nfs_vnops.c revision 41186
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 * $Id: nfs_vnops.c,v 1.113 1998/11/13 22:58:48 msmith Exp $
38 */
39
40
41/*
42 * vnode op calls for Sun NFS version 2 and 3
43 */
44
45#include "opt_inet.h"
46
47#include <sys/param.h>
48#include <sys/kernel.h>
49#include <sys/systm.h>
50#include <sys/resourcevar.h>
51#include <sys/proc.h>
52#include <sys/mount.h>
53#include <sys/buf.h>
54#include <sys/malloc.h>
55#include <sys/mbuf.h>
56#include <sys/namei.h>
57#include <sys/socket.h>
58#include <sys/vnode.h>
59#include <sys/dirent.h>
60#include <sys/fcntl.h>
61#include <sys/lockf.h>
62#include <sys/stat.h>
63#include <sys/sysctl.h>
64
65#include <vm/vm.h>
66#include <vm/vm_extern.h>
67#include <vm/vm_zone.h>
68
69#include <miscfs/fifofs/fifo.h>
70#include <miscfs/specfs/specdev.h>
71
72#include <nfs/rpcv2.h>
73#include <nfs/nfsproto.h>
74#include <nfs/nfs.h>
75#include <nfs/nfsnode.h>
76#include <nfs/nfsmount.h>
77#include <nfs/xdr_subs.h>
78#include <nfs/nfsm_subs.h>
79#include <nfs/nqnfs.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 __P((struct vop_read_args *));
99static int	nfsspec_write __P((struct vop_write_args *));
100static int	nfsfifo_read __P((struct vop_read_args *));
101static int	nfsfifo_write __P((struct vop_write_args *));
102static int	nfsspec_close __P((struct vop_close_args *));
103static int	nfsfifo_close __P((struct vop_close_args *));
104#define nfs_poll vop_nopoll
105static int	nfs_flush __P((struct vnode *,struct ucred *,int,struct proc *,int));
106static int	nfs_setattrrpc __P((struct vnode *,struct vattr *,struct ucred *,struct proc *));
107static	int	nfs_lookup __P((struct vop_lookup_args *));
108static	int	nfs_create __P((struct vop_create_args *));
109static	int	nfs_mknod __P((struct vop_mknod_args *));
110static	int	nfs_open __P((struct vop_open_args *));
111static	int	nfs_close __P((struct vop_close_args *));
112static	int	nfs_access __P((struct vop_access_args *));
113static	int	nfs_getattr __P((struct vop_getattr_args *));
114static	int	nfs_setattr __P((struct vop_setattr_args *));
115static	int	nfs_read __P((struct vop_read_args *));
116static	int	nfs_mmap __P((struct vop_mmap_args *));
117static	int	nfs_fsync __P((struct vop_fsync_args *));
118static	int	nfs_remove __P((struct vop_remove_args *));
119static	int	nfs_link __P((struct vop_link_args *));
120static	int	nfs_rename __P((struct vop_rename_args *));
121static	int	nfs_mkdir __P((struct vop_mkdir_args *));
122static	int	nfs_rmdir __P((struct vop_rmdir_args *));
123static	int	nfs_symlink __P((struct vop_symlink_args *));
124static	int	nfs_readdir __P((struct vop_readdir_args *));
125static	int	nfs_bmap __P((struct vop_bmap_args *));
126static	int	nfs_strategy __P((struct vop_strategy_args *));
127static	int	nfs_lookitup __P((struct vnode *, const char *, int,
128			struct ucred *, struct proc *, struct nfsnode **));
129static	int	nfs_sillyrename __P((struct vnode *,struct vnode *,struct componentname *));
130static int	nfsspec_access __P((struct vop_access_args *));
131static int	nfs_readlink __P((struct vop_readlink_args *));
132static int	nfs_print __P((struct vop_print_args *));
133static int	nfs_advlock __P((struct vop_advlock_args *));
134static int	nfs_bwrite __P((struct vop_bwrite_args *));
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_abortop_desc,		(vop_t *) nfs_abortop },
142	{ &vop_access_desc,		(vop_t *) nfs_access },
143	{ &vop_advlock_desc,		(vop_t *) nfs_advlock },
144	{ &vop_bmap_desc,		(vop_t *) nfs_bmap },
145	{ &vop_bwrite_desc,		(vop_t *) nfs_bwrite },
146	{ &vop_close_desc,		(vop_t *) nfs_close },
147	{ &vop_create_desc,		(vop_t *) nfs_create },
148	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
149	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
150	{ &vop_getpages_desc,		(vop_t *) nfs_getpages },
151	{ &vop_putpages_desc,		(vop_t *) nfs_putpages },
152	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
153	{ &vop_lease_desc,		(vop_t *) vop_null },
154	{ &vop_link_desc,		(vop_t *) nfs_link },
155	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
156	{ &vop_lookup_desc,		(vop_t *) nfs_lookup },
157	{ &vop_mkdir_desc,		(vop_t *) nfs_mkdir },
158	{ &vop_mknod_desc,		(vop_t *) nfs_mknod },
159	{ &vop_mmap_desc,		(vop_t *) nfs_mmap },
160	{ &vop_open_desc,		(vop_t *) nfs_open },
161	{ &vop_poll_desc,		(vop_t *) nfs_poll },
162	{ &vop_print_desc,		(vop_t *) nfs_print },
163	{ &vop_read_desc,		(vop_t *) nfs_read },
164	{ &vop_readdir_desc,		(vop_t *) nfs_readdir },
165	{ &vop_readlink_desc,		(vop_t *) nfs_readlink },
166	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
167	{ &vop_remove_desc,		(vop_t *) nfs_remove },
168	{ &vop_rename_desc,		(vop_t *) nfs_rename },
169	{ &vop_rmdir_desc,		(vop_t *) nfs_rmdir },
170	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
171	{ &vop_strategy_desc,		(vop_t *) nfs_strategy },
172	{ &vop_symlink_desc,		(vop_t *) nfs_symlink },
173	{ &vop_write_desc,		(vop_t *) nfs_write },
174	{ NULL, NULL }
175};
176static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
177	{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
178VNODEOP_SET(nfsv2_vnodeop_opv_desc);
179
180/*
181 * Special device vnode ops
182 */
183vop_t **spec_nfsv2nodeop_p;
184static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
185	{ &vop_default_desc,		(vop_t *) spec_vnoperate },
186	{ &vop_access_desc,		(vop_t *) nfsspec_access },
187	{ &vop_close_desc,		(vop_t *) nfsspec_close },
188	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
189	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
190	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
191	{ &vop_lock_desc,		(vop_t *) vop_sharedlock },
192	{ &vop_print_desc,		(vop_t *) nfs_print },
193	{ &vop_read_desc,		(vop_t *) nfsspec_read },
194	{ &vop_reclaim_desc,		(vop_t *) nfs_reclaim },
195	{ &vop_setattr_desc,		(vop_t *) nfs_setattr },
196	{ &vop_write_desc,		(vop_t *) nfsspec_write },
197	{ NULL, NULL }
198};
199static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
200	{ &spec_nfsv2nodeop_p, nfsv2_specop_entries };
201VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
202
203vop_t **fifo_nfsv2nodeop_p;
204static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
205	{ &vop_default_desc,		(vop_t *) fifo_vnoperate },
206	{ &vop_access_desc,		(vop_t *) nfsspec_access },
207	{ &vop_close_desc,		(vop_t *) nfsfifo_close },
208	{ &vop_fsync_desc,		(vop_t *) nfs_fsync },
209	{ &vop_getattr_desc,		(vop_t *) nfs_getattr },
210	{ &vop_inactive_desc,		(vop_t *) nfs_inactive },
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_write_desc,		(vop_t *) nfsfifo_write },
217	{ NULL, NULL }
218};
219static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
220	{ &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
221VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
222
223static int	nfs_commit __P((struct vnode *vp, u_quad_t offset, int cnt,
224				struct ucred *cred, struct proc *procp));
225static int	nfs_mknodrpc __P((struct vnode *dvp, struct vnode **vpp,
226				  struct componentname *cnp,
227				  struct vattr *vap));
228static int	nfs_removerpc __P((struct vnode *dvp, const char *name,
229				   int namelen,
230				   struct ucred *cred, struct proc *proc));
231static int	nfs_renamerpc __P((struct vnode *fdvp, const char *fnameptr,
232				   int fnamelen, struct vnode *tdvp,
233				   const char *tnameptr, int tnamelen,
234				   struct ucred *cred, struct proc *proc));
235static int	nfs_renameit __P((struct vnode *sdvp,
236				  struct componentname *scnp,
237				  struct sillyrename *sp));
238
239/*
240 * Global variables
241 */
242extern u_int32_t nfs_true, nfs_false;
243extern u_int32_t nfs_xdrneg1;
244extern struct nfsstats nfsstats;
245extern nfstype nfsv3_type[9];
246struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
247struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
248int nfs_numasync = 0;
249#define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
250
251static int	nfsaccess_cache_timeout = 2;
252SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
253	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
254
255static int	nfsaccess_cache_hits;
256SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
257	   &nfsaccess_cache_hits, 0, "NFS ACCESS cache hit count");
258
259static int	nfsaccess_cache_fills;
260SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_fills, CTLFLAG_RD,
261	   &nfsaccess_cache_fills, 0, "NFS ACCESS cache fill count");
262
263/*
264 * nfs access vnode op.
265 * For nfs version 2, just return ok. File accesses may fail later.
266 * For nfs version 3, use the access rpc to check accessibility. If file modes
267 * are changed on the server, accesses might still fail later.
268 */
269static int
270nfs_access(ap)
271	struct vop_access_args /* {
272		struct vnode *a_vp;
273		int  a_mode;
274		struct ucred *a_cred;
275		struct proc *a_p;
276	} */ *ap;
277{
278	register struct vnode *vp = ap->a_vp;
279	register u_int32_t *tl;
280	register caddr_t cp;
281	register int32_t t1, t2;
282	caddr_t bpos, dpos, cp2;
283	int error = 0, attrflag;
284	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
285	u_int32_t mode, rmode, wmode;
286	int v3 = NFS_ISV3(vp);
287	struct nfsnode *np = VTONFS(vp);
288
289	/*
290	 * Disallow write attempts on filesystems mounted read-only;
291	 * unless the file is a socket, fifo, or a block or character
292	 * device resident on the filesystem.
293	 */
294	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
295		switch (vp->v_type) {
296		case VREG:
297		case VDIR:
298		case VLNK:
299			return (EROFS);
300		default:
301			break;
302		}
303	}
304	/*
305	 * For nfs v3, check to see if we have done this recently, and if
306	 * so return our cached result instead of making an ACCESS call.
307	 * If not, do an access rpc, otherwise you are stuck emulating
308	 * ufs_access() locally using the vattr. This may not be correct,
309	 * since the server may apply other access criteria such as
310	 * client uid-->server uid mapping that we do not know about.
311	 */
312	if (v3) {
313		if (ap->a_mode & VREAD)
314			mode = NFSV3ACCESS_READ;
315		else
316			mode = 0;
317		if (vp->v_type != VDIR) {
318			if (ap->a_mode & VWRITE)
319				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
320			if (ap->a_mode & VEXEC)
321				mode |= NFSV3ACCESS_EXECUTE;
322		} else {
323			if (ap->a_mode & VWRITE)
324				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
325					 NFSV3ACCESS_DELETE);
326			if (ap->a_mode & VEXEC)
327				mode |= NFSV3ACCESS_LOOKUP;
328		}
329		/* XXX safety belt, only make blanket request if caching */
330		if (nfsaccess_cache_timeout > 0) {
331			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
332				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
333				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
334		} else {
335			wmode = mode;
336		}
337
338		/*
339		 * Does our cached result allow us to give a definite yes to
340		 * this request?
341		 */
342		if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
343		    (ap->a_cred->cr_uid == np->n_modeuid) &&
344		    ((np->n_mode & mode) == mode)) {
345			nfsaccess_cache_hits++;
346		} else {
347			/*
348			 * Either a no, or a don't know.  Go to the wire.
349			 */
350			nfsstats.rpccnt[NFSPROC_ACCESS]++;
351			nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
352			nfsm_fhtom(vp, v3);
353			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
354			*tl = txdr_unsigned(wmode);
355			nfsm_request(vp, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
356			nfsm_postop_attr(vp, attrflag);
357			if (!error) {
358				nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
359				rmode = fxdr_unsigned(u_int32_t, *tl);
360				/*
361				 * The NFS V3 spec does not clarify whether or not
362				 * the returned access bits can be a superset of
363				 * the ones requested, so...
364				 */
365				if ((rmode & mode) != mode) {
366					error = EACCES;
367				} else if (nfsaccess_cache_timeout > 0) {
368					/* cache the result */
369					nfsaccess_cache_fills++;
370					np->n_mode = rmode;
371					np->n_modeuid = ap->a_cred->cr_uid;
372					np->n_modestamp = time_second;
373				}
374			}
375			nfsm_reqdone;
376		}
377		return (error);
378	} else {
379		if (error = nfsspec_access(ap))
380			return (error);
381
382		/*
383		 * Attempt to prevent a mapped root from accessing a file
384		 * which it shouldn't.  We try to read a byte from the file
385		 * if the user is root and the file is not zero length.
386		 * After calling nfsspec_access, we should have the correct
387		 * file size cached.
388		 */
389		if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
390		    && VTONFS(vp)->n_size > 0) {
391			struct iovec aiov;
392			struct uio auio;
393			char buf[1];
394
395			aiov.iov_base = buf;
396			aiov.iov_len = 1;
397			auio.uio_iov = &aiov;
398			auio.uio_iovcnt = 1;
399			auio.uio_offset = 0;
400			auio.uio_resid = 1;
401			auio.uio_segflg = UIO_SYSSPACE;
402			auio.uio_rw = UIO_READ;
403			auio.uio_procp = ap->a_p;
404
405			if (vp->v_type == VREG)
406				error = nfs_readrpc(vp, &auio, ap->a_cred);
407			else if (vp->v_type == VDIR) {
408				char* bp;
409				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
410				aiov.iov_base = bp;
411				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
412				error = nfs_readdirrpc(vp, &auio, ap->a_cred);
413				free(bp, M_TEMP);
414			} else if (vp->v_type = VLNK)
415				error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
416			else
417				error = EACCES;
418		}
419		return (error);
420	}
421}
422
423/*
424 * nfs open vnode op
425 * Check to see if the type is ok
426 * and that deletion is not in progress.
427 * For paged in text files, you will need to flush the page cache
428 * if consistency is lost.
429 */
430/* ARGSUSED */
431static int
432nfs_open(ap)
433	struct vop_open_args /* {
434		struct vnode *a_vp;
435		int  a_mode;
436		struct ucred *a_cred;
437		struct proc *a_p;
438	} */ *ap;
439{
440	register struct vnode *vp = ap->a_vp;
441	struct nfsnode *np = VTONFS(vp);
442	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
443	struct vattr vattr;
444	int error;
445
446	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
447#ifdef DIAGNOSTIC
448		printf("open eacces vtyp=%d\n",vp->v_type);
449#endif
450		return (EACCES);
451	}
452	/*
453	 * Get a valid lease. If cached data is stale, flush it.
454	 */
455	if (nmp->nm_flag & NFSMNT_NQNFS) {
456		if (NQNFS_CKINVALID(vp, np, ND_READ)) {
457		    do {
458			error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
459			    ap->a_p);
460		    } while (error == NQNFS_EXPIRED);
461		    if (error)
462			return (error);
463		    if (np->n_lrev != np->n_brev ||
464			(np->n_flag & NQNFSNONCACHE)) {
465			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
466				ap->a_p, 1)) == EINTR)
467				return (error);
468			np->n_brev = np->n_lrev;
469		    }
470		}
471	} else {
472		if (np->n_flag & NMODIFIED) {
473			if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
474				ap->a_p, 1)) == EINTR)
475				return (error);
476			np->n_attrstamp = 0;
477			if (vp->v_type == VDIR)
478				np->n_direofoffset = 0;
479			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
480			if (error)
481				return (error);
482			np->n_mtime = vattr.va_mtime.tv_sec;
483		} else {
484			error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
485			if (error)
486				return (error);
487			if (np->n_mtime != vattr.va_mtime.tv_sec) {
488				if (vp->v_type == VDIR)
489					np->n_direofoffset = 0;
490				if ((error = nfs_vinvalbuf(vp, V_SAVE,
491					ap->a_cred, ap->a_p, 1)) == EINTR)
492					return (error);
493				np->n_mtime = vattr.va_mtime.tv_sec;
494			}
495		}
496	}
497	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
498		np->n_attrstamp = 0; /* For Open/Close consistency */
499	return (0);
500}
501
502/*
503 * nfs close vnode op
504 * What an NFS client should do upon close after writing is a debatable issue.
505 * Most NFS clients push delayed writes to the server upon close, basically for
506 * two reasons:
507 * 1 - So that any write errors may be reported back to the client process
508 *     doing the close system call. By far the two most likely errors are
509 *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
510 * 2 - To put a worst case upper bound on cache inconsistency between
511 *     multiple clients for the file.
512 * There is also a consistency problem for Version 2 of the protocol w.r.t.
513 * not being able to tell if other clients are writing a file concurrently,
514 * since there is no way of knowing if the changed modify time in the reply
515 * is only due to the write for this client.
516 * (NFS Version 3 provides weak cache consistency data in the reply that
517 *  should be sufficient to detect and handle this case.)
518 *
519 * The current code does the following:
520 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
521 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
522 *                     or commit them (this satisfies 1 and 2 except for the
523 *                     case where the server crashes after this close but
524 *                     before the commit RPC, which is felt to be "good
525 *                     enough". Changing the last argument to nfs_flush() to
526 *                     a 1 would force a commit operation, if it is felt a
527 *                     commit is necessary now.
528 * for NQNFS         - do nothing now, since 2 is dealt with via leases and
529 *                     1 should be dealt with via an fsync() system call for
530 *                     cases where write errors are important.
531 */
532/* ARGSUSED */
533static int
534nfs_close(ap)
535	struct vop_close_args /* {
536		struct vnodeop_desc *a_desc;
537		struct vnode *a_vp;
538		int  a_fflag;
539		struct ucred *a_cred;
540		struct proc *a_p;
541	} */ *ap;
542{
543	register struct vnode *vp = ap->a_vp;
544	register struct nfsnode *np = VTONFS(vp);
545	int error = 0;
546
547	if (vp->v_type == VREG) {
548	    if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
549		(np->n_flag & NMODIFIED)) {
550		if (NFS_ISV3(vp)) {
551		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
552		    np->n_flag &= ~NMODIFIED;
553		} else
554		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
555		np->n_attrstamp = 0;
556	    }
557	    if (np->n_flag & NWRITEERR) {
558		np->n_flag &= ~NWRITEERR;
559		error = np->n_error;
560	    }
561	}
562	return (error);
563}
564
565/*
566 * nfs getattr call from vfs.
567 */
568static int
569nfs_getattr(ap)
570	struct vop_getattr_args /* {
571		struct vnode *a_vp;
572		struct vattr *a_vap;
573		struct ucred *a_cred;
574		struct proc *a_p;
575	} */ *ap;
576{
577	register struct vnode *vp = ap->a_vp;
578	register struct nfsnode *np = VTONFS(vp);
579	register caddr_t cp;
580	register u_int32_t *tl;
581	register int32_t t1, t2;
582	caddr_t bpos, dpos;
583	int error = 0;
584	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
585	int v3 = NFS_ISV3(vp);
586
587	/*
588	 * Update local times for special files.
589	 */
590	if (np->n_flag & (NACC | NUPD))
591		np->n_flag |= NCHG;
592	/*
593	 * First look in the cache.
594	 */
595	if (nfs_getattrcache(vp, ap->a_vap) == 0)
596		return (0);
597	nfsstats.rpccnt[NFSPROC_GETATTR]++;
598	nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
599	nfsm_fhtom(vp, v3);
600	nfsm_request(vp, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
601	if (!error) {
602		nfsm_loadattr(vp, ap->a_vap);
603	}
604	nfsm_reqdone;
605	return (error);
606}
607
608/*
609 * nfs setattr call.
610 */
611static int
612nfs_setattr(ap)
613	struct vop_setattr_args /* {
614		struct vnodeop_desc *a_desc;
615		struct vnode *a_vp;
616		struct vattr *a_vap;
617		struct ucred *a_cred;
618		struct proc *a_p;
619	} */ *ap;
620{
621	register struct vnode *vp = ap->a_vp;
622	register struct nfsnode *np = VTONFS(vp);
623	register struct vattr *vap = ap->a_vap;
624	int error = 0;
625	u_quad_t tsize;
626
627#ifndef nolint
628	tsize = (u_quad_t)0;
629#endif
630
631	/*
632	 * Setting of flags is not supported.
633	 */
634	if (vap->va_flags != VNOVAL)
635		return (EOPNOTSUPP);
636
637	/*
638	 * Disallow write attempts if the filesystem is mounted read-only.
639	 */
640  	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
641	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
642	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
643	    (vp->v_mount->mnt_flag & MNT_RDONLY))
644		return (EROFS);
645	if (vap->va_size != VNOVAL) {
646 		switch (vp->v_type) {
647 		case VDIR:
648 			return (EISDIR);
649 		case VCHR:
650 		case VBLK:
651 		case VSOCK:
652 		case VFIFO:
653			if (vap->va_mtime.tv_sec == VNOVAL &&
654			    vap->va_atime.tv_sec == VNOVAL &&
655			    vap->va_mode == (mode_t)VNOVAL &&
656			    vap->va_uid == (uid_t)VNOVAL &&
657			    vap->va_gid == (gid_t)VNOVAL)
658				return (0);
659 			vap->va_size = VNOVAL;
660 			break;
661 		default:
662			/*
663			 * Disallow write attempts if the filesystem is
664			 * mounted read-only.
665			 */
666			if (vp->v_mount->mnt_flag & MNT_RDONLY)
667				return (EROFS);
668			vnode_pager_setsize(vp, vap->va_size);
669 			if (np->n_flag & NMODIFIED) {
670 			    if (vap->va_size == 0)
671 				error = nfs_vinvalbuf(vp, 0,
672 					ap->a_cred, ap->a_p, 1);
673 			    else
674 				error = nfs_vinvalbuf(vp, V_SAVE,
675 					ap->a_cred, ap->a_p, 1);
676 			    if (error) {
677				vnode_pager_setsize(vp, np->n_size);
678 				return (error);
679			    }
680 			}
681 			tsize = np->n_size;
682 			np->n_size = np->n_vattr.va_size = vap->va_size;
683  		};
684  	} else if ((vap->va_mtime.tv_sec != VNOVAL ||
685		vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
686		vp->v_type == VREG &&
687  		(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
688		 ap->a_p, 1)) == EINTR)
689		return (error);
690	error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
691	if (error && vap->va_size != VNOVAL) {
692		np->n_size = np->n_vattr.va_size = tsize;
693		vnode_pager_setsize(vp, np->n_size);
694	}
695	return (error);
696}
697
698/*
699 * Do an nfs setattr rpc.
700 */
701static int
702nfs_setattrrpc(vp, vap, cred, procp)
703	register struct vnode *vp;
704	register struct vattr *vap;
705	struct ucred *cred;
706	struct proc *procp;
707{
708	register struct nfsv2_sattr *sp;
709	register caddr_t cp;
710	register int32_t t1, t2;
711	caddr_t bpos, dpos, cp2;
712	u_int32_t *tl;
713	int error = 0, wccflag = NFSV3_WCCRATTR;
714	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
715	int v3 = NFS_ISV3(vp);
716
717	nfsstats.rpccnt[NFSPROC_SETATTR]++;
718	nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
719	nfsm_fhtom(vp, v3);
720	if (v3) {
721		if (vap->va_mode != (mode_t)VNOVAL) {
722			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
723			*tl++ = nfs_true;
724			*tl = txdr_unsigned(vap->va_mode);
725		} else {
726			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
727			*tl = nfs_false;
728		}
729		if (vap->va_uid != (uid_t)VNOVAL) {
730			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
731			*tl++ = nfs_true;
732			*tl = txdr_unsigned(vap->va_uid);
733		} else {
734			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
735			*tl = nfs_false;
736		}
737		if (vap->va_gid != (gid_t)VNOVAL) {
738			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
739			*tl++ = nfs_true;
740			*tl = txdr_unsigned(vap->va_gid);
741		} else {
742			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
743			*tl = nfs_false;
744		}
745		if (vap->va_size != VNOVAL) {
746			nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
747			*tl++ = nfs_true;
748			txdr_hyper(&vap->va_size, tl);
749		} else {
750			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
751			*tl = nfs_false;
752		}
753		if (vap->va_atime.tv_sec != VNOVAL) {
754			if (vap->va_atime.tv_sec != time_second) {
755				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
756				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
757				txdr_nfsv3time(&vap->va_atime, tl);
758			} else {
759				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
760				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
761			}
762		} else {
763			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
764			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
765		}
766		if (vap->va_mtime.tv_sec != VNOVAL) {
767			if (vap->va_mtime.tv_sec != time_second) {
768				nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
769				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
770				txdr_nfsv3time(&vap->va_mtime, tl);
771			} else {
772				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
773				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
774			}
775		} else {
776			nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
777			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
778		}
779		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
780		*tl = nfs_false;
781	} else {
782		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
783		if (vap->va_mode == (mode_t)VNOVAL)
784			sp->sa_mode = nfs_xdrneg1;
785		else
786			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
787		if (vap->va_uid == (uid_t)VNOVAL)
788			sp->sa_uid = nfs_xdrneg1;
789		else
790			sp->sa_uid = txdr_unsigned(vap->va_uid);
791		if (vap->va_gid == (gid_t)VNOVAL)
792			sp->sa_gid = nfs_xdrneg1;
793		else
794			sp->sa_gid = txdr_unsigned(vap->va_gid);
795		sp->sa_size = txdr_unsigned(vap->va_size);
796		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
797		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
798	}
799	nfsm_request(vp, NFSPROC_SETATTR, procp, cred);
800	if (v3) {
801		nfsm_wcc_data(vp, wccflag);
802	} else
803		nfsm_loadattr(vp, (struct vattr *)0);
804	nfsm_reqdone;
805	return (error);
806}
807
808/*
809 * nfs lookup call, one step at a time...
810 * First look in cache
811 * If not found, unlock the directory nfsnode and do the rpc
812 */
813static int
814nfs_lookup(ap)
815	struct vop_lookup_args /* {
816		struct vnodeop_desc *a_desc;
817		struct vnode *a_dvp;
818		struct vnode **a_vpp;
819		struct componentname *a_cnp;
820	} */ *ap;
821{
822	struct componentname *cnp = ap->a_cnp;
823	struct vnode *dvp = ap->a_dvp;
824	struct vnode **vpp = ap->a_vpp;
825	int flags = cnp->cn_flags;
826	struct vnode *newvp;
827	u_int32_t *tl;
828	caddr_t cp;
829	int32_t t1, t2;
830	struct nfsmount *nmp;
831	caddr_t bpos, dpos, cp2;
832	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
833	long len;
834	nfsfh_t *fhp;
835	struct nfsnode *np;
836	int lockparent, wantparent, error = 0, attrflag, fhsize;
837	int v3 = NFS_ISV3(dvp);
838	struct proc *p = cnp->cn_proc;
839
840	*vpp = NULLVP;
841	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
842	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
843		return (EROFS);
844	if (dvp->v_type != VDIR)
845		return (ENOTDIR);
846	lockparent = flags & LOCKPARENT;
847	wantparent = flags & (LOCKPARENT|WANTPARENT);
848	nmp = VFSTONFS(dvp->v_mount);
849	np = VTONFS(dvp);
850	if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
851		struct vattr vattr;
852		int vpid;
853
854		if (error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, p)) {
855			*vpp = NULLVP;
856			return (error);
857		}
858
859		newvp = *vpp;
860		vpid = newvp->v_id;
861		/*
862		 * See the comment starting `Step through' in ufs/ufs_lookup.c
863		 * for an explanation of the locking protocol
864		 */
865		if (dvp == newvp) {
866			VREF(newvp);
867			error = 0;
868		} else if (flags & ISDOTDOT) {
869			VOP_UNLOCK(dvp, 0, p);
870			error = vget(newvp, LK_EXCLUSIVE, p);
871			if (!error && lockparent && (flags & ISLASTCN))
872				error = vn_lock(dvp, LK_EXCLUSIVE, p);
873		} else {
874			error = vget(newvp, LK_EXCLUSIVE, p);
875			if (!lockparent || error || !(flags & ISLASTCN))
876				VOP_UNLOCK(dvp, 0, p);
877		}
878		if (!error) {
879			if (vpid == newvp->v_id) {
880			   if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, p)
881			    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
882				nfsstats.lookupcache_hits++;
883				if (cnp->cn_nameiop != LOOKUP &&
884				    (flags & ISLASTCN))
885					cnp->cn_flags |= SAVENAME;
886				return (0);
887			   }
888			   cache_purge(newvp);
889			}
890			vput(newvp);
891			if (lockparent && dvp != newvp && (flags & ISLASTCN))
892				VOP_UNLOCK(dvp, 0, p);
893		}
894		error = vn_lock(dvp, LK_EXCLUSIVE, p);
895		*vpp = NULLVP;
896		if (error)
897			return (error);
898	}
899	error = 0;
900	newvp = NULLVP;
901	nfsstats.lookupcache_misses++;
902	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
903	len = cnp->cn_namelen;
904	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
905		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
906	nfsm_fhtom(dvp, v3);
907	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
908	nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
909	if (error) {
910		nfsm_postop_attr(dvp, attrflag);
911		m_freem(mrep);
912		goto nfsmout;
913	}
914	nfsm_getfh(fhp, fhsize, v3);
915
916	/*
917	 * Handle RENAME case...
918	 */
919	if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
920		if (NFS_CMPFH(np, fhp, fhsize)) {
921			m_freem(mrep);
922			return (EISDIR);
923		}
924		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
925		if (error) {
926			m_freem(mrep);
927			return (error);
928		}
929		newvp = NFSTOV(np);
930		if (v3) {
931			nfsm_postop_attr(newvp, attrflag);
932			nfsm_postop_attr(dvp, attrflag);
933		} else
934			nfsm_loadattr(newvp, (struct vattr *)0);
935		*vpp = newvp;
936		m_freem(mrep);
937		cnp->cn_flags |= SAVENAME;
938		if (!lockparent)
939			VOP_UNLOCK(dvp, 0, p);
940		return (0);
941	}
942
943	if (flags & ISDOTDOT) {
944		VOP_UNLOCK(dvp, 0, p);
945		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
946		if (error) {
947			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
948			return (error);
949		}
950		newvp = NFSTOV(np);
951		if (lockparent && (flags & ISLASTCN) &&
952		    (error = vn_lock(dvp, LK_EXCLUSIVE, p))) {
953		    	vput(newvp);
954			return (error);
955		}
956	} else if (NFS_CMPFH(np, fhp, fhsize)) {
957		VREF(dvp);
958		newvp = dvp;
959	} else {
960		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
961		if (error) {
962			m_freem(mrep);
963			return (error);
964		}
965		if (!lockparent || !(flags & ISLASTCN))
966			VOP_UNLOCK(dvp, 0, p);
967		newvp = NFSTOV(np);
968	}
969	if (v3) {
970		nfsm_postop_attr(newvp, attrflag);
971		nfsm_postop_attr(dvp, attrflag);
972	} else
973		nfsm_loadattr(newvp, (struct vattr *)0);
974	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
975		cnp->cn_flags |= SAVENAME;
976	if ((cnp->cn_flags & MAKEENTRY) &&
977	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
978		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
979		cache_enter(dvp, newvp, cnp);
980	}
981	*vpp = newvp;
982	nfsm_reqdone;
983	if (error) {
984		if (newvp != NULLVP) {
985			vrele(newvp);
986			*vpp = NULLVP;
987		}
988		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
989		    (flags & ISLASTCN) && error == ENOENT) {
990			if (!lockparent)
991				VOP_UNLOCK(dvp, 0, p);
992			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
993				error = EROFS;
994			else
995				error = EJUSTRETURN;
996		}
997		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
998			cnp->cn_flags |= SAVENAME;
999	}
1000	return (error);
1001}
1002
1003/*
1004 * nfs read call.
1005 * Just call nfs_bioread() to do the work.
1006 */
1007static int
1008nfs_read(ap)
1009	struct vop_read_args /* {
1010		struct vnode *a_vp;
1011		struct uio *a_uio;
1012		int  a_ioflag;
1013		struct ucred *a_cred;
1014	} */ *ap;
1015{
1016	register struct vnode *vp = ap->a_vp;
1017
1018	if (vp->v_type != VREG)
1019		return (EPERM);
1020	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1021}
1022
1023/*
1024 * nfs readlink call
1025 */
1026static int
1027nfs_readlink(ap)
1028	struct vop_readlink_args /* {
1029		struct vnode *a_vp;
1030		struct uio *a_uio;
1031		struct ucred *a_cred;
1032	} */ *ap;
1033{
1034	register struct vnode *vp = ap->a_vp;
1035
1036	if (vp->v_type != VLNK)
1037		return (EINVAL);
1038	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1039}
1040
1041/*
1042 * Do a readlink rpc.
1043 * Called by nfs_doio() from below the buffer cache.
1044 */
1045int
1046nfs_readlinkrpc(vp, uiop, cred)
1047	register struct vnode *vp;
1048	struct uio *uiop;
1049	struct ucred *cred;
1050{
1051	register u_int32_t *tl;
1052	register caddr_t cp;
1053	register int32_t t1, t2;
1054	caddr_t bpos, dpos, cp2;
1055	int error = 0, len, attrflag;
1056	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1057	int v3 = NFS_ISV3(vp);
1058
1059	nfsstats.rpccnt[NFSPROC_READLINK]++;
1060	nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1061	nfsm_fhtom(vp, v3);
1062	nfsm_request(vp, NFSPROC_READLINK, uiop->uio_procp, cred);
1063	if (v3)
1064		nfsm_postop_attr(vp, attrflag);
1065	if (!error) {
1066		nfsm_strsiz(len, NFS_MAXPATHLEN);
1067		nfsm_mtouio(uiop, len);
1068	}
1069	nfsm_reqdone;
1070	return (error);
1071}
1072
1073/*
1074 * nfs read rpc call
1075 * Ditto above
1076 */
1077int
1078nfs_readrpc(vp, uiop, cred)
1079	register struct vnode *vp;
1080	struct uio *uiop;
1081	struct ucred *cred;
1082{
1083	register u_int32_t *tl;
1084	register caddr_t cp;
1085	register int32_t t1, t2;
1086	caddr_t bpos, dpos, cp2;
1087	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1088	struct nfsmount *nmp;
1089	int error = 0, len, retlen, tsiz, eof, attrflag;
1090	int v3 = NFS_ISV3(vp);
1091
1092#ifndef nolint
1093	eof = 0;
1094#endif
1095	nmp = VFSTONFS(vp->v_mount);
1096	tsiz = uiop->uio_resid;
1097	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1098		return (EFBIG);
1099	while (tsiz > 0) {
1100		nfsstats.rpccnt[NFSPROC_READ]++;
1101		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1102		nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1103		nfsm_fhtom(vp, v3);
1104		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1105		if (v3) {
1106			txdr_hyper(&uiop->uio_offset, tl);
1107			*(tl + 2) = txdr_unsigned(len);
1108		} else {
1109			*tl++ = txdr_unsigned(uiop->uio_offset);
1110			*tl++ = txdr_unsigned(len);
1111			*tl = 0;
1112		}
1113		nfsm_request(vp, NFSPROC_READ, uiop->uio_procp, cred);
1114		if (v3) {
1115			nfsm_postop_attr(vp, attrflag);
1116			if (error) {
1117				m_freem(mrep);
1118				goto nfsmout;
1119			}
1120			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1121			eof = fxdr_unsigned(int, *(tl + 1));
1122		} else
1123			nfsm_loadattr(vp, (struct vattr *)0);
1124		nfsm_strsiz(retlen, nmp->nm_rsize);
1125		nfsm_mtouio(uiop, retlen);
1126		m_freem(mrep);
1127		tsiz -= retlen;
1128		if (v3) {
1129			if (eof || retlen == 0)
1130				tsiz = 0;
1131		} else if (retlen < len)
1132			tsiz = 0;
1133	}
1134nfsmout:
1135	return (error);
1136}
1137
1138/*
1139 * nfs write call
1140 */
1141int
1142nfs_writerpc(vp, uiop, cred, iomode, must_commit)
1143	register struct vnode *vp;
1144	register struct uio *uiop;
1145	struct ucred *cred;
1146	int *iomode, *must_commit;
1147{
1148	register u_int32_t *tl;
1149	register caddr_t cp;
1150	register int32_t t1, t2, backup;
1151	caddr_t bpos, dpos, cp2;
1152	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1153	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1154	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1155	int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1156
1157#ifndef DIAGNOSTIC
1158	if (uiop->uio_iovcnt != 1)
1159		panic("nfs: writerpc iovcnt > 1");
1160#endif
1161	*must_commit = 0;
1162	tsiz = uiop->uio_resid;
1163	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1164		return (EFBIG);
1165	while (tsiz > 0) {
1166		nfsstats.rpccnt[NFSPROC_WRITE]++;
1167		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1168		nfsm_reqhead(vp, NFSPROC_WRITE,
1169			NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1170		nfsm_fhtom(vp, v3);
1171		if (v3) {
1172			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1173			txdr_hyper(&uiop->uio_offset, tl);
1174			tl += 2;
1175			*tl++ = txdr_unsigned(len);
1176			*tl++ = txdr_unsigned(*iomode);
1177			*tl = txdr_unsigned(len);
1178		} else {
1179			register u_int32_t x;
1180
1181			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1182			/* Set both "begin" and "current" to non-garbage. */
1183			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1184			*tl++ = x;	/* "begin offset" */
1185			*tl++ = x;	/* "current offset" */
1186			x = txdr_unsigned(len);
1187			*tl++ = x;	/* total to this offset */
1188			*tl = x;	/* size of this write */
1189		}
1190		nfsm_uiotom(uiop, len);
1191		nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp, cred);
1192		if (v3) {
1193			wccflag = NFSV3_WCCCHK;
1194			nfsm_wcc_data(vp, wccflag);
1195			if (!error) {
1196				nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1197					+ NFSX_V3WRITEVERF);
1198				rlen = fxdr_unsigned(int, *tl++);
1199				if (rlen == 0) {
1200					error = NFSERR_IO;
1201					m_freem(mrep);
1202					break;
1203				} else if (rlen < len) {
1204					backup = len - rlen;
1205					uiop->uio_iov->iov_base -= backup;
1206					uiop->uio_iov->iov_len += backup;
1207					uiop->uio_offset -= backup;
1208					uiop->uio_resid += backup;
1209					len = rlen;
1210				}
1211				commit = fxdr_unsigned(int, *tl++);
1212
1213				/*
1214				 * Return the lowest committment level
1215				 * obtained by any of the RPCs.
1216				 */
1217				if (committed == NFSV3WRITE_FILESYNC)
1218					committed = commit;
1219				else if (committed == NFSV3WRITE_DATASYNC &&
1220					commit == NFSV3WRITE_UNSTABLE)
1221					committed = commit;
1222				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1223				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1224					NFSX_V3WRITEVERF);
1225				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1226				} else if (bcmp((caddr_t)tl,
1227				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1228				    *must_commit = 1;
1229				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1230					NFSX_V3WRITEVERF);
1231				}
1232			}
1233		} else
1234		    nfsm_loadattr(vp, (struct vattr *)0);
1235		if (wccflag)
1236		    VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1237		m_freem(mrep);
1238		if (error)
1239			break;
1240		tsiz -= len;
1241	}
1242nfsmout:
1243	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1244		committed = NFSV3WRITE_FILESYNC;
1245	*iomode = committed;
1246	if (error)
1247		uiop->uio_resid = tsiz;
1248	return (error);
1249}
1250
1251/*
1252 * nfs mknod rpc
1253 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1254 * mode set to specify the file type and the size field for rdev.
1255 */
1256static int
1257nfs_mknodrpc(dvp, vpp, cnp, vap)
1258	register struct vnode *dvp;
1259	register struct vnode **vpp;
1260	register struct componentname *cnp;
1261	register struct vattr *vap;
1262{
1263	register struct nfsv2_sattr *sp;
1264	register struct nfsv3_sattr *sp3;
1265	register u_int32_t *tl;
1266	register caddr_t cp;
1267	register int32_t t1, t2;
1268	struct vnode *newvp = (struct vnode *)0;
1269	struct nfsnode *np = (struct nfsnode *)0;
1270	struct vattr vattr;
1271	char *cp2;
1272	caddr_t bpos, dpos;
1273	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1274	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1275	u_int32_t rdev;
1276	int v3 = NFS_ISV3(dvp);
1277
1278	if (vap->va_type == VCHR || vap->va_type == VBLK)
1279		rdev = txdr_unsigned(vap->va_rdev);
1280	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1281		rdev = nfs_xdrneg1;
1282	else {
1283		VOP_ABORTOP(dvp, cnp);
1284		return (EOPNOTSUPP);
1285	}
1286	if (error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_proc)) {
1287		VOP_ABORTOP(dvp, cnp);
1288		return (error);
1289	}
1290	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1291	nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1292		+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1293	nfsm_fhtom(dvp, v3);
1294	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1295	if (v3) {
1296		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3SRVSATTR);
1297		*tl++ = vtonfsv3_type(vap->va_type);
1298		sp3 = (struct nfsv3_sattr *)tl;
1299		nfsm_v3sattr(sp3, vap);
1300		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1301			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1302			*tl++ = txdr_unsigned(major(vap->va_rdev));
1303			*tl = txdr_unsigned(minor(vap->va_rdev));
1304		}
1305	} else {
1306		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1307		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1308		sp->sa_uid = nfs_xdrneg1;
1309		sp->sa_gid = nfs_xdrneg1;
1310		sp->sa_size = rdev;
1311		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1312		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1313	}
1314	nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1315	if (!error) {
1316		nfsm_mtofh(dvp, newvp, v3, gotvp);
1317		if (!gotvp) {
1318			if (newvp) {
1319				vput(newvp);
1320				newvp = (struct vnode *)0;
1321			}
1322			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1323			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1324			if (!error)
1325				newvp = NFSTOV(np);
1326		}
1327	}
1328	if (v3)
1329		nfsm_wcc_data(dvp, wccflag);
1330	nfsm_reqdone;
1331	if (error) {
1332		if (newvp)
1333			vput(newvp);
1334	} else {
1335		if (cnp->cn_flags & MAKEENTRY)
1336			cache_enter(dvp, newvp, cnp);
1337		*vpp = newvp;
1338	}
1339	zfree(namei_zone, cnp->cn_pnbuf);
1340	VTONFS(dvp)->n_flag |= NMODIFIED;
1341	if (!wccflag)
1342		VTONFS(dvp)->n_attrstamp = 0;
1343	return (error);
1344}
1345
1346/*
1347 * nfs mknod vop
1348 * just call nfs_mknodrpc() to do the work.
1349 */
1350/* ARGSUSED */
1351static int
1352nfs_mknod(ap)
1353	struct vop_mknod_args /* {
1354		struct vnode *a_dvp;
1355		struct vnode **a_vpp;
1356		struct componentname *a_cnp;
1357		struct vattr *a_vap;
1358	} */ *ap;
1359{
1360	struct vnode *newvp;
1361	int error;
1362
1363	error = nfs_mknodrpc(ap->a_dvp, &newvp, ap->a_cnp, ap->a_vap);
1364	if (!error)
1365		vput(newvp);
1366	return (error);
1367}
1368
1369static u_long create_verf;
1370/*
1371 * nfs file create call
1372 */
1373static int
1374nfs_create(ap)
1375	struct vop_create_args /* {
1376		struct vnode *a_dvp;
1377		struct vnode **a_vpp;
1378		struct componentname *a_cnp;
1379		struct vattr *a_vap;
1380	} */ *ap;
1381{
1382	register struct vnode *dvp = ap->a_dvp;
1383	register struct vattr *vap = ap->a_vap;
1384	register struct componentname *cnp = ap->a_cnp;
1385	register struct nfsv2_sattr *sp;
1386	register struct nfsv3_sattr *sp3;
1387	register u_int32_t *tl;
1388	register caddr_t cp;
1389	register int32_t t1, t2;
1390	struct nfsnode *np = (struct nfsnode *)0;
1391	struct vnode *newvp = (struct vnode *)0;
1392	caddr_t bpos, dpos, cp2;
1393	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1394	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1395	struct vattr vattr;
1396	int v3 = NFS_ISV3(dvp);
1397
1398	/*
1399	 * Oops, not for me..
1400	 */
1401	if (vap->va_type == VSOCK)
1402		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1403
1404	if (error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_proc)) {
1405		VOP_ABORTOP(dvp, cnp);
1406		return (error);
1407	}
1408	if (vap->va_vaflags & VA_EXCLUSIVE)
1409		fmode |= O_EXCL;
1410again:
1411	nfsstats.rpccnt[NFSPROC_CREATE]++;
1412	nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1413		nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1414	nfsm_fhtom(dvp, v3);
1415	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1416	if (v3) {
1417		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1418		if (fmode & O_EXCL) {
1419		    *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1420		    nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1421#ifdef INET
1422		    if (!TAILQ_EMPTY(&in_ifaddrhead))
1423			*tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
1424		    else
1425#endif
1426			*tl++ = create_verf;
1427		    *tl = ++create_verf;
1428		} else {
1429		    *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1430		    nfsm_build(tl, u_int32_t *, NFSX_V3SRVSATTR);
1431		    sp3 = (struct nfsv3_sattr *)tl;
1432		    nfsm_v3sattr(sp3, vap);
1433		}
1434	} else {
1435		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1436		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1437		sp->sa_uid = nfs_xdrneg1;
1438		sp->sa_gid = nfs_xdrneg1;
1439		sp->sa_size = 0;
1440		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1441		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1442	}
1443	nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1444	if (!error) {
1445		nfsm_mtofh(dvp, newvp, v3, gotvp);
1446		if (!gotvp) {
1447			if (newvp) {
1448				vput(newvp);
1449				newvp = (struct vnode *)0;
1450			}
1451			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1452			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1453			if (!error)
1454				newvp = NFSTOV(np);
1455		}
1456	}
1457	if (v3)
1458		nfsm_wcc_data(dvp, wccflag);
1459	nfsm_reqdone;
1460	if (error) {
1461		if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1462			fmode &= ~O_EXCL;
1463			goto again;
1464		}
1465		if (newvp)
1466			vput(newvp);
1467	} else if (v3 && (fmode & O_EXCL))
1468		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1469	if (!error) {
1470		if (cnp->cn_flags & MAKEENTRY)
1471			cache_enter(dvp, newvp, cnp);
1472		*ap->a_vpp = newvp;
1473	}
1474	zfree(namei_zone, cnp->cn_pnbuf);
1475	VTONFS(dvp)->n_flag |= NMODIFIED;
1476	if (!wccflag)
1477		VTONFS(dvp)->n_attrstamp = 0;
1478	return (error);
1479}
1480
1481/*
1482 * nfs file remove call
1483 * To try and make nfs semantics closer to ufs semantics, a file that has
1484 * other processes using the vnode is renamed instead of removed and then
1485 * removed later on the last close.
1486 * - If v_usecount > 1
1487 *	  If a rename is not already in the works
1488 *	     call nfs_sillyrename() to set it up
1489 *     else
1490 *	  do the remove rpc
1491 */
1492static int
1493nfs_remove(ap)
1494	struct vop_remove_args /* {
1495		struct vnodeop_desc *a_desc;
1496		struct vnode * a_dvp;
1497		struct vnode * a_vp;
1498		struct componentname * a_cnp;
1499	} */ *ap;
1500{
1501	register struct vnode *vp = ap->a_vp;
1502	register struct vnode *dvp = ap->a_dvp;
1503	register struct componentname *cnp = ap->a_cnp;
1504	register struct nfsnode *np = VTONFS(vp);
1505	int error = 0;
1506	struct vattr vattr;
1507
1508#ifndef DIAGNOSTIC
1509	if ((cnp->cn_flags & HASBUF) == 0)
1510		panic("nfs_remove: no name");
1511	if (vp->v_usecount < 1)
1512		panic("nfs_remove: bad v_usecount");
1513#endif
1514	if (vp->v_type == VDIR)
1515		error = EPERM;
1516	else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1517	    VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1518	    vattr.va_nlink > 1)) {
1519		/*
1520		 * Purge the name cache so that the chance of a lookup for
1521		 * the name succeeding while the remove is in progress is
1522		 * minimized. Without node locking it can still happen, such
1523		 * that an I/O op returns ESTALE, but since you get this if
1524		 * another host removes the file..
1525		 */
1526		cache_purge(vp);
1527		/*
1528		 * throw away biocache buffers, mainly to avoid
1529		 * unnecessary delayed writes later.
1530		 */
1531		error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1532		/* Do the rpc */
1533		if (error != EINTR)
1534			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1535				cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1536		/*
1537		 * Kludge City: If the first reply to the remove rpc is lost..
1538		 *   the reply to the retransmitted request will be ENOENT
1539		 *   since the file was in fact removed
1540		 *   Therefore, we cheat and return success.
1541		 */
1542		if (error == ENOENT)
1543			error = 0;
1544	} else if (!np->n_sillyrename)
1545		error = nfs_sillyrename(dvp, vp, cnp);
1546	zfree(namei_zone, cnp->cn_pnbuf);
1547	np->n_attrstamp = 0;
1548	return (error);
1549}
1550
1551/*
1552 * nfs file remove rpc called from nfs_inactive
1553 */
1554int
1555nfs_removeit(sp)
1556	register struct sillyrename *sp;
1557{
1558
1559	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1560		(struct proc *)0));
1561}
1562
1563/*
1564 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1565 */
1566static int
1567nfs_removerpc(dvp, name, namelen, cred, proc)
1568	register struct vnode *dvp;
1569	const char *name;
1570	int namelen;
1571	struct ucred *cred;
1572	struct proc *proc;
1573{
1574	register u_int32_t *tl;
1575	register caddr_t cp;
1576	register int32_t t1, t2;
1577	caddr_t bpos, dpos, cp2;
1578	int error = 0, wccflag = NFSV3_WCCRATTR;
1579	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1580	int v3 = NFS_ISV3(dvp);
1581
1582	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1583	nfsm_reqhead(dvp, NFSPROC_REMOVE,
1584		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1585	nfsm_fhtom(dvp, v3);
1586	nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1587	nfsm_request(dvp, NFSPROC_REMOVE, proc, cred);
1588	if (v3)
1589		nfsm_wcc_data(dvp, wccflag);
1590	nfsm_reqdone;
1591	VTONFS(dvp)->n_flag |= NMODIFIED;
1592	if (!wccflag)
1593		VTONFS(dvp)->n_attrstamp = 0;
1594	return (error);
1595}
1596
1597/*
1598 * nfs file rename call
1599 */
1600static int
1601nfs_rename(ap)
1602	struct vop_rename_args  /* {
1603		struct vnode *a_fdvp;
1604		struct vnode *a_fvp;
1605		struct componentname *a_fcnp;
1606		struct vnode *a_tdvp;
1607		struct vnode *a_tvp;
1608		struct componentname *a_tcnp;
1609	} */ *ap;
1610{
1611	register struct vnode *fvp = ap->a_fvp;
1612	register struct vnode *tvp = ap->a_tvp;
1613	register struct vnode *fdvp = ap->a_fdvp;
1614	register struct vnode *tdvp = ap->a_tdvp;
1615	register struct componentname *tcnp = ap->a_tcnp;
1616	register struct componentname *fcnp = ap->a_fcnp;
1617	int error;
1618
1619#ifndef DIAGNOSTIC
1620	if ((tcnp->cn_flags & HASBUF) == 0 ||
1621	    (fcnp->cn_flags & HASBUF) == 0)
1622		panic("nfs_rename: no name");
1623#endif
1624	/* Check for cross-device rename */
1625	if ((fvp->v_mount != tdvp->v_mount) ||
1626	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1627		error = EXDEV;
1628		goto out;
1629	}
1630
1631	/*
1632	 * If the tvp exists and is in use, sillyrename it before doing the
1633	 * rename of the new file over it.
1634	 * XXX Can't sillyrename a directory.
1635	 */
1636	if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1637		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1638		vput(tvp);
1639		tvp = NULL;
1640	}
1641
1642	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1643		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1644		tcnp->cn_proc);
1645
1646	if (fvp->v_type == VDIR) {
1647		if (tvp != NULL && tvp->v_type == VDIR)
1648			cache_purge(tdvp);
1649		cache_purge(fdvp);
1650	}
1651
1652out:
1653	if (tdvp == tvp)
1654		vrele(tdvp);
1655	else
1656		vput(tdvp);
1657	if (tvp)
1658		vput(tvp);
1659	vrele(fdvp);
1660	vrele(fvp);
1661	/*
1662	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1663	 */
1664	if (error == ENOENT)
1665		error = 0;
1666	return (error);
1667}
1668
1669/*
1670 * nfs file rename rpc called from nfs_remove() above
1671 */
1672static int
1673nfs_renameit(sdvp, scnp, sp)
1674	struct vnode *sdvp;
1675	struct componentname *scnp;
1676	register struct sillyrename *sp;
1677{
1678	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1679		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1680}
1681
1682/*
1683 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1684 */
1685static int
1686nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1687	register struct vnode *fdvp;
1688	const char *fnameptr;
1689	int fnamelen;
1690	register struct vnode *tdvp;
1691	const char *tnameptr;
1692	int tnamelen;
1693	struct ucred *cred;
1694	struct proc *proc;
1695{
1696	register u_int32_t *tl;
1697	register caddr_t cp;
1698	register int32_t t1, t2;
1699	caddr_t bpos, dpos, cp2;
1700	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1701	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1702	int v3 = NFS_ISV3(fdvp);
1703
1704	nfsstats.rpccnt[NFSPROC_RENAME]++;
1705	nfsm_reqhead(fdvp, NFSPROC_RENAME,
1706		(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1707		nfsm_rndup(tnamelen));
1708	nfsm_fhtom(fdvp, v3);
1709	nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1710	nfsm_fhtom(tdvp, v3);
1711	nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1712	nfsm_request(fdvp, NFSPROC_RENAME, proc, cred);
1713	if (v3) {
1714		nfsm_wcc_data(fdvp, fwccflag);
1715		nfsm_wcc_data(tdvp, twccflag);
1716	}
1717	nfsm_reqdone;
1718	VTONFS(fdvp)->n_flag |= NMODIFIED;
1719	VTONFS(tdvp)->n_flag |= NMODIFIED;
1720	if (!fwccflag)
1721		VTONFS(fdvp)->n_attrstamp = 0;
1722	if (!twccflag)
1723		VTONFS(tdvp)->n_attrstamp = 0;
1724	return (error);
1725}
1726
1727/*
1728 * nfs hard link create call
1729 */
1730static int
1731nfs_link(ap)
1732	struct vop_link_args /* {
1733		struct vnode *a_tdvp;
1734		struct vnode *a_vp;
1735		struct componentname *a_cnp;
1736	} */ *ap;
1737{
1738	register struct vnode *vp = ap->a_vp;
1739	register struct vnode *tdvp = ap->a_tdvp;
1740	register struct componentname *cnp = ap->a_cnp;
1741	register u_int32_t *tl;
1742	register caddr_t cp;
1743	register int32_t t1, t2;
1744	caddr_t bpos, dpos, cp2;
1745	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1746	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1747	int v3;
1748
1749	if (vp->v_mount != tdvp->v_mount) {
1750		VOP_ABORTOP(tdvp, cnp);
1751		vput(tdvp);
1752		return (EXDEV);
1753	}
1754
1755	/*
1756	 * Push all writes to the server, so that the attribute cache
1757	 * doesn't get "out of sync" with the server.
1758	 * XXX There should be a better way!
1759	 */
1760	VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_proc);
1761
1762	v3 = NFS_ISV3(vp);
1763	nfsstats.rpccnt[NFSPROC_LINK]++;
1764	nfsm_reqhead(vp, NFSPROC_LINK,
1765		NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1766	nfsm_fhtom(vp, v3);
1767	nfsm_fhtom(tdvp, v3);
1768	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1769	nfsm_request(vp, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1770	if (v3) {
1771		nfsm_postop_attr(vp, attrflag);
1772		nfsm_wcc_data(tdvp, wccflag);
1773	}
1774	nfsm_reqdone;
1775	zfree(namei_zone, cnp->cn_pnbuf);
1776	VTONFS(tdvp)->n_flag |= NMODIFIED;
1777	if (!attrflag)
1778		VTONFS(vp)->n_attrstamp = 0;
1779	if (!wccflag)
1780		VTONFS(tdvp)->n_attrstamp = 0;
1781	/*
1782	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1783	 */
1784	if (error == EEXIST)
1785		error = 0;
1786	return (error);
1787}
1788
1789/*
1790 * nfs symbolic link create call
1791 */
1792static int
1793nfs_symlink(ap)
1794	struct vop_symlink_args /* {
1795		struct vnode *a_dvp;
1796		struct vnode **a_vpp;
1797		struct componentname *a_cnp;
1798		struct vattr *a_vap;
1799		char *a_target;
1800	} */ *ap;
1801{
1802	register struct vnode *dvp = ap->a_dvp;
1803	register struct vattr *vap = ap->a_vap;
1804	register struct componentname *cnp = ap->a_cnp;
1805	register struct nfsv2_sattr *sp;
1806	register struct nfsv3_sattr *sp3;
1807	register u_int32_t *tl;
1808	register caddr_t cp;
1809	register int32_t t1, t2;
1810	caddr_t bpos, dpos, cp2;
1811	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1812	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1813	struct vnode *newvp = (struct vnode *)0;
1814	int v3 = NFS_ISV3(dvp);
1815
1816	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1817	slen = strlen(ap->a_target);
1818	nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1819	    nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1820	nfsm_fhtom(dvp, v3);
1821	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1822	if (v3) {
1823		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1824		nfsm_v3sattr(sp3, vap);
1825	}
1826	nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1827	if (!v3) {
1828		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1829		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1830		sp->sa_uid = nfs_xdrneg1;
1831		sp->sa_gid = nfs_xdrneg1;
1832		sp->sa_size = nfs_xdrneg1;
1833		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1834		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1835	}
1836	nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
1837	if (v3) {
1838		if (!error)
1839			nfsm_mtofh(dvp, newvp, v3, gotvp);
1840		nfsm_wcc_data(dvp, wccflag);
1841	}
1842	nfsm_reqdone;
1843	if (newvp)
1844		vput(newvp);
1845	zfree(namei_zone, cnp->cn_pnbuf);
1846	VTONFS(dvp)->n_flag |= NMODIFIED;
1847	if (!wccflag)
1848		VTONFS(dvp)->n_attrstamp = 0;
1849	/*
1850	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1851	 */
1852	if (error == EEXIST)
1853		error = 0;
1854	return (error);
1855}
1856
1857/*
1858 * nfs make dir call
1859 */
1860static int
1861nfs_mkdir(ap)
1862	struct vop_mkdir_args /* {
1863		struct vnode *a_dvp;
1864		struct vnode **a_vpp;
1865		struct componentname *a_cnp;
1866		struct vattr *a_vap;
1867	} */ *ap;
1868{
1869	register struct vnode *dvp = ap->a_dvp;
1870	register struct vattr *vap = ap->a_vap;
1871	register struct componentname *cnp = ap->a_cnp;
1872	register struct nfsv2_sattr *sp;
1873	register struct nfsv3_sattr *sp3;
1874	register u_int32_t *tl;
1875	register caddr_t cp;
1876	register int32_t t1, t2;
1877	register int len;
1878	struct nfsnode *np = (struct nfsnode *)0;
1879	struct vnode *newvp = (struct vnode *)0;
1880	caddr_t bpos, dpos, cp2;
1881	int error = 0, wccflag = NFSV3_WCCRATTR;
1882	int gotvp = 0;
1883	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1884	struct vattr vattr;
1885	int v3 = NFS_ISV3(dvp);
1886
1887	if (error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_proc)) {
1888		VOP_ABORTOP(dvp, cnp);
1889		return (error);
1890	}
1891	len = cnp->cn_namelen;
1892	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1893	nfsm_reqhead(dvp, NFSPROC_MKDIR,
1894	  NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1895	nfsm_fhtom(dvp, v3);
1896	nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1897	if (v3) {
1898		nfsm_build(sp3, struct nfsv3_sattr *, NFSX_V3SRVSATTR);
1899		nfsm_v3sattr(sp3, vap);
1900	} else {
1901		nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1902		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1903		sp->sa_uid = nfs_xdrneg1;
1904		sp->sa_gid = nfs_xdrneg1;
1905		sp->sa_size = nfs_xdrneg1;
1906		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1907		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1908	}
1909	nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
1910	if (!error)
1911		nfsm_mtofh(dvp, newvp, v3, gotvp);
1912	if (v3)
1913		nfsm_wcc_data(dvp, wccflag);
1914	nfsm_reqdone;
1915	VTONFS(dvp)->n_flag |= NMODIFIED;
1916	if (!wccflag)
1917		VTONFS(dvp)->n_attrstamp = 0;
1918	/*
1919	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1920	 * if we can succeed in looking up the directory.
1921	 */
1922	if (error == EEXIST || (!error && !gotvp)) {
1923		if (newvp) {
1924			vrele(newvp);
1925			newvp = (struct vnode *)0;
1926		}
1927		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1928			cnp->cn_proc, &np);
1929		if (!error) {
1930			newvp = NFSTOV(np);
1931			if (newvp->v_type != VDIR)
1932				error = EEXIST;
1933		}
1934	}
1935	if (error) {
1936		if (newvp)
1937			vrele(newvp);
1938	} else
1939		*ap->a_vpp = newvp;
1940	zfree(namei_zone, cnp->cn_pnbuf);
1941	return (error);
1942}
1943
1944/*
1945 * nfs remove directory call
1946 */
1947static int
1948nfs_rmdir(ap)
1949	struct vop_rmdir_args /* {
1950		struct vnode *a_dvp;
1951		struct vnode *a_vp;
1952		struct componentname *a_cnp;
1953	} */ *ap;
1954{
1955	register struct vnode *vp = ap->a_vp;
1956	register struct vnode *dvp = ap->a_dvp;
1957	register struct componentname *cnp = ap->a_cnp;
1958	register u_int32_t *tl;
1959	register caddr_t cp;
1960	register int32_t t1, t2;
1961	caddr_t bpos, dpos, cp2;
1962	int error = 0, wccflag = NFSV3_WCCRATTR;
1963	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1964	int v3 = NFS_ISV3(dvp);
1965
1966	if (dvp == vp)
1967		return (EINVAL);
1968	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1969	nfsm_reqhead(dvp, NFSPROC_RMDIR,
1970		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1971	nfsm_fhtom(dvp, v3);
1972	nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1973	nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
1974	if (v3)
1975		nfsm_wcc_data(dvp, wccflag);
1976	nfsm_reqdone;
1977	zfree(namei_zone, cnp->cn_pnbuf);
1978	VTONFS(dvp)->n_flag |= NMODIFIED;
1979	if (!wccflag)
1980		VTONFS(dvp)->n_attrstamp = 0;
1981	cache_purge(dvp);
1982	cache_purge(vp);
1983	/*
1984	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1985	 */
1986	if (error == ENOENT)
1987		error = 0;
1988	return (error);
1989}
1990
1991/*
1992 * nfs readdir call
1993 */
1994static int
1995nfs_readdir(ap)
1996	struct vop_readdir_args /* {
1997		struct vnode *a_vp;
1998		struct uio *a_uio;
1999		struct ucred *a_cred;
2000	} */ *ap;
2001{
2002	register struct vnode *vp = ap->a_vp;
2003	register struct nfsnode *np = VTONFS(vp);
2004	register struct uio *uio = ap->a_uio;
2005	int tresid, error;
2006	struct vattr vattr;
2007
2008	if (vp->v_type != VDIR)
2009		return (EPERM);
2010	/*
2011	 * First, check for hit on the EOF offset cache
2012	 */
2013	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2014	    (np->n_flag & NMODIFIED) == 0) {
2015		if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
2016			if (NQNFS_CKCACHABLE(vp, ND_READ)) {
2017				nfsstats.direofcache_hits++;
2018				return (0);
2019			}
2020		} else if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_procp) == 0 &&
2021			np->n_mtime == vattr.va_mtime.tv_sec) {
2022			nfsstats.direofcache_hits++;
2023			return (0);
2024		}
2025	}
2026
2027	/*
2028	 * Call nfs_bioread() to do the real work.
2029	 */
2030	tresid = uio->uio_resid;
2031	error = nfs_bioread(vp, uio, 0, ap->a_cred, 0);
2032
2033	if (!error && uio->uio_resid == tresid)
2034		nfsstats.direofcache_misses++;
2035	return (error);
2036}
2037
2038/*
2039 * Readdir rpc call.
2040 * Called from below the buffer cache by nfs_doio().
2041 */
2042int
2043nfs_readdirrpc(vp, uiop, cred)
2044	struct vnode *vp;
2045	register struct uio *uiop;
2046	struct ucred *cred;
2047
2048{
2049	register int len, left;
2050	register struct dirent *dp;
2051	register u_int32_t *tl;
2052	register caddr_t cp;
2053	register int32_t t1, t2;
2054	register nfsuint64 *cookiep;
2055	caddr_t bpos, dpos, cp2;
2056	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2057	nfsuint64 cookie;
2058	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2059	struct nfsnode *dnp = VTONFS(vp);
2060	u_quad_t fileno;
2061	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2062	int attrflag;
2063	int v3 = NFS_ISV3(vp);
2064
2065#ifndef nolint
2066	dp = (struct dirent *)0;
2067#endif
2068#ifndef DIAGNOSTIC
2069	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (NFS_DIRBLKSIZ - 1)) ||
2070		(uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2071		panic("nfs readdirrpc bad uio");
2072#endif
2073
2074	/*
2075	 * If there is no cookie, assume directory was stale.
2076	 */
2077	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2078	if (cookiep)
2079		cookie = *cookiep;
2080	else
2081		return (NFSERR_BAD_COOKIE);
2082	/*
2083	 * Loop around doing readdir rpc's of size nm_readdirsize
2084	 * truncated to a multiple of DIRBLKSIZ.
2085	 * The stopping criteria is EOF or buffer full.
2086	 */
2087	while (more_dirs && bigenough) {
2088		nfsstats.rpccnt[NFSPROC_READDIR]++;
2089		nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2090			NFSX_READDIR(v3));
2091		nfsm_fhtom(vp, v3);
2092		if (v3) {
2093			nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2094			*tl++ = cookie.nfsuquad[0];
2095			*tl++ = cookie.nfsuquad[1];
2096			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2097			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2098		} else {
2099			nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2100			*tl++ = cookie.nfsuquad[0];
2101		}
2102		*tl = txdr_unsigned(nmp->nm_readdirsize);
2103		nfsm_request(vp, NFSPROC_READDIR, uiop->uio_procp, cred);
2104		if (v3) {
2105			nfsm_postop_attr(vp, attrflag);
2106			if (!error) {
2107				nfsm_dissect(tl, u_int32_t *,
2108				    2 * NFSX_UNSIGNED);
2109				dnp->n_cookieverf.nfsuquad[0] = *tl++;
2110				dnp->n_cookieverf.nfsuquad[1] = *tl;
2111			} else {
2112				m_freem(mrep);
2113				goto nfsmout;
2114			}
2115		}
2116		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2117		more_dirs = fxdr_unsigned(int, *tl);
2118
2119		/* loop thru the dir entries, doctoring them to 4bsd form */
2120		while (more_dirs && bigenough) {
2121			if (v3) {
2122				nfsm_dissect(tl, u_int32_t *,
2123				    3 * NFSX_UNSIGNED);
2124				fxdr_hyper(tl, &fileno);
2125				len = fxdr_unsigned(int, *(tl + 2));
2126			} else {
2127				nfsm_dissect(tl, u_int32_t *,
2128				    2 * NFSX_UNSIGNED);
2129				fileno = fxdr_unsigned(u_quad_t, *tl++);
2130				len = fxdr_unsigned(int, *tl);
2131			}
2132			if (len <= 0 || len > NFS_MAXNAMLEN) {
2133				error = EBADRPC;
2134				m_freem(mrep);
2135				goto nfsmout;
2136			}
2137			tlen = nfsm_rndup(len);
2138			if (tlen == len)
2139				tlen += 4;	/* To ensure null termination */
2140			left = DIRBLKSIZ - blksiz;
2141			if ((tlen + DIRHDSIZ) > left) {
2142				dp->d_reclen += left;
2143				uiop->uio_iov->iov_base += left;
2144				uiop->uio_iov->iov_len -= left;
2145				uiop->uio_offset += left;
2146				uiop->uio_resid -= left;
2147				blksiz = 0;
2148			}
2149			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2150				bigenough = 0;
2151			if (bigenough) {
2152				dp = (struct dirent *)uiop->uio_iov->iov_base;
2153				dp->d_fileno = (int)fileno;
2154				dp->d_namlen = len;
2155				dp->d_reclen = tlen + DIRHDSIZ;
2156				dp->d_type = DT_UNKNOWN;
2157				blksiz += dp->d_reclen;
2158				if (blksiz == DIRBLKSIZ)
2159					blksiz = 0;
2160				uiop->uio_offset += DIRHDSIZ;
2161				uiop->uio_resid -= DIRHDSIZ;
2162				uiop->uio_iov->iov_base += DIRHDSIZ;
2163				uiop->uio_iov->iov_len -= DIRHDSIZ;
2164				nfsm_mtouio(uiop, len);
2165				cp = uiop->uio_iov->iov_base;
2166				tlen -= len;
2167				*cp = '\0';	/* null terminate */
2168				uiop->uio_iov->iov_base += tlen;
2169				uiop->uio_iov->iov_len -= tlen;
2170				uiop->uio_offset += tlen;
2171				uiop->uio_resid -= tlen;
2172			} else
2173				nfsm_adv(nfsm_rndup(len));
2174			if (v3) {
2175				nfsm_dissect(tl, u_int32_t *,
2176				    3 * NFSX_UNSIGNED);
2177			} else {
2178				nfsm_dissect(tl, u_int32_t *,
2179				    2 * NFSX_UNSIGNED);
2180			}
2181			if (bigenough) {
2182				cookie.nfsuquad[0] = *tl++;
2183				if (v3)
2184					cookie.nfsuquad[1] = *tl++;
2185			} else if (v3)
2186				tl += 2;
2187			else
2188				tl++;
2189			more_dirs = fxdr_unsigned(int, *tl);
2190		}
2191		/*
2192		 * If at end of rpc data, get the eof boolean
2193		 */
2194		if (!more_dirs) {
2195			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2196			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2197		}
2198		m_freem(mrep);
2199	}
2200	/*
2201	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2202	 * by increasing d_reclen for the last record.
2203	 */
2204	if (blksiz > 0) {
2205		left = DIRBLKSIZ - blksiz;
2206		dp->d_reclen += left;
2207		uiop->uio_iov->iov_base += left;
2208		uiop->uio_iov->iov_len -= left;
2209		uiop->uio_offset += left;
2210		uiop->uio_resid -= left;
2211	}
2212
2213	/*
2214	 * We are now either at the end of the directory or have filled the
2215	 * block.
2216	 */
2217	if (bigenough)
2218		dnp->n_direofoffset = uiop->uio_offset;
2219	else {
2220		if (uiop->uio_resid > 0)
2221			printf("EEK! readdirrpc resid > 0\n");
2222		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2223		*cookiep = cookie;
2224	}
2225nfsmout:
2226	return (error);
2227}
2228
2229/*
2230 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2231 */
2232int
2233nfs_readdirplusrpc(vp, uiop, cred)
2234	struct vnode *vp;
2235	register struct uio *uiop;
2236	struct ucred *cred;
2237{
2238	register int len, left;
2239	register struct dirent *dp;
2240	register u_int32_t *tl;
2241	register caddr_t cp;
2242	register int32_t t1, t2;
2243	register struct vnode *newvp;
2244	register nfsuint64 *cookiep;
2245	caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2246	struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2247	struct nameidata nami, *ndp = &nami;
2248	struct componentname *cnp = &ndp->ni_cnd;
2249	nfsuint64 cookie;
2250	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2251	struct nfsnode *dnp = VTONFS(vp), *np;
2252	nfsfh_t *fhp;
2253	u_quad_t fileno;
2254	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2255	int attrflag, fhsize;
2256
2257#ifndef nolint
2258	dp = (struct dirent *)0;
2259#endif
2260#ifndef DIAGNOSTIC
2261	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2262		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2263		panic("nfs readdirplusrpc bad uio");
2264#endif
2265	ndp->ni_dvp = vp;
2266	newvp = NULLVP;
2267
2268	/*
2269	 * If there is no cookie, assume directory was stale.
2270	 */
2271	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2272	if (cookiep)
2273		cookie = *cookiep;
2274	else
2275		return (NFSERR_BAD_COOKIE);
2276	/*
2277	 * Loop around doing readdir rpc's of size nm_readdirsize
2278	 * truncated to a multiple of DIRBLKSIZ.
2279	 * The stopping criteria is EOF or buffer full.
2280	 */
2281	while (more_dirs && bigenough) {
2282		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2283		nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2284			NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2285		nfsm_fhtom(vp, 1);
2286 		nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2287		*tl++ = cookie.nfsuquad[0];
2288		*tl++ = cookie.nfsuquad[1];
2289		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2290		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2291		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2292		*tl = txdr_unsigned(nmp->nm_rsize);
2293		nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2294		nfsm_postop_attr(vp, attrflag);
2295		if (error) {
2296			m_freem(mrep);
2297			goto nfsmout;
2298		}
2299		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2300		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2301		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2302		more_dirs = fxdr_unsigned(int, *tl);
2303
2304		/* loop thru the dir entries, doctoring them to 4bsd form */
2305		while (more_dirs && bigenough) {
2306			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2307			fxdr_hyper(tl, &fileno);
2308			len = fxdr_unsigned(int, *(tl + 2));
2309			if (len <= 0 || len > NFS_MAXNAMLEN) {
2310				error = EBADRPC;
2311				m_freem(mrep);
2312				goto nfsmout;
2313			}
2314			tlen = nfsm_rndup(len);
2315			if (tlen == len)
2316				tlen += 4;	/* To ensure null termination*/
2317			left = DIRBLKSIZ - blksiz;
2318			if ((tlen + DIRHDSIZ) > left) {
2319				dp->d_reclen += left;
2320				uiop->uio_iov->iov_base += left;
2321				uiop->uio_iov->iov_len -= left;
2322				uiop->uio_offset += left;
2323				uiop->uio_resid -= left;
2324				blksiz = 0;
2325			}
2326			if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2327				bigenough = 0;
2328			if (bigenough) {
2329				dp = (struct dirent *)uiop->uio_iov->iov_base;
2330				dp->d_fileno = (int)fileno;
2331				dp->d_namlen = len;
2332				dp->d_reclen = tlen + DIRHDSIZ;
2333				dp->d_type = DT_UNKNOWN;
2334				blksiz += dp->d_reclen;
2335				if (blksiz == DIRBLKSIZ)
2336					blksiz = 0;
2337				uiop->uio_offset += DIRHDSIZ;
2338				uiop->uio_resid -= DIRHDSIZ;
2339				uiop->uio_iov->iov_base += DIRHDSIZ;
2340				uiop->uio_iov->iov_len -= DIRHDSIZ;
2341				cnp->cn_nameptr = uiop->uio_iov->iov_base;
2342				cnp->cn_namelen = len;
2343				nfsm_mtouio(uiop, len);
2344				cp = uiop->uio_iov->iov_base;
2345				tlen -= len;
2346				*cp = '\0';
2347				uiop->uio_iov->iov_base += tlen;
2348				uiop->uio_iov->iov_len -= tlen;
2349				uiop->uio_offset += tlen;
2350				uiop->uio_resid -= tlen;
2351			} else
2352				nfsm_adv(nfsm_rndup(len));
2353			nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2354			if (bigenough) {
2355				cookie.nfsuquad[0] = *tl++;
2356				cookie.nfsuquad[1] = *tl++;
2357			} else
2358				tl += 2;
2359
2360			/*
2361			 * Since the attributes are before the file handle
2362			 * (sigh), we must skip over the attributes and then
2363			 * come back and get them.
2364			 */
2365			attrflag = fxdr_unsigned(int, *tl);
2366			if (attrflag) {
2367			    dpossav1 = dpos;
2368			    mdsav1 = md;
2369			    nfsm_adv(NFSX_V3FATTR);
2370			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2371			    doit = fxdr_unsigned(int, *tl);
2372			    if (doit) {
2373				nfsm_getfh(fhp, fhsize, 1);
2374				if (NFS_CMPFH(dnp, fhp, fhsize)) {
2375				    VREF(vp);
2376				    newvp = vp;
2377				    np = dnp;
2378				} else {
2379				    error = nfs_nget(vp->v_mount, fhp,
2380					fhsize, &np);
2381				    if (error)
2382					doit = 0;
2383				    else
2384					newvp = NFSTOV(np);
2385				}
2386			    }
2387			    if (doit) {
2388				dpossav2 = dpos;
2389				dpos = dpossav1;
2390				mdsav2 = md;
2391				md = mdsav1;
2392				nfsm_loadattr(newvp, (struct vattr *)0);
2393				dpos = dpossav2;
2394				md = mdsav2;
2395				dp->d_type =
2396				    IFTODT(VTTOIF(np->n_vattr.va_type));
2397				ndp->ni_vp = newvp;
2398				cnp->cn_hash = 0;
2399				for (cp = cnp->cn_nameptr, i = 1; i <= len;
2400				    i++, cp++)
2401				    cnp->cn_hash += (unsigned char)*cp * i;
2402			        cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2403			    }
2404			} else {
2405			    /* Just skip over the file handle */
2406			    nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2407			    i = fxdr_unsigned(int, *tl);
2408			    nfsm_adv(nfsm_rndup(i));
2409			}
2410			if (newvp != NULLVP) {
2411			    vrele(newvp);
2412			    newvp = NULLVP;
2413			}
2414			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2415			more_dirs = fxdr_unsigned(int, *tl);
2416		}
2417		/*
2418		 * If at end of rpc data, get the eof boolean
2419		 */
2420		if (!more_dirs) {
2421			nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2422			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2423		}
2424		m_freem(mrep);
2425	}
2426	/*
2427	 * Fill last record, iff any, out to a multiple of NFS_DIRBLKSIZ
2428	 * by increasing d_reclen for the last record.
2429	 */
2430	if (blksiz > 0) {
2431		left = DIRBLKSIZ - blksiz;
2432		dp->d_reclen += left;
2433		uiop->uio_iov->iov_base += left;
2434		uiop->uio_iov->iov_len -= left;
2435		uiop->uio_offset += left;
2436		uiop->uio_resid -= left;
2437	}
2438
2439	/*
2440	 * We are now either at the end of the directory or have filled the
2441	 * block.
2442	 */
2443	if (bigenough)
2444		dnp->n_direofoffset = uiop->uio_offset;
2445	else {
2446		if (uiop->uio_resid > 0)
2447			printf("EEK! readdirplusrpc resid > 0\n");
2448		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2449		*cookiep = cookie;
2450	}
2451nfsmout:
2452	if (newvp != NULLVP) {
2453	        if (newvp == vp)
2454			vrele(newvp);
2455		else
2456			vput(newvp);
2457		newvp = NULLVP;
2458	}
2459	return (error);
2460}
2461
2462/*
2463 * Silly rename. To make the NFS filesystem that is stateless look a little
2464 * more like the "ufs" a remove of an active vnode is translated to a rename
2465 * to a funny looking filename that is removed by nfs_inactive on the
2466 * nfsnode. There is the potential for another process on a different client
2467 * to create the same funny name between the nfs_lookitup() fails and the
2468 * nfs_rename() completes, but...
2469 */
2470static int
2471nfs_sillyrename(dvp, vp, cnp)
2472	struct vnode *dvp, *vp;
2473	struct componentname *cnp;
2474{
2475	register struct sillyrename *sp;
2476	struct nfsnode *np;
2477	int error;
2478	short pid;
2479
2480	cache_purge(dvp);
2481	np = VTONFS(vp);
2482#ifndef DIAGNOSTIC
2483	if (vp->v_type == VDIR)
2484		panic("nfs: sillyrename dir");
2485#endif
2486	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2487		M_NFSREQ, M_WAITOK);
2488	sp->s_cred = crdup(cnp->cn_cred);
2489	sp->s_dvp = dvp;
2490	VREF(dvp);
2491
2492	/* Fudge together a funny name */
2493	pid = cnp->cn_proc->p_pid;
2494	sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
2495
2496	/* Try lookitups until we get one that isn't there */
2497	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2498		cnp->cn_proc, (struct nfsnode **)0) == 0) {
2499		sp->s_name[4]++;
2500		if (sp->s_name[4] > 'z') {
2501			error = EINVAL;
2502			goto bad;
2503		}
2504	}
2505	error = nfs_renameit(dvp, cnp, sp);
2506	if (error)
2507		goto bad;
2508	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2509		cnp->cn_proc, &np);
2510	np->n_sillyrename = sp;
2511	return (0);
2512bad:
2513	vrele(sp->s_dvp);
2514	crfree(sp->s_cred);
2515	free((caddr_t)sp, M_NFSREQ);
2516	return (error);
2517}
2518
2519/*
2520 * Look up a file name and optionally either update the file handle or
2521 * allocate an nfsnode, depending on the value of npp.
2522 * npp == NULL	--> just do the lookup
2523 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2524 *			handled too
2525 * *npp != NULL --> update the file handle in the vnode
2526 */
2527static int
2528nfs_lookitup(dvp, name, len, cred, procp, npp)
2529	register struct vnode *dvp;
2530	const char *name;
2531	int len;
2532	struct ucred *cred;
2533	struct proc *procp;
2534	struct nfsnode **npp;
2535{
2536	register u_int32_t *tl;
2537	register caddr_t cp;
2538	register int32_t t1, t2;
2539	struct vnode *newvp = (struct vnode *)0;
2540	struct nfsnode *np, *dnp = VTONFS(dvp);
2541	caddr_t bpos, dpos, cp2;
2542	int error = 0, fhlen, attrflag;
2543	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2544	nfsfh_t *nfhp;
2545	int v3 = NFS_ISV3(dvp);
2546
2547	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2548	nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2549		NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2550	nfsm_fhtom(dvp, v3);
2551	nfsm_strtom(name, len, NFS_MAXNAMLEN);
2552	nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred);
2553	if (npp && !error) {
2554		nfsm_getfh(nfhp, fhlen, v3);
2555		if (*npp) {
2556		    np = *npp;
2557		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2558			free((caddr_t)np->n_fhp, M_NFSBIGFH);
2559			np->n_fhp = &np->n_fh;
2560		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2561			np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2562		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2563		    np->n_fhsize = fhlen;
2564		    newvp = NFSTOV(np);
2565		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2566		    VREF(dvp);
2567		    newvp = dvp;
2568		} else {
2569		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2570		    if (error) {
2571			m_freem(mrep);
2572			return (error);
2573		    }
2574		    newvp = NFSTOV(np);
2575		}
2576		if (v3) {
2577			nfsm_postop_attr(newvp, attrflag);
2578			if (!attrflag && *npp == NULL) {
2579				m_freem(mrep);
2580				if (newvp == dvp)
2581					vrele(newvp);
2582				else
2583					vput(newvp);
2584				return (ENOENT);
2585			}
2586		} else
2587			nfsm_loadattr(newvp, (struct vattr *)0);
2588	}
2589	nfsm_reqdone;
2590	if (npp && *npp == NULL) {
2591		if (error) {
2592			if (newvp)
2593				if (newvp == dvp)
2594					vrele(newvp);
2595				else
2596					vput(newvp);
2597		} else
2598			*npp = np;
2599	}
2600	return (error);
2601}
2602
2603/*
2604 * Nfs Version 3 commit rpc
2605 */
2606static int
2607nfs_commit(vp, offset, cnt, cred, procp)
2608	register struct vnode *vp;
2609	u_quad_t offset;
2610	int cnt;
2611	struct ucred *cred;
2612	struct proc *procp;
2613{
2614	register caddr_t cp;
2615	register u_int32_t *tl;
2616	register int32_t t1, t2;
2617	register struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2618	caddr_t bpos, dpos, cp2;
2619	int error = 0, wccflag = NFSV3_WCCRATTR;
2620	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2621
2622	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2623		return (0);
2624	nfsstats.rpccnt[NFSPROC_COMMIT]++;
2625	nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2626	nfsm_fhtom(vp, 1);
2627	nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2628	txdr_hyper(&offset, tl);
2629	tl += 2;
2630	*tl = txdr_unsigned(cnt);
2631	nfsm_request(vp, NFSPROC_COMMIT, procp, cred);
2632	nfsm_wcc_data(vp, wccflag);
2633	if (!error) {
2634		nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2635		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2636			NFSX_V3WRITEVERF)) {
2637			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2638				NFSX_V3WRITEVERF);
2639			error = NFSERR_STALEWRITEVERF;
2640		}
2641	}
2642	nfsm_reqdone;
2643	return (error);
2644}
2645
2646/*
2647 * Kludge City..
2648 * - make nfs_bmap() essentially a no-op that does no translation
2649 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2650 *   (Maybe I could use the process's page mapping, but I was concerned that
2651 *    Kernel Write might not be enabled and also figured copyout() would do
2652 *    a lot more work than bcopy() and also it currently happens in the
2653 *    context of the swapper process (2).
2654 */
2655static int
2656nfs_bmap(ap)
2657	struct vop_bmap_args /* {
2658		struct vnode *a_vp;
2659		daddr_t  a_bn;
2660		struct vnode **a_vpp;
2661		daddr_t *a_bnp;
2662		int *a_runp;
2663		int *a_runb;
2664	} */ *ap;
2665{
2666	register struct vnode *vp = ap->a_vp;
2667
2668	if (ap->a_vpp != NULL)
2669		*ap->a_vpp = vp;
2670	if (ap->a_bnp != NULL)
2671		*ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2672	if (ap->a_runp != NULL)
2673		*ap->a_runp = 0;
2674	if (ap->a_runb != NULL)
2675		*ap->a_runb = 0;
2676	return (0);
2677}
2678
2679/*
2680 * Strategy routine.
2681 * For async requests when nfsiod(s) are running, queue the request by
2682 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2683 * request.
2684 */
2685static int
2686nfs_strategy(ap)
2687	struct vop_strategy_args *ap;
2688{
2689	register struct buf *bp = ap->a_bp;
2690	struct ucred *cr;
2691	struct proc *p;
2692	int error = 0;
2693
2694	if (bp->b_flags & B_PHYS)
2695		panic("nfs physio");
2696	if (bp->b_flags & B_ASYNC)
2697		p = (struct proc *)0;
2698	else
2699		p = curproc;	/* XXX */
2700	if (bp->b_flags & B_READ)
2701		cr = bp->b_rcred;
2702	else
2703		cr = bp->b_wcred;
2704	/*
2705	 * If the op is asynchronous and an i/o daemon is waiting
2706	 * queue the request, wake it up and wait for completion
2707	 * otherwise just do it ourselves.
2708	 */
2709	if ((bp->b_flags & B_ASYNC) == 0 ||
2710		nfs_asyncio(bp, NOCRED))
2711		error = nfs_doio(bp, cr, p);
2712	return (error);
2713}
2714
2715/*
2716 * Mmap a file
2717 *
2718 * NB Currently unsupported.
2719 */
2720/* ARGSUSED */
2721static int
2722nfs_mmap(ap)
2723	struct vop_mmap_args /* {
2724		struct vnode *a_vp;
2725		int  a_fflags;
2726		struct ucred *a_cred;
2727		struct proc *a_p;
2728	} */ *ap;
2729{
2730
2731	return (EINVAL);
2732}
2733
2734/*
2735 * fsync vnode op. Just call nfs_flush() with commit == 1.
2736 */
2737/* ARGSUSED */
2738static int
2739nfs_fsync(ap)
2740	struct vop_fsync_args /* {
2741		struct vnodeop_desc *a_desc;
2742		struct vnode * a_vp;
2743		struct ucred * a_cred;
2744		int  a_waitfor;
2745		struct proc * a_p;
2746	} */ *ap;
2747{
2748
2749	return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_p, 1));
2750}
2751
2752/*
2753 * Flush all the blocks associated with a vnode.
2754 * 	Walk through the buffer pool and push any dirty pages
2755 *	associated with the vnode.
2756 */
2757static int
2758nfs_flush(vp, cred, waitfor, p, commit)
2759	register struct vnode *vp;
2760	struct ucred *cred;
2761	int waitfor;
2762	struct proc *p;
2763	int commit;
2764{
2765	register struct nfsnode *np = VTONFS(vp);
2766	register struct buf *bp;
2767	register int i;
2768	struct buf *nbp;
2769	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2770	int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2771	int passone = 1;
2772	u_quad_t off, endoff, toff;
2773	struct ucred* wcred = NULL;
2774	struct buf **bvec = NULL;
2775#ifndef NFS_COMMITBVECSIZ
2776#define NFS_COMMITBVECSIZ	20
2777#endif
2778	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2779	int bvecsize = 0, bveccount;
2780
2781	if (nmp->nm_flag & NFSMNT_INT)
2782		slpflag = PCATCH;
2783	if (!commit)
2784		passone = 0;
2785	/*
2786	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2787	 * server, but nas not been committed to stable storage on the server
2788	 * yet. On the first pass, the byte range is worked out and the commit
2789	 * rpc is done. On the second pass, nfs_writebp() is called to do the
2790	 * job.
2791	 */
2792again:
2793	off = (u_quad_t)-1;
2794	endoff = 0;
2795	bvecpos = 0;
2796	if (NFS_ISV3(vp) && commit) {
2797		s = splbio();
2798		/*
2799		 * Count up how many buffers waiting for a commit.
2800		 */
2801		bveccount = 0;
2802		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2803			nbp = TAILQ_NEXT(bp, b_vnbufs);
2804			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2805			    == (B_DELWRI | B_NEEDCOMMIT))
2806				bveccount++;
2807		}
2808		/*
2809		 * Allocate space to remember the list of bufs to commit.  It is
2810		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2811		 * If we can't get memory (for whatever reason), we will end up
2812		 * committing the buffers one-by-one in the loop below.
2813		 */
2814		if (bveccount > NFS_COMMITBVECSIZ) {
2815			if (bvec != NULL && bvec != bvec_on_stack)
2816				free(bvec, M_TEMP);
2817			bvec = (struct buf **)
2818				malloc(bveccount * sizeof(struct buf *),
2819				       M_TEMP, M_NOWAIT);
2820			if (bvec == NULL) {
2821				bvec = bvec_on_stack;
2822				bvecsize = NFS_COMMITBVECSIZ;
2823			} else
2824				bvecsize = bveccount;
2825		} else {
2826			bvec = bvec_on_stack;
2827			bvecsize = NFS_COMMITBVECSIZ;
2828		}
2829		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2830			nbp = TAILQ_NEXT(bp, b_vnbufs);
2831			if (bvecpos >= bvecsize)
2832				break;
2833			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2834				!= (B_DELWRI | B_NEEDCOMMIT))
2835				continue;
2836			bremfree(bp);
2837			/*
2838			 * Work out if all buffers are using the same cred
2839			 * so we can deal with them all with one commit.
2840			 */
2841			if (wcred == NULL)
2842				wcred = bp->b_wcred;
2843			else if (wcred != bp->b_wcred)
2844				wcred = NOCRED;
2845			bp->b_flags |= (B_BUSY | B_WRITEINPROG);
2846			vfs_busy_pages(bp, 1);
2847			/*
2848			 * A list of these buffers is kept so that the
2849			 * second loop knows which buffers have actually
2850			 * been committed. This is necessary, since there
2851			 * may be a race between the commit rpc and new
2852			 * uncommitted writes on the file.
2853			 */
2854			bvec[bvecpos++] = bp;
2855			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2856				bp->b_dirtyoff;
2857			if (toff < off)
2858				off = toff;
2859			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2860			if (toff > endoff)
2861				endoff = toff;
2862		}
2863		splx(s);
2864	}
2865	if (bvecpos > 0) {
2866		/*
2867		 * Commit data on the server, as required.
2868		 * If all bufs are using the same wcred, then use that with
2869		 * one call for all of them, otherwise commit each one
2870		 * separately.
2871		 */
2872		if (wcred != NOCRED)
2873			retv = nfs_commit(vp, off, (int)(endoff - off),
2874					  wcred, p);
2875		else {
2876			retv = 0;
2877			for (i = 0; i < bvecpos; i++) {
2878				off_t off, size;
2879				bp = bvec[i];
2880				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2881					bp->b_dirtyoff;
2882				size = (u_quad_t)(bp->b_dirtyend
2883						  - bp->b_dirtyoff);
2884				retv = nfs_commit(vp, off, (int)size,
2885						  bp->b_wcred, p);
2886				if (retv) break;
2887			}
2888		}
2889
2890		if (retv == NFSERR_STALEWRITEVERF)
2891			nfs_clearcommit(vp->v_mount);
2892		/*
2893		 * Now, either mark the blocks I/O done or mark the
2894		 * blocks dirty, depending on whether the commit
2895		 * succeeded.
2896		 */
2897		for (i = 0; i < bvecpos; i++) {
2898			bp = bvec[i];
2899			bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG);
2900			if (retv) {
2901			    vfs_unbusy_pages(bp);
2902			    brelse(bp);
2903			} else {
2904			    s = splbio();	/* XXX check this positionning */
2905			    vp->v_numoutput++;
2906			    bp->b_flags |= B_ASYNC;
2907			    if (bp->b_flags & B_DELWRI) {
2908				--numdirtybuffers;
2909			    	if (needsbuffer) {
2910					vfs_bio_need_satisfy();
2911				}
2912			    }
2913			    bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
2914			    bp->b_dirtyoff = bp->b_dirtyend = 0;
2915			    reassignbuf(bp, vp);
2916			    splx(s);
2917			    biodone(bp);
2918			}
2919		}
2920	}
2921
2922	/*
2923	 * Start/do any write(s) that are required.
2924	 */
2925loop:
2926	s = splbio();
2927	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2928		nbp = TAILQ_NEXT(bp, b_vnbufs);
2929		if (bp->b_flags & B_BUSY) {
2930			if (waitfor != MNT_WAIT || passone)
2931				continue;
2932			bp->b_flags |= B_WANTED;
2933			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
2934				"nfsfsync", slptimeo);
2935			splx(s);
2936			if (error) {
2937			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) {
2938				error = EINTR;
2939				goto done;
2940			    }
2941			    if (slpflag == PCATCH) {
2942				slpflag = 0;
2943				slptimeo = 2 * hz;
2944			    }
2945			}
2946			goto loop;
2947		}
2948		if ((bp->b_flags & B_DELWRI) == 0)
2949			panic("nfs_fsync: not dirty");
2950		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT))
2951			continue;
2952		bremfree(bp);
2953		if (passone || !commit)
2954		    bp->b_flags |= (B_BUSY|B_ASYNC);
2955		else
2956		    bp->b_flags |= (B_BUSY|B_ASYNC|B_WRITEINPROG|B_NEEDCOMMIT);
2957		splx(s);
2958		VOP_BWRITE(bp);
2959		goto loop;
2960	}
2961	splx(s);
2962	if (passone) {
2963		passone = 0;
2964		goto again;
2965	}
2966	if (waitfor == MNT_WAIT) {
2967		while (vp->v_numoutput) {
2968			vp->v_flag |= VBWAIT;
2969			error = tsleep((caddr_t)&vp->v_numoutput,
2970				slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
2971			if (error) {
2972			    if (nfs_sigintr(nmp, (struct nfsreq *)0, p)) {
2973				error = EINTR;
2974				goto done;
2975			    }
2976			    if (slpflag == PCATCH) {
2977				slpflag = 0;
2978				slptimeo = 2 * hz;
2979			    }
2980			}
2981		}
2982		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
2983			goto loop;
2984		}
2985	}
2986	if (np->n_flag & NWRITEERR) {
2987		error = np->n_error;
2988		np->n_flag &= ~NWRITEERR;
2989	}
2990done:
2991	if (bvec != NULL && bvec != bvec_on_stack)
2992		free(bvec, M_TEMP);
2993	return (error);
2994}
2995
2996/*
2997 * NFS advisory byte-level locks.
2998 * Currently unsupported.
2999 */
3000static int
3001nfs_advlock(ap)
3002	struct vop_advlock_args /* {
3003		struct vnode *a_vp;
3004		caddr_t  a_id;
3005		int  a_op;
3006		struct flock *a_fl;
3007		int  a_flags;
3008	} */ *ap;
3009{
3010	register struct nfsnode *np = VTONFS(ap->a_vp);
3011
3012	/*
3013	 * The following kludge is to allow diskless support to work
3014	 * until a real NFS lockd is implemented. Basically, just pretend
3015	 * that this is a local lock.
3016	 */
3017	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3018}
3019
3020/*
3021 * Print out the contents of an nfsnode.
3022 */
3023static int
3024nfs_print(ap)
3025	struct vop_print_args /* {
3026		struct vnode *a_vp;
3027	} */ *ap;
3028{
3029	register struct vnode *vp = ap->a_vp;
3030	register struct nfsnode *np = VTONFS(vp);
3031
3032	printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3033		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3034	if (vp->v_type == VFIFO)
3035		fifo_printinfo(vp);
3036	printf("\n");
3037	return (0);
3038}
3039
3040/*
3041 * Just call nfs_writebp() with the force argument set to 1.
3042 */
3043static int
3044nfs_bwrite(ap)
3045	struct vop_bwrite_args /* {
3046		struct vnode *a_bp;
3047	} */ *ap;
3048{
3049
3050	return (nfs_writebp(ap->a_bp, 1));
3051}
3052
3053/*
3054 * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3055 * the force flag is one and it also handles the B_NEEDCOMMIT flag.
3056 */
3057int
3058nfs_writebp(bp, force)
3059	register struct buf *bp;
3060	int force;
3061{
3062	int s;
3063	register int oldflags = bp->b_flags, retv = 1;
3064	off_t off;
3065
3066	if(!(bp->b_flags & B_BUSY))
3067		panic("bwrite: buffer is not busy???");
3068
3069	if (bp->b_flags & B_INVAL)
3070		bp->b_flags |= B_INVAL | B_NOCACHE;
3071
3072	if (bp->b_flags & B_DELWRI) {
3073		--numdirtybuffers;
3074		if (needsbuffer)
3075			vfs_bio_need_satisfy();
3076	}
3077	s = splbio(); /* XXX check if needed */
3078	bp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI);
3079
3080	if ((oldflags & (B_ASYNC|B_DELWRI)) == (B_ASYNC|B_DELWRI)) {
3081		reassignbuf(bp, bp->b_vp);
3082	}
3083
3084	bp->b_vp->v_numoutput++;
3085	curproc->p_stats->p_ru.ru_oublock++;
3086	splx(s);
3087
3088	/*
3089	 * If B_NEEDCOMMIT is set, a commit rpc may do the trick. If not
3090	 * an actual write will have to be scheduled via. VOP_STRATEGY().
3091	 * If B_WRITEINPROG is already set, then push it with a write anyhow.
3092	 */
3093	vfs_busy_pages(bp, 1);
3094	if ((oldflags & (B_NEEDCOMMIT | B_WRITEINPROG)) == B_NEEDCOMMIT) {
3095		off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
3096		bp->b_flags |= B_WRITEINPROG;
3097		retv = nfs_commit(bp->b_vp, off, bp->b_dirtyend-bp->b_dirtyoff,
3098			bp->b_wcred, bp->b_proc);
3099		bp->b_flags &= ~B_WRITEINPROG;
3100		if (!retv) {
3101			bp->b_dirtyoff = bp->b_dirtyend = 0;
3102			bp->b_flags &= ~B_NEEDCOMMIT;
3103			biodone(bp);
3104		} else if (retv == NFSERR_STALEWRITEVERF)
3105			nfs_clearcommit(bp->b_vp->v_mount);
3106	}
3107	if (retv) {
3108		if (force)
3109			bp->b_flags |= B_WRITEINPROG;
3110		VOP_STRATEGY(bp->b_vp, bp);
3111	}
3112
3113	if( (oldflags & B_ASYNC) == 0) {
3114		int rtval = biowait(bp);
3115
3116		if (oldflags & B_DELWRI) {
3117			s = splbio();
3118			reassignbuf(bp, bp->b_vp);
3119			splx(s);
3120		}
3121
3122		brelse(bp);
3123		return (rtval);
3124	}
3125
3126	return (0);
3127}
3128
3129/*
3130 * nfs special file access vnode op.
3131 * Essentially just get vattr and then imitate iaccess() since the device is
3132 * local to the client.
3133 */
3134static int
3135nfsspec_access(ap)
3136	struct vop_access_args /* {
3137		struct vnode *a_vp;
3138		int  a_mode;
3139		struct ucred *a_cred;
3140		struct proc *a_p;
3141	} */ *ap;
3142{
3143	register struct vattr *vap;
3144	register gid_t *gp;
3145	register struct ucred *cred = ap->a_cred;
3146	struct vnode *vp = ap->a_vp;
3147	mode_t mode = ap->a_mode;
3148	struct vattr vattr;
3149	register int i;
3150	int error;
3151
3152	/*
3153	 * Disallow write attempts on filesystems mounted read-only;
3154	 * unless the file is a socket, fifo, or a block or character
3155	 * device resident on the filesystem.
3156	 */
3157	if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3158		switch (vp->v_type) {
3159		case VREG:
3160		case VDIR:
3161		case VLNK:
3162			return (EROFS);
3163		default:
3164			break;
3165		}
3166	}
3167	/*
3168	 * If you're the super-user,
3169	 * you always get access.
3170	 */
3171	if (cred->cr_uid == 0)
3172		return (0);
3173	vap = &vattr;
3174	error = VOP_GETATTR(vp, vap, cred, ap->a_p);
3175	if (error)
3176		return (error);
3177	/*
3178	 * Access check is based on only one of owner, group, public.
3179	 * If not owner, then check group. If not a member of the
3180	 * group, then check public access.
3181	 */
3182	if (cred->cr_uid != vap->va_uid) {
3183		mode >>= 3;
3184		gp = cred->cr_groups;
3185		for (i = 0; i < cred->cr_ngroups; i++, gp++)
3186			if (vap->va_gid == *gp)
3187				goto found;
3188		mode >>= 3;
3189found:
3190		;
3191	}
3192	error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3193	return (error);
3194}
3195
3196/*
3197 * Read wrapper for special devices.
3198 */
3199static int
3200nfsspec_read(ap)
3201	struct vop_read_args /* {
3202		struct vnode *a_vp;
3203		struct uio *a_uio;
3204		int  a_ioflag;
3205		struct ucred *a_cred;
3206	} */ *ap;
3207{
3208	register struct nfsnode *np = VTONFS(ap->a_vp);
3209
3210	/*
3211	 * Set access flag.
3212	 */
3213	np->n_flag |= NACC;
3214	getnanotime(&np->n_atim);
3215	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3216}
3217
3218/*
3219 * Write wrapper for special devices.
3220 */
3221static int
3222nfsspec_write(ap)
3223	struct vop_write_args /* {
3224		struct vnode *a_vp;
3225		struct uio *a_uio;
3226		int  a_ioflag;
3227		struct ucred *a_cred;
3228	} */ *ap;
3229{
3230	register struct nfsnode *np = VTONFS(ap->a_vp);
3231
3232	/*
3233	 * Set update flag.
3234	 */
3235	np->n_flag |= NUPD;
3236	getnanotime(&np->n_mtim);
3237	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3238}
3239
3240/*
3241 * Close wrapper for special devices.
3242 *
3243 * Update the times on the nfsnode then do device close.
3244 */
3245static int
3246nfsspec_close(ap)
3247	struct vop_close_args /* {
3248		struct vnode *a_vp;
3249		int  a_fflag;
3250		struct ucred *a_cred;
3251		struct proc *a_p;
3252	} */ *ap;
3253{
3254	register struct vnode *vp = ap->a_vp;
3255	register struct nfsnode *np = VTONFS(vp);
3256	struct vattr vattr;
3257
3258	if (np->n_flag & (NACC | NUPD)) {
3259		np->n_flag |= NCHG;
3260		if (vp->v_usecount == 1 &&
3261		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3262			VATTR_NULL(&vattr);
3263			if (np->n_flag & NACC)
3264				vattr.va_atime = np->n_atim;
3265			if (np->n_flag & NUPD)
3266				vattr.va_mtime = np->n_mtim;
3267			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3268		}
3269	}
3270	return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3271}
3272
3273/*
3274 * Read wrapper for fifos.
3275 */
3276static int
3277nfsfifo_read(ap)
3278	struct vop_read_args /* {
3279		struct vnode *a_vp;
3280		struct uio *a_uio;
3281		int  a_ioflag;
3282		struct ucred *a_cred;
3283	} */ *ap;
3284{
3285	register struct nfsnode *np = VTONFS(ap->a_vp);
3286
3287	/*
3288	 * Set access flag.
3289	 */
3290	np->n_flag |= NACC;
3291	getnanotime(&np->n_atim);
3292	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3293}
3294
3295/*
3296 * Write wrapper for fifos.
3297 */
3298static int
3299nfsfifo_write(ap)
3300	struct vop_write_args /* {
3301		struct vnode *a_vp;
3302		struct uio *a_uio;
3303		int  a_ioflag;
3304		struct ucred *a_cred;
3305	} */ *ap;
3306{
3307	register struct nfsnode *np = VTONFS(ap->a_vp);
3308
3309	/*
3310	 * Set update flag.
3311	 */
3312	np->n_flag |= NUPD;
3313	getnanotime(&np->n_mtim);
3314	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3315}
3316
3317/*
3318 * Close wrapper for fifos.
3319 *
3320 * Update the times on the nfsnode then do fifo close.
3321 */
3322static int
3323nfsfifo_close(ap)
3324	struct vop_close_args /* {
3325		struct vnode *a_vp;
3326		int  a_fflag;
3327		struct ucred *a_cred;
3328		struct proc *a_p;
3329	} */ *ap;
3330{
3331	register struct vnode *vp = ap->a_vp;
3332	register struct nfsnode *np = VTONFS(vp);
3333	struct vattr vattr;
3334	struct timespec ts;
3335
3336	if (np->n_flag & (NACC | NUPD)) {
3337		getnanotime(&ts);
3338		if (np->n_flag & NACC)
3339			np->n_atim = ts;
3340		if (np->n_flag & NUPD)
3341			np->n_mtim = ts;
3342		np->n_flag |= NCHG;
3343		if (vp->v_usecount == 1 &&
3344		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3345			VATTR_NULL(&vattr);
3346			if (np->n_flag & NACC)
3347				vattr.va_atime = np->n_atim;
3348			if (np->n_flag & NUPD)
3349				vattr.va_mtime = np->n_mtim;
3350			(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3351		}
3352	}
3353	return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3354}
3355