xref: /dragonfly/sys/vfs/nfs/nfs_vnops.c (revision e7d467f4)
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  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon 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/nlookup.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/conf.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_extern.h>
69 
70 #include <sys/buf2.h>
71 
72 #include <vfs/fifofs/fifo.h>
73 #include <vfs/ufs/dir.h>
74 
75 #undef DIRBLKSIZ
76 
77 #include "rpcv2.h"
78 #include "nfsproto.h"
79 #include "nfs.h"
80 #include "nfsmount.h"
81 #include "nfsnode.h"
82 #include "xdr_subs.h"
83 #include "nfsm_subs.h"
84 
85 #include <net/if.h>
86 #include <netinet/in.h>
87 #include <netinet/in_var.h>
88 
89 #include <sys/thread2.h>
90 
91 /* Defs */
92 #define	TRUE	1
93 #define	FALSE	0
94 
95 static int	nfsfifo_read (struct vop_read_args *);
96 static int	nfsfifo_write (struct vop_write_args *);
97 static int	nfsfifo_close (struct vop_close_args *);
98 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
99 static	int	nfs_lookup (struct vop_old_lookup_args *);
100 static	int	nfs_create (struct vop_old_create_args *);
101 static	int	nfs_mknod (struct vop_old_mknod_args *);
102 static	int	nfs_open (struct vop_open_args *);
103 static	int	nfs_close (struct vop_close_args *);
104 static	int	nfs_access (struct vop_access_args *);
105 static	int	nfs_getattr (struct vop_getattr_args *);
106 static	int	nfs_setattr (struct vop_setattr_args *);
107 static	int	nfs_read (struct vop_read_args *);
108 static	int	nfs_mmap (struct vop_mmap_args *);
109 static	int	nfs_fsync (struct vop_fsync_args *);
110 static	int	nfs_remove (struct vop_old_remove_args *);
111 static	int	nfs_link (struct vop_old_link_args *);
112 static	int	nfs_rename (struct vop_old_rename_args *);
113 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
114 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
115 static	int	nfs_symlink (struct vop_old_symlink_args *);
116 static	int	nfs_readdir (struct vop_readdir_args *);
117 static	int	nfs_bmap (struct vop_bmap_args *);
118 static	int	nfs_strategy (struct vop_strategy_args *);
119 static	int	nfs_lookitup (struct vnode *, const char *, int,
120 			struct ucred *, struct thread *, struct nfsnode **);
121 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
122 static int	nfs_laccess (struct vop_access_args *);
123 static int	nfs_readlink (struct vop_readlink_args *);
124 static int	nfs_print (struct vop_print_args *);
125 static int	nfs_advlock (struct vop_advlock_args *);
126 
127 static	int	nfs_nresolve (struct vop_nresolve_args *);
128 /*
129  * Global vfs data structures for nfs
130  */
131 struct vop_ops nfsv2_vnode_vops = {
132 	.vop_default =		vop_defaultop,
133 	.vop_access =		nfs_access,
134 	.vop_advlock =		nfs_advlock,
135 	.vop_bmap =		nfs_bmap,
136 	.vop_close =		nfs_close,
137 	.vop_old_create =	nfs_create,
138 	.vop_fsync =		nfs_fsync,
139 	.vop_getattr =		nfs_getattr,
140 	.vop_getpages =		vop_stdgetpages,
141 	.vop_putpages =		vop_stdputpages,
142 	.vop_inactive =		nfs_inactive,
143 	.vop_old_link =		nfs_link,
144 	.vop_old_lookup =	nfs_lookup,
145 	.vop_old_mkdir =	nfs_mkdir,
146 	.vop_old_mknod =	nfs_mknod,
147 	.vop_mmap =		nfs_mmap,
148 	.vop_open =		nfs_open,
149 	.vop_print =		nfs_print,
150 	.vop_read =		nfs_read,
151 	.vop_readdir =		nfs_readdir,
152 	.vop_readlink =		nfs_readlink,
153 	.vop_reclaim =		nfs_reclaim,
154 	.vop_old_remove =	nfs_remove,
155 	.vop_old_rename =	nfs_rename,
156 	.vop_old_rmdir =	nfs_rmdir,
157 	.vop_setattr =		nfs_setattr,
158 	.vop_strategy =		nfs_strategy,
159 	.vop_old_symlink =	nfs_symlink,
160 	.vop_write =		nfs_write,
161 	.vop_nresolve =		nfs_nresolve
162 };
163 
164 /*
165  * Special device vnode ops
166  */
167 struct vop_ops nfsv2_spec_vops = {
168 	.vop_default =		vop_defaultop,
169 	.vop_access =		nfs_laccess,
170 	.vop_close =		nfs_close,
171 	.vop_fsync =		nfs_fsync,
172 	.vop_getattr =		nfs_getattr,
173 	.vop_inactive =		nfs_inactive,
174 	.vop_print =		nfs_print,
175 	.vop_read =		vop_stdnoread,
176 	.vop_reclaim =		nfs_reclaim,
177 	.vop_setattr =		nfs_setattr,
178 	.vop_write =		vop_stdnowrite
179 };
180 
181 struct vop_ops nfsv2_fifo_vops = {
182 	.vop_default =		fifo_vnoperate,
183 	.vop_access =		nfs_laccess,
184 	.vop_close =		nfsfifo_close,
185 	.vop_fsync =		nfs_fsync,
186 	.vop_getattr =		nfs_getattr,
187 	.vop_inactive =		nfs_inactive,
188 	.vop_print =		nfs_print,
189 	.vop_read =		nfsfifo_read,
190 	.vop_reclaim =		nfs_reclaim,
191 	.vop_setattr =		nfs_setattr,
192 	.vop_write =		nfsfifo_write
193 };
194 
195 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
196 				  struct componentname *cnp,
197 				  struct vattr *vap);
198 static int	nfs_removerpc (struct vnode *dvp, const char *name,
199 				   int namelen,
200 				   struct ucred *cred, struct thread *td);
201 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
202 				   int fnamelen, struct vnode *tdvp,
203 				   const char *tnameptr, int tnamelen,
204 				   struct ucred *cred, struct thread *td);
205 static int	nfs_renameit (struct vnode *sdvp,
206 				  struct componentname *scnp,
207 				  struct sillyrename *sp);
208 
209 SYSCTL_DECL(_vfs_nfs);
210 
211 static int nfs_flush_on_rename = 1;
212 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
213 	   &nfs_flush_on_rename, 0, "flush fvp prior to rename");
214 static int nfs_flush_on_hlink = 0;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
216 	   &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
217 
218 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
220 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
221 
222 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
224 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
225 
226 static int	nfspos_cache_timeout = NFS_MINATTRTIMO;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
228 	   &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
229 
230 static int	nfsv3_commit_on_close = 0;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
232 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
233 #if 0
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
235 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
236 
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
238 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
239 #endif
240 
241 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
242 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
243 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
244 
245 /*
246  * Returns whether a name component is a degenerate '.' or '..'.
247  */
248 static __inline
249 int
250 nlcdegenerate(struct nlcomponent *nlc)
251 {
252 	if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
253 		return(1);
254 	if (nlc->nlc_namelen == 2 &&
255 	    nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
256 		return(1);
257 	return(0);
258 }
259 
260 static int
261 nfs3_access_otw(struct vnode *vp, int wmode,
262 		struct thread *td, struct ucred *cred)
263 {
264 	struct nfsnode *np = VTONFS(vp);
265 	int attrflag;
266 	int error = 0;
267 	u_int32_t *tl;
268 	u_int32_t rmode;
269 	struct nfsm_info info;
270 
271 	info.mrep = NULL;
272 	info.v3 = 1;
273 
274 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
275 	nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
276 		     NFSX_FH(info.v3) + NFSX_UNSIGNED);
277 	ERROROUT(nfsm_fhtom(&info, vp));
278 	tl = nfsm_build(&info, NFSX_UNSIGNED);
279 	*tl = txdr_unsigned(wmode);
280 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
281 	ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
282 	if (error == 0) {
283 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
284 		rmode = fxdr_unsigned(u_int32_t, *tl);
285 		np->n_mode = rmode;
286 		np->n_modeuid = cred->cr_uid;
287 		np->n_modestamp = mycpu->gd_time_seconds;
288 	}
289 	m_freem(info.mrep);
290 	info.mrep = NULL;
291 nfsmout:
292 	return error;
293 }
294 
295 /*
296  * nfs access vnode op.
297  * For nfs version 2, just return ok. File accesses may fail later.
298  * For nfs version 3, use the access rpc to check accessibility. If file modes
299  * are changed on the server, accesses might still fail later.
300  *
301  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
302  */
303 static int
304 nfs_access(struct vop_access_args *ap)
305 {
306 	struct ucred *cred;
307 	struct vnode *vp = ap->a_vp;
308 	thread_t td = curthread;
309 	int error = 0;
310 	u_int32_t mode, wmode;
311 	struct nfsnode *np = VTONFS(vp);
312 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
313 	int v3 = NFS_ISV3(vp);
314 
315 	lwkt_gettoken(&nmp->nm_token);
316 
317 	/*
318 	 * Disallow write attempts on filesystems mounted read-only;
319 	 * unless the file is a socket, fifo, or a block or character
320 	 * device resident on the filesystem.
321 	 */
322 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
323 		switch (vp->v_type) {
324 		case VREG:
325 		case VDIR:
326 		case VLNK:
327 			lwkt_reltoken(&nmp->nm_token);
328 			return (EROFS);
329 		default:
330 			break;
331 		}
332 	}
333 
334 	/*
335 	 * The NFS protocol passes only the effective uid/gid over the wire but
336 	 * we need to check access against real ids if AT_EACCESS not set.
337 	 * Handle this case by cloning the credentials and setting the
338 	 * effective ids to the real ones.
339 	 */
340 	if (ap->a_flags & AT_EACCESS) {
341 		cred = crhold(ap->a_cred);
342 	} else {
343 		cred = crdup(ap->a_cred);
344 		cred->cr_uid = cred->cr_ruid;
345 		cred->cr_gid = cred->cr_rgid;
346 	}
347 
348 	/*
349 	 * For nfs v3, check to see if we have done this recently, and if
350 	 * so return our cached result instead of making an ACCESS call.
351 	 * If not, do an access rpc, otherwise you are stuck emulating
352 	 * ufs_access() locally using the vattr. This may not be correct,
353 	 * since the server may apply other access criteria such as
354 	 * client uid-->server uid mapping that we do not know about.
355 	 */
356 	if (v3) {
357 		if (ap->a_mode & VREAD)
358 			mode = NFSV3ACCESS_READ;
359 		else
360 			mode = 0;
361 		if (vp->v_type != VDIR) {
362 			if (ap->a_mode & VWRITE)
363 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
364 			if (ap->a_mode & VEXEC)
365 				mode |= NFSV3ACCESS_EXECUTE;
366 		} else {
367 			if (ap->a_mode & VWRITE)
368 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
369 					 NFSV3ACCESS_DELETE);
370 			if (ap->a_mode & VEXEC)
371 				mode |= NFSV3ACCESS_LOOKUP;
372 		}
373 		/* XXX safety belt, only make blanket request if caching */
374 		if (nfsaccess_cache_timeout > 0) {
375 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
376 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
377 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
378 		} else {
379 			wmode = mode;
380 		}
381 
382 		/*
383 		 * Does our cached result allow us to give a definite yes to
384 		 * this request?
385 		 */
386 		if (np->n_modestamp &&
387 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
388 		   (cred->cr_uid == np->n_modeuid) &&
389 		   ((np->n_mode & mode) == mode)) {
390 			nfsstats.accesscache_hits++;
391 		} else {
392 			/*
393 			 * Either a no, or a don't know.  Go to the wire.
394 			 */
395 			nfsstats.accesscache_misses++;
396 		        error = nfs3_access_otw(vp, wmode, td, cred);
397 			if (!error) {
398 				if ((np->n_mode & mode) != mode) {
399 					error = EACCES;
400 				}
401 			}
402 		}
403 	} else {
404 		if ((error = nfs_laccess(ap)) != 0) {
405 			crfree(cred);
406 			lwkt_reltoken(&nmp->nm_token);
407 			return (error);
408 		}
409 
410 		/*
411 		 * Attempt to prevent a mapped root from accessing a file
412 		 * which it shouldn't.  We try to read a byte from the file
413 		 * if the user is root and the file is not zero length.
414 		 * After calling nfs_laccess, we should have the correct
415 		 * file size cached.
416 		 */
417 		if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
418 		    && VTONFS(vp)->n_size > 0) {
419 			struct iovec aiov;
420 			struct uio auio;
421 			char buf[1];
422 
423 			aiov.iov_base = buf;
424 			aiov.iov_len = 1;
425 			auio.uio_iov = &aiov;
426 			auio.uio_iovcnt = 1;
427 			auio.uio_offset = 0;
428 			auio.uio_resid = 1;
429 			auio.uio_segflg = UIO_SYSSPACE;
430 			auio.uio_rw = UIO_READ;
431 			auio.uio_td = td;
432 
433 			if (vp->v_type == VREG) {
434 				error = nfs_readrpc_uio(vp, &auio);
435 			} else if (vp->v_type == VDIR) {
436 				char* bp;
437 				bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
438 				aiov.iov_base = bp;
439 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
440 				error = nfs_readdirrpc_uio(vp, &auio);
441 				kfree(bp, M_TEMP);
442 			} else if (vp->v_type == VLNK) {
443 				error = nfs_readlinkrpc_uio(vp, &auio);
444 			} else {
445 				error = EACCES;
446 			}
447 		}
448 	}
449 	/*
450 	 * [re]record creds for reading and/or writing if access
451 	 * was granted.  Assume the NFS server will grant read access
452 	 * for execute requests.
453 	 */
454 	if (error == 0) {
455 		if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
456 			crhold(cred);
457 			if (np->n_rucred)
458 				crfree(np->n_rucred);
459 			np->n_rucred = cred;
460 		}
461 		if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
462 			crhold(cred);
463 			if (np->n_wucred)
464 				crfree(np->n_wucred);
465 			np->n_wucred = cred;
466 		}
467 	}
468 	lwkt_reltoken(&nmp->nm_token);
469 	crfree(cred);
470 	return(error);
471 }
472 
473 /*
474  * nfs open vnode op
475  * Check to see if the type is ok
476  * and that deletion is not in progress.
477  * For paged in text files, you will need to flush the page cache
478  * if consistency is lost.
479  *
480  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
481  *	    struct file *a_fp)
482  */
483 /* ARGSUSED */
484 static int
485 nfs_open(struct vop_open_args *ap)
486 {
487 	struct vnode *vp = ap->a_vp;
488 	struct nfsnode *np = VTONFS(vp);
489 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
490 	struct vattr vattr;
491 	int error;
492 
493 	lwkt_gettoken(&nmp->nm_token);
494 
495 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
496 #ifdef DIAGNOSTIC
497 		kprintf("open eacces vtyp=%d\n",vp->v_type);
498 #endif
499 		lwkt_reltoken(&nmp->nm_token);
500 		return (EOPNOTSUPP);
501 	}
502 
503 	/*
504 	 * Save valid creds for reading and writing for later RPCs.
505 	 */
506 	if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
507 		crhold(ap->a_cred);
508 		if (np->n_rucred)
509 			crfree(np->n_rucred);
510 		np->n_rucred = ap->a_cred;
511 	}
512 	if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
513 		crhold(ap->a_cred);
514 		if (np->n_wucred)
515 			crfree(np->n_wucred);
516 		np->n_wucred = ap->a_cred;
517 	}
518 
519 	/*
520 	 * Clear the attribute cache only if opening with write access.  It
521 	 * is unclear if we should do this at all here, but we certainly
522 	 * should not clear the cache unconditionally simply because a file
523 	 * is being opened.
524 	 */
525 	if (ap->a_mode & FWRITE)
526 		np->n_attrstamp = 0;
527 
528 	/*
529 	 * For normal NFS, reconcile changes made locally verses
530 	 * changes made remotely.  Note that VOP_GETATTR only goes
531 	 * to the wire if the cached attribute has timed out or been
532 	 * cleared.
533 	 *
534 	 * If local modifications have been made clear the attribute
535 	 * cache to force an attribute and modified time check.  If
536 	 * GETATTR detects that the file has been changed by someone
537 	 * other then us it will set NRMODIFIED.
538 	 *
539 	 * If we are opening a directory and local changes have been
540 	 * made we have to invalidate the cache in order to ensure
541 	 * that we get the most up-to-date information from the
542 	 * server.  XXX
543 	 */
544 	if (np->n_flag & NLMODIFIED) {
545 		np->n_attrstamp = 0;
546 		if (vp->v_type == VDIR) {
547 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
548 			if (error == EINTR)
549 				return (error);
550 			nfs_invaldir(vp);
551 		}
552 	}
553 	error = VOP_GETATTR(vp, &vattr);
554 	if (error) {
555 		lwkt_reltoken(&nmp->nm_token);
556 		return (error);
557 	}
558 	if (np->n_flag & NRMODIFIED) {
559 		if (vp->v_type == VDIR)
560 			nfs_invaldir(vp);
561 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
562 		if (error == EINTR) {
563 			lwkt_reltoken(&nmp->nm_token);
564 			return (error);
565 		}
566 		np->n_flag &= ~NRMODIFIED;
567 	}
568 	error = vop_stdopen(ap);
569 	lwkt_reltoken(&nmp->nm_token);
570 
571 	return error;
572 }
573 
574 /*
575  * nfs close vnode op
576  * What an NFS client should do upon close after writing is a debatable issue.
577  * Most NFS clients push delayed writes to the server upon close, basically for
578  * two reasons:
579  * 1 - So that any write errors may be reported back to the client process
580  *     doing the close system call. By far the two most likely errors are
581  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582  * 2 - To put a worst case upper bound on cache inconsistency between
583  *     multiple clients for the file.
584  * There is also a consistency problem for Version 2 of the protocol w.r.t.
585  * not being able to tell if other clients are writing a file concurrently,
586  * since there is no way of knowing if the changed modify time in the reply
587  * is only due to the write for this client.
588  * (NFS Version 3 provides weak cache consistency data in the reply that
589  *  should be sufficient to detect and handle this case.)
590  *
591  * The current code does the following:
592  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594  *                     or commit them (this satisfies 1 and 2 except for the
595  *                     case where the server crashes after this close but
596  *                     before the commit RPC, which is felt to be "good
597  *                     enough". Changing the last argument to nfs_flush() to
598  *                     a 1 would force a commit operation, if it is felt a
599  *                     commit is necessary now.
600  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
601  *                     1 should be dealt with via an fsync() system call for
602  *                     cases where write errors are important.
603  *
604  * nfs_close(struct vnode *a_vp, int a_fflag)
605  */
606 /* ARGSUSED */
607 static int
608 nfs_close(struct vop_close_args *ap)
609 {
610 	struct vnode *vp = ap->a_vp;
611 	struct nfsnode *np = VTONFS(vp);
612 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
613 	int error = 0;
614 	thread_t td = curthread;
615 
616 	lwkt_gettoken(&nmp->nm_token);
617 
618 	if (vp->v_type == VREG) {
619 	    if (np->n_flag & NLMODIFIED) {
620 		if (NFS_ISV3(vp)) {
621 		    /*
622 		     * Under NFSv3 we have dirty buffers to dispose of.  We
623 		     * must flush them to the NFS server.  We have the option
624 		     * of waiting all the way through the commit rpc or just
625 		     * waiting for the initial write.  The default is to only
626 		     * wait through the initial write so the data is in the
627 		     * server's cache, which is roughly similar to the state
628 		     * a standard disk subsystem leaves the file in on close().
629 		     *
630 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
631 		     * potential races with other processes, and certainly
632 		     * cannot clear it if we don't commit.
633 		     */
634 		    int cm = nfsv3_commit_on_close ? 1 : 0;
635 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
636 		    /* np->n_flag &= ~NLMODIFIED; */
637 		} else {
638 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
639 		}
640 		np->n_attrstamp = 0;
641 	    }
642 	    if (np->n_flag & NWRITEERR) {
643 		np->n_flag &= ~NWRITEERR;
644 		error = np->n_error;
645 	    }
646 	}
647 	vop_stdclose(ap);
648 	lwkt_reltoken(&nmp->nm_token);
649 
650 	return (error);
651 }
652 
653 /*
654  * nfs getattr call from vfs.
655  *
656  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
657  */
658 static int
659 nfs_getattr(struct vop_getattr_args *ap)
660 {
661 	struct vnode *vp = ap->a_vp;
662 	struct nfsnode *np = VTONFS(vp);
663 	struct nfsmount *nmp;
664 	int error = 0;
665 	thread_t td = curthread;
666 	struct nfsm_info info;
667 
668 	info.mrep = NULL;
669 	info.v3 = NFS_ISV3(vp);
670 	nmp = VFSTONFS(vp->v_mount);
671 
672 	lwkt_gettoken(&nmp->nm_token);
673 
674 	/*
675 	 * Update local times for special files.
676 	 */
677 	if (np->n_flag & (NACC | NUPD))
678 		np->n_flag |= NCHG;
679 	/*
680 	 * First look in the cache.
681 	 */
682 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
683 		goto done;
684 
685 	if (info.v3 && nfsaccess_cache_timeout > 0) {
686 		nfsstats.accesscache_misses++;
687 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
688 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
689 			goto done;
690 	}
691 
692 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
693 	nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
694 	ERROROUT(nfsm_fhtom(&info, vp));
695 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
696 				nfs_vpcred(vp, ND_CHECK), &error));
697 	if (error == 0) {
698 		ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
699 	}
700 	m_freem(info.mrep);
701 	info.mrep = NULL;
702 done:
703 	/*
704 	 * NFS doesn't support chflags flags.  If the nfs mount was
705 	 * made -o cache set the UF_CACHE bit for swapcache.
706 	 */
707 	if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
708 		ap->a_vap->va_flags |= UF_CACHE;
709 nfsmout:
710 	lwkt_reltoken(&nmp->nm_token);
711 	return (error);
712 }
713 
714 /*
715  * nfs setattr call.
716  *
717  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
718  */
719 static int
720 nfs_setattr(struct vop_setattr_args *ap)
721 {
722 	struct vnode *vp = ap->a_vp;
723 	struct nfsnode *np = VTONFS(vp);
724 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
725 	struct vattr *vap = ap->a_vap;
726 	int error = 0;
727 	off_t tsize;
728 	thread_t td = curthread;
729 
730 #ifndef nolint
731 	tsize = (off_t)0;
732 #endif
733 	/*
734 	 * Setting of flags is not supported.
735 	 */
736 	if (vap->va_flags != VNOVAL)
737 		return (EOPNOTSUPP);
738 
739 	/*
740 	 * Disallow write attempts if the filesystem is mounted read-only.
741 	 */
742   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
743 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
744 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
745 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
746 		return (EROFS);
747 
748 	lwkt_gettoken(&nmp->nm_token);
749 
750 	if (vap->va_size != VNOVAL) {
751 		/*
752 		 * truncation requested
753 		 */
754  		switch (vp->v_type) {
755  		case VDIR:
756 			lwkt_reltoken(&nmp->nm_token);
757  			return (EISDIR);
758  		case VCHR:
759  		case VBLK:
760  		case VSOCK:
761  		case VFIFO:
762 			if (vap->va_mtime.tv_sec == VNOVAL &&
763 			    vap->va_atime.tv_sec == VNOVAL &&
764 			    vap->va_mode == (mode_t)VNOVAL &&
765 			    vap->va_uid == (uid_t)VNOVAL &&
766 			    vap->va_gid == (gid_t)VNOVAL) {
767 				lwkt_reltoken(&nmp->nm_token);
768 				return (0);
769 			}
770  			vap->va_size = VNOVAL;
771  			break;
772  		default:
773 			/*
774 			 * Disallow write attempts if the filesystem is
775 			 * mounted read-only.
776 			 */
777 			if (vp->v_mount->mnt_flag & MNT_RDONLY) {
778 				lwkt_reltoken(&nmp->nm_token);
779 				return (EROFS);
780 			}
781 
782 			tsize = np->n_size;
783 again:
784 			error = nfs_meta_setsize(vp, td, vap->va_size, 0);
785 
786 #if 0
787  			if (np->n_flag & NLMODIFIED) {
788  			    if (vap->va_size == 0)
789  				error = nfs_vinvalbuf(vp, 0, 1);
790  			    else
791  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
792  			}
793 #endif
794 			/*
795 			 * note: this loop case almost always happens at
796 			 * least once per truncation.
797 			 */
798 			if (error == 0 && np->n_size != vap->va_size)
799 				goto again;
800 			np->n_vattr.va_size = vap->va_size;
801 			break;
802 		}
803 	} else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
804 		/*
805 		 * What to do.  If we are modifying the mtime we lose
806 		 * mtime detection of changes made by the server or other
807 		 * clients.  But programs like rsync/rdist/cpdup are going
808 		 * to call utimes a lot.  We don't want to piecemeal sync.
809 		 *
810 		 * For now sync if any prior remote changes were detected,
811 		 * but allow us to lose track of remote changes made during
812 		 * the utimes operation.
813 		 */
814 		if (np->n_flag & NRMODIFIED)
815 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
816 		if (error == EINTR)
817 			return (error);
818 		if (error == 0) {
819 			if (vap->va_mtime.tv_sec != VNOVAL) {
820 				np->n_mtime = vap->va_mtime.tv_sec;
821 			}
822 		}
823 	}
824 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
825 
826 	/*
827 	 * Sanity check if a truncation was issued.  This should only occur
828 	 * if multiple processes are racing on the same file.
829 	 */
830 	if (error == 0 && vap->va_size != VNOVAL &&
831 	    np->n_size != vap->va_size) {
832 		kprintf("NFS ftruncate: server disagrees on the file size: "
833 			"%jd/%jd/%jd\n",
834 			(intmax_t)tsize,
835 			(intmax_t)vap->va_size,
836 			(intmax_t)np->n_size);
837 		goto again;
838 	}
839 	if (error && vap->va_size != VNOVAL) {
840 		np->n_size = np->n_vattr.va_size = tsize;
841 		nfs_meta_setsize(vp, td, np->n_size, 0);
842 	}
843 	lwkt_reltoken(&nmp->nm_token);
844 
845 	return (error);
846 }
847 
848 /*
849  * Do an nfs setattr rpc.
850  */
851 static int
852 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
853 	       struct ucred *cred, struct thread *td)
854 {
855 	struct nfsv2_sattr *sp;
856 	struct nfsnode *np = VTONFS(vp);
857 	u_int32_t *tl;
858 	int error = 0, wccflag = NFSV3_WCCRATTR;
859 	struct nfsm_info info;
860 
861 	info.mrep = NULL;
862 	info.v3 = NFS_ISV3(vp);
863 
864 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
865 	nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
866 		     NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
867 	ERROROUT(nfsm_fhtom(&info, vp));
868 	if (info.v3) {
869 		nfsm_v3attrbuild(&info, vap, TRUE);
870 		tl = nfsm_build(&info, NFSX_UNSIGNED);
871 		*tl = nfs_false;
872 	} else {
873 		sp = nfsm_build(&info, NFSX_V2SATTR);
874 		if (vap->va_mode == (mode_t)VNOVAL)
875 			sp->sa_mode = nfs_xdrneg1;
876 		else
877 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
878 		if (vap->va_uid == (uid_t)VNOVAL)
879 			sp->sa_uid = nfs_xdrneg1;
880 		else
881 			sp->sa_uid = txdr_unsigned(vap->va_uid);
882 		if (vap->va_gid == (gid_t)VNOVAL)
883 			sp->sa_gid = nfs_xdrneg1;
884 		else
885 			sp->sa_gid = txdr_unsigned(vap->va_gid);
886 		sp->sa_size = txdr_unsigned(vap->va_size);
887 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
888 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
889 	}
890 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
891 	if (info.v3) {
892 		np->n_modestamp = 0;
893 		ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
894 	} else {
895 		ERROROUT(nfsm_loadattr(&info, vp, NULL));
896 	}
897 	m_freem(info.mrep);
898 	info.mrep = NULL;
899 nfsmout:
900 	return (error);
901 }
902 
903 static
904 void
905 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
906 {
907 	if (nctimeout == 0)
908 		nctimeout = 1;
909 	else
910 		nctimeout *= hz;
911 	cache_setvp(nch, vp);
912 	cache_settimeout(nch, nctimeout);
913 }
914 
915 /*
916  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
917  * nfs_lookup() until all remaining new api calls are implemented.
918  *
919  * Resolve a namecache entry.  This function is passed a locked ncp and
920  * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
921  */
922 static int
923 nfs_nresolve(struct vop_nresolve_args *ap)
924 {
925 	struct thread *td = curthread;
926 	struct namecache *ncp;
927 	struct nfsmount *nmp;
928 	struct nfsnode *np;
929 	struct vnode *dvp;
930 	struct vnode *nvp;
931 	nfsfh_t *fhp;
932 	int attrflag;
933 	int fhsize;
934 	int error;
935 	int tmp_error;
936 	int len;
937 	struct nfsm_info info;
938 
939 	dvp = ap->a_dvp;
940 	nmp = VFSTONFS(dvp->v_mount);
941 
942 	lwkt_gettoken(&nmp->nm_token);
943 
944 	if ((error = vget(dvp, LK_SHARED)) != 0) {
945 		lwkt_reltoken(&nmp->nm_token);
946 		return (error);
947 	}
948 
949 	info.mrep = NULL;
950 	info.v3 = NFS_ISV3(dvp);
951 
952 	nvp = NULL;
953 	nfsstats.lookupcache_misses++;
954 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
955 	ncp = ap->a_nch->ncp;
956 	len = ncp->nc_nlen;
957 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
958 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
959 	ERROROUT(nfsm_fhtom(&info, dvp));
960 	ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
961 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
962 				ap->a_cred, &error));
963 	if (error) {
964 		/*
965 		 * Cache negatve lookups to reduce NFS traffic, but use
966 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
967 		 * XXX we should add a namecache flag for no-caching
968 		 * to uncache the negative hit as soon as possible, but
969 		 * we cannot simply destroy the entry because it is used
970 		 * as a placeholder by the caller.
971 		 *
972 		 * The refactored nfs code will overwrite a non-zero error
973 		 * with 0 when we use ERROROUT(), so don't here.
974 		 */
975 		if (error == ENOENT)
976 			nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
977 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
978 					     NFS_LATTR_NOSHRINK);
979 		if (tmp_error) {
980 			error = tmp_error;
981 			goto nfsmout;
982 		}
983 		m_freem(info.mrep);
984 		info.mrep = NULL;
985 		goto nfsmout;
986 	}
987 
988 	/*
989 	 * Success, get the file handle, do various checks, and load
990 	 * post-operation data from the reply packet.  Theoretically
991 	 * we should never be looking up "." so, theoretically, we
992 	 * should never get the same file handle as our directory.  But
993 	 * we check anyway. XXX
994 	 *
995 	 * Note that no timeout is set for the positive cache hit.  We
996 	 * assume, theoretically, that ESTALE returns will be dealt with
997 	 * properly to handle NFS races and in anycase we cannot depend
998 	 * on a timeout to deal with NFS open/create/excl issues so instead
999 	 * of a bad hack here the rest of the NFS client code needs to do
1000 	 * the right thing.
1001 	 */
1002 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1003 
1004 	np = VTONFS(dvp);
1005 	if (NFS_CMPFH(np, fhp, fhsize)) {
1006 		vref(dvp);
1007 		nvp = dvp;
1008 	} else {
1009 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, NULL);
1010 		if (error) {
1011 			m_freem(info.mrep);
1012 			info.mrep = NULL;
1013 			vput(dvp);
1014 			lwkt_reltoken(&nmp->nm_token);
1015 			return (error);
1016 		}
1017 		nvp = NFSTOV(np);
1018 	}
1019 	if (info.v3) {
1020 		ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1021 					  NFS_LATTR_NOSHRINK));
1022 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1023 					  NFS_LATTR_NOSHRINK));
1024 	} else {
1025 		ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1026 	}
1027 	nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1028 	m_freem(info.mrep);
1029 	info.mrep = NULL;
1030 nfsmout:
1031 	lwkt_reltoken(&nmp->nm_token);
1032 	vput(dvp);
1033 	if (nvp) {
1034 		if (nvp == dvp)
1035 			vrele(nvp);
1036 		else
1037 			vput(nvp);
1038 	}
1039 	return (error);
1040 }
1041 
1042 /*
1043  * 'cached' nfs directory lookup
1044  *
1045  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1046  *
1047  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1048  *	      struct componentname *a_cnp)
1049  */
1050 static int
1051 nfs_lookup(struct vop_old_lookup_args *ap)
1052 {
1053 	struct componentname *cnp = ap->a_cnp;
1054 	struct vnode *dvp = ap->a_dvp;
1055 	struct vnode **vpp = ap->a_vpp;
1056 	int flags = cnp->cn_flags;
1057 	struct vnode *newvp;
1058 	struct vnode *notvp;
1059 	struct nfsmount *nmp;
1060 	long len;
1061 	nfsfh_t *fhp;
1062 	struct nfsnode *np;
1063 	int lockparent, wantparent, attrflag, fhsize;
1064 	int error;
1065 	int tmp_error;
1066 	struct nfsm_info info;
1067 
1068 	info.mrep = NULL;
1069 	info.v3 = NFS_ISV3(dvp);
1070 	error = 0;
1071 
1072 	notvp = (cnp->cn_flags & CNP_NOTVP) ? cnp->cn_notvp : NULL;
1073 
1074 	/*
1075 	 * Read-only mount check and directory check.
1076 	 */
1077 	*vpp = NULLVP;
1078 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1079 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1080 		return (EROFS);
1081 
1082 	if (dvp->v_type != VDIR)
1083 		return (ENOTDIR);
1084 
1085 	/*
1086 	 * Look it up in the cache.  Note that ENOENT is only returned if we
1087 	 * previously entered a negative hit (see later on).  The additional
1088 	 * nfsneg_cache_timeout check causes previously cached results to
1089 	 * be instantly ignored if the negative caching is turned off.
1090 	 */
1091 	lockparent = flags & CNP_LOCKPARENT;
1092 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1093 	nmp = VFSTONFS(dvp->v_mount);
1094 	np = VTONFS(dvp);
1095 
1096 	lwkt_gettoken(&nmp->nm_token);
1097 
1098 	/*
1099 	 * Go to the wire.
1100 	 */
1101 	error = 0;
1102 	newvp = NULLVP;
1103 	nfsstats.lookupcache_misses++;
1104 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1105 	len = cnp->cn_namelen;
1106 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1107 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1108 	ERROROUT(nfsm_fhtom(&info, dvp));
1109 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1110 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1111 				cnp->cn_cred, &error));
1112 	if (error) {
1113 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1114 					     NFS_LATTR_NOSHRINK);
1115 		if (tmp_error) {
1116 			error = tmp_error;
1117 			goto nfsmout;
1118 		}
1119 
1120 		m_freem(info.mrep);
1121 		info.mrep = NULL;
1122 		goto nfsmout;
1123 	}
1124 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1125 
1126 	/*
1127 	 * Handle RENAME case...
1128 	 */
1129 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1130 		if (NFS_CMPFH(np, fhp, fhsize)) {
1131 			m_freem(info.mrep);
1132 			info.mrep = NULL;
1133 			lwkt_reltoken(&nmp->nm_token);
1134 			return (EISDIR);
1135 		}
1136 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1137 		if (error) {
1138 			m_freem(info.mrep);
1139 			info.mrep = NULL;
1140 			lwkt_reltoken(&nmp->nm_token);
1141 			return (error);
1142 		}
1143 		newvp = NFSTOV(np);
1144 		if (info.v3) {
1145 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1146 						  NFS_LATTR_NOSHRINK));
1147 			ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1148 						  NFS_LATTR_NOSHRINK));
1149 		} else {
1150 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1151 		}
1152 		*vpp = newvp;
1153 		m_freem(info.mrep);
1154 		info.mrep = NULL;
1155 		if (!lockparent) {
1156 			vn_unlock(dvp);
1157 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1158 		}
1159 		lwkt_reltoken(&nmp->nm_token);
1160 		return (0);
1161 	}
1162 
1163 	if (flags & CNP_ISDOTDOT) {
1164 		vn_unlock(dvp);
1165 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1166 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1167 		if (error) {
1168 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1169 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1170 			lwkt_reltoken(&nmp->nm_token);
1171 			return (error); /* NOTE: return error from nget */
1172 		}
1173 		newvp = NFSTOV(np);
1174 		if (lockparent) {
1175 			error = vn_lock(dvp, LK_EXCLUSIVE);
1176 			if (error) {
1177 				vput(newvp);
1178 				lwkt_reltoken(&nmp->nm_token);
1179 				return (error);
1180 			}
1181 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1182 		}
1183 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1184 		vref(dvp);
1185 		newvp = dvp;
1186 	} else {
1187 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1188 		if (error) {
1189 			m_freem(info.mrep);
1190 			info.mrep = NULL;
1191 			lwkt_reltoken(&nmp->nm_token);
1192 			return (error);
1193 		}
1194 		if (!lockparent) {
1195 			vn_unlock(dvp);
1196 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1197 		}
1198 		newvp = NFSTOV(np);
1199 	}
1200 	if (info.v3) {
1201 		ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1202 					  NFS_LATTR_NOSHRINK));
1203 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1204 					  NFS_LATTR_NOSHRINK));
1205 	} else {
1206 		ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1207 	}
1208 #if 0
1209 	/* XXX MOVE TO nfs_nremove() */
1210 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1211 	    cnp->cn_nameiop != NAMEI_DELETE) {
1212 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1213 	}
1214 #endif
1215 	*vpp = newvp;
1216 	m_freem(info.mrep);
1217 	info.mrep = NULL;
1218 nfsmout:
1219 	if (error) {
1220 		if (newvp != NULLVP) {
1221 			vrele(newvp);
1222 			*vpp = NULLVP;
1223 		}
1224 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1225 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1226 		    error == ENOENT) {
1227 			if (!lockparent) {
1228 				vn_unlock(dvp);
1229 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1230 			}
1231 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1232 				error = EROFS;
1233 			else
1234 				error = EJUSTRETURN;
1235 		}
1236 	}
1237 	lwkt_reltoken(&nmp->nm_token);
1238 	return (error);
1239 }
1240 
1241 /*
1242  * nfs read call.
1243  * Just call nfs_bioread() to do the work.
1244  *
1245  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1246  *	    struct ucred *a_cred)
1247  */
1248 static int
1249 nfs_read(struct vop_read_args *ap)
1250 {
1251 	struct vnode *vp = ap->a_vp;
1252 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1253 	int error;
1254 
1255 	lwkt_gettoken(&nmp->nm_token);
1256 	error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1257 	lwkt_reltoken(&nmp->nm_token);
1258 
1259 	return error;
1260 }
1261 
1262 /*
1263  * nfs readlink call
1264  *
1265  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1266  */
1267 static int
1268 nfs_readlink(struct vop_readlink_args *ap)
1269 {
1270 	struct vnode *vp = ap->a_vp;
1271 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1272 	int error;
1273 
1274 	if (vp->v_type != VLNK)
1275 		return (EINVAL);
1276 
1277 	lwkt_gettoken(&nmp->nm_token);
1278 	error = nfs_bioread(vp, ap->a_uio, 0);
1279 	lwkt_reltoken(&nmp->nm_token);
1280 
1281 	return error;
1282 }
1283 
1284 /*
1285  * Do a readlink rpc.
1286  * Called by nfs_doio() from below the buffer cache.
1287  */
1288 int
1289 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1290 {
1291 	int error = 0, len, attrflag;
1292 	struct nfsm_info info;
1293 
1294 	info.mrep = NULL;
1295 	info.v3 = NFS_ISV3(vp);
1296 
1297 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1298 	nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1299 	ERROROUT(nfsm_fhtom(&info, vp));
1300 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1301 				nfs_vpcred(vp, ND_CHECK), &error));
1302 	if (info.v3) {
1303 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1304 					  NFS_LATTR_NOSHRINK));
1305 	}
1306 	if (!error) {
1307 		NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1308 		if (len == NFS_MAXPATHLEN) {
1309 			struct nfsnode *np = VTONFS(vp);
1310 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1311 				len = np->n_size;
1312 		}
1313 		ERROROUT(nfsm_mtouio(&info, uiop, len));
1314 	}
1315 	m_freem(info.mrep);
1316 	info.mrep = NULL;
1317 nfsmout:
1318 	return (error);
1319 }
1320 
1321 /*
1322  * nfs synchronous read rpc using UIO
1323  */
1324 int
1325 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1326 {
1327 	u_int32_t *tl;
1328 	struct nfsmount *nmp;
1329 	int error = 0, len, retlen, tsiz, eof, attrflag;
1330 	struct nfsm_info info;
1331 	off_t tmp_off;
1332 
1333 	info.mrep = NULL;
1334 	info.v3 = NFS_ISV3(vp);
1335 
1336 #ifndef nolint
1337 	eof = 0;
1338 #endif
1339 	nmp = VFSTONFS(vp->v_mount);
1340 
1341 	tsiz = uiop->uio_resid;
1342 	tmp_off = uiop->uio_offset + tsiz;
1343 	if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1344 		return (EFBIG);
1345 	tmp_off = uiop->uio_offset;
1346 	while (tsiz > 0) {
1347 		nfsstats.rpccnt[NFSPROC_READ]++;
1348 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1349 		nfsm_reqhead(&info, vp, NFSPROC_READ,
1350 			     NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1351 		ERROROUT(nfsm_fhtom(&info, vp));
1352 		tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1353 		if (info.v3) {
1354 			txdr_hyper(uiop->uio_offset, tl);
1355 			*(tl + 2) = txdr_unsigned(len);
1356 		} else {
1357 			*tl++ = txdr_unsigned(uiop->uio_offset);
1358 			*tl++ = txdr_unsigned(len);
1359 			*tl = 0;
1360 		}
1361 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1362 					nfs_vpcred(vp, ND_READ), &error));
1363 		if (info.v3) {
1364 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1365 						 NFS_LATTR_NOSHRINK));
1366 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1367 			eof = fxdr_unsigned(int, *(tl + 1));
1368 		} else {
1369 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1370 		}
1371 		NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1372 		ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1373 		m_freem(info.mrep);
1374 		info.mrep = NULL;
1375 
1376 		/*
1377 		 * Handle short-read from server (NFSv3).  If EOF is not
1378 		 * flagged (and no error occurred), but retlen is less
1379 		 * then the request size, we must zero-fill the remainder.
1380 		 */
1381 		if (retlen < len && info.v3 && eof == 0) {
1382 			ERROROUT(uiomovez(len - retlen, uiop));
1383 			retlen = len;
1384 		}
1385 		tsiz -= retlen;
1386 
1387 		/*
1388 		 * Terminate loop on EOF or zero-length read.
1389 		 *
1390 		 * For NFSv2 a short-read indicates EOF, not zero-fill,
1391 		 * and also terminates the loop.
1392 		 */
1393 		if (info.v3) {
1394 			if (eof || retlen == 0)
1395 				tsiz = 0;
1396 		} else if (retlen < len) {
1397 			tsiz = 0;
1398 		}
1399 	}
1400 nfsmout:
1401 	return (error);
1402 }
1403 
1404 /*
1405  * nfs write call
1406  */
1407 int
1408 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1409 		 int *iomode, int *must_commit)
1410 {
1411 	u_int32_t *tl;
1412 	int32_t backup;
1413 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1414 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1415 	int  committed = NFSV3WRITE_FILESYNC;
1416 	struct nfsm_info info;
1417 
1418 	info.mrep = NULL;
1419 	info.v3 = NFS_ISV3(vp);
1420 
1421 #ifndef DIAGNOSTIC
1422 	if (uiop->uio_iovcnt != 1)
1423 		panic("nfs: writerpc iovcnt > 1");
1424 #endif
1425 	*must_commit = 0;
1426 	tsiz = uiop->uio_resid;
1427 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1428 		return (EFBIG);
1429 	while (tsiz > 0) {
1430 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1431 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1432 		nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1433 			     NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1434 			     nfsm_rndup(len));
1435 		ERROROUT(nfsm_fhtom(&info, vp));
1436 		if (info.v3) {
1437 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1438 			txdr_hyper(uiop->uio_offset, tl);
1439 			tl += 2;
1440 			*tl++ = txdr_unsigned(len);
1441 			*tl++ = txdr_unsigned(*iomode);
1442 			*tl = txdr_unsigned(len);
1443 		} else {
1444 			u_int32_t x;
1445 
1446 			tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1447 			/* Set both "begin" and "current" to non-garbage. */
1448 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1449 			*tl++ = x;	/* "begin offset" */
1450 			*tl++ = x;	/* "current offset" */
1451 			x = txdr_unsigned(len);
1452 			*tl++ = x;	/* total to this offset */
1453 			*tl = x;	/* size of this write */
1454 		}
1455 		ERROROUT(nfsm_uiotom(&info, uiop, len));
1456 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1457 					nfs_vpcred(vp, ND_WRITE), &error));
1458 		if (info.v3) {
1459 			/*
1460 			 * The write RPC returns a before and after mtime.  The
1461 			 * nfsm_wcc_data() macro checks the before n_mtime
1462 			 * against the before time and stores the after time
1463 			 * in the nfsnode's cached vattr and n_mtime field.
1464 			 * The NRMODIFIED bit will be set if the before
1465 			 * time did not match the original mtime.
1466 			 */
1467 			wccflag = NFSV3_WCCCHK;
1468 			ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1469 			if (error == 0) {
1470 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1471 				rlen = fxdr_unsigned(int, *tl++);
1472 				if (rlen == 0) {
1473 					error = NFSERR_IO;
1474 					m_freem(info.mrep);
1475 					info.mrep = NULL;
1476 					break;
1477 				} else if (rlen < len) {
1478 					backup = len - rlen;
1479 					uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1480 					uiop->uio_iov->iov_len += backup;
1481 					uiop->uio_offset -= backup;
1482 					uiop->uio_resid += backup;
1483 					len = rlen;
1484 				}
1485 				commit = fxdr_unsigned(int, *tl++);
1486 
1487 				/*
1488 				 * Return the lowest committment level
1489 				 * obtained by any of the RPCs.
1490 				 */
1491 				if (committed == NFSV3WRITE_FILESYNC)
1492 					committed = commit;
1493 				else if (committed == NFSV3WRITE_DATASYNC &&
1494 					commit == NFSV3WRITE_UNSTABLE)
1495 					committed = commit;
1496 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1497 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1498 					NFSX_V3WRITEVERF);
1499 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1500 				} else if (bcmp((caddr_t)tl,
1501 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1502 				    *must_commit = 1;
1503 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1504 					NFSX_V3WRITEVERF);
1505 				}
1506 			}
1507 		} else {
1508 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1509 		}
1510 		m_freem(info.mrep);
1511 		info.mrep = NULL;
1512 		if (error)
1513 			break;
1514 		tsiz -= len;
1515 	}
1516 nfsmout:
1517 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1518 		committed = NFSV3WRITE_FILESYNC;
1519 	*iomode = committed;
1520 	if (error)
1521 		uiop->uio_resid = tsiz;
1522 	return (error);
1523 }
1524 
1525 /*
1526  * nfs mknod rpc
1527  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1528  * mode set to specify the file type and the size field for rdev.
1529  */
1530 static int
1531 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1532 	     struct vattr *vap)
1533 {
1534 	struct nfsv2_sattr *sp;
1535 	u_int32_t *tl;
1536 	struct vnode *newvp = NULL;
1537 	struct nfsnode *np = NULL;
1538 	struct vattr vattr;
1539 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1540 	int rmajor, rminor;
1541 	struct nfsm_info info;
1542 
1543 	info.mrep = NULL;
1544 	info.v3 = NFS_ISV3(dvp);
1545 
1546 	if (vap->va_type == VCHR || vap->va_type == VBLK) {
1547 		rmajor = txdr_unsigned(vap->va_rmajor);
1548 		rminor = txdr_unsigned(vap->va_rminor);
1549 	} else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1550 		rmajor = nfs_xdrneg1;
1551 		rminor = nfs_xdrneg1;
1552 	} else {
1553 		return (EOPNOTSUPP);
1554 	}
1555 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1556 		return (error);
1557 	}
1558 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1559 	nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1560 		     NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1561 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1562 	ERROROUT(nfsm_fhtom(&info, dvp));
1563 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1564 			     NFS_MAXNAMLEN));
1565 	if (info.v3) {
1566 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1567 		*tl++ = vtonfsv3_type(vap->va_type);
1568 		nfsm_v3attrbuild(&info, vap, FALSE);
1569 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1570 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1571 			*tl++ = txdr_unsigned(vap->va_rmajor);
1572 			*tl = txdr_unsigned(vap->va_rminor);
1573 		}
1574 	} else {
1575 		sp = nfsm_build(&info, NFSX_V2SATTR);
1576 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1577 		sp->sa_uid = nfs_xdrneg1;
1578 		sp->sa_gid = nfs_xdrneg1;
1579 		sp->sa_size = makeudev(rmajor, rminor);
1580 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1581 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1582 	}
1583 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1584 				cnp->cn_cred, &error));
1585 	if (!error) {
1586 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1587 		if (!gotvp) {
1588 			if (newvp) {
1589 				vput(newvp);
1590 				newvp = NULL;
1591 			}
1592 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1593 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1594 			if (!error)
1595 				newvp = NFSTOV(np);
1596 		}
1597 	}
1598 	if (info.v3) {
1599 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1600 	}
1601 	m_freem(info.mrep);
1602 	info.mrep = NULL;
1603 nfsmout:
1604 	if (error) {
1605 		if (newvp)
1606 			vput(newvp);
1607 	} else {
1608 		*vpp = newvp;
1609 	}
1610 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1611 	if (!wccflag)
1612 		VTONFS(dvp)->n_attrstamp = 0;
1613 	return (error);
1614 }
1615 
1616 /*
1617  * nfs mknod vop
1618  * just call nfs_mknodrpc() to do the work.
1619  *
1620  * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1621  *	     struct componentname *a_cnp, struct vattr *a_vap)
1622  */
1623 /* ARGSUSED */
1624 static int
1625 nfs_mknod(struct vop_old_mknod_args *ap)
1626 {
1627 	struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1628 	int error;
1629 
1630 	lwkt_gettoken(&nmp->nm_token);
1631 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1632 	lwkt_reltoken(&nmp->nm_token);
1633 
1634 	return error;
1635 }
1636 
1637 static u_long create_verf;
1638 /*
1639  * nfs file create call
1640  *
1641  * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1642  *	      struct componentname *a_cnp, struct vattr *a_vap)
1643  */
1644 static int
1645 nfs_create(struct vop_old_create_args *ap)
1646 {
1647 	struct vnode *dvp = ap->a_dvp;
1648 	struct vattr *vap = ap->a_vap;
1649 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1650 	struct componentname *cnp = ap->a_cnp;
1651 	struct nfsv2_sattr *sp;
1652 	u_int32_t *tl;
1653 	struct nfsnode *np = NULL;
1654 	struct vnode *newvp = NULL;
1655 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1656 	struct vattr vattr;
1657 	struct nfsm_info info;
1658 
1659 	info.mrep = NULL;
1660 	info.v3 = NFS_ISV3(dvp);
1661 	lwkt_gettoken(&nmp->nm_token);
1662 
1663 	/*
1664 	 * Oops, not for me..
1665 	 */
1666 	if (vap->va_type == VSOCK) {
1667 		error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1668 		lwkt_reltoken(&nmp->nm_token);
1669 		return error;
1670 	}
1671 
1672 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1673 		lwkt_reltoken(&nmp->nm_token);
1674 		return (error);
1675 	}
1676 	if (vap->va_vaflags & VA_EXCLUSIVE)
1677 		fmode |= O_EXCL;
1678 again:
1679 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1680 	nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1681 		     NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1682 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1683 	ERROROUT(nfsm_fhtom(&info, dvp));
1684 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1685 			     NFS_MAXNAMLEN));
1686 	if (info.v3) {
1687 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1688 		if (fmode & O_EXCL) {
1689 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1690 			tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1691 #ifdef INET
1692 			if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1693 				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1694 			else
1695 #endif
1696 				*tl++ = create_verf;
1697 			*tl = ++create_verf;
1698 		} else {
1699 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1700 			nfsm_v3attrbuild(&info, vap, FALSE);
1701 		}
1702 	} else {
1703 		sp = nfsm_build(&info, NFSX_V2SATTR);
1704 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1705 		sp->sa_uid = nfs_xdrneg1;
1706 		sp->sa_gid = nfs_xdrneg1;
1707 		sp->sa_size = 0;
1708 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1709 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1710 	}
1711 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1712 				cnp->cn_cred, &error));
1713 	if (error == 0) {
1714 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1715 		if (!gotvp) {
1716 			if (newvp) {
1717 				vput(newvp);
1718 				newvp = NULL;
1719 			}
1720 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1721 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1722 			if (!error)
1723 				newvp = NFSTOV(np);
1724 		}
1725 	}
1726 	if (info.v3) {
1727 		if (error == 0)
1728 			error = nfsm_wcc_data(&info, dvp, &wccflag);
1729 		else
1730 			(void)nfsm_wcc_data(&info, dvp, &wccflag);
1731 	}
1732 	m_freem(info.mrep);
1733 	info.mrep = NULL;
1734 nfsmout:
1735 	if (error) {
1736 		if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1737 			KKASSERT(newvp == NULL);
1738 			fmode &= ~O_EXCL;
1739 			goto again;
1740 		}
1741 	} else if (info.v3 && (fmode & O_EXCL)) {
1742 		/*
1743 		 * We are normally called with only a partially initialized
1744 		 * VAP.  Since the NFSv3 spec says that server may use the
1745 		 * file attributes to store the verifier, the spec requires
1746 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1747 		 * in atime, but we can't really assume that all servers will
1748 		 * so we ensure that our SETATTR sets both atime and mtime.
1749 		 */
1750 		if (vap->va_mtime.tv_sec == VNOVAL)
1751 			vfs_timestamp(&vap->va_mtime);
1752 		if (vap->va_atime.tv_sec == VNOVAL)
1753 			vap->va_atime = vap->va_mtime;
1754 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1755 	}
1756 	if (error == 0) {
1757 		/*
1758 		 * The new np may have enough info for access
1759 		 * checks, make sure rucred and wucred are
1760 		 * initialized for read and write rpc's.
1761 		 */
1762 		np = VTONFS(newvp);
1763 		if (np->n_rucred == NULL)
1764 			np->n_rucred = crhold(cnp->cn_cred);
1765 		if (np->n_wucred == NULL)
1766 			np->n_wucred = crhold(cnp->cn_cred);
1767 		*ap->a_vpp = newvp;
1768 	} else if (newvp) {
1769 		vput(newvp);
1770 	}
1771 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1772 	if (!wccflag)
1773 		VTONFS(dvp)->n_attrstamp = 0;
1774 	lwkt_reltoken(&nmp->nm_token);
1775 	return (error);
1776 }
1777 
1778 /*
1779  * nfs file remove call
1780  * To try and make nfs semantics closer to ufs semantics, a file that has
1781  * other processes using the vnode is renamed instead of removed and then
1782  * removed later on the last close.
1783  * - If v_sysref.refcnt > 1
1784  *	  If a rename is not already in the works
1785  *	     call nfs_sillyrename() to set it up
1786  *     else
1787  *	  do the remove rpc
1788  *
1789  * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1790  *	      struct componentname *a_cnp)
1791  */
1792 static int
1793 nfs_remove(struct vop_old_remove_args *ap)
1794 {
1795 	struct vnode *vp = ap->a_vp;
1796 	struct vnode *dvp = ap->a_dvp;
1797 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1798 	struct componentname *cnp = ap->a_cnp;
1799 	struct nfsnode *np = VTONFS(vp);
1800 	int error = 0;
1801 	struct vattr vattr;
1802 
1803 	lwkt_gettoken(&nmp->nm_token);
1804 #ifndef DIAGNOSTIC
1805 	if (vp->v_sysref.refcnt < 1)
1806 		panic("nfs_remove: bad v_sysref.refcnt");
1807 #endif
1808 	if (vp->v_type == VDIR) {
1809 		error = EPERM;
1810 	} else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1811 		   VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1812 		/*
1813 		 * throw away biocache buffers, mainly to avoid
1814 		 * unnecessary delayed writes later.
1815 		 */
1816 		error = nfs_vinvalbuf(vp, 0, 1);
1817 		/* Do the rpc */
1818 		if (error != EINTR)
1819 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1820 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1821 		/*
1822 		 * Kludge City: If the first reply to the remove rpc is lost..
1823 		 *   the reply to the retransmitted request will be ENOENT
1824 		 *   since the file was in fact removed
1825 		 *   Therefore, we cheat and return success.
1826 		 */
1827 		if (error == ENOENT)
1828 			error = 0;
1829 	} else if (!np->n_sillyrename) {
1830 		error = nfs_sillyrename(dvp, vp, cnp);
1831 	}
1832 	np->n_attrstamp = 0;
1833 	lwkt_reltoken(&nmp->nm_token);
1834 
1835 	return (error);
1836 }
1837 
1838 /*
1839  * nfs file remove rpc called from nfs_inactive
1840  */
1841 int
1842 nfs_removeit(struct sillyrename *sp)
1843 {
1844 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1845 		sp->s_cred, NULL));
1846 }
1847 
1848 /*
1849  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1850  */
1851 static int
1852 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1853 	      struct ucred *cred, struct thread *td)
1854 {
1855 	int error = 0, wccflag = NFSV3_WCCRATTR;
1856 	struct nfsm_info info;
1857 
1858 	info.mrep = NULL;
1859 	info.v3 = NFS_ISV3(dvp);
1860 
1861 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1862 	nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1863 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1864 	ERROROUT(nfsm_fhtom(&info, dvp));
1865 	ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1866 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1867 	if (info.v3) {
1868 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1869 	}
1870 	m_freem(info.mrep);
1871 	info.mrep = NULL;
1872 nfsmout:
1873 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1874 	if (!wccflag)
1875 		VTONFS(dvp)->n_attrstamp = 0;
1876 	return (error);
1877 }
1878 
1879 /*
1880  * nfs file rename call
1881  *
1882  * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1883  *	      struct componentname *a_fcnp, struct vnode *a_tdvp,
1884  *	      struct vnode *a_tvp, struct componentname *a_tcnp)
1885  */
1886 static int
1887 nfs_rename(struct vop_old_rename_args *ap)
1888 {
1889 	struct vnode *fvp = ap->a_fvp;
1890 	struct vnode *tvp = ap->a_tvp;
1891 	struct vnode *fdvp = ap->a_fdvp;
1892 	struct vnode *tdvp = ap->a_tdvp;
1893 	struct componentname *tcnp = ap->a_tcnp;
1894 	struct componentname *fcnp = ap->a_fcnp;
1895 	struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1896 	int error;
1897 
1898 	lwkt_gettoken(&nmp->nm_token);
1899 
1900 	/* Check for cross-device rename */
1901 	if ((fvp->v_mount != tdvp->v_mount) ||
1902 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1903 		error = EXDEV;
1904 		goto out;
1905 	}
1906 
1907 	/*
1908 	 * We shouldn't have to flush fvp on rename for most server-side
1909 	 * filesystems as the file handle should not change.  Unfortunately
1910 	 * the inode for some filesystems (msdosfs) might be tied to the
1911 	 * file name or directory position so to be completely safe
1912 	 * vfs.nfs.flush_on_rename is set by default.  Clear to improve
1913 	 * performance.
1914 	 *
1915 	 * We must flush tvp on rename because it might become stale on the
1916 	 * server after the rename.
1917 	 */
1918 	if (nfs_flush_on_rename)
1919 	    VOP_FSYNC(fvp, MNT_WAIT, 0);
1920 	if (tvp)
1921 	    VOP_FSYNC(tvp, MNT_WAIT, 0);
1922 
1923 	/*
1924 	 * If the tvp exists and is in use, sillyrename it before doing the
1925 	 * rename of the new file over it.
1926 	 *
1927 	 * XXX Can't sillyrename a directory.
1928 	 *
1929 	 * We do not attempt to do any namecache purges in this old API
1930 	 * routine.  The new API compat functions have access to the actual
1931 	 * namecache structures and will do it for us.
1932 	 */
1933 	if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1934 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1935 		vput(tvp);
1936 		tvp = NULL;
1937 	} else if (tvp) {
1938 		;
1939 	}
1940 
1941 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1942 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1943 		tcnp->cn_td);
1944 
1945 out:
1946 	lwkt_reltoken(&nmp->nm_token);
1947 	if (tdvp == tvp)
1948 		vrele(tdvp);
1949 	else
1950 		vput(tdvp);
1951 	if (tvp)
1952 		vput(tvp);
1953 	vrele(fdvp);
1954 	vrele(fvp);
1955 	/*
1956 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1957 	 */
1958 	if (error == ENOENT)
1959 		error = 0;
1960 	return (error);
1961 }
1962 
1963 /*
1964  * nfs file rename rpc called from nfs_remove() above
1965  */
1966 static int
1967 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1968 	     struct sillyrename *sp)
1969 {
1970 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1971 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1972 }
1973 
1974 /*
1975  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1976  */
1977 static int
1978 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1979 	      struct vnode *tdvp, const char *tnameptr, int tnamelen,
1980 	      struct ucred *cred, struct thread *td)
1981 {
1982 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1983 	struct nfsm_info info;
1984 
1985 	info.mrep = NULL;
1986 	info.v3 = NFS_ISV3(fdvp);
1987 
1988 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1989 	nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1990 		    (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1991 		    nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1992 	ERROROUT(nfsm_fhtom(&info, fdvp));
1993 	ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1994 	ERROROUT(nfsm_fhtom(&info, tdvp));
1995 	ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1996 	NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1997 	if (info.v3) {
1998 		ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1999 		ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
2000 	}
2001 	m_freem(info.mrep);
2002 	info.mrep = NULL;
2003 nfsmout:
2004 	VTONFS(fdvp)->n_flag |= NLMODIFIED;
2005 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2006 	if (!fwccflag)
2007 		VTONFS(fdvp)->n_attrstamp = 0;
2008 	if (!twccflag)
2009 		VTONFS(tdvp)->n_attrstamp = 0;
2010 	return (error);
2011 }
2012 
2013 /*
2014  * nfs hard link create call
2015  *
2016  * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2017  *	    struct componentname *a_cnp)
2018  */
2019 static int
2020 nfs_link(struct vop_old_link_args *ap)
2021 {
2022 	struct vnode *vp = ap->a_vp;
2023 	struct vnode *tdvp = ap->a_tdvp;
2024 	struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2025 	struct componentname *cnp = ap->a_cnp;
2026 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2027 	struct nfsm_info info;
2028 
2029 	if (vp->v_mount != tdvp->v_mount) {
2030 		return (EXDEV);
2031 	}
2032 	lwkt_gettoken(&nmp->nm_token);
2033 
2034 	/*
2035 	 * The attribute cache may get out of sync with the server on link.
2036 	 * Pushing writes to the server before handle was inherited from
2037 	 * long long ago and it is unclear if we still need to do this.
2038 	 * Defaults to off.
2039 	 */
2040 	if (nfs_flush_on_hlink)
2041 		VOP_FSYNC(vp, MNT_WAIT, 0);
2042 
2043 	info.mrep = NULL;
2044 	info.v3 = NFS_ISV3(vp);
2045 
2046 	nfsstats.rpccnt[NFSPROC_LINK]++;
2047 	nfsm_reqhead(&info, vp, NFSPROC_LINK,
2048 		     NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2049 		     nfsm_rndup(cnp->cn_namelen));
2050 	ERROROUT(nfsm_fhtom(&info, vp));
2051 	ERROROUT(nfsm_fhtom(&info, tdvp));
2052 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2053 			     NFS_MAXNAMLEN));
2054 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2055 				cnp->cn_cred, &error));
2056 	if (info.v3) {
2057 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2058 					 NFS_LATTR_NOSHRINK));
2059 		ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2060 	}
2061 	m_freem(info.mrep);
2062 	info.mrep = NULL;
2063 nfsmout:
2064 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2065 	if (!attrflag)
2066 		VTONFS(vp)->n_attrstamp = 0;
2067 	if (!wccflag)
2068 		VTONFS(tdvp)->n_attrstamp = 0;
2069 	/*
2070 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2071 	 */
2072 	if (error == EEXIST)
2073 		error = 0;
2074 	lwkt_reltoken(&nmp->nm_token);
2075 	return (error);
2076 }
2077 
2078 /*
2079  * nfs symbolic link create call
2080  *
2081  * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2082  *		struct componentname *a_cnp, struct vattr *a_vap,
2083  *		char *a_target)
2084  */
2085 static int
2086 nfs_symlink(struct vop_old_symlink_args *ap)
2087 {
2088 	struct vnode *dvp = ap->a_dvp;
2089 	struct vattr *vap = ap->a_vap;
2090 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2091 	struct componentname *cnp = ap->a_cnp;
2092 	struct nfsv2_sattr *sp;
2093 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2094 	struct vnode *newvp = NULL;
2095 	struct nfsm_info info;
2096 
2097 	info.mrep = NULL;
2098 	info.v3 = NFS_ISV3(dvp);
2099 	lwkt_gettoken(&nmp->nm_token);
2100 
2101 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2102 	slen = strlen(ap->a_target);
2103 	nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2104 		     NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2105 		     nfsm_rndup(cnp->cn_namelen) +
2106 		     nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2107 	ERROROUT(nfsm_fhtom(&info, dvp));
2108 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2109 			     NFS_MAXNAMLEN));
2110 	if (info.v3) {
2111 		nfsm_v3attrbuild(&info, vap, FALSE);
2112 	}
2113 	ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2114 	if (info.v3 == 0) {
2115 		sp = nfsm_build(&info, NFSX_V2SATTR);
2116 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2117 		sp->sa_uid = nfs_xdrneg1;
2118 		sp->sa_gid = nfs_xdrneg1;
2119 		sp->sa_size = nfs_xdrneg1;
2120 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2121 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2122 	}
2123 
2124 	/*
2125 	 * Issue the NFS request and get the rpc response.
2126 	 *
2127 	 * Only NFSv3 responses returning an error of 0 actually return
2128 	 * a file handle that can be converted into newvp without having
2129 	 * to do an extra lookup rpc.
2130 	 */
2131 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2132 				cnp->cn_cred, &error));
2133 	if (info.v3) {
2134 		if (error == 0) {
2135 		       ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2136 		}
2137 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2138 	}
2139 
2140 	/*
2141 	 * out code jumps -> here, mrep is also freed.
2142 	 */
2143 
2144 	m_freem(info.mrep);
2145 	info.mrep = NULL;
2146 nfsmout:
2147 
2148 	/*
2149 	 * If we get an EEXIST error, silently convert it to no-error
2150 	 * in case of an NFS retry.
2151 	 */
2152 	if (error == EEXIST)
2153 		error = 0;
2154 
2155 	/*
2156 	 * If we do not have (or no longer have) an error, and we could
2157 	 * not extract the newvp from the response due to the request being
2158 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
2159 	 * to obtain a newvp to return.
2160 	 */
2161 	if (error == 0 && newvp == NULL) {
2162 		struct nfsnode *np = NULL;
2163 
2164 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2165 				     cnp->cn_cred, cnp->cn_td, &np);
2166 		if (!error)
2167 			newvp = NFSTOV(np);
2168 	}
2169 	if (error) {
2170 		if (newvp)
2171 			vput(newvp);
2172 	} else {
2173 		*ap->a_vpp = newvp;
2174 	}
2175 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2176 	if (!wccflag)
2177 		VTONFS(dvp)->n_attrstamp = 0;
2178 	lwkt_reltoken(&nmp->nm_token);
2179 
2180 	return (error);
2181 }
2182 
2183 /*
2184  * nfs make dir call
2185  *
2186  * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2187  *	     struct componentname *a_cnp, struct vattr *a_vap)
2188  */
2189 static int
2190 nfs_mkdir(struct vop_old_mkdir_args *ap)
2191 {
2192 	struct vnode *dvp = ap->a_dvp;
2193 	struct vattr *vap = ap->a_vap;
2194 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2195 	struct componentname *cnp = ap->a_cnp;
2196 	struct nfsv2_sattr *sp;
2197 	struct nfsnode *np = NULL;
2198 	struct vnode *newvp = NULL;
2199 	struct vattr vattr;
2200 	int error = 0, wccflag = NFSV3_WCCRATTR;
2201 	int gotvp = 0;
2202 	int len;
2203 	struct nfsm_info info;
2204 
2205 	info.mrep = NULL;
2206 	info.v3 = NFS_ISV3(dvp);
2207 	lwkt_gettoken(&nmp->nm_token);
2208 
2209 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2210 		lwkt_reltoken(&nmp->nm_token);
2211 		return (error);
2212 	}
2213 	len = cnp->cn_namelen;
2214 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2215 	nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2216 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2217 		     nfsm_rndup(len) + NFSX_SATTR(info.v3));
2218 	ERROROUT(nfsm_fhtom(&info, dvp));
2219 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2220 	if (info.v3) {
2221 		nfsm_v3attrbuild(&info, vap, FALSE);
2222 	} else {
2223 		sp = nfsm_build(&info, NFSX_V2SATTR);
2224 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2225 		sp->sa_uid = nfs_xdrneg1;
2226 		sp->sa_gid = nfs_xdrneg1;
2227 		sp->sa_size = nfs_xdrneg1;
2228 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2229 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2230 	}
2231 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2232 		    cnp->cn_cred, &error));
2233 	if (error == 0) {
2234 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2235 	}
2236 	if (info.v3) {
2237 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2238 	}
2239 	m_freem(info.mrep);
2240 	info.mrep = NULL;
2241 nfsmout:
2242 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2243 	if (!wccflag)
2244 		VTONFS(dvp)->n_attrstamp = 0;
2245 	/*
2246 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2247 	 * if we can succeed in looking up the directory.
2248 	 */
2249 	if (error == EEXIST || (!error && !gotvp)) {
2250 		if (newvp) {
2251 			vrele(newvp);
2252 			newvp = NULL;
2253 		}
2254 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2255 			cnp->cn_td, &np);
2256 		if (!error) {
2257 			newvp = NFSTOV(np);
2258 			if (newvp->v_type != VDIR)
2259 				error = EEXIST;
2260 		}
2261 	}
2262 	if (error) {
2263 		if (newvp)
2264 			vrele(newvp);
2265 	} else {
2266 		*ap->a_vpp = newvp;
2267 	}
2268 	lwkt_reltoken(&nmp->nm_token);
2269 	return (error);
2270 }
2271 
2272 /*
2273  * nfs remove directory call
2274  *
2275  * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2276  *	     struct componentname *a_cnp)
2277  */
2278 static int
2279 nfs_rmdir(struct vop_old_rmdir_args *ap)
2280 {
2281 	struct vnode *vp = ap->a_vp;
2282 	struct vnode *dvp = ap->a_dvp;
2283 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2284 	struct componentname *cnp = ap->a_cnp;
2285 	int error = 0, wccflag = NFSV3_WCCRATTR;
2286 	struct nfsm_info info;
2287 
2288 	info.mrep = NULL;
2289 	info.v3 = NFS_ISV3(dvp);
2290 
2291 	if (dvp == vp)
2292 		return (EINVAL);
2293 
2294 	lwkt_gettoken(&nmp->nm_token);
2295 
2296 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2297 	nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2298 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2299 		     nfsm_rndup(cnp->cn_namelen));
2300 	ERROROUT(nfsm_fhtom(&info, dvp));
2301 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2302 		 NFS_MAXNAMLEN));
2303 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2304 				cnp->cn_cred, &error));
2305 	if (info.v3) {
2306 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2307 	}
2308 	m_freem(info.mrep);
2309 	info.mrep = NULL;
2310 nfsmout:
2311 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2312 	if (!wccflag)
2313 		VTONFS(dvp)->n_attrstamp = 0;
2314 	/*
2315 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2316 	 */
2317 	if (error == ENOENT)
2318 		error = 0;
2319 	lwkt_reltoken(&nmp->nm_token);
2320 
2321 	return (error);
2322 }
2323 
2324 /*
2325  * nfs readdir call
2326  *
2327  * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2328  */
2329 static int
2330 nfs_readdir(struct vop_readdir_args *ap)
2331 {
2332 	struct vnode *vp = ap->a_vp;
2333 	struct nfsnode *np = VTONFS(vp);
2334 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2335 	struct uio *uio = ap->a_uio;
2336 	int tresid, error;
2337 	struct vattr vattr;
2338 
2339 	if (vp->v_type != VDIR)
2340 		return (EPERM);
2341 
2342 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2343 		return (error);
2344 
2345 	lwkt_gettoken(&nmp->nm_token);
2346 
2347 	/*
2348 	 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2349 	 * and then check that is still valid, or if this is an NQNFS mount
2350 	 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR().  Note that
2351 	 * VOP_GETATTR() does not necessarily go to the wire.
2352 	 */
2353 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2354 	    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2355 		if (VOP_GETATTR(vp, &vattr) == 0 &&
2356 		    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2357 		) {
2358 			nfsstats.direofcache_hits++;
2359 			goto done;
2360 		}
2361 	}
2362 
2363 	/*
2364 	 * Call nfs_bioread() to do the real work.  nfs_bioread() does its
2365 	 * own cache coherency checks so we do not have to.
2366 	 */
2367 	tresid = uio->uio_resid;
2368 	error = nfs_bioread(vp, uio, 0);
2369 
2370 	if (!error && uio->uio_resid == tresid)
2371 		nfsstats.direofcache_misses++;
2372 done:
2373 	lwkt_reltoken(&nmp->nm_token);
2374 	vn_unlock(vp);
2375 
2376 	return (error);
2377 }
2378 
2379 /*
2380  * Readdir rpc call.  nfs_bioread->nfs_doio->nfs_readdirrpc.
2381  *
2382  * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2383  * offset/block and converts the nfs formatted directory entries for userland
2384  * consumption as well as deals with offsets into the middle of blocks.
2385  * nfs_doio only deals with logical blocks.  In particular, uio_offset will
2386  * be block-bounded.  It must convert to cookies for the actual RPC.
2387  */
2388 int
2389 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2390 {
2391 	int len, left;
2392 	struct nfs_dirent *dp = NULL;
2393 	u_int32_t *tl;
2394 	nfsuint64 *cookiep;
2395 	caddr_t cp;
2396 	nfsuint64 cookie;
2397 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2398 	struct nfsnode *dnp = VTONFS(vp);
2399 	u_quad_t fileno;
2400 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2401 	int attrflag;
2402 	struct nfsm_info info;
2403 
2404 	info.mrep = NULL;
2405 	info.v3 = NFS_ISV3(vp);
2406 
2407 #ifndef DIAGNOSTIC
2408 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2409 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2410 		panic("nfs readdirrpc bad uio");
2411 #endif
2412 
2413 	/*
2414 	 * If there is no cookie, assume directory was stale.
2415 	 */
2416 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2417 	if (cookiep)
2418 		cookie = *cookiep;
2419 	else
2420 		return (NFSERR_BAD_COOKIE);
2421 	/*
2422 	 * Loop around doing readdir rpc's of size nm_readdirsize
2423 	 * truncated to a multiple of DIRBLKSIZ.
2424 	 * The stopping criteria is EOF or buffer full.
2425 	 */
2426 	while (more_dirs && bigenough) {
2427 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2428 		nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2429 			     NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2430 		ERROROUT(nfsm_fhtom(&info, vp));
2431 		if (info.v3) {
2432 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2433 			*tl++ = cookie.nfsuquad[0];
2434 			*tl++ = cookie.nfsuquad[1];
2435 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2436 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2437 		} else {
2438 			/*
2439 			 * WARNING!  HAMMER DIRECTORIES WILL NOT WORK WELL
2440 			 * WITH NFSv2!!!  There's nothing I can really do
2441 			 * about it other than to hope the server supports
2442 			 * rdirplus w/NFSv2.
2443 			 */
2444 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2445 			*tl++ = cookie.nfsuquad[0];
2446 		}
2447 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2448 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2449 					uiop->uio_td,
2450 					nfs_vpcred(vp, ND_READ), &error));
2451 		if (info.v3) {
2452 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2453 						  NFS_LATTR_NOSHRINK));
2454 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2455 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2456 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2457 		}
2458 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2459 		more_dirs = fxdr_unsigned(int, *tl);
2460 
2461 		/* loop thru the dir entries, converting them to std form */
2462 		while (more_dirs && bigenough) {
2463 			if (info.v3) {
2464 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2465 				fileno = fxdr_hyper(tl);
2466 				len = fxdr_unsigned(int, *(tl + 2));
2467 			} else {
2468 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2469 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2470 				len = fxdr_unsigned(int, *tl);
2471 			}
2472 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2473 				error = EBADRPC;
2474 				m_freem(info.mrep);
2475 				info.mrep = NULL;
2476 				goto nfsmout;
2477 			}
2478 
2479 			/*
2480 			 * len is the number of bytes in the path element
2481 			 * name, not including the \0 termination.
2482 			 *
2483 			 * tlen is the number of bytes w have to reserve for
2484 			 * the path element name.
2485 			 */
2486 			tlen = nfsm_rndup(len);
2487 			if (tlen == len)
2488 				tlen += 4;	/* To ensure null termination */
2489 
2490 			/*
2491 			 * If the entry would cross a DIRBLKSIZ boundary,
2492 			 * extend the previous nfs_dirent to cover the
2493 			 * remaining space.
2494 			 */
2495 			left = DIRBLKSIZ - blksiz;
2496 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2497 				dp->nfs_reclen += left;
2498 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2499 				uiop->uio_iov->iov_len -= left;
2500 				uiop->uio_offset += left;
2501 				uiop->uio_resid -= left;
2502 				blksiz = 0;
2503 			}
2504 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2505 				bigenough = 0;
2506 			if (bigenough) {
2507 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2508 				dp->nfs_ino = fileno;
2509 				dp->nfs_namlen = len;
2510 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2511 				dp->nfs_type = DT_UNKNOWN;
2512 				blksiz += dp->nfs_reclen;
2513 				if (blksiz == DIRBLKSIZ)
2514 					blksiz = 0;
2515 				uiop->uio_offset += sizeof(struct nfs_dirent);
2516 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2517 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2518 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2519 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2520 
2521 				/*
2522 				 * The uiop has advanced by nfs_dirent + len
2523 				 * but really needs to advance by
2524 				 * nfs_dirent + tlen
2525 				 */
2526 				cp = uiop->uio_iov->iov_base;
2527 				tlen -= len;
2528 				*cp = '\0';	/* null terminate */
2529 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2530 				uiop->uio_iov->iov_len -= tlen;
2531 				uiop->uio_offset += tlen;
2532 				uiop->uio_resid -= tlen;
2533 			} else {
2534 				/*
2535 				 * NFS strings must be rounded up (nfsm_myouio
2536 				 * handled that in the bigenough case).
2537 				 */
2538 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2539 			}
2540 			if (info.v3) {
2541 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2542 			} else {
2543 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2544 			}
2545 
2546 			/*
2547 			 * If we were able to accomodate the last entry,
2548 			 * get the cookie for the next one.  Otherwise
2549 			 * hold-over the cookie for the one we were not
2550 			 * able to accomodate.
2551 			 */
2552 			if (bigenough) {
2553 				cookie.nfsuquad[0] = *tl++;
2554 				if (info.v3)
2555 					cookie.nfsuquad[1] = *tl++;
2556 			} else if (info.v3) {
2557 				tl += 2;
2558 			} else {
2559 				tl++;
2560 			}
2561 			more_dirs = fxdr_unsigned(int, *tl);
2562 		}
2563 		/*
2564 		 * If at end of rpc data, get the eof boolean
2565 		 */
2566 		if (!more_dirs) {
2567 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2568 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2569 		}
2570 		m_freem(info.mrep);
2571 		info.mrep = NULL;
2572 	}
2573 	/*
2574 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2575 	 * by increasing d_reclen for the last record.
2576 	 */
2577 	if (blksiz > 0) {
2578 		left = DIRBLKSIZ - blksiz;
2579 		dp->nfs_reclen += left;
2580 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2581 		uiop->uio_iov->iov_len -= left;
2582 		uiop->uio_offset += left;
2583 		uiop->uio_resid -= left;
2584 	}
2585 
2586 	if (bigenough) {
2587 		/*
2588 		 * We hit the end of the directory, update direofoffset.
2589 		 */
2590 		dnp->n_direofoffset = uiop->uio_offset;
2591 	} else {
2592 		/*
2593 		 * There is more to go, insert the link cookie so the
2594 		 * next block can be read.
2595 		 */
2596 		if (uiop->uio_resid > 0)
2597 			kprintf("EEK! readdirrpc resid > 0\n");
2598 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2599 		*cookiep = cookie;
2600 	}
2601 nfsmout:
2602 	return (error);
2603 }
2604 
2605 /*
2606  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2607  */
2608 int
2609 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2610 {
2611 	int len, left;
2612 	struct nfs_dirent *dp;
2613 	u_int32_t *tl;
2614 	struct vnode *newvp;
2615 	nfsuint64 *cookiep;
2616 	caddr_t dpossav1, dpossav2;
2617 	caddr_t cp;
2618 	struct mbuf *mdsav1, *mdsav2;
2619 	nfsuint64 cookie;
2620 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2621 	struct nfsnode *dnp = VTONFS(vp), *np;
2622 	nfsfh_t *fhp;
2623 	u_quad_t fileno;
2624 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2625 	int attrflag, fhsize;
2626 	struct nchandle nch;
2627 	struct nchandle dnch;
2628 	struct nlcomponent nlc;
2629 	struct nfsm_info info;
2630 
2631 	info.mrep = NULL;
2632 	info.v3 = 1;
2633 
2634 #ifndef nolint
2635 	dp = NULL;
2636 #endif
2637 #ifndef DIAGNOSTIC
2638 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2639 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2640 		panic("nfs readdirplusrpc bad uio");
2641 #endif
2642 	/*
2643 	 * Obtain the namecache record for the directory so we have something
2644 	 * to use as a basis for creating the entries.  This function will
2645 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2646 	 * from the tree and cannot be used for upward traversals, and the
2647 	 * ncp may be unnamed.  Note that other unrelated operations may
2648 	 * cause the ncp to be named at any time.
2649 	 *
2650 	 * We have to lock the ncp to prevent a lock order reversal when
2651 	 * rdirplus does nlookups of the children, because the vnode is
2652 	 * locked and has to stay that way.
2653 	 */
2654 	cache_fromdvp(vp, NULL, 0, &dnch);
2655 	bzero(&nlc, sizeof(nlc));
2656 	newvp = NULLVP;
2657 
2658 	/*
2659 	 * If there is no cookie, assume directory was stale.
2660 	 */
2661 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2662 	if (cookiep) {
2663 		cookie = *cookiep;
2664 	} else {
2665 		if (dnch.ncp)
2666 			cache_drop(&dnch);
2667 		return (NFSERR_BAD_COOKIE);
2668 	}
2669 
2670 	/*
2671 	 * Loop around doing readdir rpc's of size nm_readdirsize
2672 	 * truncated to a multiple of DIRBLKSIZ.
2673 	 * The stopping criteria is EOF or buffer full.
2674 	 */
2675 	while (more_dirs && bigenough) {
2676 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2677 		nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2678 			     NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2679 		ERROROUT(nfsm_fhtom(&info, vp));
2680 		tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2681 		*tl++ = cookie.nfsuquad[0];
2682 		*tl++ = cookie.nfsuquad[1];
2683 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2684 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2685 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2686 		*tl = txdr_unsigned(nmp->nm_rsize);
2687 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2688 					uiop->uio_td,
2689 					nfs_vpcred(vp, ND_READ), &error));
2690 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2691 					  NFS_LATTR_NOSHRINK));
2692 		NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2693 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2694 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2695 		more_dirs = fxdr_unsigned(int, *tl);
2696 
2697 		/* loop thru the dir entries, doctoring them to 4bsd form */
2698 		while (more_dirs && bigenough) {
2699 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2700 			fileno = fxdr_hyper(tl);
2701 			len = fxdr_unsigned(int, *(tl + 2));
2702 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2703 				error = EBADRPC;
2704 				m_freem(info.mrep);
2705 				info.mrep = NULL;
2706 				goto nfsmout;
2707 			}
2708 			tlen = nfsm_rndup(len);
2709 			if (tlen == len)
2710 				tlen += 4;	/* To ensure null termination*/
2711 			left = DIRBLKSIZ - blksiz;
2712 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2713 				dp->nfs_reclen += left;
2714 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2715 				uiop->uio_iov->iov_len -= left;
2716 				uiop->uio_offset += left;
2717 				uiop->uio_resid -= left;
2718 				blksiz = 0;
2719 			}
2720 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2721 				bigenough = 0;
2722 			if (bigenough) {
2723 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2724 				dp->nfs_ino = fileno;
2725 				dp->nfs_namlen = len;
2726 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2727 				dp->nfs_type = DT_UNKNOWN;
2728 				blksiz += dp->nfs_reclen;
2729 				if (blksiz == DIRBLKSIZ)
2730 					blksiz = 0;
2731 				uiop->uio_offset += sizeof(struct nfs_dirent);
2732 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2733 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2734 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2735 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2736 				nlc.nlc_namelen = len;
2737 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2738 				cp = uiop->uio_iov->iov_base;
2739 				tlen -= len;
2740 				*cp = '\0';
2741 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2742 				uiop->uio_iov->iov_len -= tlen;
2743 				uiop->uio_offset += tlen;
2744 				uiop->uio_resid -= tlen;
2745 			} else {
2746 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2747 			}
2748 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2749 			if (bigenough) {
2750 				cookie.nfsuquad[0] = *tl++;
2751 				cookie.nfsuquad[1] = *tl++;
2752 			} else {
2753 				tl += 2;
2754 			}
2755 
2756 			/*
2757 			 * Since the attributes are before the file handle
2758 			 * (sigh), we must skip over the attributes and then
2759 			 * come back and get them.
2760 			 */
2761 			attrflag = fxdr_unsigned(int, *tl);
2762 			if (attrflag) {
2763 			    dpossav1 = info.dpos;
2764 			    mdsav1 = info.md;
2765 			    ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2766 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2767 			    doit = fxdr_unsigned(int, *tl);
2768 			    if (doit) {
2769 				NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2770 			    }
2771 			    if (doit && bigenough && !nlcdegenerate(&nlc) &&
2772 				!NFS_CMPFH(dnp, fhp, fhsize)
2773 			    ) {
2774 				if (dnch.ncp) {
2775 #if 0
2776 				    kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2777 					nlc.nlc_namelen, nlc.nlc_namelen,
2778 					nlc.nlc_nameptr);
2779 #endif
2780 				    /*
2781 				     * This is a bit hokey but there isn't
2782 				     * much we can do about it.  We can't
2783 				     * hold the directory vp locked while
2784 				     * doing lookups and gets.
2785 				     */
2786 				    nch = cache_nlookup_nonblock(&dnch, &nlc);
2787 				    if (nch.ncp == NULL)
2788 					goto rdfail;
2789 				    cache_setunresolved(&nch);
2790 				    error = nfs_nget_nonblock(vp->v_mount, fhp,
2791 							      fhsize, &np,
2792 							      NULL);
2793 				    if (error) {
2794 					cache_put(&nch);
2795 					goto rdfail;
2796 				    }
2797 				    newvp = NFSTOV(np);
2798 				    dpossav2 = info.dpos;
2799 				    info.dpos = dpossav1;
2800 				    mdsav2 = info.md;
2801 				    info.md = mdsav1;
2802 				    ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2803 				    info.dpos = dpossav2;
2804 				    info.md = mdsav2;
2805 				    dp->nfs_type =
2806 					    IFTODT(VTTOIF(np->n_vattr.va_type));
2807 				    nfs_cache_setvp(&nch, newvp,
2808 						    nfspos_cache_timeout);
2809 				    vput(newvp);
2810 				    newvp = NULLVP;
2811 				    cache_put(&nch);
2812 				} else {
2813 rdfail:
2814 				    ;
2815 #if 0
2816 				    kprintf("Warning: NFS/rddirplus, "
2817 					    "UNABLE TO ENTER %*.*s\n",
2818 					nlc.nlc_namelen, nlc.nlc_namelen,
2819 					nlc.nlc_nameptr);
2820 #endif
2821 				}
2822 			    }
2823 			} else {
2824 			    /* Just skip over the file handle */
2825 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2826 			    i = fxdr_unsigned(int, *tl);
2827 			    ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2828 			}
2829 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2830 			more_dirs = fxdr_unsigned(int, *tl);
2831 		}
2832 		/*
2833 		 * If at end of rpc data, get the eof boolean
2834 		 */
2835 		if (!more_dirs) {
2836 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2837 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2838 		}
2839 		m_freem(info.mrep);
2840 		info.mrep = NULL;
2841 	}
2842 	/*
2843 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2844 	 * by increasing d_reclen for the last record.
2845 	 */
2846 	if (blksiz > 0) {
2847 		left = DIRBLKSIZ - blksiz;
2848 		dp->nfs_reclen += left;
2849 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2850 		uiop->uio_iov->iov_len -= left;
2851 		uiop->uio_offset += left;
2852 		uiop->uio_resid -= left;
2853 	}
2854 
2855 	/*
2856 	 * We are now either at the end of the directory or have filled the
2857 	 * block.
2858 	 */
2859 	if (bigenough) {
2860 		dnp->n_direofoffset = uiop->uio_offset;
2861 	} else {
2862 		if (uiop->uio_resid > 0)
2863 			kprintf("EEK! readdirplusrpc resid > 0\n");
2864 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2865 		*cookiep = cookie;
2866 	}
2867 nfsmout:
2868 	if (newvp != NULLVP) {
2869 	        if (newvp == vp)
2870 			vrele(newvp);
2871 		else
2872 			vput(newvp);
2873 		newvp = NULLVP;
2874 	}
2875 	if (dnch.ncp)
2876 		cache_drop(&dnch);
2877 	return (error);
2878 }
2879 
2880 /*
2881  * Silly rename. To make the NFS filesystem that is stateless look a little
2882  * more like the "ufs" a remove of an active vnode is translated to a rename
2883  * to a funny looking filename that is removed by nfs_inactive on the
2884  * nfsnode. There is the potential for another process on a different client
2885  * to create the same funny name between the nfs_lookitup() fails and the
2886  * nfs_rename() completes, but...
2887  */
2888 static int
2889 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2890 {
2891 	struct sillyrename *sp;
2892 	struct nfsnode *np;
2893 	int error;
2894 
2895 	/*
2896 	 * We previously purged dvp instead of vp.  I don't know why, it
2897 	 * completely destroys performance.  We can't do it anyway with the
2898 	 * new VFS API since we would be breaking the namecache topology.
2899 	 */
2900 	cache_purge(vp);	/* XXX */
2901 	np = VTONFS(vp);
2902 #ifndef DIAGNOSTIC
2903 	if (vp->v_type == VDIR)
2904 		panic("nfs: sillyrename dir");
2905 #endif
2906 	sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
2907 	sp->s_cred = crdup(cnp->cn_cred);
2908 	sp->s_dvp = dvp;
2909 	vref(dvp);
2910 
2911 	/* Fudge together a funny name */
2912 	sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2913 				(int)(intptr_t)cnp->cn_td);
2914 
2915 	/* Try lookitups until we get one that isn't there */
2916 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2917 		cnp->cn_td, NULL) == 0) {
2918 		sp->s_name[4]++;
2919 		if (sp->s_name[4] > 'z') {
2920 			error = EINVAL;
2921 			goto bad;
2922 		}
2923 	}
2924 	error = nfs_renameit(dvp, cnp, sp);
2925 	if (error)
2926 		goto bad;
2927 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2928 		cnp->cn_td, &np);
2929 	np->n_sillyrename = sp;
2930 	return (0);
2931 bad:
2932 	vrele(sp->s_dvp);
2933 	crfree(sp->s_cred);
2934 	kfree((caddr_t)sp, M_NFSREQ);
2935 	return (error);
2936 }
2937 
2938 /*
2939  * Look up a file name and optionally either update the file handle or
2940  * allocate an nfsnode, depending on the value of npp.
2941  * npp == NULL	--> just do the lookup
2942  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2943  *			handled too
2944  * *npp != NULL --> update the file handle in the vnode
2945  */
2946 static int
2947 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2948 	     struct thread *td, struct nfsnode **npp)
2949 {
2950 	struct vnode *newvp = NULL;
2951 	struct nfsnode *np, *dnp = VTONFS(dvp);
2952 	int error = 0, fhlen, attrflag;
2953 	nfsfh_t *nfhp;
2954 	struct nfsm_info info;
2955 
2956 	info.mrep = NULL;
2957 	info.v3 = NFS_ISV3(dvp);
2958 
2959 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2960 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2961 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2962 	ERROROUT(nfsm_fhtom(&info, dvp));
2963 	ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2964 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2965 	if (npp && !error) {
2966 		NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2967 		if (*npp) {
2968 		    np = *npp;
2969 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2970 			kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2971 			np->n_fhp = &np->n_fh;
2972 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2973 			np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2974 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2975 		    np->n_fhsize = fhlen;
2976 		    newvp = NFSTOV(np);
2977 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2978 		    vref(dvp);
2979 		    newvp = dvp;
2980 		} else {
2981 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, NULL);
2982 		    if (error) {
2983 			m_freem(info.mrep);
2984 			info.mrep = NULL;
2985 			return (error);
2986 		    }
2987 		    newvp = NFSTOV(np);
2988 		}
2989 		if (info.v3) {
2990 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2991 						  NFS_LATTR_NOSHRINK));
2992 			if (!attrflag && *npp == NULL) {
2993 				m_freem(info.mrep);
2994 				info.mrep = NULL;
2995 				if (newvp == dvp)
2996 					vrele(newvp);
2997 				else
2998 					vput(newvp);
2999 				return (ENOENT);
3000 			}
3001 		} else {
3002 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3003 		}
3004 	}
3005 	m_freem(info.mrep);
3006 	info.mrep = NULL;
3007 nfsmout:
3008 	if (npp && *npp == NULL) {
3009 		if (error) {
3010 			if (newvp) {
3011 				if (newvp == dvp)
3012 					vrele(newvp);
3013 				else
3014 					vput(newvp);
3015 			}
3016 		} else
3017 			*npp = np;
3018 	}
3019 	return (error);
3020 }
3021 
3022 /*
3023  * Nfs Version 3 commit rpc
3024  *
3025  * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3026  * involved.
3027  */
3028 int
3029 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3030 {
3031 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3032 	int error = 0, wccflag = NFSV3_WCCRATTR;
3033 	struct nfsm_info info;
3034 	u_int32_t *tl;
3035 
3036 	info.mrep = NULL;
3037 	info.v3 = 1;
3038 
3039 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3040 		return (0);
3041 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3042 	nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3043 	ERROROUT(nfsm_fhtom(&info, vp));
3044 	tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3045 	txdr_hyper(offset, tl);
3046 	tl += 2;
3047 	*tl = txdr_unsigned(cnt);
3048 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3049 				nfs_vpcred(vp, ND_WRITE), &error));
3050 	ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3051 	if (!error) {
3052 		NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3053 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3054 			NFSX_V3WRITEVERF)) {
3055 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3056 				NFSX_V3WRITEVERF);
3057 			error = NFSERR_STALEWRITEVERF;
3058 		}
3059 	}
3060 	m_freem(info.mrep);
3061 	info.mrep = NULL;
3062 nfsmout:
3063 	return (error);
3064 }
3065 
3066 /*
3067  * Kludge City..
3068  * - make nfs_bmap() essentially a no-op that does no translation
3069  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3070  *   (Maybe I could use the process's page mapping, but I was concerned that
3071  *    Kernel Write might not be enabled and also figured copyout() would do
3072  *    a lot more work than bcopy() and also it currently happens in the
3073  *    context of the swapper process (2).
3074  *
3075  * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3076  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
3077  */
3078 static int
3079 nfs_bmap(struct vop_bmap_args *ap)
3080 {
3081 	/* no token lock required */
3082 	if (ap->a_doffsetp != NULL)
3083 		*ap->a_doffsetp = ap->a_loffset;
3084 	if (ap->a_runp != NULL)
3085 		*ap->a_runp = 0;
3086 	if (ap->a_runb != NULL)
3087 		*ap->a_runb = 0;
3088 	return (0);
3089 }
3090 
3091 /*
3092  * Strategy routine.
3093  */
3094 static int
3095 nfs_strategy(struct vop_strategy_args *ap)
3096 {
3097 	struct bio *bio = ap->a_bio;
3098 	struct bio *nbio;
3099 	struct buf *bp __debugvar = bio->bio_buf;
3100 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3101 	struct thread *td;
3102 	int error;
3103 
3104 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
3105 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
3106 	KASSERT(BUF_REFCNT(bp) > 0,
3107 		("nfs_strategy: buffer %p not locked", bp));
3108 
3109 	if (bio->bio_flags & BIO_SYNC)
3110 		td = curthread;	/* XXX */
3111 	else
3112 		td = NULL;
3113 
3114 	lwkt_gettoken(&nmp->nm_token);
3115 
3116         /*
3117 	 * We probably don't need to push an nbio any more since no
3118 	 * block conversion is required due to the use of 64 bit byte
3119 	 * offsets, but do it anyway.
3120 	 *
3121 	 * NOTE: When NFS callers itself via this strategy routines and
3122 	 *	 sets up a synchronous I/O, it expects the I/O to run
3123 	 *	 synchronously (its bio_done routine just assumes it),
3124 	 *	 so for now we have to honor the bit.
3125          */
3126 	nbio = push_bio(bio);
3127 	nbio->bio_offset = bio->bio_offset;
3128 	nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3129 
3130 	/*
3131 	 * If the op is asynchronous and an i/o daemon is waiting
3132 	 * queue the request, wake it up and wait for completion
3133 	 * otherwise just do it ourselves.
3134 	 */
3135 	if (bio->bio_flags & BIO_SYNC) {
3136 		error = nfs_doio(ap->a_vp, nbio, td);
3137 	} else {
3138 		nfs_asyncio(ap->a_vp, nbio);
3139 		error = 0;
3140 	}
3141 	lwkt_reltoken(&nmp->nm_token);
3142 
3143 	return (error);
3144 }
3145 
3146 /*
3147  * Mmap a file
3148  *
3149  * NB Currently unsupported.
3150  *
3151  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3152  */
3153 /* ARGSUSED */
3154 static int
3155 nfs_mmap(struct vop_mmap_args *ap)
3156 {
3157 	/* no token lock required */
3158 	return (EINVAL);
3159 }
3160 
3161 /*
3162  * fsync vnode op. Just call nfs_flush() with commit == 1.
3163  *
3164  * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3165  */
3166 /* ARGSUSED */
3167 static int
3168 nfs_fsync(struct vop_fsync_args *ap)
3169 {
3170 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3171 	int error;
3172 
3173 	lwkt_gettoken(&nmp->nm_token);
3174 	error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3175 	lwkt_reltoken(&nmp->nm_token);
3176 
3177 	return error;
3178 }
3179 
3180 /*
3181  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
3182  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
3183  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
3184  * set the buffer contains data that has already been written to the server
3185  * and which now needs a commit RPC.
3186  *
3187  * If commit is 0 we only take one pass and only flush buffers containing new
3188  * dirty data.
3189  *
3190  * If commit is 1 we take two passes, issuing a commit RPC in the second
3191  * pass.
3192  *
3193  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3194  * to completely flush all pending data.
3195  *
3196  * Note that the RB_SCAN code properly handles the case where the
3197  * callback might block and directly or indirectly (another thread) cause
3198  * the RB tree to change.
3199  */
3200 
3201 #ifndef NFS_COMMITBVECSIZ
3202 #define NFS_COMMITBVECSIZ	16
3203 #endif
3204 
3205 struct nfs_flush_info {
3206 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3207 	struct thread *td;
3208 	struct vnode *vp;
3209 	int waitfor;
3210 	int slpflag;
3211 	int slptimeo;
3212 	int loops;
3213 	struct buf *bvary[NFS_COMMITBVECSIZ];
3214 	int bvsize;
3215 	off_t beg_off;
3216 	off_t end_off;
3217 };
3218 
3219 static int nfs_flush_bp(struct buf *bp, void *data);
3220 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3221 
3222 int
3223 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3224 {
3225 	struct nfsnode *np = VTONFS(vp);
3226 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3227 	struct nfs_flush_info info;
3228 	int error;
3229 
3230 	bzero(&info, sizeof(info));
3231 	info.td = td;
3232 	info.vp = vp;
3233 	info.waitfor = waitfor;
3234 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3235 	info.loops = 0;
3236 	lwkt_gettoken(&vp->v_token);
3237 
3238 	do {
3239 		/*
3240 		 * Flush mode
3241 		 */
3242 		info.mode = NFI_FLUSHNEW;
3243 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3244 				nfs_flush_bp, &info);
3245 
3246 		/*
3247 		 * Take a second pass if committing and no error occured.
3248 		 * Clean up any left over collection (whether an error
3249 		 * occurs or not).
3250 		 */
3251 		if (commit && error == 0) {
3252 			info.mode = NFI_COMMIT;
3253 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3254 					nfs_flush_bp, &info);
3255 			if (info.bvsize)
3256 				error = nfs_flush_docommit(&info, error);
3257 		}
3258 
3259 		/*
3260 		 * Wait for pending I/O to complete before checking whether
3261 		 * any further dirty buffers exist.
3262 		 */
3263 		while (waitfor == MNT_WAIT &&
3264 		       bio_track_active(&vp->v_track_write)) {
3265 			error = bio_track_wait(&vp->v_track_write,
3266 					       info.slpflag, info.slptimeo);
3267 			if (error) {
3268 				/*
3269 				 * We have to be able to break out if this
3270 				 * is an 'intr' mount.
3271 				 */
3272 				if (nfs_sigintr(nmp, NULL, td)) {
3273 					error = -EINTR;
3274 					break;
3275 				}
3276 
3277 				/*
3278 				 * Since we do not process pending signals,
3279 				 * once we get a PCATCH our tsleep() will no
3280 				 * longer sleep, switch to a fixed timeout
3281 				 * instead.
3282 				 */
3283 				if (info.slpflag == PCATCH) {
3284 					info.slpflag = 0;
3285 					info.slptimeo = 2 * hz;
3286 				}
3287 				error = 0;
3288 			}
3289 		}
3290 		++info.loops;
3291 		/*
3292 		 * Loop if we are flushing synchronous as well as committing,
3293 		 * and dirty buffers are still present.  Otherwise we might livelock.
3294 		 */
3295 	} while (waitfor == MNT_WAIT && commit &&
3296 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3297 
3298 	/*
3299 	 * The callbacks have to return a negative error to terminate the
3300 	 * RB scan.
3301 	 */
3302 	if (error < 0)
3303 		error = -error;
3304 
3305 	/*
3306 	 * Deal with any error collection
3307 	 */
3308 	if (np->n_flag & NWRITEERR) {
3309 		error = np->n_error;
3310 		np->n_flag &= ~NWRITEERR;
3311 	}
3312 	lwkt_reltoken(&vp->v_token);
3313 	return (error);
3314 }
3315 
3316 static
3317 int
3318 nfs_flush_bp(struct buf *bp, void *data)
3319 {
3320 	struct nfs_flush_info *info = data;
3321 	int lkflags;
3322 	int error;
3323 	off_t toff;
3324 
3325 	error = 0;
3326 	switch(info->mode) {
3327 	case NFI_FLUSHNEW:
3328 		error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3329 		if (error && info->loops && info->waitfor == MNT_WAIT) {
3330 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3331 			if (error) {
3332 				lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3333 				if (info->slpflag & PCATCH)
3334 					lkflags |= LK_PCATCH;
3335 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3336 						     info->slptimeo);
3337 			}
3338 		}
3339 
3340 		/*
3341 		 * Ignore locking errors
3342 		 */
3343 		if (error) {
3344 			error = 0;
3345 			break;
3346 		}
3347 
3348 		/*
3349 		 * The buffer may have changed out from under us, even if
3350 		 * we did not block (MPSAFE).  Check again now that it is
3351 		 * locked.
3352 		 */
3353 		if (bp->b_vp == info->vp &&
3354 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3355 			bremfree(bp);
3356 			bawrite(bp);
3357 		} else {
3358 			BUF_UNLOCK(bp);
3359 		}
3360 		break;
3361 	case NFI_COMMIT:
3362 		/*
3363 		 * Only process buffers in need of a commit which we can
3364 		 * immediately lock.  This may prevent a buffer from being
3365 		 * committed, but the normal flush loop will block on the
3366 		 * same buffer so we shouldn't get into an endless loop.
3367 		 */
3368 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3369 		    (B_DELWRI | B_NEEDCOMMIT)) {
3370 			break;
3371 		}
3372 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3373 			break;
3374 
3375 		/*
3376 		 * We must recheck after successfully locking the buffer.
3377 		 */
3378 		if (bp->b_vp != info->vp ||
3379 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3380 		    (B_DELWRI | B_NEEDCOMMIT)) {
3381 			BUF_UNLOCK(bp);
3382 			break;
3383 		}
3384 
3385 		/*
3386 		 * NOTE: storing the bp in the bvary[] basically sets
3387 		 * it up for a commit operation.
3388 		 *
3389 		 * We must call vfs_busy_pages() now so the commit operation
3390 		 * is interlocked with user modifications to memory mapped
3391 		 * pages.  The b_dirtyoff/b_dirtyend range is not correct
3392 		 * until after the pages have been busied.
3393 		 *
3394 		 * Note: to avoid loopback deadlocks, we do not
3395 		 * assign b_runningbufspace.
3396 		 */
3397 		bremfree(bp);
3398 		bp->b_cmd = BUF_CMD_WRITE;
3399 		vfs_busy_pages(bp->b_vp, bp);
3400 		info->bvary[info->bvsize] = bp;
3401 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3402 		if (info->bvsize == 0 || toff < info->beg_off)
3403 			info->beg_off = toff;
3404 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3405 		if (info->bvsize == 0 || toff > info->end_off)
3406 			info->end_off = toff;
3407 		++info->bvsize;
3408 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3409 			error = nfs_flush_docommit(info, 0);
3410 			KKASSERT(info->bvsize == 0);
3411 		}
3412 	}
3413 	return (error);
3414 }
3415 
3416 static
3417 int
3418 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3419 {
3420 	struct vnode *vp;
3421 	struct buf *bp;
3422 	off_t bytes;
3423 	int retv;
3424 	int i;
3425 
3426 	vp = info->vp;
3427 
3428 	if (info->bvsize > 0) {
3429 		/*
3430 		 * Commit data on the server, as required.  Note that
3431 		 * nfs_commit will use the vnode's cred for the commit.
3432 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3433 		 */
3434 		bytes = info->end_off - info->beg_off;
3435 		if (bytes > 0x40000000)
3436 			bytes = 0x40000000;
3437 		if (error) {
3438 			retv = -error;
3439 		} else {
3440 			retv = nfs_commitrpc_uio(vp, info->beg_off,
3441 						 (int)bytes, info->td);
3442 			if (retv == NFSERR_STALEWRITEVERF)
3443 				nfs_clearcommit(vp->v_mount);
3444 		}
3445 
3446 		/*
3447 		 * Now, either mark the blocks I/O done or mark the
3448 		 * blocks dirty, depending on whether the commit
3449 		 * succeeded.
3450 		 */
3451 		for (i = 0; i < info->bvsize; ++i) {
3452 			bp = info->bvary[i];
3453 			if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3454 				/*
3455 				 * Either an error or the original
3456 				 * vfs_busy_pages() cleared B_NEEDCOMMIT
3457 				 * due to finding new dirty VM pages in
3458 				 * the buffer.
3459 				 *
3460 				 * Leave B_DELWRI intact.
3461 				 */
3462 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3463 				vfs_unbusy_pages(bp);
3464 				bp->b_cmd = BUF_CMD_DONE;
3465 				bqrelse(bp);
3466 			} else {
3467 				/*
3468 				 * Success, remove B_DELWRI ( bundirty() ).
3469 				 *
3470 				 * b_dirtyoff/b_dirtyend seem to be NFS
3471 				 * specific.  We should probably move that
3472 				 * into bundirty(). XXX
3473 				 *
3474 				 * We are faking an I/O write, we have to
3475 				 * start the transaction in order to
3476 				 * immediately biodone() it.
3477 				 */
3478 				bundirty(bp);
3479 				bp->b_flags &= ~B_ERROR;
3480 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3481 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3482 				biodone(&bp->b_bio1);
3483 			}
3484 		}
3485 		info->bvsize = 0;
3486 	}
3487 	return (error);
3488 }
3489 
3490 /*
3491  * NFS advisory byte-level locks.
3492  * Currently unsupported.
3493  *
3494  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3495  *		int a_flags)
3496  */
3497 static int
3498 nfs_advlock(struct vop_advlock_args *ap)
3499 {
3500 	struct nfsnode *np = VTONFS(ap->a_vp);
3501 
3502 	/* no token lock currently required */
3503 	/*
3504 	 * The following kludge is to allow diskless support to work
3505 	 * until a real NFS lockd is implemented. Basically, just pretend
3506 	 * that this is a local lock.
3507 	 */
3508 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3509 }
3510 
3511 /*
3512  * Print out the contents of an nfsnode.
3513  *
3514  * nfs_print(struct vnode *a_vp)
3515  */
3516 static int
3517 nfs_print(struct vop_print_args *ap)
3518 {
3519 	struct vnode *vp = ap->a_vp;
3520 	struct nfsnode *np = VTONFS(vp);
3521 
3522 	kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3523 		(long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3524 	if (vp->v_type == VFIFO)
3525 		fifo_printinfo(vp);
3526 	kprintf("\n");
3527 	return (0);
3528 }
3529 
3530 /*
3531  * nfs special file access vnode op.
3532  *
3533  * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3534  */
3535 static int
3536 nfs_laccess(struct vop_access_args *ap)
3537 {
3538 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3539 	struct vattr vattr;
3540 	int error;
3541 
3542 	lwkt_gettoken(&nmp->nm_token);
3543 	error = VOP_GETATTR(ap->a_vp, &vattr);
3544 	if (error == 0) {
3545 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3546 					  vattr.va_mode, 0);
3547 	}
3548 	lwkt_reltoken(&nmp->nm_token);
3549 
3550 	return (error);
3551 }
3552 
3553 /*
3554  * Read wrapper for fifos.
3555  *
3556  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3557  *		struct ucred *a_cred)
3558  */
3559 static int
3560 nfsfifo_read(struct vop_read_args *ap)
3561 {
3562 	struct nfsnode *np = VTONFS(ap->a_vp);
3563 
3564 	/* no token access required */
3565 	/*
3566 	 * Set access flag.
3567 	 */
3568 	np->n_flag |= NACC;
3569 	getnanotime(&np->n_atim);
3570 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3571 }
3572 
3573 /*
3574  * Write wrapper for fifos.
3575  *
3576  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3577  *		 struct ucred *a_cred)
3578  */
3579 static int
3580 nfsfifo_write(struct vop_write_args *ap)
3581 {
3582 	struct nfsnode *np = VTONFS(ap->a_vp);
3583 
3584 	/* no token access required */
3585 	/*
3586 	 * Set update flag.
3587 	 */
3588 	np->n_flag |= NUPD;
3589 	getnanotime(&np->n_mtim);
3590 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3591 }
3592 
3593 /*
3594  * Close wrapper for fifos.
3595  *
3596  * Update the times on the nfsnode then do fifo close.
3597  *
3598  * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3599  */
3600 static int
3601 nfsfifo_close(struct vop_close_args *ap)
3602 {
3603 	struct vnode *vp = ap->a_vp;
3604 	struct nfsnode *np = VTONFS(vp);
3605 	struct vattr vattr;
3606 	struct timespec ts;
3607 
3608 	/* no token access required */
3609 
3610 	if (np->n_flag & (NACC | NUPD)) {
3611 		getnanotime(&ts);
3612 		if (np->n_flag & NACC)
3613 			np->n_atim = ts;
3614 		if (np->n_flag & NUPD)
3615 			np->n_mtim = ts;
3616 		np->n_flag |= NCHG;
3617 		if (vp->v_sysref.refcnt == 1 &&
3618 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3619 			VATTR_NULL(&vattr);
3620 			if (np->n_flag & NACC)
3621 				vattr.va_atime = np->n_atim;
3622 			if (np->n_flag & NUPD)
3623 				vattr.va_mtime = np->n_mtim;
3624 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3625 		}
3626 	}
3627 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3628 }
3629 
3630