xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 7aa383846770374466b1dcb2cefd71bde9acf463)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_inet.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/jail.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 
70 
71 #include <fs/nfs/nfsport.h>
72 #include <fs/nfsclient/nfsnode.h>
73 #include <fs/nfsclient/nfsmount.h>
74 #include <fs/nfsclient/nfs.h>
75 #include <fs/nfsclient/nfs_lock.h>
76 
77 #include <net/if.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 
81 /* Defs */
82 #define	TRUE	1
83 #define	FALSE	0
84 
85 extern struct nfsstats newnfsstats;
86 MALLOC_DECLARE(M_NEWNFSREQ);
87 vop_advlock_t	*ncl_advlock_p = ncl_dolock;
88 
89 /*
90  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91  * calls are not in getblk() and brelse() so that they would not be necessary
92  * here.
93  */
94 #ifndef B_VMIO
95 #define	vfs_busy_pages(bp, f)
96 #endif
97 
98 static vop_read_t	nfsfifo_read;
99 static vop_write_t	nfsfifo_write;
100 static vop_close_t	nfsfifo_close;
101 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 		    struct thread *);
103 static vop_lookup_t	nfs_lookup;
104 static vop_create_t	nfs_create;
105 static vop_mknod_t	nfs_mknod;
106 static vop_open_t	nfs_open;
107 static vop_close_t	nfs_close;
108 static vop_access_t	nfs_access;
109 static vop_getattr_t	nfs_getattr;
110 static vop_setattr_t	nfs_setattr;
111 static vop_read_t	nfs_read;
112 static vop_fsync_t	nfs_fsync;
113 static vop_remove_t	nfs_remove;
114 static vop_link_t	nfs_link;
115 static vop_rename_t	nfs_rename;
116 static vop_mkdir_t	nfs_mkdir;
117 static vop_rmdir_t	nfs_rmdir;
118 static vop_symlink_t	nfs_symlink;
119 static vop_readdir_t	nfs_readdir;
120 static vop_strategy_t	nfs_strategy;
121 static vop_lock1_t	nfs_lock1;
122 static	int	nfs_lookitup(struct vnode *, char *, int,
123 		    struct ucred *, struct thread *, struct nfsnode **);
124 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
125 		    struct componentname *);
126 static vop_access_t	nfsspec_access;
127 static vop_readlink_t	nfs_readlink;
128 static vop_print_t	nfs_print;
129 static vop_advlock_t	nfs_advlock;
130 static vop_advlockasync_t nfs_advlockasync;
131 static vop_getacl_t nfs_getacl;
132 static vop_setacl_t nfs_setacl;
133 
134 /*
135  * Global vfs data structures for nfs
136  */
137 struct vop_vector newnfs_vnodeops = {
138 	.vop_default =		&default_vnodeops,
139 	.vop_access =		nfs_access,
140 	.vop_advlock =		nfs_advlock,
141 	.vop_advlockasync =	nfs_advlockasync,
142 	.vop_close =		nfs_close,
143 	.vop_create =		nfs_create,
144 	.vop_fsync =		nfs_fsync,
145 	.vop_getattr =		nfs_getattr,
146 	.vop_getpages =		ncl_getpages,
147 	.vop_putpages =		ncl_putpages,
148 	.vop_inactive =		ncl_inactive,
149 	.vop_link =		nfs_link,
150 	.vop_lock1 = 		nfs_lock1,
151 	.vop_lookup =		nfs_lookup,
152 	.vop_mkdir =		nfs_mkdir,
153 	.vop_mknod =		nfs_mknod,
154 	.vop_open =		nfs_open,
155 	.vop_print =		nfs_print,
156 	.vop_read =		nfs_read,
157 	.vop_readdir =		nfs_readdir,
158 	.vop_readlink =		nfs_readlink,
159 	.vop_reclaim =		ncl_reclaim,
160 	.vop_remove =		nfs_remove,
161 	.vop_rename =		nfs_rename,
162 	.vop_rmdir =		nfs_rmdir,
163 	.vop_setattr =		nfs_setattr,
164 	.vop_strategy =		nfs_strategy,
165 	.vop_symlink =		nfs_symlink,
166 	.vop_write =		ncl_write,
167 	.vop_getacl =		nfs_getacl,
168 	.vop_setacl =		nfs_setacl,
169 };
170 
171 struct vop_vector newnfs_fifoops = {
172 	.vop_default =		&fifo_specops,
173 	.vop_access =		nfsspec_access,
174 	.vop_close =		nfsfifo_close,
175 	.vop_fsync =		nfs_fsync,
176 	.vop_getattr =		nfs_getattr,
177 	.vop_inactive =		ncl_inactive,
178 	.vop_print =		nfs_print,
179 	.vop_read =		nfsfifo_read,
180 	.vop_reclaim =		ncl_reclaim,
181 	.vop_setattr =		nfs_setattr,
182 	.vop_write =		nfsfifo_write,
183 };
184 
185 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
186     struct componentname *cnp, struct vattr *vap);
187 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
188     int namelen, struct ucred *cred, struct thread *td);
189 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
190     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
191     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
192 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
193     struct componentname *scnp, struct sillyrename *sp);
194 
195 /*
196  * Global variables
197  */
198 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
199 
200 SYSCTL_DECL(_vfs_newnfs);
201 
202 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
203 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
204 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
205 
206 static int	nfs_prime_access_cache = 0;
207 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
208 	   &nfs_prime_access_cache, 0,
209 	   "Prime NFS ACCESS cache when fetching attributes");
210 
211 static int	newnfs_commit_on_close = 0;
212 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
213     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
214 
215 static int	nfs_clean_pages_on_close = 1;
216 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
217 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
218 
219 int newnfs_directio_enable = 0;
220 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
221 	   &newnfs_directio_enable, 0, "Enable NFS directio");
222 
223 /*
224  * This sysctl allows other processes to mmap a file that has been opened
225  * O_DIRECT by a process.  In general, having processes mmap the file while
226  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
227  * this by default to prevent DoS attacks - to prevent a malicious user from
228  * opening up files O_DIRECT preventing other users from mmap'ing these
229  * files.  "Protected" environments where stricter consistency guarantees are
230  * required can disable this knob.  The process that opened the file O_DIRECT
231  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
232  * meaningful.
233  */
234 int newnfs_directio_allow_mmap = 1;
235 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
236 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
237 
238 #if 0
239 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
240 	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
241 
242 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
243 	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
244 #endif
245 
246 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
247 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
248 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
249 
250 /*
251  * SMP Locking Note :
252  * The list of locks after the description of the lock is the ordering
253  * of other locks acquired with the lock held.
254  * np->n_mtx : Protects the fields in the nfsnode.
255        VM Object Lock
256        VI_MTX (acquired indirectly)
257  * nmp->nm_mtx : Protects the fields in the nfsmount.
258        rep->r_mtx
259  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
260  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
261        nmp->nm_mtx
262        rep->r_mtx
263  * rep->r_mtx : Protects the fields in an nfsreq.
264  */
265 
266 static int
267 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
268     struct ucred *cred, u_int32_t *retmode)
269 {
270 	int error = 0, attrflag, i, lrupos;
271 	u_int32_t rmode;
272 	struct nfsnode *np = VTONFS(vp);
273 	struct nfsvattr nfsva;
274 
275 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
276 	    &rmode, NULL);
277 	if (attrflag)
278 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
279 	if (!error) {
280 		lrupos = 0;
281 		mtx_lock(&np->n_mtx);
282 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
283 			if (np->n_accesscache[i].uid == cred->cr_uid) {
284 				np->n_accesscache[i].mode = rmode;
285 				np->n_accesscache[i].stamp = time_second;
286 				break;
287 			}
288 			if (i > 0 && np->n_accesscache[i].stamp <
289 			    np->n_accesscache[lrupos].stamp)
290 				lrupos = i;
291 		}
292 		if (i == NFS_ACCESSCACHESIZE) {
293 			np->n_accesscache[lrupos].uid = cred->cr_uid;
294 			np->n_accesscache[lrupos].mode = rmode;
295 			np->n_accesscache[lrupos].stamp = time_second;
296 		}
297 		mtx_unlock(&np->n_mtx);
298 		if (retmode != NULL)
299 			*retmode = rmode;
300 	} else if (NFS_ISV4(vp)) {
301 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
302 	}
303 	return (error);
304 }
305 
306 /*
307  * nfs access vnode op.
308  * For nfs version 2, just return ok. File accesses may fail later.
309  * For nfs version 3, use the access rpc to check accessibility. If file modes
310  * are changed on the server, accesses might still fail later.
311  */
312 static int
313 nfs_access(struct vop_access_args *ap)
314 {
315 	struct vnode *vp = ap->a_vp;
316 	int error = 0, i, gotahit;
317 	u_int32_t mode, wmode, rmode;
318 	int v34 = NFS_ISV34(vp);
319 	struct nfsnode *np = VTONFS(vp);
320 
321 	/*
322 	 * Disallow write attempts on filesystems mounted read-only;
323 	 * unless the file is a socket, fifo, or a block or character
324 	 * device resident on the filesystem.
325 	 */
326 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
327 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
328 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
329 		switch (vp->v_type) {
330 		case VREG:
331 		case VDIR:
332 		case VLNK:
333 			return (EROFS);
334 		default:
335 			break;
336 		}
337 	}
338 	/*
339 	 * For nfs v3 or v4, check to see if we have done this recently, and if
340 	 * so return our cached result instead of making an ACCESS call.
341 	 * If not, do an access rpc, otherwise you are stuck emulating
342 	 * ufs_access() locally using the vattr. This may not be correct,
343 	 * since the server may apply other access criteria such as
344 	 * client uid-->server uid mapping that we do not know about.
345 	 */
346 	if (v34) {
347 		if (ap->a_accmode & VREAD)
348 			mode = NFSACCESS_READ;
349 		else
350 			mode = 0;
351 		if (vp->v_type != VDIR) {
352 			if (ap->a_accmode & VWRITE)
353 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
354 			if (ap->a_accmode & VAPPEND)
355 				mode |= NFSACCESS_EXTEND;
356 			if (ap->a_accmode & VEXEC)
357 				mode |= NFSACCESS_EXECUTE;
358 			if (ap->a_accmode & VDELETE)
359 				mode |= NFSACCESS_DELETE;
360 		} else {
361 			if (ap->a_accmode & VWRITE)
362 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
363 			if (ap->a_accmode & VAPPEND)
364 				mode |= NFSACCESS_EXTEND;
365 			if (ap->a_accmode & VEXEC)
366 				mode |= NFSACCESS_LOOKUP;
367 			if (ap->a_accmode & VDELETE)
368 				mode |= NFSACCESS_DELETE;
369 			if (ap->a_accmode & VDELETE_CHILD)
370 				mode |= NFSACCESS_MODIFY;
371 		}
372 		/* XXX safety belt, only make blanket request if caching */
373 		if (nfsaccess_cache_timeout > 0) {
374 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
375 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
376 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
377 		} else {
378 			wmode = mode;
379 		}
380 
381 		/*
382 		 * Does our cached result allow us to give a definite yes to
383 		 * this request?
384 		 */
385 		gotahit = 0;
386 		mtx_lock(&np->n_mtx);
387 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
388 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
389 			    if (time_second < (np->n_accesscache[i].stamp
390 				+ nfsaccess_cache_timeout) &&
391 				(np->n_accesscache[i].mode & mode) == mode) {
392 				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
393 				gotahit = 1;
394 			    }
395 			    break;
396 			}
397 		}
398 		mtx_unlock(&np->n_mtx);
399 		if (gotahit == 0) {
400 			/*
401 			 * Either a no, or a don't know.  Go to the wire.
402 			 */
403 			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
404 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
405 			    ap->a_cred, &rmode);
406 			if (!error &&
407 			    (rmode & mode) != mode)
408 				error = EACCES;
409 		}
410 		return (error);
411 	} else {
412 		if ((error = nfsspec_access(ap)) != 0) {
413 			return (error);
414 		}
415 		/*
416 		 * Attempt to prevent a mapped root from accessing a file
417 		 * which it shouldn't.  We try to read a byte from the file
418 		 * if the user is root and the file is not zero length.
419 		 * After calling nfsspec_access, we should have the correct
420 		 * file size cached.
421 		 */
422 		mtx_lock(&np->n_mtx);
423 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
424 		    && VTONFS(vp)->n_size > 0) {
425 			struct iovec aiov;
426 			struct uio auio;
427 			char buf[1];
428 
429 			mtx_unlock(&np->n_mtx);
430 			aiov.iov_base = buf;
431 			aiov.iov_len = 1;
432 			auio.uio_iov = &aiov;
433 			auio.uio_iovcnt = 1;
434 			auio.uio_offset = 0;
435 			auio.uio_resid = 1;
436 			auio.uio_segflg = UIO_SYSSPACE;
437 			auio.uio_rw = UIO_READ;
438 			auio.uio_td = ap->a_td;
439 
440 			if (vp->v_type == VREG)
441 				error = ncl_readrpc(vp, &auio, ap->a_cred);
442 			else if (vp->v_type == VDIR) {
443 				char* bp;
444 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
445 				aiov.iov_base = bp;
446 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
447 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
448 				    ap->a_td);
449 				free(bp, M_TEMP);
450 			} else if (vp->v_type == VLNK)
451 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
452 			else
453 				error = EACCES;
454 		} else
455 			mtx_unlock(&np->n_mtx);
456 		return (error);
457 	}
458 }
459 
460 
461 /*
462  * nfs open vnode op
463  * Check to see if the type is ok
464  * and that deletion is not in progress.
465  * For paged in text files, you will need to flush the page cache
466  * if consistency is lost.
467  */
468 /* ARGSUSED */
469 static int
470 nfs_open(struct vop_open_args *ap)
471 {
472 	struct vnode *vp = ap->a_vp;
473 	struct nfsnode *np = VTONFS(vp);
474 	struct vattr vattr;
475 	int error;
476 	int fmode = ap->a_mode;
477 
478 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
479 		return (EOPNOTSUPP);
480 
481 	/*
482 	 * For NFSv4, we need to do the Open Op before cache validation,
483 	 * so that we conform to RFC3530 Sec. 9.3.1.
484 	 */
485 	if (NFS_ISV4(vp)) {
486 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
487 		if (error) {
488 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
489 			    (gid_t)0);
490 			return (error);
491 		}
492 	}
493 
494 	/*
495 	 * Now, if this Open will be doing reading, re-validate/flush the
496 	 * cache, so that Close/Open coherency is maintained.
497 	 */
498 	if ((fmode & FREAD) && (!NFS_ISV4(vp) || nfscl_mustflush(vp))) {
499 		mtx_lock(&np->n_mtx);
500 		if (np->n_flag & NMODIFIED) {
501 			mtx_unlock(&np->n_mtx);
502 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
503 			if (error == EINTR || error == EIO) {
504 				if (NFS_ISV4(vp))
505 					(void) nfsrpc_close(vp, 0, ap->a_td);
506 				return (error);
507 			}
508 			mtx_lock(&np->n_mtx);
509 			np->n_attrstamp = 0;
510 			if (vp->v_type == VDIR)
511 				np->n_direofoffset = 0;
512 			mtx_unlock(&np->n_mtx);
513 			error = VOP_GETATTR(vp, &vattr, ap->a_cred);
514 			if (error) {
515 				if (NFS_ISV4(vp))
516 					(void) nfsrpc_close(vp, 0, ap->a_td);
517 				return (error);
518 			}
519 			mtx_lock(&np->n_mtx);
520 			np->n_mtime = vattr.va_mtime;
521 			if (NFS_ISV4(vp))
522 				np->n_change = vattr.va_filerev;
523 			mtx_unlock(&np->n_mtx);
524 		} else {
525 			mtx_unlock(&np->n_mtx);
526 			error = VOP_GETATTR(vp, &vattr, ap->a_cred);
527 			if (error) {
528 				if (NFS_ISV4(vp))
529 					(void) nfsrpc_close(vp, 0, ap->a_td);
530 				return (error);
531 			}
532 			mtx_lock(&np->n_mtx);
533 			if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
534 			    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
535 				if (vp->v_type == VDIR)
536 					np->n_direofoffset = 0;
537 				mtx_unlock(&np->n_mtx);
538 				error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
539 				if (error == EINTR || error == EIO) {
540 					if (NFS_ISV4(vp))
541 						(void) nfsrpc_close(vp, 0,
542 						    ap->a_td);
543 					return (error);
544 				}
545 				mtx_lock(&np->n_mtx);
546 				np->n_mtime = vattr.va_mtime;
547 				if (NFS_ISV4(vp))
548 					np->n_change = vattr.va_filerev;
549 			}
550 			mtx_unlock(&np->n_mtx);
551 		}
552 	}
553 
554 	/*
555 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
556 	 */
557 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
558 		if (np->n_directio_opens == 0) {
559 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
560 			if (error) {
561 				if (NFS_ISV4(vp))
562 					(void) nfsrpc_close(vp, 0, ap->a_td);
563 				return (error);
564 			}
565 			mtx_lock(&np->n_mtx);
566 			np->n_flag |= NNONCACHE;
567 		} else {
568 			mtx_lock(&np->n_mtx);
569 		}
570 		np->n_directio_opens++;
571 		mtx_unlock(&np->n_mtx);
572 	}
573 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
574 	return (0);
575 }
576 
577 /*
578  * nfs close vnode op
579  * What an NFS client should do upon close after writing is a debatable issue.
580  * Most NFS clients push delayed writes to the server upon close, basically for
581  * two reasons:
582  * 1 - So that any write errors may be reported back to the client process
583  *     doing the close system call. By far the two most likely errors are
584  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
585  * 2 - To put a worst case upper bound on cache inconsistency between
586  *     multiple clients for the file.
587  * There is also a consistency problem for Version 2 of the protocol w.r.t.
588  * not being able to tell if other clients are writing a file concurrently,
589  * since there is no way of knowing if the changed modify time in the reply
590  * is only due to the write for this client.
591  * (NFS Version 3 provides weak cache consistency data in the reply that
592  *  should be sufficient to detect and handle this case.)
593  *
594  * The current code does the following:
595  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
596  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
597  *                     or commit them (this satisfies 1 and 2 except for the
598  *                     case where the server crashes after this close but
599  *                     before the commit RPC, which is felt to be "good
600  *                     enough". Changing the last argument to ncl_flush() to
601  *                     a 1 would force a commit operation, if it is felt a
602  *                     commit is necessary now.
603  * for NFS Version 4 - flush the dirty buffers and commit them, if
604  *		       nfscl_mustflush() says this is necessary.
605  *                     It is necessary if there is no write delegation held,
606  *                     in order to satisfy open/close coherency.
607  *                     If the file isn't cached on local stable storage,
608  *                     it may be necessary in order to detect "out of space"
609  *                     errors from the server, if the write delegation
610  *                     issued by the server doesn't allow the file to grow.
611  */
612 /* ARGSUSED */
613 static int
614 nfs_close(struct vop_close_args *ap)
615 {
616 	struct vnode *vp = ap->a_vp;
617 	struct nfsnode *np = VTONFS(vp);
618 	struct nfsvattr nfsva;
619 	struct ucred *cred;
620 	int error = 0, ret, localcred = 0;
621 	int fmode = ap->a_fflag;
622 
623 	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
624 		return (0);
625 	/*
626 	 * During shutdown, a_cred isn't valid, so just use root.
627 	 */
628 	if (ap->a_cred == NOCRED) {
629 		cred = newnfs_getcred();
630 		localcred = 1;
631 	} else {
632 		cred = ap->a_cred;
633 	}
634 	if (vp->v_type == VREG) {
635 	    /*
636 	     * Examine and clean dirty pages, regardless of NMODIFIED.
637 	     * This closes a major hole in close-to-open consistency.
638 	     * We want to push out all dirty pages (and buffers) on
639 	     * close, regardless of whether they were dirtied by
640 	     * mmap'ed writes or via write().
641 	     */
642 	    if (nfs_clean_pages_on_close && vp->v_object) {
643 		VM_OBJECT_LOCK(vp->v_object);
644 		vm_object_page_clean(vp->v_object, 0, 0, 0);
645 		VM_OBJECT_UNLOCK(vp->v_object);
646 	    }
647 	    mtx_lock(&np->n_mtx);
648 	    if (np->n_flag & NMODIFIED) {
649 		mtx_unlock(&np->n_mtx);
650 		if (NFS_ISV3(vp)) {
651 		    /*
652 		     * Under NFSv3 we have dirty buffers to dispose of.  We
653 		     * must flush them to the NFS server.  We have the option
654 		     * of waiting all the way through the commit rpc or just
655 		     * waiting for the initial write.  The default is to only
656 		     * wait through the initial write so the data is in the
657 		     * server's cache, which is roughly similar to the state
658 		     * a standard disk subsystem leaves the file in on close().
659 		     *
660 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
661 		     * potential races with other processes, and certainly
662 		     * cannot clear it if we don't commit.
663 		     * These races occur when there is no longer the old
664 		     * traditional vnode locking implemented for Vnode Ops.
665 		     */
666 		    int cm = newnfs_commit_on_close ? 1 : 0;
667 		    error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
668 		    /* np->n_flag &= ~NMODIFIED; */
669 		} else if (NFS_ISV4(vp)) {
670 			if (nfscl_mustflush(vp)) {
671 				int cm = newnfs_commit_on_close ? 1 : 0;
672 				error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
673 				    cm, 0);
674 				/*
675 				 * as above w.r.t races when clearing
676 				 * NMODIFIED.
677 				 * np->n_flag &= ~NMODIFIED;
678 				 */
679 			}
680 		} else
681 		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
682 		mtx_lock(&np->n_mtx);
683 	    }
684  	    /*
685  	     * Invalidate the attribute cache in all cases.
686  	     * An open is going to fetch fresh attrs any way, other procs
687  	     * on this node that have file open will be forced to do an
688  	     * otw attr fetch, but this is safe.
689 	     * --> A user found that their RPC count dropped by 20% when
690 	     *     this was commented out and I can't see any requirement
691 	     *     for it, so I've disabled it when negative lookups are
692 	     *     enabled. (What does this have to do with negative lookup
693 	     *     caching? Well nothing, except it was reported by the
694 	     *     same user that needed negative lookup caching and I wanted
695 	     *     there to be a way to disable it to see if it
696 	     *     is the cause of some caching/coherency issue that might
697 	     *     crop up.)
698  	     */
699 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
700 		    np->n_attrstamp = 0;
701 	    if (np->n_flag & NWRITEERR) {
702 		np->n_flag &= ~NWRITEERR;
703 		error = np->n_error;
704 	    }
705 	    mtx_unlock(&np->n_mtx);
706 	}
707 
708 	if (NFS_ISV4(vp)) {
709 		/*
710 		 * Get attributes so "change" is up to date.
711 		 */
712 		if (!error) {
713 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
714 			    NULL);
715 			if (!ret) {
716 				np->n_change = nfsva.na_filerev;
717 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
718 				    NULL, 0, 0);
719 			}
720 		}
721 
722 		/*
723 		 * and do the close.
724 		 */
725 		ret = nfsrpc_close(vp, 0, ap->a_td);
726 		if (!error && ret)
727 			error = ret;
728 		if (error)
729 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
730 			    (gid_t)0);
731 	}
732 	if (newnfs_directio_enable)
733 		KASSERT((np->n_directio_asyncwr == 0),
734 			("nfs_close: dirty unflushed (%d) directio buffers\n",
735 			 np->n_directio_asyncwr));
736 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
737 		mtx_lock(&np->n_mtx);
738 		KASSERT((np->n_directio_opens > 0),
739 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
740 		np->n_directio_opens--;
741 		if (np->n_directio_opens == 0)
742 			np->n_flag &= ~NNONCACHE;
743 		mtx_unlock(&np->n_mtx);
744 	}
745 	if (localcred)
746 		NFSFREECRED(cred);
747 	return (error);
748 }
749 
750 /*
751  * nfs getattr call from vfs.
752  */
753 static int
754 nfs_getattr(struct vop_getattr_args *ap)
755 {
756 	struct vnode *vp = ap->a_vp;
757 	struct thread *td = curthread;	/* XXX */
758 	struct nfsnode *np = VTONFS(vp);
759 	int error = 0;
760 	struct nfsvattr nfsva;
761 	struct vattr *vap = ap->a_vap;
762 	struct vattr vattr;
763 
764 	/*
765 	 * Update local times for special files.
766 	 */
767 	mtx_lock(&np->n_mtx);
768 	if (np->n_flag & (NACC | NUPD))
769 		np->n_flag |= NCHG;
770 	mtx_unlock(&np->n_mtx);
771 	/*
772 	 * First look in the cache.
773 	 */
774 	if (ncl_getattrcache(vp, &vattr) == 0) {
775 		vap->va_type = vattr.va_type;
776 		vap->va_mode = vattr.va_mode;
777 		vap->va_nlink = vattr.va_nlink;
778 		vap->va_uid = vattr.va_uid;
779 		vap->va_gid = vattr.va_gid;
780 		vap->va_fsid = vattr.va_fsid;
781 		vap->va_fileid = vattr.va_fileid;
782 		vap->va_size = vattr.va_size;
783 		vap->va_blocksize = vattr.va_blocksize;
784 		vap->va_atime = vattr.va_atime;
785 		vap->va_mtime = vattr.va_mtime;
786 		vap->va_ctime = vattr.va_ctime;
787 		vap->va_gen = vattr.va_gen;
788 		vap->va_flags = vattr.va_flags;
789 		vap->va_rdev = vattr.va_rdev;
790 		vap->va_bytes = vattr.va_bytes;
791 		vap->va_filerev = vattr.va_filerev;
792 		/*
793 		 * Get the local modify time for the case of a write
794 		 * delegation.
795 		 */
796 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
797 		return (0);
798 	}
799 
800 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
801 	    nfsaccess_cache_timeout > 0) {
802 		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
803 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
804 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
805 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
806 			return (0);
807 		}
808 	}
809 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
810 	if (!error)
811 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
812 	if (!error) {
813 		/*
814 		 * Get the local modify time for the case of a write
815 		 * delegation.
816 		 */
817 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
818 	} else if (NFS_ISV4(vp)) {
819 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
820 	}
821 	return (error);
822 }
823 
824 /*
825  * nfs setattr call.
826  */
827 static int
828 nfs_setattr(struct vop_setattr_args *ap)
829 {
830 	struct vnode *vp = ap->a_vp;
831 	struct nfsnode *np = VTONFS(vp);
832 	struct thread *td = curthread;	/* XXX */
833 	struct vattr *vap = ap->a_vap;
834 	int error = 0;
835 	u_quad_t tsize;
836 
837 #ifndef nolint
838 	tsize = (u_quad_t)0;
839 #endif
840 
841 	/*
842 	 * Setting of flags and marking of atimes are not supported.
843 	 */
844 	if (vap->va_flags != VNOVAL)
845 		return (EOPNOTSUPP);
846 
847 	/*
848 	 * Disallow write attempts if the filesystem is mounted read-only.
849 	 */
850   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
851 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
852 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
853 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
854 		return (EROFS);
855 	if (vap->va_size != VNOVAL) {
856  		switch (vp->v_type) {
857  		case VDIR:
858  			return (EISDIR);
859  		case VCHR:
860  		case VBLK:
861  		case VSOCK:
862  		case VFIFO:
863 			if (vap->va_mtime.tv_sec == VNOVAL &&
864 			    vap->va_atime.tv_sec == VNOVAL &&
865 			    vap->va_mode == (mode_t)VNOVAL &&
866 			    vap->va_uid == (uid_t)VNOVAL &&
867 			    vap->va_gid == (gid_t)VNOVAL)
868 				return (0);
869  			vap->va_size = VNOVAL;
870  			break;
871  		default:
872 			/*
873 			 * Disallow write attempts if the filesystem is
874 			 * mounted read-only.
875 			 */
876 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
877 				return (EROFS);
878 			/*
879 			 *  We run vnode_pager_setsize() early (why?),
880 			 * we must set np->n_size now to avoid vinvalbuf
881 			 * V_SAVE races that might setsize a lower
882 			 * value.
883 			 */
884 			mtx_lock(&np->n_mtx);
885 			tsize = np->n_size;
886 			mtx_unlock(&np->n_mtx);
887 			error = ncl_meta_setsize(vp, ap->a_cred, td,
888 			    vap->va_size);
889 			mtx_lock(&np->n_mtx);
890  			if (np->n_flag & NMODIFIED) {
891 			    tsize = np->n_size;
892 			    mtx_unlock(&np->n_mtx);
893  			    if (vap->va_size == 0)
894  				error = ncl_vinvalbuf(vp, 0, td, 1);
895  			    else
896  				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
897  			    if (error) {
898 				vnode_pager_setsize(vp, tsize);
899 				return (error);
900 			    }
901 			    /*
902 			     * Call nfscl_delegmodtime() to set the modify time
903 			     * locally, as required.
904 			     */
905 			    nfscl_delegmodtime(vp);
906  			} else
907 			    mtx_unlock(&np->n_mtx);
908 			/*
909 			 * np->n_size has already been set to vap->va_size
910 			 * in ncl_meta_setsize(). We must set it again since
911 			 * nfs_loadattrcache() could be called through
912 			 * ncl_meta_setsize() and could modify np->n_size.
913 			 */
914 			mtx_lock(&np->n_mtx);
915  			np->n_vattr.na_size = np->n_size = vap->va_size;
916 			mtx_unlock(&np->n_mtx);
917   		};
918   	} else {
919 		mtx_lock(&np->n_mtx);
920 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
921 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
922 			mtx_unlock(&np->n_mtx);
923 			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
924 			    (error == EINTR || error == EIO))
925 				return (error);
926 		} else
927 			mtx_unlock(&np->n_mtx);
928 	}
929 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
930 	if (error && vap->va_size != VNOVAL) {
931 		mtx_lock(&np->n_mtx);
932 		np->n_size = np->n_vattr.na_size = tsize;
933 		vnode_pager_setsize(vp, tsize);
934 		mtx_unlock(&np->n_mtx);
935 	}
936 	return (error);
937 }
938 
939 /*
940  * Do an nfs setattr rpc.
941  */
942 static int
943 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
944     struct thread *td)
945 {
946 	struct nfsnode *np = VTONFS(vp);
947 	int error, ret, attrflag, i;
948 	struct nfsvattr nfsva;
949 
950 	if (NFS_ISV34(vp)) {
951 		mtx_lock(&np->n_mtx);
952 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
953 			np->n_accesscache[i].stamp = 0;
954 		np->n_flag |= NDELEGMOD;
955 		mtx_unlock(&np->n_mtx);
956 	}
957 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
958 	    NULL);
959 	if (attrflag) {
960 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
961 		if (ret && !error)
962 			error = ret;
963 	}
964 	if (error && NFS_ISV4(vp))
965 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
966 	return (error);
967 }
968 
969 /*
970  * nfs lookup call, one step at a time...
971  * First look in cache
972  * If not found, unlock the directory nfsnode and do the rpc
973  */
974 static int
975 nfs_lookup(struct vop_lookup_args *ap)
976 {
977 	struct componentname *cnp = ap->a_cnp;
978 	struct vnode *dvp = ap->a_dvp;
979 	struct vnode **vpp = ap->a_vpp;
980 	struct mount *mp = dvp->v_mount;
981 	int flags = cnp->cn_flags;
982 	struct vnode *newvp;
983 	struct nfsmount *nmp;
984 	struct nfsnode *np, *newnp;
985 	int error = 0, attrflag, dattrflag, ltype;
986 	struct thread *td = cnp->cn_thread;
987 	struct nfsfh *nfhp;
988 	struct nfsvattr dnfsva, nfsva;
989 	struct vattr vattr;
990 	time_t dmtime;
991 
992 	*vpp = NULLVP;
993 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
994 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
995 		return (EROFS);
996 	if (dvp->v_type != VDIR)
997 		return (ENOTDIR);
998 	nmp = VFSTONFS(mp);
999 	np = VTONFS(dvp);
1000 
1001 	/* For NFSv4, wait until any remove is done. */
1002 	mtx_lock(&np->n_mtx);
1003 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1004 		np->n_flag |= NREMOVEWANT;
1005 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1006 	}
1007 	mtx_unlock(&np->n_mtx);
1008 
1009 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1010 		return (error);
1011 	error = cache_lookup(dvp, vpp, cnp);
1012 	if (error > 0 && error != ENOENT)
1013 		return (error);
1014 	if (error == -1) {
1015 		/*
1016 		 * We only accept a positive hit in the cache if the
1017 		 * change time of the file matches our cached copy.
1018 		 * Otherwise, we discard the cache entry and fallback
1019 		 * to doing a lookup RPC.
1020 		 *
1021 		 * To better handle stale file handles and attributes,
1022 		 * clear the attribute cache of this node if it is a
1023 		 * leaf component, part of an open() call, and not
1024 		 * locally modified before fetching the attributes.
1025 		 * This should allow stale file handles to be detected
1026 		 * here where we can fall back to a LOOKUP RPC to
1027 		 * recover rather than having nfs_open() detect the
1028 		 * stale file handle and failing open(2) with ESTALE.
1029 		 */
1030 		newvp = *vpp;
1031 		newnp = VTONFS(newvp);
1032 		if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1033 		    !(newnp->n_flag & NMODIFIED)) {
1034 			mtx_lock(&newnp->n_mtx);
1035 			newnp->n_attrstamp = 0;
1036 			mtx_unlock(&newnp->n_mtx);
1037 		}
1038 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1039 		    (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1040 		    vattr.va_ctime.tv_sec == newnp->n_ctime)) {
1041 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1042 			if (cnp->cn_nameiop != LOOKUP &&
1043 			    (flags & ISLASTCN))
1044 				cnp->cn_flags |= SAVENAME;
1045 			return (0);
1046 		}
1047 		cache_purge(newvp);
1048 		if (dvp != newvp)
1049 			vput(newvp);
1050 		else
1051 			vrele(newvp);
1052 		*vpp = NULLVP;
1053 	} else if (error == ENOENT) {
1054 		if (dvp->v_iflag & VI_DOOMED)
1055 			return (ENOENT);
1056 		/*
1057 		 * We only accept a negative hit in the cache if the
1058 		 * modification time of the parent directory matches
1059 		 * our cached copy.  Otherwise, we discard all of the
1060 		 * negative cache entries for this directory. We also
1061 		 * only trust -ve cache entries for less than
1062 		 * nm_negative_namecache_timeout seconds.
1063 		 */
1064 		if ((u_int)(ticks - np->n_dmtime_ticks) <
1065 		    (nmp->nm_negnametimeo * hz) &&
1066 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1067 		    vattr.va_mtime.tv_sec == np->n_dmtime) {
1068 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1069 			return (ENOENT);
1070 		}
1071 		cache_purge_negative(dvp);
1072 		mtx_lock(&np->n_mtx);
1073 		np->n_dmtime = 0;
1074 		mtx_unlock(&np->n_mtx);
1075 	}
1076 
1077 	/*
1078 	 * Cache the modification time of the parent directory in case
1079 	 * the lookup fails and results in adding the first negative
1080 	 * name cache entry for the directory.  Since this is reading
1081 	 * a single time_t, don't bother with locking.  The
1082 	 * modification time may be a bit stale, but it must be read
1083 	 * before performing the lookup RPC to prevent a race where
1084 	 * another lookup updates the timestamp on the directory after
1085 	 * the lookup RPC has been performed on the server but before
1086 	 * n_dmtime is set at the end of this function.
1087 	 */
1088 	dmtime = np->n_vattr.na_mtime.tv_sec;
1089 	error = 0;
1090 	newvp = NULLVP;
1091 	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1092 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1093 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1094 	    NULL);
1095 	if (dattrflag)
1096 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1097 	if (error) {
1098 		if (newvp != NULLVP) {
1099 			vput(newvp);
1100 			*vpp = NULLVP;
1101 		}
1102 
1103 		if (error != ENOENT) {
1104 			if (NFS_ISV4(dvp))
1105 				error = nfscl_maperr(td, error, (uid_t)0,
1106 				    (gid_t)0);
1107 			return (error);
1108 		}
1109 
1110 		/* The requested file was not found. */
1111 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1112 		    (flags & ISLASTCN)) {
1113 			/*
1114 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1115 			 * VWRITE) here instead of just checking
1116 			 * MNT_RDONLY.
1117 			 */
1118 			if (mp->mnt_flag & MNT_RDONLY)
1119 				return (EROFS);
1120 			cnp->cn_flags |= SAVENAME;
1121 			return (EJUSTRETURN);
1122 		}
1123 
1124 		if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1125 			/*
1126 			 * Maintain n_dmtime as the modification time
1127 			 * of the parent directory when the oldest -ve
1128 			 * name cache entry for this directory was
1129 			 * added.  If a -ve cache entry has already
1130 			 * been added with a newer modification time
1131 			 * by a concurrent lookup, then don't bother
1132 			 * adding a cache entry.  The modification
1133 			 * time of the directory might have changed
1134 			 * due to the file this lookup failed to find
1135 			 * being created.  In that case a subsequent
1136 			 * lookup would incorrectly use the entry
1137 			 * added here instead of doing an extra
1138 			 * lookup.
1139 			 */
1140 			mtx_lock(&np->n_mtx);
1141 			if (np->n_dmtime <= dmtime) {
1142 				if (np->n_dmtime == 0) {
1143 					np->n_dmtime = dmtime;
1144 					np->n_dmtime_ticks = ticks;
1145 				}
1146 				mtx_unlock(&np->n_mtx);
1147 				cache_enter(dvp, NULL, cnp);
1148 			} else
1149 				mtx_unlock(&np->n_mtx);
1150 		}
1151 		return (ENOENT);
1152 	}
1153 
1154 	/*
1155 	 * Handle RENAME case...
1156 	 */
1157 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1158 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1159 			FREE((caddr_t)nfhp, M_NFSFH);
1160 			return (EISDIR);
1161 		}
1162 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1163 		if (error)
1164 			return (error);
1165 		newvp = NFSTOV(np);
1166 		if (attrflag)
1167 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1168 			    0, 1);
1169 		*vpp = newvp;
1170 		cnp->cn_flags |= SAVENAME;
1171 		return (0);
1172 	}
1173 
1174 	if (flags & ISDOTDOT) {
1175 		ltype = VOP_ISLOCKED(dvp);
1176 		error = vfs_busy(mp, MBF_NOWAIT);
1177 		if (error != 0) {
1178 			vfs_ref(mp);
1179 			VOP_UNLOCK(dvp, 0);
1180 			error = vfs_busy(mp, 0);
1181 			vn_lock(dvp, ltype | LK_RETRY);
1182 			vfs_rel(mp);
1183 			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1184 				vfs_unbusy(mp);
1185 				error = ENOENT;
1186 			}
1187 			if (error != 0)
1188 				return (error);
1189 		}
1190 		VOP_UNLOCK(dvp, 0);
1191 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1192 		if (error == 0)
1193 			newvp = NFSTOV(np);
1194 		vfs_unbusy(mp);
1195 		if (newvp != dvp)
1196 			vn_lock(dvp, ltype | LK_RETRY);
1197 		if (dvp->v_iflag & VI_DOOMED) {
1198 			if (error == 0) {
1199 				if (newvp == dvp)
1200 					vrele(newvp);
1201 				else
1202 					vput(newvp);
1203 			}
1204 			error = ENOENT;
1205 		}
1206 		if (error != 0)
1207 			return (error);
1208 		if (attrflag)
1209 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1210 			    0, 1);
1211 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1212 		FREE((caddr_t)nfhp, M_NFSFH);
1213 		VREF(dvp);
1214 		newvp = dvp;
1215 		if (attrflag)
1216 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1217 			    0, 1);
1218 	} else {
1219 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1220 		if (error)
1221 			return (error);
1222 		newvp = NFSTOV(np);
1223 		if (attrflag)
1224 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1225 			    0, 1);
1226 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1227 		    !(np->n_flag & NMODIFIED)) {
1228 			/*
1229 			 * Flush the attribute cache when opening a
1230 			 * leaf node to ensure that fresh attributes
1231 			 * are fetched in nfs_open() since we did not
1232 			 * fetch attributes from the LOOKUP reply.
1233 			 */
1234 			mtx_lock(&np->n_mtx);
1235 			np->n_attrstamp = 0;
1236 			mtx_unlock(&np->n_mtx);
1237 		}
1238 	}
1239 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1240 		cnp->cn_flags |= SAVENAME;
1241 	if ((cnp->cn_flags & MAKEENTRY) &&
1242 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1243 		np->n_ctime = np->n_vattr.na_vattr.va_ctime.tv_sec;
1244 		cache_enter(dvp, newvp, cnp);
1245 	}
1246 	*vpp = newvp;
1247 	return (0);
1248 }
1249 
1250 /*
1251  * nfs read call.
1252  * Just call ncl_bioread() to do the work.
1253  */
1254 static int
1255 nfs_read(struct vop_read_args *ap)
1256 {
1257 	struct vnode *vp = ap->a_vp;
1258 
1259 	switch (vp->v_type) {
1260 	case VREG:
1261 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1262 	case VDIR:
1263 		return (EISDIR);
1264 	default:
1265 		return (EOPNOTSUPP);
1266 	}
1267 }
1268 
1269 /*
1270  * nfs readlink call
1271  */
1272 static int
1273 nfs_readlink(struct vop_readlink_args *ap)
1274 {
1275 	struct vnode *vp = ap->a_vp;
1276 
1277 	if (vp->v_type != VLNK)
1278 		return (EINVAL);
1279 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1280 }
1281 
1282 /*
1283  * Do a readlink rpc.
1284  * Called by ncl_doio() from below the buffer cache.
1285  */
1286 int
1287 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1288 {
1289 	int error, ret, attrflag;
1290 	struct nfsvattr nfsva;
1291 
1292 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1293 	    &attrflag, NULL);
1294 	if (attrflag) {
1295 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1296 		if (ret && !error)
1297 			error = ret;
1298 	}
1299 	if (error && NFS_ISV4(vp))
1300 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1301 	return (error);
1302 }
1303 
1304 /*
1305  * nfs read rpc call
1306  * Ditto above
1307  */
1308 int
1309 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1310 {
1311 	int error, ret, attrflag;
1312 	struct nfsvattr nfsva;
1313 
1314 	error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1315 	    NULL);
1316 	if (attrflag) {
1317 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1318 		if (ret && !error)
1319 			error = ret;
1320 	}
1321 	if (error && NFS_ISV4(vp))
1322 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1323 	return (error);
1324 }
1325 
1326 /*
1327  * nfs write call
1328  */
1329 int
1330 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1331     int *iomode, int *must_commit, int called_from_strategy)
1332 {
1333 	struct nfsvattr nfsva;
1334 	int error = 0, attrflag, ret;
1335 	u_char verf[NFSX_VERF];
1336 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1337 
1338 	*must_commit = 0;
1339 	error = nfsrpc_write(vp, uiop, iomode, verf, cred,
1340 	    uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1341 	NFSLOCKMNT(nmp);
1342 	if (!error && NFSHASWRITEVERF(nmp) &&
1343 	    NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1344 		*must_commit = 1;
1345 		NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1346 	}
1347 	NFSUNLOCKMNT(nmp);
1348 	if (attrflag) {
1349 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1350 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1351 			    1);
1352 		else
1353 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1354 			    1);
1355 		if (ret && !error)
1356 			error = ret;
1357 	}
1358 	if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1359 		*iomode = NFSWRITE_FILESYNC;
1360 	if (error && NFS_ISV4(vp))
1361 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1362 	return (error);
1363 }
1364 
1365 /*
1366  * nfs mknod rpc
1367  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1368  * mode set to specify the file type and the size field for rdev.
1369  */
1370 static int
1371 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1372     struct vattr *vap)
1373 {
1374 	struct nfsvattr nfsva, dnfsva;
1375 	struct vnode *newvp = NULL;
1376 	struct nfsnode *np = NULL, *dnp;
1377 	struct nfsfh *nfhp;
1378 	struct vattr vattr;
1379 	int error = 0, attrflag, dattrflag;
1380 	u_int32_t rdev;
1381 
1382 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1383 		rdev = vap->va_rdev;
1384 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1385 		rdev = 0xffffffff;
1386 	else
1387 		return (EOPNOTSUPP);
1388 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1389 		return (error);
1390 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1391 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1392 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1393 	if (!error) {
1394 		if (!nfhp)
1395 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1396 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1397 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1398 			    NULL);
1399 		if (nfhp)
1400 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1401 			    cnp->cn_thread, &np, NULL);
1402 	}
1403 	if (dattrflag)
1404 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1405 	if (!error) {
1406 		newvp = NFSTOV(np);
1407 		if (attrflag)
1408 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1409 			    0, 1);
1410 	}
1411 	if (!error) {
1412 		if ((cnp->cn_flags & MAKEENTRY))
1413 			cache_enter(dvp, newvp, cnp);
1414 		*vpp = newvp;
1415 	} else if (NFS_ISV4(dvp)) {
1416 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1417 		    vap->va_gid);
1418 	}
1419 	dnp = VTONFS(dvp);
1420 	mtx_lock(&dnp->n_mtx);
1421 	dnp->n_flag |= NMODIFIED;
1422 	if (!dattrflag)
1423 		dnp->n_attrstamp = 0;
1424 	mtx_unlock(&dnp->n_mtx);
1425 	return (error);
1426 }
1427 
1428 /*
1429  * nfs mknod vop
1430  * just call nfs_mknodrpc() to do the work.
1431  */
1432 /* ARGSUSED */
1433 static int
1434 nfs_mknod(struct vop_mknod_args *ap)
1435 {
1436 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1437 }
1438 
1439 static struct mtx nfs_cverf_mtx;
1440 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1441     MTX_DEF);
1442 
1443 static nfsquad_t
1444 nfs_get_cverf(void)
1445 {
1446 	static nfsquad_t cverf;
1447 	nfsquad_t ret;
1448 	static int cverf_initialized = 0;
1449 
1450 	mtx_lock(&nfs_cverf_mtx);
1451 	if (cverf_initialized == 0) {
1452 		cverf.lval[0] = arc4random();
1453 		cverf.lval[1] = arc4random();
1454 		cverf_initialized = 1;
1455 	} else
1456 		cverf.qval++;
1457 	ret = cverf;
1458 	mtx_unlock(&nfs_cverf_mtx);
1459 
1460 	return (ret);
1461 }
1462 
1463 /*
1464  * nfs file create call
1465  */
1466 static int
1467 nfs_create(struct vop_create_args *ap)
1468 {
1469 	struct vnode *dvp = ap->a_dvp;
1470 	struct vattr *vap = ap->a_vap;
1471 	struct componentname *cnp = ap->a_cnp;
1472 	struct nfsnode *np = NULL, *dnp;
1473 	struct vnode *newvp = NULL;
1474 	struct nfsmount *nmp;
1475 	struct nfsvattr dnfsva, nfsva;
1476 	struct nfsfh *nfhp;
1477 	nfsquad_t cverf;
1478 	int error = 0, attrflag, dattrflag, fmode = 0;
1479 	struct vattr vattr;
1480 
1481 	/*
1482 	 * Oops, not for me..
1483 	 */
1484 	if (vap->va_type == VSOCK)
1485 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1486 
1487 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1488 		return (error);
1489 	if (vap->va_vaflags & VA_EXCLUSIVE)
1490 		fmode |= O_EXCL;
1491 	dnp = VTONFS(dvp);
1492 	nmp = VFSTONFS(vnode_mount(dvp));
1493 again:
1494 	/* For NFSv4, wait until any remove is done. */
1495 	mtx_lock(&dnp->n_mtx);
1496 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1497 		dnp->n_flag |= NREMOVEWANT;
1498 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1499 	}
1500 	mtx_unlock(&dnp->n_mtx);
1501 
1502 	cverf = nfs_get_cverf();
1503 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1504 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1505 	    &nfhp, &attrflag, &dattrflag, NULL);
1506 	if (!error) {
1507 		if (nfhp == NULL)
1508 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1509 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1510 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1511 			    NULL);
1512 		if (nfhp != NULL)
1513 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1514 			    cnp->cn_thread, &np, NULL);
1515 	}
1516 	if (dattrflag)
1517 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1518 	if (!error) {
1519 		newvp = NFSTOV(np);
1520 		if (attrflag)
1521 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1522 			    0, 1);
1523 	}
1524 	if (error) {
1525 		if (newvp != NULL) {
1526 			vrele(newvp);
1527 			newvp = NULL;
1528 		}
1529 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1530 		    error == NFSERR_NOTSUPP) {
1531 			fmode &= ~O_EXCL;
1532 			goto again;
1533 		}
1534 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1535 		if (nfscl_checksattr(vap, &nfsva)) {
1536 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1537 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1538 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1539 			    vap->va_gid != (gid_t)VNOVAL)) {
1540 				/* try again without setting uid/gid */
1541 				vap->va_uid = (uid_t)VNOVAL;
1542 				vap->va_gid = (uid_t)VNOVAL;
1543 				error = nfsrpc_setattr(newvp, vap, NULL,
1544 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1545 				    &attrflag, NULL);
1546 			}
1547 			if (attrflag)
1548 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1549 				    NULL, 0, 1);
1550 		}
1551 	}
1552 	if (!error) {
1553 		if (cnp->cn_flags & MAKEENTRY)
1554 			cache_enter(dvp, newvp, cnp);
1555 		*ap->a_vpp = newvp;
1556 	} else if (NFS_ISV4(dvp)) {
1557 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1558 		    vap->va_gid);
1559 	}
1560 	mtx_lock(&dnp->n_mtx);
1561 	dnp->n_flag |= NMODIFIED;
1562 	if (!dattrflag)
1563 		dnp->n_attrstamp = 0;
1564 	mtx_unlock(&dnp->n_mtx);
1565 	return (error);
1566 }
1567 
1568 /*
1569  * nfs file remove call
1570  * To try and make nfs semantics closer to ufs semantics, a file that has
1571  * other processes using the vnode is renamed instead of removed and then
1572  * removed later on the last close.
1573  * - If v_usecount > 1
1574  *	  If a rename is not already in the works
1575  *	     call nfs_sillyrename() to set it up
1576  *     else
1577  *	  do the remove rpc
1578  */
1579 static int
1580 nfs_remove(struct vop_remove_args *ap)
1581 {
1582 	struct vnode *vp = ap->a_vp;
1583 	struct vnode *dvp = ap->a_dvp;
1584 	struct componentname *cnp = ap->a_cnp;
1585 	struct nfsnode *np = VTONFS(vp);
1586 	int error = 0;
1587 	struct vattr vattr;
1588 
1589 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1590 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1591 	if (vp->v_type == VDIR)
1592 		error = EPERM;
1593 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1594 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1595 	    vattr.va_nlink > 1)) {
1596 		/*
1597 		 * Purge the name cache so that the chance of a lookup for
1598 		 * the name succeeding while the remove is in progress is
1599 		 * minimized. Without node locking it can still happen, such
1600 		 * that an I/O op returns ESTALE, but since you get this if
1601 		 * another host removes the file..
1602 		 */
1603 		cache_purge(vp);
1604 		/*
1605 		 * throw away biocache buffers, mainly to avoid
1606 		 * unnecessary delayed writes later.
1607 		 */
1608 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1609 		/* Do the rpc */
1610 		if (error != EINTR && error != EIO)
1611 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1612 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1613 		/*
1614 		 * Kludge City: If the first reply to the remove rpc is lost..
1615 		 *   the reply to the retransmitted request will be ENOENT
1616 		 *   since the file was in fact removed
1617 		 *   Therefore, we cheat and return success.
1618 		 */
1619 		if (error == ENOENT)
1620 			error = 0;
1621 	} else if (!np->n_sillyrename)
1622 		error = nfs_sillyrename(dvp, vp, cnp);
1623 	np->n_attrstamp = 0;
1624 	return (error);
1625 }
1626 
1627 /*
1628  * nfs file remove rpc called from nfs_inactive
1629  */
1630 int
1631 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1632 {
1633 	/*
1634 	 * Make sure that the directory vnode is still valid.
1635 	 * XXX we should lock sp->s_dvp here.
1636 	 */
1637 	if (sp->s_dvp->v_type == VBAD)
1638 		return (0);
1639 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1640 	    sp->s_cred, NULL));
1641 }
1642 
1643 /*
1644  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1645  */
1646 static int
1647 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1648     int namelen, struct ucred *cred, struct thread *td)
1649 {
1650 	struct nfsvattr dnfsva;
1651 	struct nfsnode *dnp = VTONFS(dvp);
1652 	int error = 0, dattrflag;
1653 
1654 	mtx_lock(&dnp->n_mtx);
1655 	dnp->n_flag |= NREMOVEINPROG;
1656 	mtx_unlock(&dnp->n_mtx);
1657 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1658 	    &dattrflag, NULL);
1659 	mtx_lock(&dnp->n_mtx);
1660 	if ((dnp->n_flag & NREMOVEWANT)) {
1661 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1662 		mtx_unlock(&dnp->n_mtx);
1663 		wakeup((caddr_t)dnp);
1664 	} else {
1665 		dnp->n_flag &= ~NREMOVEINPROG;
1666 		mtx_unlock(&dnp->n_mtx);
1667 	}
1668 	if (dattrflag)
1669 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1670 	mtx_lock(&dnp->n_mtx);
1671 	dnp->n_flag |= NMODIFIED;
1672 	if (!dattrflag)
1673 		dnp->n_attrstamp = 0;
1674 	mtx_unlock(&dnp->n_mtx);
1675 	if (error && NFS_ISV4(dvp))
1676 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1677 	return (error);
1678 }
1679 
1680 /*
1681  * nfs file rename call
1682  */
1683 static int
1684 nfs_rename(struct vop_rename_args *ap)
1685 {
1686 	struct vnode *fvp = ap->a_fvp;
1687 	struct vnode *tvp = ap->a_tvp;
1688 	struct vnode *fdvp = ap->a_fdvp;
1689 	struct vnode *tdvp = ap->a_tdvp;
1690 	struct componentname *tcnp = ap->a_tcnp;
1691 	struct componentname *fcnp = ap->a_fcnp;
1692 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1693 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1694 	struct nfsv4node *newv4 = NULL;
1695 	int error;
1696 
1697 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1698 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1699 	/* Check for cross-device rename */
1700 	if ((fvp->v_mount != tdvp->v_mount) ||
1701 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1702 		error = EXDEV;
1703 		goto out;
1704 	}
1705 
1706 	if (fvp == tvp) {
1707 		ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1708 		error = 0;
1709 		goto out;
1710 	}
1711 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1712 		goto out;
1713 
1714 	/*
1715 	 * We have to flush B_DELWRI data prior to renaming
1716 	 * the file.  If we don't, the delayed-write buffers
1717 	 * can be flushed out later after the file has gone stale
1718 	 * under NFSV3.  NFSV2 does not have this problem because
1719 	 * ( as far as I can tell ) it flushes dirty buffers more
1720 	 * often.
1721 	 *
1722 	 * Skip the rename operation if the fsync fails, this can happen
1723 	 * due to the server's volume being full, when we pushed out data
1724 	 * that was written back to our cache earlier. Not checking for
1725 	 * this condition can result in potential (silent) data loss.
1726 	 */
1727 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1728 	VOP_UNLOCK(fvp, 0);
1729 	if (!error && tvp)
1730 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1731 	if (error)
1732 		goto out;
1733 
1734 	/*
1735 	 * If the tvp exists and is in use, sillyrename it before doing the
1736 	 * rename of the new file over it.
1737 	 * XXX Can't sillyrename a directory.
1738 	 */
1739 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1740 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1741 		vput(tvp);
1742 		tvp = NULL;
1743 	}
1744 
1745 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1746 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1747 	    tcnp->cn_thread);
1748 
1749 	if (!error) {
1750 		/*
1751 		 * For NFSv4, check to see if it is the same name and
1752 		 * replace the name, if it is different.
1753 		 */
1754 		MALLOC(newv4, struct nfsv4node *,
1755 		    sizeof (struct nfsv4node) +
1756 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1757 		    M_NFSV4NODE, M_WAITOK);
1758 		mtx_lock(&tdnp->n_mtx);
1759 		mtx_lock(&fnp->n_mtx);
1760 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1761 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1762 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1763 		      tcnp->cn_namelen) ||
1764 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1765 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1766 			tdnp->n_fhp->nfh_len))) {
1767 #ifdef notdef
1768 { char nnn[100]; int nnnl;
1769 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1770 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1771 nnn[nnnl] = '\0';
1772 printf("ren replace=%s\n",nnn);
1773 }
1774 #endif
1775 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1776 			fnp->n_v4 = newv4;
1777 			newv4 = NULL;
1778 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1779 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1780 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1781 			    tdnp->n_fhp->nfh_len);
1782 			NFSBCOPY(tcnp->cn_nameptr,
1783 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1784 		}
1785 		mtx_unlock(&tdnp->n_mtx);
1786 		mtx_unlock(&fnp->n_mtx);
1787 		if (newv4 != NULL)
1788 			FREE((caddr_t)newv4, M_NFSV4NODE);
1789 	}
1790 
1791 	if (fvp->v_type == VDIR) {
1792 		if (tvp != NULL && tvp->v_type == VDIR)
1793 			cache_purge(tdvp);
1794 		cache_purge(fdvp);
1795 	}
1796 
1797 out:
1798 	if (tdvp == tvp)
1799 		vrele(tdvp);
1800 	else
1801 		vput(tdvp);
1802 	if (tvp)
1803 		vput(tvp);
1804 	vrele(fdvp);
1805 	vrele(fvp);
1806 	/*
1807 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1808 	 */
1809 	if (error == ENOENT)
1810 		error = 0;
1811 	return (error);
1812 }
1813 
1814 /*
1815  * nfs file rename rpc called from nfs_remove() above
1816  */
1817 static int
1818 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1819     struct sillyrename *sp)
1820 {
1821 
1822 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1823 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1824 	    scnp->cn_thread));
1825 }
1826 
1827 /*
1828  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1829  */
1830 static int
1831 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1832     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1833     int tnamelen, struct ucred *cred, struct thread *td)
1834 {
1835 	struct nfsvattr fnfsva, tnfsva;
1836 	struct nfsnode *fdnp = VTONFS(fdvp);
1837 	struct nfsnode *tdnp = VTONFS(tdvp);
1838 	int error = 0, fattrflag, tattrflag;
1839 
1840 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1841 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1842 	    &tattrflag, NULL, NULL);
1843 	mtx_lock(&fdnp->n_mtx);
1844 	fdnp->n_flag |= NMODIFIED;
1845 	mtx_unlock(&fdnp->n_mtx);
1846 	mtx_lock(&tdnp->n_mtx);
1847 	tdnp->n_flag |= NMODIFIED;
1848 	mtx_unlock(&tdnp->n_mtx);
1849 	if (fattrflag)
1850 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1851 	else
1852 		fdnp->n_attrstamp = 0;
1853 	if (tattrflag)
1854 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1855 	else
1856 		tdnp->n_attrstamp = 0;
1857 	if (error && NFS_ISV4(fdvp))
1858 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1859 	return (error);
1860 }
1861 
1862 /*
1863  * nfs hard link create call
1864  */
1865 static int
1866 nfs_link(struct vop_link_args *ap)
1867 {
1868 	struct vnode *vp = ap->a_vp;
1869 	struct vnode *tdvp = ap->a_tdvp;
1870 	struct componentname *cnp = ap->a_cnp;
1871 	struct nfsnode *tdnp;
1872 	struct nfsvattr nfsva, dnfsva;
1873 	int error = 0, attrflag, dattrflag;
1874 
1875 	if (vp->v_mount != tdvp->v_mount) {
1876 		return (EXDEV);
1877 	}
1878 
1879 	/*
1880 	 * Push all writes to the server, so that the attribute cache
1881 	 * doesn't get "out of sync" with the server.
1882 	 * XXX There should be a better way!
1883 	 */
1884 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1885 
1886 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1887 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1888 	    &dattrflag, NULL);
1889 	tdnp = VTONFS(tdvp);
1890 	mtx_lock(&tdnp->n_mtx);
1891 	tdnp->n_flag |= NMODIFIED;
1892 	mtx_unlock(&tdnp->n_mtx);
1893 	if (attrflag)
1894 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1895 	else
1896 		VTONFS(vp)->n_attrstamp = 0;
1897 	if (dattrflag)
1898 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1899 	else
1900 		tdnp->n_attrstamp = 0;
1901 	/*
1902 	 * If negative lookup caching is enabled, I might as well
1903 	 * add an entry for this node. Not necessary for correctness,
1904 	 * but if negative caching is enabled, then the system
1905 	 * must care about lookup caching hit rate, so...
1906 	 */
1907 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1908 	    (cnp->cn_flags & MAKEENTRY))
1909 		cache_enter(tdvp, vp, cnp);
1910 	if (error && NFS_ISV4(vp))
1911 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1912 		    (gid_t)0);
1913 	return (error);
1914 }
1915 
1916 /*
1917  * nfs symbolic link create call
1918  */
1919 static int
1920 nfs_symlink(struct vop_symlink_args *ap)
1921 {
1922 	struct vnode *dvp = ap->a_dvp;
1923 	struct vattr *vap = ap->a_vap;
1924 	struct componentname *cnp = ap->a_cnp;
1925 	struct nfsvattr nfsva, dnfsva;
1926 	struct nfsfh *nfhp;
1927 	struct nfsnode *np = NULL, *dnp;
1928 	struct vnode *newvp = NULL;
1929 	int error = 0, attrflag, dattrflag, ret;
1930 
1931 	vap->va_type = VLNK;
1932 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1933 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1934 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1935 	if (nfhp) {
1936 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1937 		    &np, NULL);
1938 		if (!ret)
1939 			newvp = NFSTOV(np);
1940 		else if (!error)
1941 			error = ret;
1942 	}
1943 	if (newvp != NULL) {
1944 		if (attrflag)
1945 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1946 			    0, 1);
1947 	} else if (!error) {
1948 		/*
1949 		 * If we do not have an error and we could not extract the
1950 		 * newvp from the response due to the request being NFSv2, we
1951 		 * have to do a lookup in order to obtain a newvp to return.
1952 		 */
1953 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1954 		    cnp->cn_cred, cnp->cn_thread, &np);
1955 		if (!error)
1956 			newvp = NFSTOV(np);
1957 	}
1958 	if (error) {
1959 		if (newvp)
1960 			vput(newvp);
1961 		if (NFS_ISV4(dvp))
1962 			error = nfscl_maperr(cnp->cn_thread, error,
1963 			    vap->va_uid, vap->va_gid);
1964 	} else {
1965 		/*
1966 		 * If negative lookup caching is enabled, I might as well
1967 		 * add an entry for this node. Not necessary for correctness,
1968 		 * but if negative caching is enabled, then the system
1969 		 * must care about lookup caching hit rate, so...
1970 		 */
1971 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1972 		    (cnp->cn_flags & MAKEENTRY))
1973 			cache_enter(dvp, newvp, cnp);
1974 		*ap->a_vpp = newvp;
1975 	}
1976 
1977 	dnp = VTONFS(dvp);
1978 	mtx_lock(&dnp->n_mtx);
1979 	dnp->n_flag |= NMODIFIED;
1980 	mtx_unlock(&dnp->n_mtx);
1981 	if (dattrflag)
1982 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1983 	else
1984 		dnp->n_attrstamp = 0;
1985 	return (error);
1986 }
1987 
1988 /*
1989  * nfs make dir call
1990  */
1991 static int
1992 nfs_mkdir(struct vop_mkdir_args *ap)
1993 {
1994 	struct vnode *dvp = ap->a_dvp;
1995 	struct vattr *vap = ap->a_vap;
1996 	struct componentname *cnp = ap->a_cnp;
1997 	struct nfsnode *np = NULL, *dnp;
1998 	struct vnode *newvp = NULL;
1999 	struct vattr vattr;
2000 	struct nfsfh *nfhp;
2001 	struct nfsvattr nfsva, dnfsva;
2002 	int error = 0, attrflag, dattrflag, ret;
2003 
2004 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
2005 		return (error);
2006 	vap->va_type = VDIR;
2007 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2008 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2009 	    &attrflag, &dattrflag, NULL);
2010 	dnp = VTONFS(dvp);
2011 	mtx_lock(&dnp->n_mtx);
2012 	dnp->n_flag |= NMODIFIED;
2013 	mtx_unlock(&dnp->n_mtx);
2014 	if (dattrflag)
2015 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2016 	else
2017 		dnp->n_attrstamp = 0;
2018 	if (nfhp) {
2019 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2020 		    &np, NULL);
2021 		if (!ret) {
2022 			newvp = NFSTOV(np);
2023 			if (attrflag)
2024 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2025 				NULL, 0, 1);
2026 		} else if (!error)
2027 			error = ret;
2028 	}
2029 	if (!error && newvp == NULL) {
2030 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2031 		    cnp->cn_cred, cnp->cn_thread, &np);
2032 		if (!error) {
2033 			newvp = NFSTOV(np);
2034 			if (newvp->v_type != VDIR)
2035 				error = EEXIST;
2036 		}
2037 	}
2038 	if (error) {
2039 		if (newvp)
2040 			vput(newvp);
2041 		if (NFS_ISV4(dvp))
2042 			error = nfscl_maperr(cnp->cn_thread, error,
2043 			    vap->va_uid, vap->va_gid);
2044 	} else {
2045 		/*
2046 		 * If negative lookup caching is enabled, I might as well
2047 		 * add an entry for this node. Not necessary for correctness,
2048 		 * but if negative caching is enabled, then the system
2049 		 * must care about lookup caching hit rate, so...
2050 		 */
2051 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2052 		    (cnp->cn_flags & MAKEENTRY))
2053 			cache_enter(dvp, newvp, cnp);
2054 		*ap->a_vpp = newvp;
2055 	}
2056 	return (error);
2057 }
2058 
2059 /*
2060  * nfs remove directory call
2061  */
2062 static int
2063 nfs_rmdir(struct vop_rmdir_args *ap)
2064 {
2065 	struct vnode *vp = ap->a_vp;
2066 	struct vnode *dvp = ap->a_dvp;
2067 	struct componentname *cnp = ap->a_cnp;
2068 	struct nfsnode *dnp;
2069 	struct nfsvattr dnfsva;
2070 	int error, dattrflag;
2071 
2072 	if (dvp == vp)
2073 		return (EINVAL);
2074 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2075 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2076 	dnp = VTONFS(dvp);
2077 	mtx_lock(&dnp->n_mtx);
2078 	dnp->n_flag |= NMODIFIED;
2079 	mtx_unlock(&dnp->n_mtx);
2080 	if (dattrflag)
2081 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2082 	else
2083 		dnp->n_attrstamp = 0;
2084 
2085 	cache_purge(dvp);
2086 	cache_purge(vp);
2087 	if (error && NFS_ISV4(dvp))
2088 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2089 		    (gid_t)0);
2090 	/*
2091 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2092 	 */
2093 	if (error == ENOENT)
2094 		error = 0;
2095 	return (error);
2096 }
2097 
2098 /*
2099  * nfs readdir call
2100  */
2101 static int
2102 nfs_readdir(struct vop_readdir_args *ap)
2103 {
2104 	struct vnode *vp = ap->a_vp;
2105 	struct nfsnode *np = VTONFS(vp);
2106 	struct uio *uio = ap->a_uio;
2107 	int tresid, error = 0;
2108 	struct vattr vattr;
2109 
2110 	if (vp->v_type != VDIR)
2111 		return(EPERM);
2112 
2113 	/*
2114 	 * First, check for hit on the EOF offset cache
2115 	 */
2116 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2117 	    (np->n_flag & NMODIFIED) == 0) {
2118 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2119 			mtx_lock(&np->n_mtx);
2120 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2121 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2122 				mtx_unlock(&np->n_mtx);
2123 				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2124 				return (0);
2125 			} else
2126 				mtx_unlock(&np->n_mtx);
2127 		}
2128 	}
2129 
2130 	/*
2131 	 * Call ncl_bioread() to do the real work.
2132 	 */
2133 	tresid = uio->uio_resid;
2134 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2135 
2136 	if (!error && uio->uio_resid == tresid)
2137 		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2138 	return (error);
2139 }
2140 
2141 /*
2142  * Readdir rpc call.
2143  * Called from below the buffer cache by ncl_doio().
2144  */
2145 int
2146 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2147     struct thread *td)
2148 {
2149 	struct nfsvattr nfsva;
2150 	nfsuint64 *cookiep, cookie;
2151 	struct nfsnode *dnp = VTONFS(vp);
2152 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2153 	int error = 0, eof, attrflag;
2154 
2155 	KASSERT(uiop->uio_iovcnt == 1 &&
2156 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2157 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2158 	    ("nfs readdirrpc bad uio"));
2159 
2160 	/*
2161 	 * If there is no cookie, assume directory was stale.
2162 	 */
2163 	ncl_dircookie_lock(dnp);
2164 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2165 	if (cookiep) {
2166 		cookie = *cookiep;
2167 		ncl_dircookie_unlock(dnp);
2168 	} else {
2169 		ncl_dircookie_unlock(dnp);
2170 		return (NFSERR_BAD_COOKIE);
2171 	}
2172 
2173 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2174 		(void)ncl_fsinfo(nmp, vp, cred, td);
2175 
2176 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2177 	    &attrflag, &eof, NULL);
2178 	if (attrflag)
2179 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2180 
2181 	if (!error) {
2182 		/*
2183 		 * We are now either at the end of the directory or have filled
2184 		 * the block.
2185 		 */
2186 		if (eof)
2187 			dnp->n_direofoffset = uiop->uio_offset;
2188 		else {
2189 			if (uiop->uio_resid > 0)
2190 				ncl_printf("EEK! readdirrpc resid > 0\n");
2191 			ncl_dircookie_lock(dnp);
2192 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2193 			*cookiep = cookie;
2194 			ncl_dircookie_unlock(dnp);
2195 		}
2196 	} else if (NFS_ISV4(vp)) {
2197 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2198 	}
2199 	return (error);
2200 }
2201 
2202 /*
2203  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2204  */
2205 int
2206 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2207     struct thread *td)
2208 {
2209 	struct nfsvattr nfsva;
2210 	nfsuint64 *cookiep, cookie;
2211 	struct nfsnode *dnp = VTONFS(vp);
2212 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2213 	int error = 0, attrflag, eof;
2214 
2215 	KASSERT(uiop->uio_iovcnt == 1 &&
2216 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2217 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2218 	    ("nfs readdirplusrpc bad uio"));
2219 
2220 	/*
2221 	 * If there is no cookie, assume directory was stale.
2222 	 */
2223 	ncl_dircookie_lock(dnp);
2224 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2225 	if (cookiep) {
2226 		cookie = *cookiep;
2227 		ncl_dircookie_unlock(dnp);
2228 	} else {
2229 		ncl_dircookie_unlock(dnp);
2230 		return (NFSERR_BAD_COOKIE);
2231 	}
2232 
2233 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2234 		(void)ncl_fsinfo(nmp, vp, cred, td);
2235 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2236 	    &attrflag, &eof, NULL);
2237 	if (attrflag)
2238 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2239 
2240 	if (!error) {
2241 		/*
2242 		 * We are now either at end of the directory or have filled the
2243 		 * the block.
2244 		 */
2245 		if (eof)
2246 			dnp->n_direofoffset = uiop->uio_offset;
2247 		else {
2248 			if (uiop->uio_resid > 0)
2249 				ncl_printf("EEK! readdirplusrpc resid > 0\n");
2250 			ncl_dircookie_lock(dnp);
2251 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2252 			*cookiep = cookie;
2253 			ncl_dircookie_unlock(dnp);
2254 		}
2255 	} else if (NFS_ISV4(vp)) {
2256 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2257 	}
2258 	return (error);
2259 }
2260 
2261 /*
2262  * Silly rename. To make the NFS filesystem that is stateless look a little
2263  * more like the "ufs" a remove of an active vnode is translated to a rename
2264  * to a funny looking filename that is removed by nfs_inactive on the
2265  * nfsnode. There is the potential for another process on a different client
2266  * to create the same funny name between the nfs_lookitup() fails and the
2267  * nfs_rename() completes, but...
2268  */
2269 static int
2270 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2271 {
2272 	struct sillyrename *sp;
2273 	struct nfsnode *np;
2274 	int error;
2275 	short pid;
2276 	unsigned int lticks;
2277 
2278 	cache_purge(dvp);
2279 	np = VTONFS(vp);
2280 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2281 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2282 	    M_NEWNFSREQ, M_WAITOK);
2283 	sp->s_cred = crhold(cnp->cn_cred);
2284 	sp->s_dvp = dvp;
2285 	VREF(dvp);
2286 
2287 	/*
2288 	 * Fudge together a funny name.
2289 	 * Changing the format of the funny name to accomodate more
2290 	 * sillynames per directory.
2291 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2292 	 * CPU ticks since boot.
2293 	 */
2294 	pid = cnp->cn_thread->td_proc->p_pid;
2295 	lticks = (unsigned int)ticks;
2296 	for ( ; ; ) {
2297 		sp->s_namlen = sprintf(sp->s_name,
2298 				       ".nfs.%08x.%04x4.4", lticks,
2299 				       pid);
2300 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2301 				 cnp->cn_thread, NULL))
2302 			break;
2303 		lticks++;
2304 	}
2305 	error = nfs_renameit(dvp, vp, cnp, sp);
2306 	if (error)
2307 		goto bad;
2308 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2309 		cnp->cn_thread, &np);
2310 	np->n_sillyrename = sp;
2311 	return (0);
2312 bad:
2313 	vrele(sp->s_dvp);
2314 	crfree(sp->s_cred);
2315 	free((caddr_t)sp, M_NEWNFSREQ);
2316 	return (error);
2317 }
2318 
2319 /*
2320  * Look up a file name and optionally either update the file handle or
2321  * allocate an nfsnode, depending on the value of npp.
2322  * npp == NULL	--> just do the lookup
2323  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2324  *			handled too
2325  * *npp != NULL --> update the file handle in the vnode
2326  */
2327 static int
2328 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2329     struct thread *td, struct nfsnode **npp)
2330 {
2331 	struct vnode *newvp = NULL, *vp;
2332 	struct nfsnode *np, *dnp = VTONFS(dvp);
2333 	struct nfsfh *nfhp, *onfhp;
2334 	struct nfsvattr nfsva, dnfsva;
2335 	struct componentname cn;
2336 	int error = 0, attrflag, dattrflag;
2337 	u_int hash;
2338 
2339 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2340 	    &nfhp, &attrflag, &dattrflag, NULL);
2341 	if (dattrflag)
2342 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2343 	if (npp && !error) {
2344 		if (*npp != NULL) {
2345 		    np = *npp;
2346 		    vp = NFSTOV(np);
2347 		    /*
2348 		     * For NFSv4, check to see if it is the same name and
2349 		     * replace the name, if it is different.
2350 		     */
2351 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2352 			(np->n_v4->n4_namelen != len ||
2353 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2354 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2355 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2356 			 dnp->n_fhp->nfh_len))) {
2357 #ifdef notdef
2358 { char nnn[100]; int nnnl;
2359 nnnl = (len < 100) ? len : 99;
2360 bcopy(name, nnn, nnnl);
2361 nnn[nnnl] = '\0';
2362 printf("replace=%s\n",nnn);
2363 }
2364 #endif
2365 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2366 			    MALLOC(np->n_v4, struct nfsv4node *,
2367 				sizeof (struct nfsv4node) +
2368 				dnp->n_fhp->nfh_len + len - 1,
2369 				M_NFSV4NODE, M_WAITOK);
2370 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2371 			    np->n_v4->n4_namelen = len;
2372 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2373 				dnp->n_fhp->nfh_len);
2374 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2375 		    }
2376 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2377 			FNV1_32_INIT);
2378 		    onfhp = np->n_fhp;
2379 		    /*
2380 		     * Rehash node for new file handle.
2381 		     */
2382 		    vfs_hash_rehash(vp, hash);
2383 		    np->n_fhp = nfhp;
2384 		    if (onfhp != NULL)
2385 			FREE((caddr_t)onfhp, M_NFSFH);
2386 		    newvp = NFSTOV(np);
2387 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2388 		    FREE((caddr_t)nfhp, M_NFSFH);
2389 		    VREF(dvp);
2390 		    newvp = dvp;
2391 		} else {
2392 		    cn.cn_nameptr = name;
2393 		    cn.cn_namelen = len;
2394 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2395 			&np, NULL);
2396 		    if (error)
2397 			return (error);
2398 		    newvp = NFSTOV(np);
2399 		}
2400 		if (!attrflag && *npp == NULL) {
2401 			vrele(newvp);
2402 			return (ENOENT);
2403 		}
2404 		if (attrflag)
2405 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2406 			    0, 1);
2407 	}
2408 	if (npp && *npp == NULL) {
2409 		if (error) {
2410 			if (newvp) {
2411 				if (newvp == dvp)
2412 					vrele(newvp);
2413 				else
2414 					vput(newvp);
2415 			}
2416 		} else
2417 			*npp = np;
2418 	}
2419 	if (error && NFS_ISV4(dvp))
2420 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2421 	return (error);
2422 }
2423 
2424 /*
2425  * Nfs Version 3 and 4 commit rpc
2426  */
2427 int
2428 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2429    struct thread *td)
2430 {
2431 	struct nfsvattr nfsva;
2432 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2433 	int error, attrflag;
2434 	u_char verf[NFSX_VERF];
2435 
2436 	mtx_lock(&nmp->nm_mtx);
2437 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2438 		mtx_unlock(&nmp->nm_mtx);
2439 		return (0);
2440 	}
2441 	mtx_unlock(&nmp->nm_mtx);
2442 	error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2443 	    &attrflag, NULL);
2444 	if (!error) {
2445 		if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2446 			NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2447 			error = NFSERR_STALEWRITEVERF;
2448 		}
2449 		if (!error && attrflag)
2450 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2451 			    0, 1);
2452 	} else if (NFS_ISV4(vp)) {
2453 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2454 	}
2455 	return (error);
2456 }
2457 
2458 /*
2459  * Strategy routine.
2460  * For async requests when nfsiod(s) are running, queue the request by
2461  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2462  * request.
2463  */
2464 static int
2465 nfs_strategy(struct vop_strategy_args *ap)
2466 {
2467 	struct buf *bp = ap->a_bp;
2468 	struct ucred *cr;
2469 
2470 	KASSERT(!(bp->b_flags & B_DONE),
2471 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2472 	BUF_ASSERT_HELD(bp);
2473 
2474 	if (bp->b_iocmd == BIO_READ)
2475 		cr = bp->b_rcred;
2476 	else
2477 		cr = bp->b_wcred;
2478 
2479 	/*
2480 	 * If the op is asynchronous and an i/o daemon is waiting
2481 	 * queue the request, wake it up and wait for completion
2482 	 * otherwise just do it ourselves.
2483 	 */
2484 	if ((bp->b_flags & B_ASYNC) == 0 ||
2485 	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2486 		(void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2487 	return (0);
2488 }
2489 
2490 /*
2491  * fsync vnode op. Just call ncl_flush() with commit == 1.
2492  */
2493 /* ARGSUSED */
2494 static int
2495 nfs_fsync(struct vop_fsync_args *ap)
2496 {
2497 	return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2498 }
2499 
2500 /*
2501  * Flush all the blocks associated with a vnode.
2502  * 	Walk through the buffer pool and push any dirty pages
2503  *	associated with the vnode.
2504  * If the called_from_renewthread argument is TRUE, it has been called
2505  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2506  * waiting for a buffer write to complete.
2507  */
2508 int
2509 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2510     int commit, int called_from_renewthread)
2511 {
2512 	struct nfsnode *np = VTONFS(vp);
2513 	struct buf *bp;
2514 	int i;
2515 	struct buf *nbp;
2516 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2517 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2518 	int passone = 1, trycnt = 0;
2519 	u_quad_t off, endoff, toff;
2520 	struct ucred* wcred = NULL;
2521 	struct buf **bvec = NULL;
2522 	struct bufobj *bo;
2523 #ifndef NFS_COMMITBVECSIZ
2524 #define	NFS_COMMITBVECSIZ	20
2525 #endif
2526 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2527 	int bvecsize = 0, bveccount;
2528 
2529 	if (called_from_renewthread != 0)
2530 		slptimeo = hz;
2531 	if (nmp->nm_flag & NFSMNT_INT)
2532 		slpflag = NFS_PCATCH;
2533 	if (!commit)
2534 		passone = 0;
2535 	bo = &vp->v_bufobj;
2536 	/*
2537 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2538 	 * server, but has not been committed to stable storage on the server
2539 	 * yet. On the first pass, the byte range is worked out and the commit
2540 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2541 	 * job.
2542 	 */
2543 again:
2544 	off = (u_quad_t)-1;
2545 	endoff = 0;
2546 	bvecpos = 0;
2547 	if (NFS_ISV34(vp) && commit) {
2548 		if (bvec != NULL && bvec != bvec_on_stack)
2549 			free(bvec, M_TEMP);
2550 		/*
2551 		 * Count up how many buffers waiting for a commit.
2552 		 */
2553 		bveccount = 0;
2554 		BO_LOCK(bo);
2555 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2556 			if (!BUF_ISLOCKED(bp) &&
2557 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2558 				== (B_DELWRI | B_NEEDCOMMIT))
2559 				bveccount++;
2560 		}
2561 		/*
2562 		 * Allocate space to remember the list of bufs to commit.  It is
2563 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2564 		 * If we can't get memory (for whatever reason), we will end up
2565 		 * committing the buffers one-by-one in the loop below.
2566 		 */
2567 		if (bveccount > NFS_COMMITBVECSIZ) {
2568 			/*
2569 			 * Release the vnode interlock to avoid a lock
2570 			 * order reversal.
2571 			 */
2572 			BO_UNLOCK(bo);
2573 			bvec = (struct buf **)
2574 				malloc(bveccount * sizeof(struct buf *),
2575 				       M_TEMP, M_NOWAIT);
2576 			BO_LOCK(bo);
2577 			if (bvec == NULL) {
2578 				bvec = bvec_on_stack;
2579 				bvecsize = NFS_COMMITBVECSIZ;
2580 			} else
2581 				bvecsize = bveccount;
2582 		} else {
2583 			bvec = bvec_on_stack;
2584 			bvecsize = NFS_COMMITBVECSIZ;
2585 		}
2586 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2587 			if (bvecpos >= bvecsize)
2588 				break;
2589 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2590 				nbp = TAILQ_NEXT(bp, b_bobufs);
2591 				continue;
2592 			}
2593 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2594 			    (B_DELWRI | B_NEEDCOMMIT)) {
2595 				BUF_UNLOCK(bp);
2596 				nbp = TAILQ_NEXT(bp, b_bobufs);
2597 				continue;
2598 			}
2599 			BO_UNLOCK(bo);
2600 			bremfree(bp);
2601 			/*
2602 			 * Work out if all buffers are using the same cred
2603 			 * so we can deal with them all with one commit.
2604 			 *
2605 			 * NOTE: we are not clearing B_DONE here, so we have
2606 			 * to do it later on in this routine if we intend to
2607 			 * initiate I/O on the bp.
2608 			 *
2609 			 * Note: to avoid loopback deadlocks, we do not
2610 			 * assign b_runningbufspace.
2611 			 */
2612 			if (wcred == NULL)
2613 				wcred = bp->b_wcred;
2614 			else if (wcred != bp->b_wcred)
2615 				wcred = NOCRED;
2616 			vfs_busy_pages(bp, 1);
2617 
2618 			BO_LOCK(bo);
2619 			/*
2620 			 * bp is protected by being locked, but nbp is not
2621 			 * and vfs_busy_pages() may sleep.  We have to
2622 			 * recalculate nbp.
2623 			 */
2624 			nbp = TAILQ_NEXT(bp, b_bobufs);
2625 
2626 			/*
2627 			 * A list of these buffers is kept so that the
2628 			 * second loop knows which buffers have actually
2629 			 * been committed. This is necessary, since there
2630 			 * may be a race between the commit rpc and new
2631 			 * uncommitted writes on the file.
2632 			 */
2633 			bvec[bvecpos++] = bp;
2634 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2635 				bp->b_dirtyoff;
2636 			if (toff < off)
2637 				off = toff;
2638 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2639 			if (toff > endoff)
2640 				endoff = toff;
2641 		}
2642 		BO_UNLOCK(bo);
2643 	}
2644 	if (bvecpos > 0) {
2645 		/*
2646 		 * Commit data on the server, as required.
2647 		 * If all bufs are using the same wcred, then use that with
2648 		 * one call for all of them, otherwise commit each one
2649 		 * separately.
2650 		 */
2651 		if (wcred != NOCRED)
2652 			retv = ncl_commit(vp, off, (int)(endoff - off),
2653 					  wcred, td);
2654 		else {
2655 			retv = 0;
2656 			for (i = 0; i < bvecpos; i++) {
2657 				off_t off, size;
2658 				bp = bvec[i];
2659 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2660 					bp->b_dirtyoff;
2661 				size = (u_quad_t)(bp->b_dirtyend
2662 						  - bp->b_dirtyoff);
2663 				retv = ncl_commit(vp, off, (int)size,
2664 						  bp->b_wcred, td);
2665 				if (retv) break;
2666 			}
2667 		}
2668 
2669 		if (retv == NFSERR_STALEWRITEVERF)
2670 			ncl_clearcommit(vp->v_mount);
2671 
2672 		/*
2673 		 * Now, either mark the blocks I/O done or mark the
2674 		 * blocks dirty, depending on whether the commit
2675 		 * succeeded.
2676 		 */
2677 		for (i = 0; i < bvecpos; i++) {
2678 			bp = bvec[i];
2679 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2680 			if (retv) {
2681 				/*
2682 				 * Error, leave B_DELWRI intact
2683 				 */
2684 				vfs_unbusy_pages(bp);
2685 				brelse(bp);
2686 			} else {
2687 				/*
2688 				 * Success, remove B_DELWRI ( bundirty() ).
2689 				 *
2690 				 * b_dirtyoff/b_dirtyend seem to be NFS
2691 				 * specific.  We should probably move that
2692 				 * into bundirty(). XXX
2693 				 */
2694 				bufobj_wref(bo);
2695 				bp->b_flags |= B_ASYNC;
2696 				bundirty(bp);
2697 				bp->b_flags &= ~B_DONE;
2698 				bp->b_ioflags &= ~BIO_ERROR;
2699 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2700 				bufdone(bp);
2701 			}
2702 		}
2703 	}
2704 
2705 	/*
2706 	 * Start/do any write(s) that are required.
2707 	 */
2708 loop:
2709 	BO_LOCK(bo);
2710 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2711 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2712 			if (waitfor != MNT_WAIT || passone)
2713 				continue;
2714 
2715 			error = BUF_TIMELOCK(bp,
2716 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2717 			    BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2718 			if (error == 0) {
2719 				BUF_UNLOCK(bp);
2720 				goto loop;
2721 			}
2722 			if (error == ENOLCK) {
2723 				error = 0;
2724 				goto loop;
2725 			}
2726 			if (called_from_renewthread != 0) {
2727 				/*
2728 				 * Return EIO so the flush will be retried
2729 				 * later.
2730 				 */
2731 				error = EIO;
2732 				goto done;
2733 			}
2734 			if (newnfs_sigintr(nmp, td)) {
2735 				error = EINTR;
2736 				goto done;
2737 			}
2738 			if (slpflag & PCATCH) {
2739 				slpflag = 0;
2740 				slptimeo = 2 * hz;
2741 			}
2742 			goto loop;
2743 		}
2744 		if ((bp->b_flags & B_DELWRI) == 0)
2745 			panic("nfs_fsync: not dirty");
2746 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2747 			BUF_UNLOCK(bp);
2748 			continue;
2749 		}
2750 		BO_UNLOCK(bo);
2751 		bremfree(bp);
2752 		if (passone || !commit)
2753 		    bp->b_flags |= B_ASYNC;
2754 		else
2755 		    bp->b_flags |= B_ASYNC;
2756 		bwrite(bp);
2757 		if (newnfs_sigintr(nmp, td)) {
2758 			error = EINTR;
2759 			goto done;
2760 		}
2761 		goto loop;
2762 	}
2763 	if (passone) {
2764 		passone = 0;
2765 		BO_UNLOCK(bo);
2766 		goto again;
2767 	}
2768 	if (waitfor == MNT_WAIT) {
2769 		while (bo->bo_numoutput) {
2770 			error = bufobj_wwait(bo, slpflag, slptimeo);
2771 			if (error) {
2772 			    BO_UNLOCK(bo);
2773 			    if (called_from_renewthread != 0) {
2774 				/*
2775 				 * Return EIO so that the flush will be
2776 				 * retried later.
2777 				 */
2778 				error = EIO;
2779 				goto done;
2780 			    }
2781 			    error = newnfs_sigintr(nmp, td);
2782 			    if (error)
2783 				goto done;
2784 			    if (slpflag & PCATCH) {
2785 				slpflag = 0;
2786 				slptimeo = 2 * hz;
2787 			    }
2788 			    BO_LOCK(bo);
2789 			}
2790 		}
2791 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2792 			BO_UNLOCK(bo);
2793 			goto loop;
2794 		}
2795 		/*
2796 		 * Wait for all the async IO requests to drain
2797 		 */
2798 		BO_UNLOCK(bo);
2799 		mtx_lock(&np->n_mtx);
2800 		while (np->n_directio_asyncwr > 0) {
2801 			np->n_flag |= NFSYNCWAIT;
2802 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
2803 			    &np->n_mtx, slpflag | (PRIBIO + 1),
2804 			    "nfsfsync", 0);
2805 			if (error) {
2806 				if (newnfs_sigintr(nmp, td)) {
2807 					mtx_unlock(&np->n_mtx);
2808 					error = EINTR;
2809 					goto done;
2810 				}
2811 			}
2812 		}
2813 		mtx_unlock(&np->n_mtx);
2814 	} else
2815 		BO_UNLOCK(bo);
2816 	mtx_lock(&np->n_mtx);
2817 	if (np->n_flag & NWRITEERR) {
2818 		error = np->n_error;
2819 		np->n_flag &= ~NWRITEERR;
2820 	}
2821   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2822 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2823   		np->n_flag &= ~NMODIFIED;
2824 	mtx_unlock(&np->n_mtx);
2825 done:
2826 	if (bvec != NULL && bvec != bvec_on_stack)
2827 		free(bvec, M_TEMP);
2828 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2829 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2830 	     np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2831 		/* try, try again... */
2832 		passone = 1;
2833 		wcred = NULL;
2834 		bvec = NULL;
2835 		bvecsize = 0;
2836 printf("try%d\n", trycnt);
2837 		goto again;
2838 	}
2839 	return (error);
2840 }
2841 
2842 /*
2843  * NFS advisory byte-level locks.
2844  */
2845 static int
2846 nfs_advlock(struct vop_advlock_args *ap)
2847 {
2848 	struct vnode *vp = ap->a_vp;
2849 	struct ucred *cred;
2850 	struct nfsnode *np = VTONFS(ap->a_vp);
2851 	struct proc *p = (struct proc *)ap->a_id;
2852 	struct thread *td = curthread;	/* XXX */
2853 	struct vattr va;
2854 	int ret, error = EOPNOTSUPP;
2855 	u_quad_t size;
2856 
2857 	if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2858 		cred = p->p_ucred;
2859 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2860 		if (vp->v_iflag & VI_DOOMED) {
2861 			VOP_UNLOCK(vp, 0);
2862 			return (EBADF);
2863 		}
2864 
2865 		/*
2866 		 * If this is unlocking a write locked region, flush and
2867 		 * commit them before unlocking. This is required by
2868 		 * RFC3530 Sec. 9.3.2.
2869 		 */
2870 		if (ap->a_op == F_UNLCK &&
2871 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
2872 			(void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2873 
2874 		/*
2875 		 * Loop around doing the lock op, while a blocking lock
2876 		 * must wait for the lock op to succeed.
2877 		 */
2878 		do {
2879 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2880 			    ap->a_fl, 0, cred, td);
2881 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2882 			    ap->a_op == F_SETLK) {
2883 				VOP_UNLOCK(vp, 0);
2884 				error = nfs_catnap(PZERO | PCATCH, ret,
2885 				    "ncladvl");
2886 				if (error)
2887 					return (EINTR);
2888 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2889 				if (vp->v_iflag & VI_DOOMED) {
2890 					VOP_UNLOCK(vp, 0);
2891 					return (EBADF);
2892 				}
2893 			}
2894 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2895 		     ap->a_op == F_SETLK);
2896 		if (ret == NFSERR_DENIED) {
2897 			VOP_UNLOCK(vp, 0);
2898 			return (EAGAIN);
2899 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2900 			VOP_UNLOCK(vp, 0);
2901 			return (ret);
2902 		} else if (ret != 0) {
2903 			VOP_UNLOCK(vp, 0);
2904 			return (EACCES);
2905 		}
2906 
2907 		/*
2908 		 * Now, if we just got a lock, invalidate data in the buffer
2909 		 * cache, as required, so that the coherency conforms with
2910 		 * RFC3530 Sec. 9.3.2.
2911 		 */
2912 		if (ap->a_op == F_SETLK) {
2913 			if ((np->n_flag & NMODIFIED) == 0) {
2914 				np->n_attrstamp = 0;
2915 				ret = VOP_GETATTR(vp, &va, cred);
2916 			}
2917 			if ((np->n_flag & NMODIFIED) || ret ||
2918 			    np->n_change != va.va_filerev) {
2919 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2920 				np->n_attrstamp = 0;
2921 				ret = VOP_GETATTR(vp, &va, cred);
2922 				if (!ret) {
2923 					np->n_mtime = va.va_mtime;
2924 					np->n_change = va.va_filerev;
2925 				}
2926 			}
2927 		}
2928 		VOP_UNLOCK(vp, 0);
2929 		return (0);
2930 	} else if (!NFS_ISV4(vp)) {
2931 		error = vn_lock(vp, LK_SHARED);
2932 		if (error)
2933 			return (error);
2934 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2935 			size = VTONFS(vp)->n_size;
2936 			VOP_UNLOCK(vp, 0);
2937 			error = lf_advlock(ap, &(vp->v_lockf), size);
2938 		} else {
2939 			if (ncl_advlock_p)
2940 				error = ncl_advlock_p(ap);
2941 			else
2942 				error = ENOLCK;
2943 		}
2944 	}
2945 	return (error);
2946 }
2947 
2948 /*
2949  * NFS advisory byte-level locks.
2950  */
2951 static int
2952 nfs_advlockasync(struct vop_advlockasync_args *ap)
2953 {
2954 	struct vnode *vp = ap->a_vp;
2955 	u_quad_t size;
2956 	int error;
2957 
2958 	if (NFS_ISV4(vp))
2959 		return (EOPNOTSUPP);
2960 	error = vn_lock(vp, LK_SHARED);
2961 	if (error)
2962 		return (error);
2963 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2964 		size = VTONFS(vp)->n_size;
2965 		VOP_UNLOCK(vp, 0);
2966 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
2967 	} else {
2968 		VOP_UNLOCK(vp, 0);
2969 		error = EOPNOTSUPP;
2970 	}
2971 	return (error);
2972 }
2973 
2974 /*
2975  * Print out the contents of an nfsnode.
2976  */
2977 static int
2978 nfs_print(struct vop_print_args *ap)
2979 {
2980 	struct vnode *vp = ap->a_vp;
2981 	struct nfsnode *np = VTONFS(vp);
2982 
2983 	ncl_printf("\tfileid %ld fsid 0x%x",
2984 	   np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2985 	if (vp->v_type == VFIFO)
2986 		fifo_printinfo(vp);
2987 	printf("\n");
2988 	return (0);
2989 }
2990 
2991 /*
2992  * This is the "real" nfs::bwrite(struct buf*).
2993  * We set B_CACHE if this is a VMIO buffer.
2994  */
2995 int
2996 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
2997 {
2998 	int s;
2999 	int oldflags = bp->b_flags;
3000 #if 0
3001 	int retv = 1;
3002 	off_t off;
3003 #endif
3004 
3005 	BUF_ASSERT_HELD(bp);
3006 
3007 	if (bp->b_flags & B_INVAL) {
3008 		brelse(bp);
3009 		return(0);
3010 	}
3011 
3012 	bp->b_flags |= B_CACHE;
3013 
3014 	/*
3015 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3016 	 */
3017 
3018 	s = splbio();
3019 	bundirty(bp);
3020 	bp->b_flags &= ~B_DONE;
3021 	bp->b_ioflags &= ~BIO_ERROR;
3022 	bp->b_iocmd = BIO_WRITE;
3023 
3024 	bufobj_wref(bp->b_bufobj);
3025 	curthread->td_ru.ru_oublock++;
3026 	splx(s);
3027 
3028 	/*
3029 	 * Note: to avoid loopback deadlocks, we do not
3030 	 * assign b_runningbufspace.
3031 	 */
3032 	vfs_busy_pages(bp, 1);
3033 
3034 	BUF_KERNPROC(bp);
3035 	bp->b_iooffset = dbtob(bp->b_blkno);
3036 	bstrategy(bp);
3037 
3038 	if( (oldflags & B_ASYNC) == 0) {
3039 		int rtval = bufwait(bp);
3040 
3041 		if (oldflags & B_DELWRI) {
3042 			s = splbio();
3043 			reassignbuf(bp);
3044 			splx(s);
3045 		}
3046 		brelse(bp);
3047 		return (rtval);
3048 	}
3049 
3050 	return (0);
3051 }
3052 
3053 /*
3054  * nfs special file access vnode op.
3055  * Essentially just get vattr and then imitate iaccess() since the device is
3056  * local to the client.
3057  */
3058 static int
3059 nfsspec_access(struct vop_access_args *ap)
3060 {
3061 	struct vattr *vap;
3062 	struct ucred *cred = ap->a_cred;
3063 	struct vnode *vp = ap->a_vp;
3064 	accmode_t accmode = ap->a_accmode;
3065 	struct vattr vattr;
3066 	int error;
3067 
3068 	/*
3069 	 * Disallow write attempts on filesystems mounted read-only;
3070 	 * unless the file is a socket, fifo, or a block or character
3071 	 * device resident on the filesystem.
3072 	 */
3073 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3074 		switch (vp->v_type) {
3075 		case VREG:
3076 		case VDIR:
3077 		case VLNK:
3078 			return (EROFS);
3079 		default:
3080 			break;
3081 		}
3082 	}
3083 	vap = &vattr;
3084 	error = VOP_GETATTR(vp, vap, cred);
3085 	if (error)
3086 		goto out;
3087 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3088 	    accmode, cred, NULL);
3089 out:
3090 	return error;
3091 }
3092 
3093 /*
3094  * Read wrapper for fifos.
3095  */
3096 static int
3097 nfsfifo_read(struct vop_read_args *ap)
3098 {
3099 	struct nfsnode *np = VTONFS(ap->a_vp);
3100 	int error;
3101 
3102 	/*
3103 	 * Set access flag.
3104 	 */
3105 	mtx_lock(&np->n_mtx);
3106 	np->n_flag |= NACC;
3107 	getnanotime(&np->n_atim);
3108 	mtx_unlock(&np->n_mtx);
3109 	error = fifo_specops.vop_read(ap);
3110 	return error;
3111 }
3112 
3113 /*
3114  * Write wrapper for fifos.
3115  */
3116 static int
3117 nfsfifo_write(struct vop_write_args *ap)
3118 {
3119 	struct nfsnode *np = VTONFS(ap->a_vp);
3120 
3121 	/*
3122 	 * Set update flag.
3123 	 */
3124 	mtx_lock(&np->n_mtx);
3125 	np->n_flag |= NUPD;
3126 	getnanotime(&np->n_mtim);
3127 	mtx_unlock(&np->n_mtx);
3128 	return(fifo_specops.vop_write(ap));
3129 }
3130 
3131 /*
3132  * Close wrapper for fifos.
3133  *
3134  * Update the times on the nfsnode then do fifo close.
3135  */
3136 static int
3137 nfsfifo_close(struct vop_close_args *ap)
3138 {
3139 	struct vnode *vp = ap->a_vp;
3140 	struct nfsnode *np = VTONFS(vp);
3141 	struct vattr vattr;
3142 	struct timespec ts;
3143 
3144 	mtx_lock(&np->n_mtx);
3145 	if (np->n_flag & (NACC | NUPD)) {
3146 		getnanotime(&ts);
3147 		if (np->n_flag & NACC)
3148 			np->n_atim = ts;
3149 		if (np->n_flag & NUPD)
3150 			np->n_mtim = ts;
3151 		np->n_flag |= NCHG;
3152 		if (vrefcnt(vp) == 1 &&
3153 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3154 			VATTR_NULL(&vattr);
3155 			if (np->n_flag & NACC)
3156 				vattr.va_atime = np->n_atim;
3157 			if (np->n_flag & NUPD)
3158 				vattr.va_mtime = np->n_mtim;
3159 			mtx_unlock(&np->n_mtx);
3160 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3161 			goto out;
3162 		}
3163 	}
3164 	mtx_unlock(&np->n_mtx);
3165 out:
3166 	return (fifo_specops.vop_close(ap));
3167 }
3168 
3169 /*
3170  * Just call ncl_writebp() with the force argument set to 1.
3171  *
3172  * NOTE: B_DONE may or may not be set in a_bp on call.
3173  */
3174 static int
3175 nfs_bwrite(struct buf *bp)
3176 {
3177 
3178 	return (ncl_writebp(bp, 1, curthread));
3179 }
3180 
3181 struct buf_ops buf_ops_newnfs = {
3182 	.bop_name	=	"buf_ops_nfs",
3183 	.bop_write	=	nfs_bwrite,
3184 	.bop_strategy	=	bufstrategy,
3185 	.bop_sync	=	bufsync,
3186 	.bop_bdflush	=	bufbdflush,
3187 };
3188 
3189 /*
3190  * Cloned from vop_stdlock(), and then the ugly hack added.
3191  */
3192 static int
3193 nfs_lock1(struct vop_lock1_args *ap)
3194 {
3195 	struct vnode *vp = ap->a_vp;
3196 	int error = 0;
3197 
3198 	/*
3199 	 * Since vfs_hash_get() calls vget() and it will no longer work
3200 	 * for FreeBSD8 with flags == 0, I can only think of this horrible
3201 	 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3202 	 * and then handle it here. All I want for this case is a v_usecount
3203 	 * on the vnode to use for recovery, while another thread might
3204 	 * hold a lock on the vnode. I have the other threads blocked, so
3205 	 * there isn't any race problem.
3206 	 */
3207 	if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3208 		if ((ap->a_flags & LK_INTERLOCK) == 0)
3209 			panic("ncllock1");
3210 		if ((vp->v_iflag & VI_DOOMED))
3211 			error = ENOENT;
3212 		VI_UNLOCK(vp);
3213 		return (error);
3214 	}
3215 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3216 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3217 	    ap->a_line));
3218 }
3219 
3220 static int
3221 nfs_getacl(struct vop_getacl_args *ap)
3222 {
3223 	int error;
3224 
3225 	if (ap->a_type != ACL_TYPE_NFS4)
3226 		return (EOPNOTSUPP);
3227 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3228 	    NULL);
3229 	if (error > NFSERR_STALE) {
3230 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3231 		error = EPERM;
3232 	}
3233 	return (error);
3234 }
3235 
3236 static int
3237 nfs_setacl(struct vop_setacl_args *ap)
3238 {
3239 	int error;
3240 
3241 	if (ap->a_type != ACL_TYPE_NFS4)
3242 		return (EOPNOTSUPP);
3243 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3244 	    NULL);
3245 	if (error > NFSERR_STALE) {
3246 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3247 		error = EPERM;
3248 	}
3249 	return (error);
3250 }
3251