019		return (EFBIG);
1020	while (tsiz > 0) {
1021		nfsstats.rpccnt[NFSPROC_READ]++;
1022		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1023		mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1024		mb = mreq;
1025		bpos = mtod(mb, caddr_t);
1026		nfsm_fhtom(vp, v3);
1027		tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1028		if (v3) {
1029			txdr_hyper(uiop->uio_offset, tl);
1030			*(tl + 2) = txdr_unsigned(len);
1031		} else {
1032			*tl++ = txdr_unsigned(uiop->uio_offset);
1033			*tl++ = txdr_unsigned(len);
1034			*tl = 0;
1035		}
1036		nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1037		if (v3) {
1038			nfsm_postop_attr(vp, attrflag);
1039			if (error) {
1040				m_freem(mrep);
1041				goto nfsmout;
1042			}
1043			tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1044			eof = fxdr_unsigned(int, *(tl + 1));
1045		} else
1046			nfsm_loadattr(vp, (struct vattr *)0);
1047		nfsm_strsiz(retlen, nmp->nm_rsize);
1048		nfsm_mtouio(uiop, retlen);
1049		m_freem(mrep);
1050		tsiz -= retlen;
1051		if (v3) {
1052			if (eof || retlen == 0)
1053				tsiz = 0;
1054		} else if (retlen < len)
1055			tsiz = 0;
1056	}
1057nfsmout:
1058	return (error);
1059}
1060
1061/*
1062 * nfs write call
1063 */
1064int
1065nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1066    int *iomode, int *must_commit)
1067{
1068	u_int32_t *tl;
1069	int32_t backup;
1070	caddr_t bpos, dpos;
1071	struct mbuf *mreq, *mrep, *md, *mb;
1072	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1073	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1074	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1075
1076#ifndef DIAGNOSTIC
1077	if (uiop->uio_iovcnt != 1)
1078		panic("nfs: writerpc iovcnt > 1");
1079#endif
1080	*must_commit = 0;
1081	tsiz = uiop->uio_resid;
1082	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1083		return (EFBIG);
1084	while (tsiz > 0) {
1085		nfsstats.rpccnt[NFSPROC_WRITE]++;
1086		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1087		mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1088			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1089		mb = mreq;
1090		bpos = mtod(mb, caddr_t);
1091		nfsm_fhtom(vp, v3);
1092		if (v3) {
1093			tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1094			txdr_hyper(uiop->uio_offset, tl);
1095			tl += 2;
1096			*tl++ = txdr_unsigned(len);
1097			*tl++ = txdr_unsigned(*iomode);
1098			*tl = txdr_unsigned(len);
1099		} else {
1100			u_int32_t x;
1101
1102			tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1103			/* Set both "begin" and "current" to non-garbage. */
1104			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1105			*tl++ = x;	/* "begin offset" */
1106			*tl++ = x;	/* "current offset" */
1107			x = txdr_unsigned(len);
1108			*tl++ = x;	/* total to this offset */
1109			*tl = x;	/* size of this write */
1110		}
1111		nfsm_uiotom(uiop, len);
1112		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1113		if (v3) {
1114			wccflag = NFSV3_WCCCHK;
1115			nfsm_wcc_data(vp, wccflag);
1116			if (!error) {
1117				tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1118					+ NFSX_V3WRITEVERF);
1119				rlen = fxdr_unsigned(int, *tl++);
1120				if (rlen == 0) {
1121					error = NFSERR_IO;
1122					m_freem(mrep);
1123					break;
1124				} else if (rlen < len) {
1125					backup = len - rlen;
1126					uiop->uio_iov->iov_base -= backup;
1127					uiop->uio_iov->iov_len += backup;
1128					uiop->uio_offset -= backup;
1129					uiop->uio_resid += backup;
1130					len = rlen;
1131				}
1132				commit = fxdr_unsigned(int, *tl++);
1133
1134				/*
1135				 * Return the lowest committment level
1136				 * obtained by any of the RPCs.
1137				 */
1138				if (committed == NFSV3WRITE_FILESYNC)
1139					committed = commit;
1140				else if (committed == NFSV3WRITE_DATASYNC &&
1141					commit == NFSV3WRITE_UNSTABLE)
1142					committed = commit;
1143				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1144				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1145					NFSX_V3WRITEVERF);
1146				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1147				} else if (bcmp((caddr_t)tl,
1148				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1149				    *must_commit = 1;
1150				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1151					NFSX_V3WRITEVERF);
1152				}
1153			}
1154		} else
1155		    nfsm_loadattr(vp, (struct vattr *)0);
1156		if (wccflag)
1157		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1158		m_freem(mrep);
1159		if (error)
1160			break;
1161		tsiz -= len;
1162	}
1163nfsmout:
1164	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1165		committed = NFSV3WRITE_FILESYNC;
1166	*iomode = committed;
1167	if (error)
1168		uiop->uio_resid = tsiz;
1169	return (error);
1170}
1171
1172/*
1173 * nfs mknod rpc
1174 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1175 * mode set to specify the file type and the size field for rdev.
1176 */
1177static int
1178nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1179    struct vattr *vap)
1180{
1181	struct nfsv2_sattr *sp;
1182	u_int32_t *tl;
1183	struct vnode *newvp = (struct vnode *)0;
1184	struct nfsnode *np = (struct nfsnode *)0;
1185	struct vattr vattr;
1186	caddr_t bpos, dpos;
1187	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1188	struct mbuf *mreq, *mrep, *md, *mb;
1189	u_int32_t rdev;
1190	int v3 = NFS_ISV3(dvp);
1191
1192	if (vap->va_type == VCHR || vap->va_type == VBLK)
1193		rdev = txdr_unsigned(vap->va_rdev);
1194	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1195		rdev = nfs_xdrneg1;
1196	else {
1197		return (EOPNOTSUPP);
1198	}
1199	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1200		return (error);
1201	}
1202	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1203	mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1204		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1205	mb = mreq;
1206	bpos = mtod(mb, caddr_t);
1207	nfsm_fhtom(dvp, v3);
1208	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1209	if (v3) {
1210		tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1211		*tl++ = vtonfsv3_type(vap->va_type);
1212		nfsm_v3attrbuild(vap, FALSE);
1213		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1214			tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1215			*tl++ = txdr_unsigned(umajor(vap->va_rdev));
1216			*tl = txdr_unsigned(uminor(vap->va_rdev));
1217		}
1218	} else {
1219		sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1220		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1221		sp->sa_uid = nfs_xdrneg1;
1222		sp->sa_gid = nfs_xdrneg1;
1223		sp->sa_size = rdev;
1224		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1225		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1226	}
1227	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1228	if (!error) {
1229		nfsm_mtofh(dvp, newvp, v3, gotvp);
1230		if (!gotvp) {
1231			if (newvp) {
1232				vput(newvp);
1233				newvp = (struct vnode *)0;
1234			}
1235			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1236			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1237			if (!error)
1238				newvp = NFSTOV(np);
1239		}
1240	}
1241	if (v3)
1242		nfsm_wcc_data(dvp, wccflag);
1243	m_freem(mrep);
1244nfsmout:
1245	if (error) {
1246		if (newvp)
1247			vput(newvp);
1248	} else {
1249		if (cnp->cn_flags & MAKEENTRY)
1250			cache_enter(dvp, newvp, cnp);
1251		*vpp = newvp;
1252	}
1253	VTONFS(dvp)->n_flag |= NMODIFIED;
1254	if (!wccflag)
1255		VTONFS(dvp)->n_attrstamp = 0;
1256	return (error);
1257}
1258
1259/*
1260 * nfs mknod vop
1261 * just call nfs_mknodrpc() to do the work.
1262 */
1263/* ARGSUSED */
1264static int
1265nfs_mknod(struct vop_mknod_args *ap)
1266{
1267
1268	return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1269}
1270
1271static u_long create_verf;
1272/*
1273 * nfs file create call
1274 */
1275static int
1276nfs_create(struct vop_create_args *ap)
1277{
1278	struct vnode *dvp = ap->a_dvp;
1279	struct vattr *vap = ap->a_vap;
1280	struct componentname *cnp = ap->a_cnp;
1281	struct nfsv2_sattr *sp;
1282	u_int32_t *tl;
1283	struct nfsnode *np = (struct nfsnode *)0;
1284	struct vnode *newvp = (struct vnode *)0;
1285	caddr_t bpos, dpos;
1286	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1287	struct mbuf *mreq, *mrep, *md, *mb;
1288	struct vattr vattr;
1289	int v3 = NFS_ISV3(dvp);
1290
1291	/*
1292	 * Oops, not for me..
1293	 */
1294	if (vap->va_type == VSOCK)
1295		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1296
1297	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1298		return (error);
1299	}
1300	if (vap->va_vaflags & VA_EXCLUSIVE)
1301		fmode |= O_EXCL;
1302again:
1303	nfsstats.rpccnt[NFSPROC_CREATE]++;
1304	mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1305		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1306	mb = mreq;
1307	bpos = mtod(mb, caddr_t);
1308	nfsm_fhtom(dvp, v3);
1309	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1310	if (v3) {
1311		tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1312		if (fmode & O_EXCL) {
1313			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1314			tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1315#ifdef INET
1316			if (!TAILQ_EMPTY(&in_ifaddrhead))
1317				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1318			else
1319#endif
1320				*tl++ = create_verf;
1321			*tl = ++create_verf;
1322		} else {
1323			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1324			nfsm_v3attrbuild(vap, FALSE);
1325		}
1326	} else {
1327		sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1328		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1329		sp->sa_uid = nfs_xdrneg1;
1330		sp->sa_gid = nfs_xdrneg1;
1331		sp->sa_size = 0;
1332		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1333		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1334	}
1335	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1336	if (!error) {
1337		nfsm_mtofh(dvp, newvp, v3, gotvp);
1338		if (!gotvp) {
1339			if (newvp) {
1340				vput(newvp);
1341				newvp = (struct vnode *)0;
1342			}
1343			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1344			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1345			if (!error)
1346				newvp = NFSTOV(np);
1347		}
1348	}
1349	if (v3)
1350		nfsm_wcc_data(dvp, wccflag);
1351	m_freem(mrep);
1352nfsmout:
1353	if (error) {
1354		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1355			fmode &= ~O_EXCL;
1356			goto again;
1357		}
1358		if (newvp)
1359			vput(newvp);
1360	} else if (v3 && (fmode & O_EXCL)) {
1361		/*
1362		 * We are normally called with only a partially initialized
1363		 * VAP.  Since the NFSv3 spec says that server may use the
1364		 * file attributes to store the verifier, the spec requires
1365		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1366		 * in atime, but we can't really assume that all servers will
1367		 * so we ensure that our SETATTR sets both atime and mtime.
1368		 */
1369		if (vap->va_mtime.tv_sec == VNOVAL)
1370			vfs_timestamp(&vap->va_mtime);
1371		if (vap->va_atime.tv_sec == VNOVAL)
1372			vap->va_atime = vap->va_mtime;
1373		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
1374	}
1375	if (!error) {
1376		if (cnp->cn_flags & MAKEENTRY)
1377			cache_enter(dvp, newvp, cnp);
1378		*ap->a_vpp = newvp;
1379	}
1380	VTONFS(dvp)->n_flag |= NMODIFIED;
1381	if (!wccflag)
1382		VTONFS(dvp)->n_attrstamp = 0;
1383	return (error);
1384}
1385
1386/*
1387 * nfs file remove call
1388 * To try and make nfs semantics closer to ufs semantics, a file that has
1389 * other processes using the vnode is renamed instead of removed and then
1390 * removed later on the last close.
1391 * - If v_usecount > 1
1392 *	  If a rename is not already in the works
1393 *	     call nfs_sillyrename() to set it up
1394 *     else
1395 *	  do the remove rpc
1396 */
1397static int
1398nfs_remove(struct vop_remove_args *ap)
1399{
1400	struct vnode *vp = ap->a_vp;
1401	struct vnode *dvp = ap->a_dvp;
1402	struct componentname *cnp = ap->a_cnp;
1403	struct nfsnode *np = VTONFS(vp);
1404	int error = 0;
1405	struct vattr vattr;
1406
1407#ifndef DIAGNOSTIC
1408	if ((cnp->cn_flags & HASBUF) == 0)
1409		panic("nfs_remove: no name");
1410	if (vp->v_usecount < 1)
1411		panic("nfs_remove: bad v_usecount");
1412#endif
1413	if (vp->v_type == VDIR)
1414		error = EPERM;
1415	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1416	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1417	    vattr.va_nlink > 1)) {
1418		/*
1419		 * Purge the name cache so that the chance of a lookup for
1420		 * the name succeeding while the remove is in progress is
1421		 * minimized. Without node locking it can still happen, such
1422		 * that an I/O op returns ESTALE, but since you get this if
1423		 * another host removes the file..
1424		 */
1425		cache_purge(vp);
1426		/*
1427		 * throw away biocache buffers, mainly to avoid
1428		 * unnecessary delayed writes later.
1429		 */
1430		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_thread, 1);
1431		/* Do the rpc */
1432		if (error != EINTR)
1433			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1434				cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1435		/*
1436		 * Kludge City: If the first reply to the remove rpc is lost..
1437		 *   the reply to the retransmitted request will be ENOENT
1438		 *   since the file was in fact removed
1439		 *   Therefore, we cheat and return success.
1440		 */
1441		if (error == ENOENT)
1442			error = 0;
1443	} else if (!np->n_sillyrename)
1444		error = nfs_sillyrename(dvp, vp, cnp);
1445	np->n_attrstamp = 0;
1446	return (error);
1447}
1448
1449/*
1450 * nfs file remove rpc called from nfs_inactive
1451 */
1452int
1453nfs_removeit(struct sillyrename *sp)
1454{
1455
1456	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1457		(struct thread *)0));
1458}
1459
1460/*
1461 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1462 */
1463static int
1464nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1465    struct ucred *cred, struct thread *td)
1466{
1467	u_int32_t *tl;
1468	caddr_t bpos, dpos;
1469	int error = 0, wccflag = NFSV3_WCCRATTR;
1470	struct mbuf *mreq, *mrep, *md, *mb;
1471	int v3 = NFS_ISV3(dvp);
1472
1473	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1474	mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1475		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1476	mb = mreq;
1477	bpos = mtod(mb, caddr_t);
1478	nfsm_fhtom(dvp, v3);
1479	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1480	nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1481	if (v3)
1482		nfsm_wcc_data(dvp, wccflag);
1483	m_freem(mrep);
1484nfsmout:
1485	VTONFS(dvp)->n_flag |= NMODIFIED;
1486	if (!wccflag)
1487		VTONFS(dvp)->n_attrstamp = 0;
1488	return (error);
1489}
1490
1491/*
1492 * nfs file rename call
1493 */
1494static int
1495nfs_rename(struct vop_rename_args *ap)
1496{
1497	struct vnode *fvp = ap->a_fvp;
1498	struct vnode *tvp = ap->a_tvp;
1499	struct vnode *fdvp = ap->a_fdvp;
1500	struct vnode *tdvp = ap->a_tdvp;
1501	struct componentname *tcnp = ap->a_tcnp;
1502	struct componentname *fcnp = ap->a_fcnp;
1503	int error;
1504
1505#ifndef DIAGNOSTIC
1506	if ((tcnp->cn_flags & HASBUF) == 0 ||
1507	    (fcnp->cn_flags & HASBUF) == 0)
1508		panic("nfs_rename: no name");
1509#endif
1510	/* Check for cross-device rename */
1511	if ((fvp->v_mount != tdvp->v_mount) ||
1512	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1513		error = EXDEV;
1514		goto out;
1515	}
1516
1517	/*
1518	 * We have to flush B_DELWRI data prior to renaming
1519	 * the file.  If we don't, the delayed-write buffers
1520	 * can be flushed out later after the file has gone stale
1521	 * under NFSV3.  NFSV2 does not have this problem because
1522	 * ( as far as I can tell ) it flushes dirty buffers more
1523	 * often.
1524	 */
1525
1526	VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_thread);
1527	if (tvp)
1528	    VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_thread);
1529
1530	/*
1531	 * If the tvp exists and is in use, sillyrename it before doing the
1532	 * rename of the new file over it.
1533	 * XXX Can't sillyrename a directory.
1534	 */
1535	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1536		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1537		vput(tvp);
1538		tvp = NULL;
1539	}
1540
1541	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1542		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1543		tcnp->cn_thread);
1544
1545	if (fvp->v_type == VDIR) {
1546		if (tvp != NULL && tvp->v_type == VDIR)
1547			cache_purge(tdvp);
1548		cache_purge(fdvp);
1549	}
1550
1551out:
1552	if (tdvp == tvp)
1553		vrele(tdvp);
1554	else
1555		vput(tdvp);
1556	if (tvp)
1557		vput(tvp);
1558	vrele(fdvp);
1559	vrele(fvp);
1560	/*
1561	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1562	 */
1563	if (error == ENOENT)
1564		error = 0;
1565	return (error);
1566}
1567
1568/*
1569 * nfs file rename rpc called from nfs_remove() above
1570 */
1571static int
1572nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1573    struct sillyrename *sp)
1574{
1575
1576	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1577	    sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1578}
1579
1580/*
1581 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1582 */
1583static int
1584nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1585    struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1586    struct thread *td)
1587{
1588	u_int32_t *tl;
1589	caddr_t bpos, dpos;
1590	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1591	struct mbuf *mreq, *mrep, *md, *mb;
1592	int v3 = NFS_ISV3(fdvp);
1593
1594	nfsstats.rpccnt[NFSPROC_RENAME]++;
1595	mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1596		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1597		nfsm_rndup(tnamelen));
1598	mb = mreq;
1599	bpos = mtod(mb, caddr_t);
1600	nfsm_fhtom(fdvp, v3);
1601	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1602	nfsm_fhtom(tdvp, v3);
1603	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1604	nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1605	if (v3) {
1606		nfsm_wcc_data(fdvp, fwccflag);
1607		nfsm_wcc_data(tdvp, twccflag);
1608	}
1609	m_freem(mrep);
1610nfsmout:
1611	VTONFS(fdvp)->n_flag |= NMODIFIED;
1612	VTONFS(tdvp)->n_flag |= NMODIFIED;
1613	if (!fwccflag)
1614		VTONFS(fdvp)->n_attrstamp = 0;
1615	if (!twccflag)
1616		VTONFS(tdvp)->n_attrstamp = 0;
1617	return (error);
1618}
1619
1620/*
1621 * nfs hard link create call
1622 */
1623static int
1624nfs_link(struct vop_link_args *ap)
1625{
1626	struct vnode *vp = ap->a_vp;
1627	struct vnode *tdvp = ap->a_tdvp;
1628	struct componentname *cnp = ap->a_cnp;
1629	u_int32_t *tl;
1630	caddr_t bpos, dpos;
1631	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1632	struct mbuf *mreq, *mrep, *md, *mb;
1633	int v3;
1634
1635	if (vp->v_mount != tdvp->v_mount) {
1636		return (EXDEV);
1637	}
1638
1639	/*
1640	 * Push all writes to the server, so that the attribute cache
1641	 * doesn't get "out of sync" with the server.
1642	 * XXX There should be a better way!
1643	 */
1644	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_thread);
1645
1646	v3 = NFS_ISV3(vp);
1647	nfsstats.rpccnt[NFSPROC_LINK]++;
1648	mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1649		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1650	mb = mreq;
1651	bpos = mtod(mb, caddr_t);
1652	nfsm_fhtom(vp, v3);
1653	nfsm_fhtom(tdvp, v3);
1654	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1655	nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1656	if (v3) {
1657		nfsm_postop_attr(vp, attrflag);
1658		nfsm_wcc_data(tdvp, wccflag);
1659	}
1660	m_freem(mrep);
1661nfsmout:
1662	VTONFS(tdvp)->n_flag |= NMODIFIED;
1663	if (!attrflag)
1664		VTONFS(vp)->n_attrstamp = 0;
1665	if (!wccflag)
1666		VTONFS(tdvp)->n_attrstamp = 0;
1667	/*
1668	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1669	 */
1670	if (error == EEXIST)
1671		error = 0;
1672	return (error);
1673}
1674
1675/*
1676 * nfs symbolic link create call
1677 */
1678static int
1679nfs_symlink(struct vop_symlink_args *ap)
1680{
1681	struct vnode *dvp = ap->a_dvp;
1682	struct vattr *vap = ap->a_vap;
1683	struct componentname *cnp = ap->a_cnp;
1684	struct nfsv2_sattr *sp;
1685	u_int32_t *tl;
1686	caddr_t bpos, dpos;
1687	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1688	struct mbuf *mreq, *mrep, *md, *mb;
1689	struct vnode *newvp = (struct vnode *)0;
1690	int v3 = NFS_ISV3(dvp);
1691
1692	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1693	slen = strlen(ap->a_target);
1694	mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1695	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1696	mb = mreq;
1697	bpos = mtod(mb, caddr_t);
1698	nfsm_fhtom(dvp, v3);
1699	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1700	if (v3) {
1701		nfsm_v3attrbuild(vap, FALSE);
1702	}
1703	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1704	if (!v3) {
1705		sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1706		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1707		sp->sa_uid = nfs_xdrneg1;
1708		sp->sa_gid = nfs_xdrneg1;
1709		sp->sa_size = nfs_xdrneg1;
1710		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1711		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1712	}
1713
1714	/*
1715	 * Issue the NFS request and get the rpc response.
1716	 *
1717	 * Only NFSv3 responses returning an error of 0 actually return
1718	 * a file handle that can be converted into newvp without having
1719	 * to do an extra lookup rpc.
1720	 */
1721	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
1722	if (v3) {
1723		if (error == 0)
1724			nfsm_mtofh(dvp, newvp, v3, gotvp);
1725		nfsm_wcc_data(dvp, wccflag);
1726	}
1727
1728	/*
1729	 * out code jumps -> here, mrep is also freed.
1730	 */
1731
1732	m_freem(mrep);
1733nfsmout:
1734
1735	/*
1736	 * If we get an EEXIST error, silently convert it to no-error
1737	 * in case of an NFS retry.
1738	 */
1739	if (error == EEXIST)
1740		error = 0;
1741
1742	/*
1743	 * If we do not have (or no longer have) an error, and we could
1744	 * not extract the newvp from the response due to the request being
1745	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
1746	 * to obtain a newvp to return.
1747	 */
1748	if (error == 0 && newvp == NULL) {
1749		struct nfsnode *np = NULL;
1750
1751		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1752		    cnp->cn_cred, cnp->cn_thread, &np);
1753		if (!error)
1754			newvp = NFSTOV(np);
1755	}
1756	if (error) {
1757		if (newvp)
1758			vput(newvp);
1759	} else {
1760		*ap->a_vpp = newvp;
1761	}
1762	VTONFS(dvp)->n_flag |= NMODIFIED;
1763	if (!wccflag)
1764		VTONFS(dvp)->n_attrstamp = 0;
1765	return (error);
1766}
1767
1768/*
1769 * nfs make dir call
1770 */
1771static int
1772nfs_mkdir(struct vop_mkdir_args *ap)
1773{
1774	struct vnode *dvp = ap->a_dvp;
1775	struct vattr *vap = ap->a_vap;
1776	struct componentname *cnp = ap->a_cnp;
1777	struct nfsv2_sattr *sp;
1778	u_int32_t *tl;
1779	int len;
1780	struct nfsnode *np = (struct nfsnode *)0;
1781	struct vnode *newvp = (struct vnode *)0;
1782	caddr_t bpos, dpos;
1783	int error = 0, wccflag = NFSV3_WCCRATTR;
1784	int gotvp = 0;
1785	struct mbuf *mreq, *mrep, *md, *mb;
1786	struct vattr vattr;
1787	int v3 = NFS_ISV3(dvp);
1788
1789	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1790		return (error);
1791	}
1792	len = cnp->cn_namelen;
1793	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1794	mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
1795	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1796	mb = mreq;
1797	bpos = mtod(mb, caddr_t);
1798	nfsm_fhtom(dvp, v3);
1799	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1800	if (v3) {
1801		nfsm_v3attrbuild(vap, FALSE);
1802	} else {
1803		sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1804		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1805		sp->sa_uid = nfs_xdrneg1;
1806		sp->sa_gid = nfs_xdrneg1;
1807		sp->sa_size = nfs_xdrneg1;
1808		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1809		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1810	}
1811	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
1812	if (!error)
1813		nfsm_mtofh(dvp, newvp, v3, gotvp);
1814	if (v3)
1815		nfsm_wcc_data(dvp, wccflag);
1816	m_freem(mrep);
1817nfsmout:
1818	VTONFS(dvp)->n_flag |= NMODIFIED;
1819	if (!wccflag)
1820		VTONFS(dvp)->n_attrstamp = 0;
1821	/*
1822	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1823	 * if we can succeed in looking up the directory.
1824	 */
1825	if (error == EEXIST || (!error && !gotvp)) {
1826		if (newvp) {
1827			vrele(newvp);
1828			newvp = (struct vnode *)0;
1829		}
1830		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1831			cnp->cn_thread, &np);
1832		if (!error) {
1833			newvp = NFSTOV(np);
1834			if (newvp->v_type != VDIR)
1835				error = EEXIST;
1836		}
1837	}
1838	if (error) {
1839		if (newvp)
1840			vrele(newvp);
1841	} else
1842		*ap->a_vpp = newvp;
1843	return (error);
1844}
1845
1846/*
1847 * nfs remove directory call
1848 */
1849static int
1850nfs_rmdir(struct vop_rmdir_args *ap)
1851{
1852	struct vnode *vp = ap->a_vp;
1853	struct vnode *dvp = ap->a_dvp;
1854	struct componentname *cnp = ap->a_cnp;
1855	u_int32_t *tl;
1856	caddr_t bpos, dpos;
1857	int error = 0, wccflag = NFSV3_WCCRATTR;
1858	struct mbuf *mreq, *mrep, *md, *mb;
1859	int v3 = NFS_ISV3(dvp);
1860
1861	if (dvp == vp)
1862		return (EINVAL);
1863	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1864	mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
1865		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1866	mb = mreq;
1867	bpos = mtod(mb, caddr_t);
1868	nfsm_fhtom(dvp, v3);
1869	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1870	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
1871	if (v3)
1872		nfsm_wcc_data(dvp, wccflag);
1873	m_freem(mrep);
1874nfsmout:
1875	VTONFS(dvp)->n_flag |= NMODIFIED;
1876	if (!wccflag)
1877		VTONFS(dvp)->n_attrstamp = 0;
1878	cache_purge(dvp);
1879	cache_purge(vp);
1880	/*
1881	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1882	 */
1883	if (error == ENOENT)
1884		error = 0;
1885	return (error);
1886}
1887
1888/*
1889 * nfs readdir call
1890 */
1891static int
1892nfs_readdir(struct vop_readdir_args *ap)
1893{
1894	struct vnode *vp = ap->a_vp;
1895	struct nfsnode *np = VTONFS(vp);
1896	struct uio *uio = ap->a_uio;
1897	int tresid, error;
1898	struct vattr vattr;
1899
1900	if (vp->v_type != VDIR)
1901		return (EPERM);
1902	/*
1903	 * First, check for hit on the EOF offset cache
1904	 */
1905	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1906	    (np->n_flag & NMODIFIED) == 0) {
1907		if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
1908			np->n_mtime == vattr.va_mtime.tv_sec) {
1909			nfsstats.direofcache_hits++;
1910			return (0);
1911		}
1912	}
1913
1914	/*
1915	 * Call nfs_bioread() to do the real work.
1916	 */
1917	tresid = uio->uio_resid;
1918	error = nfs_bioread(vp, uio, 0, ap->a_cred);
1919
1920	if (!error && uio->uio_resid == tresid)
1921		nfsstats.direofcache_misses++;
1922	return (error);
1923}
1924
1925/*
1926 * Readdir rpc call.
1927 * Called from below the buffer cache by nfs_doio().
1928 */
1929int
1930nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1931{
1932	int len, left;
1933	struct dirent *dp = NULL;
1934	u_int32_t *tl;
1935	caddr_t cp;
1936	nfsuint64 *cookiep;
1937	caddr_t bpos, dpos;
1938	struct mbuf *mreq, *mrep, *md, *mb;
1939	nfsuint64 cookie;
1940	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1941	struct nfsnode *dnp = VTONFS(vp);
1942	u_quad_t fileno;
1943	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
1944	int attrflag;
1945	int v3 = NFS_ISV3(vp);
1946
1947#ifndef DIAGNOSTIC
1948	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
1949		(uiop->uio_resid & (DIRBLKSIZ - 1)))
1950		panic("nfs readdirrpc bad uio");
1951#endif
1952
1953	/*
1954	 * If there is no cookie, assume directory was stale.
1955	 */
1956	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
1957	if (cookiep)
1958		cookie = *cookiep;
1959	else
1960		return (NFSERR_BAD_COOKIE);
1961	/*
1962	 * Loop around doing readdir rpc's of size nm_readdirsize
1963	 * truncated to a multiple of DIRBLKSIZ.
1964	 * The stopping criteria is EOF or buffer full.
1965	 */
1966	while (more_dirs && bigenough) {
1967		nfsstats.rpccnt[NFSPROC_READDIR]++;
1968		mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
1969			NFSX_READDIR(v3));
1970		mb = mreq;
1971		bpos = mtod(mb, caddr_t);
1972		nfsm_fhtom(vp, v3);
1973		if (v3) {
1974			tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1975			*tl++ = cookie.nfsuquad[0];
1976			*tl++ = cookie.nfsuquad[1];
1977			*tl++ = dnp->n_cookieverf.nfsuquad[0];
1978			*tl++ = dnp->n_cookieverf.nfsuquad[1];
1979		} else {
1980			tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1981			*tl++ = cookie.nfsuquad[0];
1982		}
1983		*tl = txdr_unsigned(nmp->nm_readdirsize);
1984		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
1985		if (v3) {
1986			nfsm_postop_attr(vp, attrflag);
1987			if (!error) {
1988				tl = nfsm_dissect(u_int32_t *,
1989				    2 * NFSX_UNSIGNED);
1990				dnp->n_cookieverf.nfsuquad[0] = *tl++;
1991				dnp->n_cookieverf.nfsuquad[1] = *tl;
1992			} else {
1993				m_freem(mrep);
1994				goto nfsmout;
1995			}
1996		}
1997		tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1998		more_dirs = fxdr_unsigned(int, *tl);
1999
2000		/* loop thru the dir entries, doctoring them to 4bsd form */
2001		while (more_dirs && bigenough) {
2002			if (v3) {
2003				tl = nfsm_dissect(u_int32_t *,
2004				    3 * NFSX_UNSIGNED);
2005				fileno = fxdr_hyper(tl);
2006				len = fxdr_unsigned(int, *(tl + 2));
2007			} else {
2008				tl = nfsm_dissect(u_int32_t *,
2009				    2 * NFSX_UNSIGNED);
2010				fileno = fxdr_unsigned(u_quad_t, *tl++);
2011				len = fxdr_unsigned(int, *tl);
2012			}
2013			if (len <= 0 || len > NFS_MAXNAMLEN) {
2014				error = EBADRPC;
2015				m_freem(mrep);
2016				goto nfsmout;
2017			}
2018			tlen = nfsm_rndup(len);
2019			if (tlen == len)
2020				tlen += 4;	/* To ensure null termination */
2021			left = DIRBLKSIZ - blksiz;
2022			if ((tlen + DIRHDSIZ) > left) {
2023				dp->d_reclen += left;
2024				uiop->uio_iov->iov_base += left;
2025				uiop->uio_iov->iov_len -= left;
2026				uiop->uio_offset += left;
2027				uiop->uio_resid -= left;
2028				blksiz = 0;
2029			}
2030			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2031				bigenough = 0;
2032			if (bigenough) {
2033				dp = (struct dirent *)uiop->uio_iov->iov_base;
2034				dp->d_fileno = (int)fileno;
2035				dp->d_namlen = len;
2036				dp->d_reclen = tlen + DIRHDSIZ;
2037				dp->d_type = DT_UNKNOWN;
2038				blksiz += dp->d_reclen;
2039				if (blksiz == DIRBLKSIZ)
2040					blksiz = 0;
2041				uiop->uio_offset += DIRHDSIZ;
2042				uiop->uio_resid -= DIRHDSIZ;
2043				uiop->uio_iov->iov_base += DIRHDSIZ;
2044				uiop->uio_iov->iov_len -= DIRHDSIZ;
2045				nfsm_mtouio(uiop, len);
2046				cp = uiop->uio_iov->iov_base;
2047				tlen -= len;
2048				*cp = '\0';	/* null terminate */
2049				uiop->uio_iov->iov_base += tlen;
2050				uiop->uio_iov->iov_len -= tlen;
2051				uiop->uio_offset += tlen;
2052				uiop->uio_resid -= tlen;
2053			} else
2054				nfsm_adv(nfsm_rndup(len));
2055			if (v3) {
2056				tl = nfsm_dissect(u_int32_t *,
2057				    3 * NFSX_UNSIGNED);
2058			} else {
2059				tl = nfsm_dissect(u_int32_t *,
2060				    2 * NFSX_UNSIGNED);
2061			}
2062			if (bigenough) {
2063				cookie.nfsuquad[0] = *tl++;
2064				if (v3)
2065					cookie.nfsuquad[1] = *tl++;
2066			} else if (v3)
2067				tl += 2;
2068			else
2069				tl++;
2070			more_dirs = fxdr_unsigned(int, *tl);
2071		}
2072		/*
2073		 * If at end of rpc data, get the eof boolean
2074		 */
2075		if (!more_dirs) {
2076			tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2077			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2078		}
2079		m_freem(mrep);
2080	}
2081	/*
2082	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2083	 * by increasing d_reclen for the last record.
2084	 */
2085	if (blksiz > 0) {
2086		left = DIRBLKSIZ - blksiz;
2087		dp->d_reclen += left;
2088		uiop->uio_iov->iov_base += left;
2089		uiop->uio_iov->iov_len -= left;
2090		uiop->uio_offset += left;
2091		uiop->uio_resid -= left;
2092	}
2093
2094	/*
2095	 * We are now either at the end of the directory or have filled the
2096	 * block.
2097	 */
2098	if (bigenough)
2099		dnp->n_direofoffset = uiop->uio_offset;
2100	else {
2101		if (uiop->uio_resid > 0)
2102			printf("EEK! readdirrpc resid > 0\n");
2103		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2104		*cookiep = cookie;
2105	}
2106nfsmout:
2107	return (error);
2108}
2109
2110/*
2111 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2112 */
2113int
2114nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2115{
2116	int len, left;
2117	struct dirent *dp;
2118	u_int32_t *tl;
2119	caddr_t cp;
2120	struct vnode *newvp;
2121	nfsuint64 *cookiep;
2122	caddr_t bpos, dpos, dpossav1, dpossav2;
2123	struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2124	struct nameidata nami, *ndp = &nami;
2125	struct componentname *cnp = &ndp->ni_cnd;
2126	nfsuint64 cookie;
2127	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2128	struct nfsnode *dnp = VTONFS(vp), *np;
2129	nfsfh_t *fhp;
2130	u_quad_t fileno;
2131	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2132	int attrflag, fhsize;
2133
2134#ifndef nolint
2135	dp = (struct dirent *)0;
2136#endif
2137#ifndef DIAGNOSTIC
2138	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2139		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2140		panic("nfs readdirplusrpc bad uio");
2141#endif
2142	ndp->ni_dvp = vp;
2143	newvp = NULLVP;
2144
2145	/*
2146	 * If there is no cookie, assume directory was stale.
2147	 */
2148	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2149	if (cookiep)
2150		cookie = *cookiep;
2151	else
2152		return (NFSERR_BAD_COOKIE);
2153	/*
2154	 * Loop around doing readdir rpc's of size nm_readdirsize
2155	 * truncated to a multiple of DIRBLKSIZ.
2156	 * The stopping criteria is EOF or buffer full.
2157	 */
2158	while (more_dirs && bigenough) {
2159		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2160		mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2161			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2162		mb = mreq;
2163		bpos = mtod(mb, caddr_t);
2164		nfsm_fhtom(vp, 1);
2165 		tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2166		*tl++ = cookie.nfsuquad[0];
2167		*tl++ = cookie.nfsuquad[1];
2168		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2169		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2170		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2171		*tl = txdr_unsigned(nmp->nm_rsize);
2172		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2173		nfsm_postop_attr(vp, attrflag);
2174		if (error) {
2175			m_freem(mrep);
2176			goto nfsmout;
2177		}
2178		tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2179		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2180		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2181		more_dirs = fxdr_unsigned(int, *tl);
2182
2183		/* loop thru the dir entries, doctoring them to 4bsd form */
2184		while (more_dirs && bigenough) {
2185			tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2186			fileno = fxdr_hyper(tl);
2187			len = fxdr_unsigned(int, *(tl + 2));
2188			if (len <= 0 || len > NFS_MAXNAMLEN) {
2189				error = EBADRPC;
2190				m_freem(mrep);
2191				goto nfsmout;
2192			}
2193			tlen = nfsm_rndup(len);
2194			if (tlen == len)
2195				tlen += 4;	/* To ensure null termination*/
2196			left = DIRBLKSIZ - blksiz;
2197			if ((tlen + DIRHDSIZ) > left) {
2198				dp->d_reclen += left;
2199				uiop->uio_iov->iov_base += left;
2200				uiop->uio_iov->iov_len -= left;
2201				uiop->uio_offset += left;
2202				uiop->uio_resid -= left;
2203				blksiz = 0;
2204			}
2205			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2206				bigenough = 0;
2207			if (bigenough) {
2208				dp = (struct dirent *)uiop->uio_iov->iov_base;
2209				dp->d_fileno = (int)fileno;
2210				dp->d_namlen = len;
2211				dp->d_reclen = tlen + DIRHDSIZ;
2212				dp->d_type = DT_UNKNOWN;
2213				blksiz += dp->d_reclen;
2214				if (blksiz == DIRBLKSIZ)
2215					blksiz = 0;
2216				uiop->uio_offset += DIRHDSIZ;
2217				uiop->uio_resid -= DIRHDSIZ;
2218				uiop->uio_iov->iov_base += DIRHDSIZ;
2219				uiop->uio_iov->iov_len -= DIRHDSIZ;
2220				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2221				cnp->cn_namelen = len;
2222				nfsm_mtouio(uiop, len);
2223				cp = uiop->uio_iov->iov_base;
2224				tlen -= len;
2225				*cp = '\0';
2226				uiop->uio_iov->iov_base += tlen;
2227				uiop->uio_iov->iov_len -= tlen;
2228				uiop->uio_offset += tlen;
2229				uiop->uio_resid -= tlen;
2230			} else
2231				nfsm_adv(nfsm_rndup(len));
2232			tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2233			if (bigenough) {
2234				cookie.nfsuquad[0] = *tl++;
2235				cookie.nfsuquad[1] = *tl++;
2236			} else
2237				tl += 2;
2238
2239			/*
2240			 * Since the attributes are before the file handle
2241			 * (sigh), we must skip over the attributes and then
2242			 * come back and get them.
2243			 */
2244			attrflag = fxdr_unsigned(int, *tl);
2245			if (attrflag) {
2246			    dpossav1 = dpos;
2247			    mdsav1 = md;
2248			    nfsm_adv(NFSX_V3FATTR);
2249			    tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2250			    doit = fxdr_unsigned(int, *tl);
2251			    if (doit) {
2252				nfsm_getfh(fhp, fhsize, 1);
2253				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2254				    VREF(vp);
2255				    newvp = vp;
2256				    np = dnp;
2257				} else {
2258				    error = nfs_nget(vp->v_mount, fhp,
2259					fhsize, &np);
2260				    if (error)
2261					doit = 0;
2262				    else
2263					newvp = NFSTOV(np);
2264				}
2265			    }
2266			    if (doit && bigenough) {
2267				dpossav2 = dpos;
2268				dpos = dpossav1;
2269				mdsav2 = md;
2270				md = mdsav1;
2271				nfsm_loadattr(newvp, (struct vattr *)0);
2272				dpos = dpossav2;
2273				md = mdsav2;
2274				dp->d_type =
2275				    IFTODT(VTTOIF(np->n_vattr.va_type));
2276				ndp->ni_vp = newvp;
2277			        cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2278			    }
2279			} else {
2280			    /* Just skip over the file handle */
2281			    tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2282			    i = fxdr_unsigned(int, *tl);
2283			    nfsm_adv(nfsm_rndup(i));
2284			}
2285			if (newvp != NULLVP) {
2286			    if (newvp == vp)
2287				vrele(newvp);
2288			    else
2289				vput(newvp);
2290			    newvp = NULLVP;
2291			}
2292			tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2293			more_dirs = fxdr_unsigned(int, *tl);
2294		}
2295		/*
2296		 * If at end of rpc data, get the eof boolean
2297		 */
2298		if (!more_dirs) {
2299			tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2300			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2301		}
2302		m_freem(mrep);
2303	}
2304	/*
2305	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2306	 * by increasing d_reclen for the last record.
2307	 */
2308	if (blksiz > 0) {
2309		left = DIRBLKSIZ - blksiz;
2310		dp->d_reclen += left;
2311		uiop->uio_iov->iov_base += left;
2312		uiop->uio_iov->iov_len -= left;
2313		uiop->uio_offset += left;
2314		uiop->uio_resid -= left;
2315	}
2316
2317	/*
2318	 * We are now either at the end of the directory or have filled the
2319	 * block.
2320	 */
2321	if (bigenough)
2322		dnp->n_direofoffset = uiop->uio_offset;
2323	else {
2324		if (uiop->uio_resid > 0)
2325			printf("EEK! readdirplusrpc resid > 0\n");
2326		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2327		*cookiep = cookie;
2328	}
2329nfsmout:
2330	if (newvp != NULLVP) {
2331	        if (newvp == vp)
2332			vrele(newvp);
2333		else
2334			vput(newvp);
2335		newvp = NULLVP;
2336	}
2337	return (error);
2338}
2339
2340/*
2341 * Silly rename. To make the NFS filesystem that is stateless look a little
2342 * more like the "ufs" a remove of an active vnode is translated to a rename
2343 * to a funny looking filename that is removed by nfs_inactive on the
2344 * nfsnode. There is the potential for another process on a different client
2345 * to create the same funny name between the nfs_lookitup() fails and the
2346 * nfs_rename() completes, but...
2347 */
2348static int
2349nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2350{
2351	struct sillyrename *sp;
2352	struct nfsnode *np;
2353	int error;
2354	short pid;
2355
2356	cache_purge(dvp);
2357	np = VTONFS(vp);
2358#ifndef DIAGNOSTIC
2359	if (vp->v_type == VDIR)
2360		panic("nfs: sillyrename dir");
2361#endif
2362	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2363		M_NFSREQ, M_WAITOK);
2364	sp->s_cred = crhold(cnp->cn_cred);
2365	sp->s_dvp = dvp;
2366	VREF(dvp);
2367
2368	/* Fudge together a funny name */
2369	pid = cnp->cn_thread->td_proc->p_pid;
2370	sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
2371
2372	/* Try lookitups until we get one that isn't there */
2373	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2374		cnp->cn_thread, (struct nfsnode **)0) == 0) {
2375		sp->s_name[4]++;
2376		if (sp->s_name[4] > 'z') {
2377			error = EINVAL;
2378			goto bad;
2379		}
2380	}
2381	error = nfs_renameit(dvp, cnp, sp);
2382	if (error)
2383		goto bad;
2384	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2385		cnp->cn_thread, &np);
2386	np->n_sillyrename = sp;
2387	return (0);
2388bad:
2389	vrele(sp->s_dvp);
2390	crfree(sp->s_cred);
2391	free((caddr_t)sp, M_NFSREQ);
2392	return (error);
2393}
2394
2395/*
2396 * Look up a file name and optionally either update the file handle or
2397 * allocate an nfsnode, depending on the value of npp.
2398 * npp == NULL	--> just do the lookup
2399 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2400 *			handled too
2401 * *npp != NULL --> update the file handle in the vnode
2402 */
2403static int
2404nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2405    struct thread *td, struct nfsnode **npp)
2406{
2407	u_int32_t *tl;
2408	struct vnode *newvp = (struct vnode *)0;
2409	struct nfsnode *np, *dnp = VTONFS(dvp);
2410	caddr_t bpos, dpos;
2411	int error = 0, fhlen, attrflag;
2412	struct mbuf *mreq, *mrep, *md, *mb;
2413	nfsfh_t *nfhp;
2414	int v3 = NFS_ISV3(dvp);
2415
2416	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2417	mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2418		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2419	mb = mreq;
2420	bpos = mtod(mb, caddr_t);
2421	nfsm_fhtom(dvp, v3);
2422	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2423	nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2424	if (npp && !error) {
2425		nfsm_getfh(nfhp, fhlen, v3);
2426		if (*npp) {
2427		    np = *npp;
2428		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2429			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2430			np->n_fhp = &np->n_fh;
2431		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2432			np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2433		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2434		    np->n_fhsize = fhlen;
2435		    newvp = NFSTOV(np);
2436		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2437		    VREF(dvp);
2438		    newvp = dvp;
2439		} else {
2440		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2441		    if (error) {
2442			m_freem(mrep);
2443			return (error);
2444		    }
2445		    newvp = NFSTOV(np);
2446		}
2447		if (v3) {
2448			nfsm_postop_attr(newvp, attrflag);
2449			if (!attrflag && *npp == NULL) {
2450				m_freem(mrep);
2451				if (newvp == dvp)
2452					vrele(newvp);
2453				else
2454					vput(newvp);
2455				return (ENOENT);
2456			}
2457		} else
2458			nfsm_loadattr(newvp, (struct vattr *)0);
2459	}
2460	m_freem(mrep);
2461nfsmout:
2462	if (npp && *npp == NULL) {
2463		if (error) {
2464			if (newvp) {
2465				if (newvp == dvp)
2466					vrele(newvp);
2467				else
2468					vput(newvp);
2469			}
2470		} else
2471			*npp = np;
2472	}
2473	return (error);
2474}
2475
2476/*
2477 * Nfs Version 3 commit rpc
2478 */
2479int
2480nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2481    struct thread *td)
2482{
2483	u_int32_t *tl;
2484	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2485	caddr_t bpos, dpos;
2486	int error = 0, wccflag = NFSV3_WCCRATTR;
2487	struct mbuf *mreq, *mrep, *md, *mb;
2488
2489	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2490		return (0);
2491	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2492	mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2493	mb = mreq;
2494	bpos = mtod(mb, caddr_t);
2495	nfsm_fhtom(vp, 1);
2496	tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2497	txdr_hyper(offset, tl);
2498	tl += 2;
2499	*tl = txdr_unsigned(cnt);
2500	nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2501	nfsm_wcc_data(vp, wccflag);
2502	if (!error) {
2503		tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2504		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2505			NFSX_V3WRITEVERF)) {
2506			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2507				NFSX_V3WRITEVERF);
2508			error = NFSERR_STALEWRITEVERF;
2509		}
2510	}
2511	m_freem(mrep);
2512nfsmout:
2513	return (error);
2514}
2515
2516/*
2517 * Strategy routine.
2518 * For async requests when nfsiod(s) are running, queue the request by
2519 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2520 * request.
2521 */
2522static int
2523nfs_strategy(struct vop_strategy_args *ap)
2524{
2525	struct buf *bp = ap->a_bp;
2526	struct ucred *cr;
2527	struct thread *td;
2528	int error = 0;
2529
2530	KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2531	KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2532
2533	if (bp->b_flags & B_PHYS)
2534		panic("nfs physio");
2535
2536	if (bp->b_flags & B_ASYNC)
2537		td = (struct thread *)0;
2538	else
2539		td = curthread;	/* XXX */
2540
2541	if (bp->b_iocmd == BIO_READ)
2542		cr = bp->b_rcred;
2543	else
2544		cr = bp->b_wcred;
2545
2546	/*
2547	 * If the op is asynchronous and an i/o daemon is waiting
2548	 * queue the request, wake it up and wait for completion
2549	 * otherwise just do it ourselves.
2550	 */
2551	if ((bp->b_flags & B_ASYNC) == 0 ||
2552		nfs_asyncio(bp, NOCRED, td))
2553		error = nfs_doio(bp, cr, td);
2554	return (error);
2555}
2556
2557/*
2558 * fsync vnode op. Just call nfs_flush() with commit == 1.
2559 */
2560/* ARGSUSED */
2561static int
2562nfs_fsync(struct vop_fsync_args *ap)
2563{
2564
2565	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1));
2566}
2567
2568/*
2569 * Flush all the blocks associated with a vnode.
2570 * 	Walk through the buffer pool and push any dirty pages
2571 *	associated with the vnode.
2572 */
2573static int
2574nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct thread *td,
2575    int commit)
2576{
2577	struct nfsnode *np = VTONFS(vp);
2578	struct buf *bp;
2579	int i;
2580	struct buf *nbp;
2581	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2582	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2583	int passone = 1;
2584	u_quad_t off, endoff, toff;
2585	struct ucred* wcred = NULL;
2586	struct buf **bvec = NULL;
2587#ifndef NFS_COMMITBVECSIZ
2588#define NFS_COMMITBVECSIZ	20
2589#endif
2590	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2591	int bvecsize = 0, bveccount;
2592
2593	if (nmp->nm_flag & NFSMNT_INT)
2594		slpflag = PCATCH;
2595	if (!commit)
2596		passone = 0;
2597	/*
2598	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2599	 * server, but nas not been committed to stable storage on the server
2600	 * yet. On the first pass, the byte range is worked out and the commit
2601	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2602	 * job.
2603	 */
2604again:
2605	off = (u_quad_t)-1;
2606	endoff = 0;
2607	bvecpos = 0;
2608	if (NFS_ISV3(vp) && commit) {
2609		s = splbio();
2610		/*
2611		 * Count up how many buffers waiting for a commit.
2612		 */
2613		bveccount = 0;
2614		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2615			nbp = TAILQ_NEXT(bp, b_vnbufs);
2616			if (BUF_REFCNT(bp) == 0 &&
2617			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2618				== (B_DELWRI | B_NEEDCOMMIT))
2619				bveccount++;
2620		}
2621		/*
2622		 * Allocate space to remember the list of bufs to commit.  It is
2623		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2624		 * If we can't get memory (for whatever reason), we will end up
2625		 * committing the buffers one-by-one in the loop below.
2626		 */
2627		if (bvec != NULL && bvec != bvec_on_stack)
2628			free(bvec, M_TEMP);
2629		if (bveccount > NFS_COMMITBVECSIZ) {
2630			bvec = (struct buf **)
2631				malloc(bveccount * sizeof(struct buf *),
2632				       M_TEMP, M_NOWAIT);
2633			if (bvec == NULL) {
2634				bvec = bvec_on_stack;
2635				bvecsize = NFS_COMMITBVECSIZ;
2636			} else
2637				bvecsize = bveccount;
2638		} else {
2639			bvec = bvec_on_stack;
2640			bvecsize = NFS_COMMITBVECSIZ;
2641		}
2642		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2643			nbp = TAILQ_NEXT(bp, b_vnbufs);
2644			if (bvecpos >= bvecsize)
2645				break;
2646			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2647			    (B_DELWRI | B_NEEDCOMMIT) ||
2648			    BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
2649				continue;
2650			bremfree(bp);
2651			/*
2652			 * Work out if all buffers are using the same cred
2653			 * so we can deal with them all with one commit.
2654			 *
2655			 * NOTE: we are not clearing B_DONE here, so we have
2656			 * to do it later on in this routine if we intend to
2657			 * initiate I/O on the bp.
2658			 *
2659			 * Note: to avoid loopback deadlocks, we do not
2660			 * assign b_runningbufspace.
2661			 */
2662			if (wcred == NULL)
2663				wcred = bp->b_wcred;
2664			else if (wcred != bp->b_wcred)
2665				wcred = NOCRED;
2666			bp->b_flags |= B_WRITEINPROG;
2667			vfs_busy_pages(bp, 1);
2668
2669			/*
2670			 * bp is protected by being locked, but nbp is not
2671			 * and vfs_busy_pages() may sleep.  We have to
2672			 * recalculate nbp.
2673			 */
2674			nbp = TAILQ_NEXT(bp, b_vnbufs);
2675
2676			/*
2677			 * A list of these buffers is kept so that the
2678			 * second loop knows which buffers have actually
2679			 * been committed. This is necessary, since there
2680			 * may be a race between the commit rpc and new
2681			 * uncommitted writes on the file.
2682			 */
2683			bvec[bvecpos++] = bp;
2684			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2685				bp->b_dirtyoff;
2686			if (toff < off)
2687				off = toff;
2688			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2689			if (toff > endoff)
2690				endoff = toff;
2691		}
2692		splx(s);
2693	}
2694	if (bvecpos > 0) {
2695		/*
2696		 * Commit data on the server, as required.
2697		 * If all bufs are using the same wcred, then use that with
2698		 * one call for all of them, otherwise commit each one
2699		 * separately.
2700		 */
2701		if (wcred != NOCRED)
2702			retv = nfs_commit(vp, off, (int)(endoff - off),
2703					  wcred, td);
2704		else {
2705			retv = 0;
2706			for (i = 0; i < bvecpos; i++) {
2707				off_t off, size;
2708				bp = bvec[i];
2709				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2710					bp->b_dirtyoff;
2711				size = (u_quad_t)(bp->b_dirtyend
2712						  - bp->b_dirtyoff);
2713				retv = nfs_commit(vp, off, (int)size,
2714						  bp->b_wcred, td);
2715				if (retv) break;
2716			}
2717		}
2718
2719		if (retv == NFSERR_STALEWRITEVERF)
2720			nfs_clearcommit(vp->v_mount);
2721
2722		/*
2723		 * Now, either mark the blocks I/O done or mark the
2724		 * blocks dirty, depending on whether the commit
2725		 * succeeded.
2726		 */
2727		for (i = 0; i < bvecpos; i++) {
2728			bp = bvec[i];
2729			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
2730			if (retv) {
2731				/*
2732				 * Error, leave B_DELWRI intact
2733				 */
2734				vfs_unbusy_pages(bp);
2735				brelse(bp);
2736			} else {
2737				/*
2738				 * Success, remove B_DELWRI ( bundirty() ).
2739				 *
2740				 * b_dirtyoff/b_dirtyend seem to be NFS
2741				 * specific.  We should probably move that
2742				 * into bundirty(). XXX
2743				 */
2744				s = splbio();
2745				vp->v_numoutput++;
2746				bp->b_flags |= B_ASYNC;
2747				bundirty(bp);
2748				bp->b_flags &= ~B_DONE;
2749				bp->b_ioflags &= ~BIO_ERROR;
2750				bp->b_dirtyoff = bp->b_dirtyend = 0;
2751				splx(s);
2752				bufdone(bp);
2753			}
2754		}
2755	}
2756
2757	/*
2758	 * Start/do any write(s) that are required.
2759	 */
2760loop:
2761	s = splbio();
2762	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2763		nbp = TAILQ_NEXT(bp, b_vnbufs);
2764		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
2765			if (waitfor != MNT_WAIT || passone)
2766				continue;
2767			error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
2768			    "nfsfsync", slpflag, slptimeo);
2769			splx(s);
2770			if (error == 0)
2771				panic("nfs_fsync: inconsistent lock");
2772			if (error == ENOLCK)
2773				goto loop;
2774			if (nfs_sigintr(nmp, (struct nfsreq *)0, td->td_proc)) {
2775				error = EINTR;
2776				goto done;
2777			}
2778			if (slpflag == PCATCH) {
2779				slpflag = 0;
2780				slptimeo = 2 * hz;
2781			}
2782			goto loop;
2783		}
2784		if ((bp->b_flags & B_DELWRI) == 0)
2785			panic("nfs_fsync: not dirty");
2786		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2787			BUF_UNLOCK(bp);
2788			continue;
2789		}
2790		bremfree(bp);
2791		if (passone || !commit)
2792		    bp->b_flags |= B_ASYNC;
2793		else
2794		    bp->b_flags |= B_ASYNC | B_WRITEINPROG;
2795		splx(s);
2796		BUF_WRITE(bp);
2797		goto loop;
2798	}
2799	splx(s);
2800	if (passone) {
2801		passone = 0;
2802		goto again;
2803	}
2804	if (waitfor == MNT_WAIT) {
2805		while (vp->v_numoutput) {
2806			vp->v_flag |= VBWAIT;
2807			error = tsleep((caddr_t)&vp->v_numoutput,
2808				slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
2809			if (error) {
2810			    if (nfs_sigintr(nmp, (struct nfsreq *)0, td->td_proc)) {
2811				error = EINTR;
2812				goto done;
2813			    }
2814			    if (slpflag == PCATCH) {
2815				slpflag = 0;
2816				slptimeo = 2 * hz;
2817			    }
2818			}
2819		}
2820		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
2821			goto loop;
2822		}
2823	}
2824	if (np->n_flag & NWRITEERR) {
2825		error = np->n_error;
2826		np->n_flag &= ~NWRITEERR;
2827	}
2828done:
2829	if (bvec != NULL && bvec != bvec_on_stack)
2830		free(bvec, M_TEMP);
2831	return (error);
2832}
2833
2834/*
2835 * NFS advisory byte-level locks.
2836 */
2837static int
2838nfs_advlock(struct vop_advlock_args *ap)
2839{
2840
2841	return (nfs_dolock(ap));
2842}
2843
2844/*
2845 * Print out the contents of an nfsnode.
2846 */
2847static int
2848nfs_print(struct vop_print_args *ap)
2849{
2850	struct vnode *vp = ap->a_vp;
2851	struct nfsnode *np = VTONFS(vp);
2852
2853	printf("tag VT_NFS, fileid %ld fsid 0x%x",
2854		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2855	if (vp->v_type == VFIFO)
2856		fifo_printinfo(vp);
2857	printf("\n");
2858	return (0);
2859}
2860
2861/*
2862 * This is the "real" nfs::bwrite(struct buf*).
2863 * B_WRITEINPROG isn't set unless the force flag is one and it
2864 * handles the B_NEEDCOMMIT flag.
2865 * We set B_CACHE if this is a VMIO buffer.
2866 */
2867int
2868nfs_writebp(struct buf *bp, int force, struct thread *td)
2869{
2870	int s;
2871	int oldflags = bp->b_flags;
2872#if 0
2873	int retv = 1;
2874	off_t off;
2875#endif
2876
2877	if (BUF_REFCNT(bp) == 0)
2878		panic("bwrite: buffer is not locked???");
2879
2880	if (bp->b_flags & B_INVAL) {
2881		brelse(bp);
2882		return(0);
2883	}
2884
2885	bp->b_flags |= B_CACHE;
2886
2887	/*
2888	 * Undirty the bp.  We will redirty it later if the I/O fails.
2889	 */
2890
2891	s = splbio();
2892	bundirty(bp);
2893	bp->b_flags &= ~B_DONE;
2894	bp->b_ioflags &= ~BIO_ERROR;
2895	bp->b_iocmd = BIO_WRITE;
2896
2897	bp->b_vp->v_numoutput++;
2898	curthread->td_proc->p_stats->p_ru.ru_oublock++;
2899	splx(s);
2900
2901	/*
2902	 * Note: to avoid loopback deadlocks, we do not
2903	 * assign b_runningbufspace.
2904	 */
2905	vfs_busy_pages(bp, 1);
2906
2907	if (force)
2908		bp->b_flags |= B_WRITEINPROG;
2909	BUF_KERNPROC(bp);
2910	BUF_STRATEGY(bp);
2911
2912	if( (oldflags & B_ASYNC) == 0) {
2913		int rtval = bufwait(bp);
2914
2915		if (oldflags & B_DELWRI) {
2916			s = splbio();
2917			reassignbuf(bp, bp->b_vp);
2918			splx(s);
2919		}
2920
2921		brelse(bp);
2922		return (rtval);
2923	}
2924
2925	return (0);
2926}
2927
2928/*
2929 * nfs special file access vnode op.
2930 * Essentially just get vattr and then imitate iaccess() since the device is
2931 * local to the client.
2932 */
2933static int
2934nfsspec_access(struct vop_access_args *ap)
2935{
2936	struct vattr *vap;
2937	gid_t *gp;
2938	struct ucred *cred = ap->a_cred;
2939	struct vnode *vp = ap->a_vp;
2940	mode_t mode = ap->a_mode;
2941	struct vattr vattr;
2942	int i;
2943	int error;
2944
2945	/*
2946	 * Disallow write attempts on filesystems mounted read-only;
2947	 * unless the file is a socket, fifo, or a block or character
2948	 * device resident on the filesystem.
2949	 */
2950	if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
2951		switch (vp->v_type) {
2952		case VREG:
2953		case VDIR:
2954		case VLNK:
2955			return (EROFS);
2956		default:
2957			break;
2958		}
2959	}
2960	/*
2961	 * If you're the super-user,
2962	 * you always get access.
2963	 */
2964	if (cred->cr_uid == 0)
2965		return (0);
2966	vap = &vattr;
2967	error = VOP_GETATTR(vp, vap, cred, ap->a_td);
2968	if (error)
2969		return (error);
2970	/*
2971	 * Access check is based on only one of owner, group, public.
2972	 * If not owner, then check group. If not a member of the
2973	 * group, then check public access.
2974	 */
2975	if (cred->cr_uid != vap->va_uid) {
2976		mode >>= 3;
2977		gp = cred->cr_groups;
2978		for (i = 0; i < cred->cr_ngroups; i++, gp++)
2979			if (vap->va_gid == *gp)
2980				goto found;
2981		mode >>= 3;
2982found:
2983		;
2984	}
2985	error = (vap->va_mode & mode) == mode ? 0 : EACCES;
2986	return (error);
2987}
2988
2989/*
2990 * Read wrapper for special devices.
2991 */
2992static int
2993nfsspec_read(struct vop_read_args *ap)
2994{
2995	struct nfsnode *np = VTONFS(ap->a_vp);
2996
2997	/*
2998	 * Set access flag.
2999	 */
3000	np->n_flag |= NACC;
3001	getnanotime(&np->n_atim);
3002	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3003}
3004
3005/*
3006 * Write wrapper for special devices.
3007 */
3008static int
3009nfsspec_write(struct vop_write_args *ap)
3010{
3011	struct nfsnode *np = VTONFS(ap->a_vp);
3012
3013	/*
3014	 * Set update flag.
3015	 */
3016	np->n_flag |= NUPD;
3017	getnanotime(&np->n_mtim);
3018	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3019}
3020
3021/*
3022 * Close wrapper for special devices.
3023 *
3024 * Update the times on the nfsnode then do device close.
3025 */
3026static int
3027nfsspec_close(struct vop_close_args *ap)
3028{
3029	struct vnode *vp = ap->a_vp;
3030	struct nfsnode *np = VTONFS(vp);
3031	struct vattr vattr;
3032
3033	if (np->n_flag & (NACC | NUPD)) {
3034		np->n_flag |= NCHG;
3035		if (vp->v_usecount == 1 &&
3036		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3037			VATTR_NULL(&vattr);
3038			if (np->n_flag & NACC)
3039				vattr.va_atime = np->n_atim;
3040			if (np->n_flag & NUPD)
3041				vattr.va_mtime = np->n_mtim;
3042			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3043		}
3044	}
3045	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3046}
3047
3048/*
3049 * Read wrapper for fifos.
3050 */
3051static int
3052nfsfifo_read(struct vop_read_args *ap)
3053{
3054	struct nfsnode *np = VTONFS(ap->a_vp);
3055
3056	/*
3057	 * Set access flag.
3058	 */
3059	np->n_flag |= NACC;
3060	getnanotime(&np->n_atim);
3061	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3062}
3063
3064/*
3065 * Write wrapper for fifos.
3066 */
3067static int
3068nfsfifo_write(struct vop_write_args *ap)
3069{
3070	struct nfsnode *np = VTONFS(ap->a_vp);
3071
3072	/*
3073	 * Set update flag.
3074	 */
3075	np->n_flag |= NUPD;
3076	getnanotime(&np->n_mtim);
3077	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3078}
3079
3080/*
3081 * Close wrapper for fifos.
3082 *
3083 * Update the times on the nfsnode then do fifo close.
3084 */
3085static int
3086nfsfifo_close(struct vop_close_args *ap)
3087{
3088	struct vnode *vp = ap->a_vp;
3089	struct nfsnode *np = VTONFS(vp);
3090	struct vattr vattr;
3091	struct timespec ts;
3092
3093	if (np->n_flag & (NACC | NUPD)) {
3094		getnanotime(&ts);
3095		if (np->n_flag & NACC)
3096			np->n_atim = ts;
3097		if (np->n_flag & NUPD)
3098			np->n_mtim = ts;
3099		np->n_flag |= NCHG;
3100		if (vp->v_usecount == 1 &&
3101		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3102			VATTR_NULL(&vattr);
3103			if (np->n_flag & NACC)
3104				vattr.va_atime = np->n_atim;
3105			if (np->n_flag & NUPD)
3106				vattr.va_mtime = np->n_mtim;
3107			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3108		}
3109	}
3110	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3111}
3112